Dissertation
Three Essays on Value in Health Care
Workplace Wellness Program Return on Investment, Effectiveness of Monetary Penalties for Tobacco Cessation Non-Participation, and Physician Perceptions of Their Use of Time and Appropriateness of Care Provided
John P. Caloyeras
This document was submitted as a dissertation in August 2017 in partial fulfillment of the requirements of the doctoral degree in public policy analysis at the Pardee RAND Graduate School. The faculty committee that supervised and approved the dissertation consisted of Soeren Mattke (Chair), Hangsheng Liu, and Jill Horwitz. Partial funding of this dissertation was provided by the Southern California Permanente Medical Group, the United States Department of Labor, PepsiCo Inc., and David Richards (through a donation to the RAND Corporation).
PARDEE RAND GRADUATE SCHOOL
For more information on this publication, visit http://www.rand.org/pubs/rgs_dissertations/RGSD401.html
Published 2018 by the RAND Corporation, Santa Monica, Calif.
R® is a registered trademark
Limited Print and Electronic Distribution Rights
This document and trademark(s) contained herein are protected by law. This representation of RAND intellectual property is provided for noncommercial use only. Unauthorized posting of this publication online is prohibited. Permission is given to duplicate this document for personal use only, as long as it is unaltered and complete. Permission is required from RAND to reproduce, or reuse in another form, any of its research documents for commercial use. For information on reprint and linking permissions, please visit www.rand.org/pubs/permissions.html.
The RAND Corporation is a research organization that develops solutions to public policy challenges to help make communities throughout the world safer and more secure, healthier and more prosperous. RAND is nonprofit, nonpartisan, and committed to the public interest.
RAND’s publications do not necessarily reflect the opinions of its research clients and sponsors.
Support RAND Make a tax-deductible charitable contribution at
www.rand.org/giving/contribute
www.rand.org
iii
TABLEOFCONTENTS
Introductory Material .................................................................................................................................. iv
Abstract .................................................................................................................................................... iv
Acknowledgements ................................................................................................................................... v
Preface ...................................................................................................................................................... v
Paper 1: Managing manifest diseases, but not health risks, saved PepsiCo money over seven years ........ 1
Abstract ..................................................................................................................................................... 2
Main Text .................................................................................................................................................. 3
Exhibits .................................................................................................................................................... 16
Supplementary Materials........................................................................................................................ 19
Paper 2: Monetary penalties marginally increase participation in workplace tobacco cessation programs but also shift costs to those with lower socioeconomic status .................................................................. 41
Main Text ................................................................................................................................................ 42
Exhibits .................................................................................................................................................... 55
Supplementary Materials........................................................................................................................ 59
Paper 3: A survey of physician perceptions on use of time and the appropriateness of care provided .... 87
Abstract ................................................................................................................................................... 88
Main Text ................................................................................................................................................ 89
Exhibits .................................................................................................................................................. 100
Supplementary Materials...................................................................................................................... 105
iv
INTRODUCTORYMATERIAL
ABSTRACT
This dissertation, in a three paper format, examines three policy levers for improving health care value.
The first paper examines the return‐on‐investment of PepsiCo’s workplace wellness program,
concluding that while the program as a whole saves health care dollars, the primary drivers of savings
are those interventions focused on employees with manifest diseases as opposed to unhealthy
lifestyles. The second paper estimates the impact of a $600 per year surcharge applied to tobacco users
who do not participate in a Fortune‐100 employer’s tobacco cessation intervention. Findings indicate
the surcharge modestly increased participation in the tobacco cessation intervention. However, given
low participation rates and the characteristics of tobacco users versus non‐users, the primary effect of
the policy is to shift health care costs towards employees who on average are of lower socioeconomic
status vs. their non‐tobacco using counterparts. In the third paper, a novel physician survey is developed
to examine physician perceptions regarding their use of time and the appropriateness of care they
provide. Findings from the implementation of the survey among clinic‐based physicians within four sites
of the Southern California Permanente Medical Group indicate that the opportunity exists to improve
health care value along these two value domains, but that it is less than commonly accepted national
wisdom. Collectively, these three studies support the notion that a multi‐pronged approach—in which
many different policy levers are simultaneously pulled—is needed to improve the value of health care in
the US. In addition, results emphasize the need to carefully monitor implemented policies to both
identify unintended consequences and characterize where and among whom a given policy option is
most effective.
v
ACKNOWLEDGEMENTS
I would like to thank my dissertation committee—Soeren Mattke, Hangsheng Liu and Jill Horwitz—for
their help and support throughout the dissertation process. I owe a special thanks to Robert Brook and
Sandra Berry for their roles developing the idea for my third paper, and to Robert Brook, Sandra Berry
and Michael Kanter for mentoring me through the implementation, analysis and writing stages for my
third paper. Soeren Mattke and Hangsheng Liu repeatedly gave me tremendous support and unique
opportunities, without which my first two papers would have been impossible. Kandice Kapinos
provided much appreciated analytical wisdom with my second paper.
I am grateful to the Pardee RAND Graduate School administration and faculty and the broader RAND
community for providing the rich atmosphere that made the opportunities to work with and learn from
countless researchers possible.
My dissertation work received full or partial funding, both directly and indirectly, from a variety of
sources including PepsiCo Inc. (client and direct funder for my first paper), the US Department of Labor
(indirect, partial funder for my second paper), an anonymous RAND donor (direct, partial funder for my
third paper), and the Andrew Marshall Scholarship. An anonymous Fortune‐100 employer provided me
with full access to their data, making my second paper possible. Finally, without Michael Kanter, Nicole
Ives, Chong Kim and the Southern California Permanente Medical Group, the implementation of the
survey for my third paper would not have been possible; lastly, thank you to the many physicians who
took the time to respond to the survey and share their thoughts.
vii
PREFACE
US health care spending exceeded $3.2 trillion dollars in 2015, representing 17.8% of GDP and nearly
$10,000 per capita. However, despite spending more per capita, the US lags other countries in a variety
of basic outcomes such as life expectancy. The notion that the US spends more but doesn’t get more has
in part led to analyses of how the return per dollar spent on health care could be improved. For
example, the Institute of Medicine has estimated “waste” within the US health care system to total $750
billion per year, made up of categories such as unnecessary services, missed prevention opportunities
and inefficiently delivered services.
Using the Institute for Healthcare Improvement’s “Triple Aim” as a reference framework for where the
US health care system is trying to end up, this dissertation evaluates three policy levers for increasing
health care value, through three stand‐alone research studies. While the three papers in this
dissertation are unified by their examination of policy levers for increasing health care value, each paper
is meant to stand alone and provide distinct contributions to the literature.
The first paper evaluates the return‐on‐investment of PepsiCo’s wellness program. It was published in
Health Affairs in 2014 and earned the award of the second most read article in Health Affairs for that
year. The second paper examines the impact of a $600 penalty for smoking cessation non‐participation
in the context of a Fortune‐100 employer’s wellness program; this paper will be distilled down in
content for submission to a peer‐reviewed journal. The third paper, which investigates physician
perceptions regarding how they use their time and the appropriateness of care provided, is in the
process of being submitted to a peer‐reviewed journal.
The cross cutting themes emerging from the results of the three papers in this dissertation are as
follows. To begin, each paper identified meaningful ways health care value could be improved. Similarly,
each paper generally found the impact of the policy levers evaluated to be modest, triangulating on the
notion that there is no single solution to the US health care value problem, but rather that we must
simultaneously pull (and properly evaluate) numerous policy levers. The papers further demonstrate
that the business of evaluation is complicated: components of interventions much be disentangled and
dissected at a micro level, while also keeping a broad perspective when interpreting results and making
policy recommendations.
The need to pay attention to, and monitor for, unintended consequences emerges as a theme, keeping
in mind equity issues and the human behaviors and incentives playing out on a granular basis. Focusing
viii
behavioral interventions, both with patients and with physicians, on areas where behavior change is
easy will have, on average, the greatest chances of success. Finally, achieving the Institute for Healthcare
Improvement’s “Triple Aim” will be a marathon, not a sprint; incremental testing and evaluation of
many policy levers will be continue to be needed to help ensure the US health care system moves and
makes progress towards achieving greater value per dollar spent.
1
PAPER1:MANAGINGMANIFESTDISEASES,BUTNOTHEALTHRISKS,SAVEDPEPSICOMONEYOVERSEVENYEARS
Caloyeras JP1,2 MPhil, Liu H2 PhD, Exum E3, Broderick M3, Mattke S2 MD ScD
1. Pardee RAND Graduate School, Santa Monica, CA
2. RAND Corporation, Santa Monica, CA
3. PepsiCo, Purchase, NY
2
ABSTRACT
Workplace wellness programs are increasingly popular. Employers expect them to improve employee
health and well‐being, lower medical costs, increase productivity, and reduce absenteeism. To test
whether such expectations are warranted, we evaluated the cost impact of the lifestyle and disease
management components of PepsiCo’s wellness program, Healthy Living. We found that seven years of
continuous participation in one or both components was associated with an average reduction of $30 in
health care cost per member per month. When we looked at each component individually, we found
that the disease management component was associated with lower costs and that the lifestyle
management component was not. We estimate disease management to reduce health care costs by
$136 per member per month, driven by a 29 percent reduction in hospital admissions. Workplace
wellness programs may reduce health risks, delay or avoid the onset of chronic diseases, and lower
health care costs for employees with manifest chronic disease. But employers and policy makers should
not take for granted that the lifestyle management component of such programs can reduce health care
costs or even lead to net savings.
3
MAINTEXT
INTRODUCTION
Workplace health and wellness programs are becoming an increasingly common workplace benefit in
the United States. The recently published RAND Workplace Wellness Programs Study found that about
half of employers with at least 50 employees and more than 90 percent of those with more than 50,000
employees offered a wellness program in 2012.1 In general, wellness programs screen employees and
sometimes their dependents to identify health risks, provide interventions to address health risks and
manifest disease, and promote healthy lifestyles. A wellness program’s specific program components
(for example, lifestyle management to promote healthy living habits or disease management to help
employees manage a chronic condition or illness) and interventions (for instance, on‐site exercise
classes) vary across employers, with larger employers more likely than smaller employers to offer more‐
elaborate programs that combine a variety of components and interventions.2
The popularity of wellness programs is driven by employers’ expectation that the programs improve
employee health and well‐being, lower medical costs, increase productivity, and reduce absenteeism.
For instance, a 2011 Automatic Data Processing (ADP) survey of employers with at least 1,000
employees found the four most commonly cited reasons for offering a wellness program to be “improve
employee health,” “control health care costs,” “increase productivity,” and “reduce absenteeism” (cited
by 78 percent, 71 percent, 42 percent, and 43 percent of employers, respectively).3 Furthermore, 43
percent of employers responding to a 2012 Deloitte survey said they believed that investments in
wellness programs offered high levels of value to the overall health care system per dollar spent on
them.4
The popularity of wellness programs is expected to continue to grow. A 2011 Aon Hewitt survey found
that among employers without a health improvement or wellness program, 47 percent planned to add
such a program in 2012, with an additional 47 percent reporting they may add such a program in the
next three to five years.5 Employers with programs want to get more employees participating in them:
70 percent of employers in the Aon Hewitt survey identified increasing use of wellness programs as a
top priority.5 The Affordable Care Act also has several provisions to promote workplace wellness. For
example, section 4303 of the act establishes a technical assistance role for the Centers for Disease
Control and Prevention to provide tools and resources to assist employers with planning, implementing,
and evaluating wellness programs.
4
Employers’ optimism regarding the benefits of wellness programs is driven by countless success stories
in the popular press and trade publications and by studies in the peer‐reviewed literature that have
largely concluded that wellness programs save money and are a good bet for employers looking to lower
health care costs.
The evidence for the prevailing wisdom today—that wellness programs can reduce health care costs and
absenteeism in excess of program costs—has been established by several reviews.6,7 Those reviews’
findings were further reinforced by a recent meta‐analysis by Katherine Baicker and colleagues that
stated that health care costs fall by $3.27 and absenteeism costs fall by $2.73 for every dollar invested in
a wellness program.8
By contrast, the recently released RAND Workplace Wellness Programs Study, which pooled 362,136
employees from five employers, found that lifestyle management programs can achieve improvements
in risk factors, such as reductions in smoking, and increases in healthy behavior, such as exercise. The
study, however, did not find that lifestyle management programs achieve statistically significant
reductions in health care costs.1 Although a 2012 three‐year evaluation of the University of Minnesota’s
wellness program also found no evidence that lifestyle management lowers health care costs, the study
did find the disease management component of the program to do so.9 Neither lifestyle management
nor disease management were found to reduce absenteeism.9
Reconciling these seemingly contradictory findings requires not only asking, “Do wellness programs
work?” but also, “Which program components have which effects under which conditions?” Such an
approach is particularly important given the heterogeneity of offerings that can be subsumed under the
label “workplace wellness” and the variety of settings in which these programs are implemented.
Against this background, we assess the impact of two common wellness program components— disease
management to support employees with chronic conditions and lifestyle management to reduce
employees’ health risks—on health care cost, use, and absenteeism by individual component and for
both components together.
Our study uses two baseline years and seven program years of data from PepsiCo’s Healthy Living
program and builds upon two prior evaluations at three years into the program that showed the overall
program, but not its lifestyle management component, was associated with lower health care costs.10,11
We undertook our current study to determine whether overall cost reductions were sustainable over a
longer time period and whether the lifestyle management component would begin to contribute to the
5
savings. To our knowledge, our study is one of the longest evaluations of a comprehensive wellness
program to date in the published literature.
STUDY DATA AND METHODS
THE PEPSICO PROGRAM
PepsiCo introduced in 2003 what evolved into their Healthy Living program. Healthy Living is a wellness
program made up of numerous components that include health risk assessments, on‐site wellness
events, lifestyle management, disease management, complex care management, a 24‐7 nurse advice
line, and maternity management. All PepsiCo employees and their dependents can participate in the
Healthy Living program, except for the fewer than 10 percent of employees who are enrolled in a health
maintenance organization or receive their health coverage through their union.
Health risk assessments help employees and their dependents understand their health status and health
risks, directing those with risks, such as obesity and smoking, to lifestyle management interventions
consisting of mailed educational materials, online programs, and telephonic coaching for those with
higher risk levels. In 2011 there were five distinct lifestyle management programs: weight management,
nutrition management, fitness, stress management, and smoking cessation. Completion of a telephonic
lifestyle management program involves a series of calls with a wellness coach over a six‐month period.
Disease management is offered to employees with at least one of ten chronic conditions and focuses on
improving medication adherence and patient self‐care knowledge and abilities. The ten conditions
covered by the disease management program were asthma, coronary artery disease, atrial fibrillation,
congestive heart failure, stroke, hyperlipidemia, hypertension, diabetes, low back pain, and chronic
obstructive pulmonary disease. Completion of a disease management program typically requires six to
nine months, during which participants have a series of calls with a nurse that average fifteen to twenty‐
five minutes per call. Completion of a program occurs when the participant is successfully managing his
or her condition.
STUDY SAMPLE
We selected our sample from a pool of 67,541 unique members who were eligible for disease
management or lifestyle management, or both, representing 400,657 member years of data. We
required participants to have at least two full years of health plan and program data as well as one year
of data prior to participation. Our sample consists of 14,555 participants in disease management, 22,880
6
in lifestyle management, and 9,324 in both disease management and lifestyle management, among
whom 2,610, 17,432, and 2,162 were successfully matched to a similar eligible nonparticipant,
respectively. There are a total of 22,204 matched pairs, representing 238,724 member years. The
matched pairs in the final analytic sample have on average 6.4 years of data.
DATA
We combined PepsiCo’s health and pharmacy plan claims data with Healthy Living eligibility and
participation data for all employees and dependents for the period between September 2002 and
August 2011. These data cover two baseline years (September 2002 to August 2004) and seven program
years (September 2004 to August 2011) for the lifestyle management program and one baseline year
(September 2002 to August 2003) and eight program years (September 2003 to August 2011) for the
disease management program. Our analytic sample was restricted to employees and dependents ages
18–64 with at least two full years of enrollment in a PepsiCo health plan and one full year of data prior
to the year in which they first participated. Following common practice, we removed program years
involving pregnancy‐related care12 as well as individuals eligible for complex care management, which
targets complex, high‐cost conditions, such as terminal cancer and organ transplants, because the
course of such conditions cannot be expected to be influenced by lifestyle or disease management.
Program costs include the vendor’s per participant per year fees for lifestyle and disease management
and the health risk assessment fee per completed survey.
ANALYTIC APPROACH
Our analytic sample consisted of all employees and dependents who were invited to participate in the
lifestyle or disease management components of Healthy Living. We used those who decided to join the
program as the intervention group and those who declined as the comparison group. Differential
changes between the groups over time were used to estimate program impact, a so‐called difference‐
indifferences design.
To adjust for differences between participants and nonparticipants, we used propensity score matching
based on baseline data to balance observable variables. Propensity scores were generated based on a
multinomial probit model, using the first year of data available for each member. The dependent
variable was participation in lifestyle management or disease management, or both, and the
independent variables included age, sex, being an employee, geographic region, calendar year, health
7
plan enrollment tenure, total health care costs, emergency department visits, hospital admissions, and
comorbidities.13
One‐to‐one propensity score matching was conducted with replacement, because there were fewer
nonparticipants than participants, and was stratified by year and by program eligibility for lifestyle
management or disease management, or both. After matching, regression models were used to
estimate the effects of participation in difference‐in‐differences on the outcomes of interest. The
models included lagged variables of program participation, adjustments for time‐varying covariates as
appropriate (age, geographic region, calendar year, and comorbidities) and individual‐level fixed effects
to control for unobservable characteristics.
MEASURES
Program eligibility represented whether an employee was eligible and invited for a program via a
telephone invitation and mailed letters during a given program year. Participation represented whether
the eligible employee subsequently chose to participate, as reflected in a monthly participation status
indicator variable. We used two approaches to define program participation. Aggregate participation
was specified as an indicator for any participation in lifestyle management or disease management, or
both. Component‐specific participation indicators were used to capture the individual impact of the
lifestyle management and disease management components on our outcomes. Our outcomes of
interest were defined as health care cost per member per month, adjusted to 2012 US dollars using the
Consumer Price Index14 and hospital admissions and emergency department visits per 1,000 employee‐
years. Absenteeism was measured based on individuals’ answers to a question on the health risk
assessment about work days lost because of illness or injuries for the preceding twelve months. Work
days lost was monetized by multiplying hours lost by the average hourly wage for private, goods‐
producing industries ($34.14) from the Bureau of Labor Statistics.15
Return on investment (ROI) was calculated as the ratio of estimated reductions of health care and
absenteeism costs to program costs as outlined above over the entire seven‐year intervention period.
LIMITATIONS
As with all observational designs, our study may have produced results that suffer from bias because of
unobservable differences between intervention‐ and comparison‐group members. These may include
differential motivation to improve health, health plan or wellness program literacy, or work schedule
issues that make participation difficult. To minimize potential bias, we used propensity score matching
8
to account for observable differences between nonparticipants and participants, such as age, sex,
comorbidities, and prior health care use. We included individual‐level fixed effects in our regression
analyses to account for unobservable differences that are constant over time. Thus, to attribute our
estimates to bias, one would have to assume the existence of unobservable characteristics that vary
over time and are associated with our endpoints.
Because of a limited pool of nonparticipants, particularly among people eligible for disease management
and those eligible for both disease management and lifestyle management, our propensity score
matching did not balance all member characteristics between our participant and nonparticipant groups
(Appendix Exhibit 1A).16 However, we controlled for such unbalanced differences in our regression
models.
PepsiCo is a large employer, and its experience with disease management and lifestyle management
programs might not be generalizable to other organizations, particularly smaller ones. Employers
considering adopting a wellness program should proceed with caution. Even if the program they
implement is very similar to PepsiCo’s lifestyle management and disease management components, key
differences in program implementation, design, and promotion to employees may affect results. For
example, differences in program design and implementation might affect participation and dropout
rates and intervention effects.
We did not examine health behavior outcomes, such as exercise frequency or medication adherence.
We also did not investigate program effects on more granular endpoints, such as wellness sensitive
hospitalizations, as proposed by Gautam Gowrisankaran and colleagues, which may capture program
effects with greater accuracy.17
Lastly, we may have overstated the true ROI because our estimates of program component costs were
confined to vendor fees. Specifically, we did not have information for the following cost items that affect
ROI: the cost of PepsiCo’s program staff, the cost of employees’ time required for program participation,
and any costs generated by false positives through extended screening.
STUDY RESULTS
Looking at the lifestyle management and disease management components as a whole, we found
participation to be associated with lower health care costs. Exhibit 1 compares the cost trends of
participants with those of statistically matched nonparticipants over seven years. After the third year of
9
participation, differences in health care costs become statistically significant. We found that seven years
of continuous participation was associated with an average reduction of $30 per member per month, or
$360 annually (p < 0:01) (Exhibit 2).
However, when we broke down the effect on health care cost by program component, we found disease
management but not lifestyle management to be associated with lower costs (Exhibit 2).We estimate
disease management to reduce costs among participants by $136 per member per month, or $1,632
annually, driven by a 29 percent reduction in hospital admissions (p < 0:01).
When looking at the subset of participants that had joined both the lifestyle management and the
disease management components of the program, we estimate a reduction in health care costs of $160
per member per month, or about $1,920 per year (p < 0:01), and a 66 percent reduction in hospital
admissions (p < 0:05).
Looking again at the lifestyle management and disease management components as a whole, we found
participation to be associated with a reduction in self‐reported absenteeism of 0.1 day, or forty‐eight
minutes (in an eight‐hour workday), per year (p < 0:01). This effect is driven by lifestyle management
participation, which is associated with a reduction of 0.13 day, or sixty‐two minutes (in an eight‐hour
work day), per year (p < 0:01). The monetized impact of 0.10 and 0.13 day is estimated to be $28 and
$35, respectively. No significant effect on absenteeism was observed among disease management
participants.
Based on our analyses, we estimate that the lifestyle management and disease management
components returned an average of $0.48 and $3.78, respectively, for every dollar invested when both
health care and absenteeism impacts were included (Exhibit 3). Together, they returned $1.46 for every
dollar invested. As shown in Exhibit 3, the main driver of the positive ROI is the reduction in health care
costs associated with disease management participation.
DISCUSSION
We estimate the impact of a disease and lifestyle management program and find that disease
management is associated with decreased health care costs and net savings after seven years—a result
that confirms our previous analysis of this program after three years.10 Participation in lifestyle
management interventions is associated with a small decrease in absenteeism but has no statistically
significant effect on health care costs. These findings are not necessarily surprising: As with any
10
preventive intervention, it is often easier to achieve cost savings in people with higher baseline
spending, as we found to be the case among disease management participants. Interestingly, the
disease management participants who also joined the lifestyle management program experienced
significantly higher savings, which suggests that proper targeting can improve the financial performance
of lifestyle management programs.
Our findings are consistent with two recent publications. The RAND Workplace Wellness Programs
Study, which is the largest evaluation of workplace wellness programs conducted to date, found lifestyle
management participation to produce no statistically significant reduction in health care costs.1 A 2012
evaluation conducted by John Nyman and colleagues of the lifestyle management and disease
management components of the University of Minnesota’s wellness program used an approach similar
to that of our study and found the two components together to generate an overall ROI of 1.76—quite
similar to our ROI of 1.46.9 Additionally, the authors found the savings to be driven entirely by the
program’s disease management component, with none generated by the lifestyle management
component, further mirroring our results.9 Collectively, these findings cast doubt on the widely held
belief in a strong business case for lifestyle management that is often supported by the above meta‐
analysis of Baicker and colleagues.8
To investigate why several recent studies came to a different conclusion than those of the wellness
programs meta‐analysis, we closely reviewed the seven papers that Baicker and colleagues analyzed.
First, five papers looked at programs that operated more than twenty years ago,18–22 a time in which
smoking was permitted in offices23 and statins were just emerging.24 These factors make it likely that the
gains from lifestyle management interventions were higher twenty years ago than they are today.
Second, the studies have a variety of methodological weaknesses, such as a lack of statistical controls for
health status;18–22,25 a lack of adjustment for concomitant participation in disease management;26 and
data limitations, such as imputation of costs from self‐reported use.22 Lastly, the included populations
are not easily generalizable—one study was of retirees21 and another was of 1,000 small‐town city
employees.18
Because of the long latency between reduction of risk factors and avoided onset of chronic disease, it is
possible that longer follow‐up is required to detect savings, but it appears unlikely that the lifestyle
management component of Healthy Living will ever be able to fully offset its cost, particularly when all
net program costs incurred to date are considered. A recent analysis by Howard Bolnick and colleagues
estimated that lowering modifiable risk factors to their theoretical minima would reduce health care
11
costs of an average working‐age adult by 18.4 percent.27 In other words, under perfect conditions, a
lifestyle management program could save $876 per person per year, on average, using the 2012 average
cost of coverage in the United States.28–30 However, even effective programs obviously cannot achieve
the complete elimination of avoidable health risks. As data from the RAND Workplace Wellness
Programs Study show, programs managed to keep a quarter of smokers off nicotine and increased the
share of normal‐weight participants from 21 percent to 33 percent after three years.1 From these
numbers, we estimate that programs can realize about 10–25 percent of the theoretically possible cost
savings, or $88–$219 per year and participant, which roughly corresponds to the $157 average annual
savings estimate from the RAND study. Average annual cost per lifestyle management program
participant for PepsiCo’s program was in line with the $144 reported by Baicker and colleagues.1,8
Together, these estimates suggest that well‐executed lifestyle management programs may be
approximately cost‐neutral.
A lack of financial return does not imply that lifestyle management cannot create value. Our study finds
a significant effect on absenteeism, and both the RAND Workplace Wellness Programs Study and a
recent systematic review showed statistically significant and clinically meaningful improvements in
certain health risks among program participants, even though one needs to caution that most of the
evidence so far has been generated from a limited set of committed employers and may have limited
generalizability.31 At the same time, wellness programs may have unintended consequences in the form
of overdiagnosis and overtreatment.32–34 Further, ADP employer survey data suggest that economic
outcomes are not the only reason employers offer wellness programs: The most common reason is to
improve employee health, with “attract and retain talent” and “maintain or increase benefit offerings”
also offered as key reasons.3
Our and other recent results have several implications for employers, researchers, and policy makers.
First, the current evidence suggests that blanket claims of “wellness saves money” are not warranted,
and it underscores that any program evaluation needs to be scrutinized to understand its results.
Readers should ask which features the program offered, what their respective contribution to the
outcomes was, whether an appropriate comparison strategy was used, and how the results should be
interpreted in light of the comparison strategy. For example, the most recent evaluation of Johnson &
Johnson’s Live for Life program compared the medical cost experience of Johnson & Johnson employees
with the experience of statistically matched employees in similar firms and found annual increases to be
3.7 percent lower among Johnson & Johnson employees.12 This evaluation’s design does imply that
12
Johnson & Johnson’s overall health and wellness strategy is successful, but its comparison strategy
compares Johnson & Johnson employees to similar employees in other firms. In contrast to our analysis,
which compares program participants and statistically matched nonparticipants within one firm, the
Johnson & Johnson evaluation cannot parse out whether individual components of the company’s
wellness program or other company characteristics, or both, such as benefits design, hiring, workplace
policies,35,36 and corporate culture, are driving the results, which makes it difficult for readers to use
such results for wellness program decision making.
Second, employers need to align program configuration with their objectives. If the primary objective is
cost control, they should focus on interventions for higher‐risk employees, such as those with multiple
risk factors or manifest chronic disease. Conversely, if the objective is to improve workforce health,
investment in evidence‐based lifestyle management programs may be warranted.1
Third, employers need to carefully consider the total cost of a program before deciding what to offer
and which vendor to use. For example, the Centers for Disease Control and Prevention is advocating
greater use of awareness campaigns and wellness events as more cost‐effective approaches than
individual coaching.37 Recently issued federal rules38 allow employers to tie substantial incentives to
wellness program participation and control of risk factors.39 Given their magnitude, such incentives can
quickly change an employer’s cost‐benefit calculation. As a matter of public policy, we will need to
understand what proportions of the savings to employers stem from true improvements in health and
what proportions are the result of cost shifting to employees with health risks.40
CONCLUSION
Workplace wellness programs have the potential to reduce health risks and to delay or avoid the onset
of chronic diseases as well as to reduce health care cost in employees with manifest chronic disease. But
employers and policy makers should not take for granted that the lifestyle management component of
such programs can reduce health care costs or even lead to net savings.
ACKNOWLEDGEMENTS
This study was funded by PepsiCo.
Prior to the submission of this dissertation this paper was presented in partial form at the 2013
AcademyHealth Annual Research Meeting and was published in identical form in the journal Health
Affairs (citations provided below).
13
Prior presentations:
Caloyeras JP, Liu H, Exum E, Broderick M, Mattke S. Managing manifest diseases, but not health
risks, saved PepsiCo money over seven years. Health Affairs. 2014;33(1):124‐131.
Caloyeras JP, Liu H, Mattke S, Exum E. Long‐term impact of PepsiCo’s Healthy Living program on
medical costs, health, and absenteeism: a seven‐year evaluation. Podium Presentation.
AcademyHealth 2013 Annual Research Meeting. Baltimore, MD. June 23‐25, 2013.
NOTES
1. Mattke S, Liu H, Caloyeras JP, Huang CY, Van Busum KR, Khodyakov D, et al.Workplace WellnessPrograms Study. Santa Monica (CA): RAND Corporation; 2013. (Pub. No. RR‐254‐DOL).
2. Program components can include health promotional materials, health screening or biometrictesting, smoking cessation or weight loss interventions, access to a nurse help line, and web‐based health coaching.
3. ADP Research Institute.Why you should care about wellness programs [Internet]. Roseland (NJ):ADP Research Institute; 2012 [cited 2013 Dec 11]. Available from: http://www.adp.com/tools‐and‐resources/case‐studies‐white‐papers/~/media/White%20Papers/WellnessFinal22112.ashx
4. Deloitte Center for Health Solutions, Deloitte Consulting. 2012 Deloitte survey of US employers:opinions about the US health care system and plans for employee health benefits [Internet].New York (NY): Deloitte; 2012 Jul [cited 2013 Dec 11]. Available from:http://www.deloitte.com/assets/Dcom‐
UnitedStates/Local%20Assets/Documents/us_dchs_employee_survey_072512.pdf5. Aon Hewitt. 2012 health care survey [Internet]. Chicago (IL): Aon Hewitt; 2012 [cited 2013 Dec
11]. Available from: http://www.aon.com/attachments/human‐capital-consulting/2012_Health_Care_Survey_final.pdf
6. 6 Chapman LS. Meta‐evaluation of worksite health promotion economic return studies: 2005update. Am JHealth Promot. 2005;19(6):1–11.
7. Pelletier KR. A review and analysis of the clinical and cost‐effectiveness studies ofcomprehensive health promotion and disease management programs at the worksite: updateVIII 2008 to 2010. J Occup Environ Med. 2011;53(11):1310–31.
8. Baicker K, Cutler D, Song Z.Workplace wellness programs can generate savings. Health Aff(Millwood). 2010;29(2):304–11.
9. Nyman JA, Abraham JM, Jeffery MM, Barleen NA. The effectiveness of a health promotionprogram after 3 years: evidence from the University of Minnesota. Med Care. 2012; 50(9):772–8.
10. Liu H, Mattke S, Harris K, Weinberger S, Serxner S, Caloyeras JP, et al. Do workplace wellnessprograms reduce medical costs? Evidence from a Fortune 500 company. Inquiry. Forthcoming2014.
14
11. Liu H, Harris KM, Weinberger S, Serxner S, Mattke S, Exum E. Effect of an employer‐sponsored health and wellness program on medical cost and utilization. Popul Health Manag. 2013;16(1):1–6.
12. Henke RM, Goetzel RZ, McHugh J, Isaac F. Recent experience in health promotion at Johnson & Johnson: lower health spending, strong return on investment. Health Aff (Millwood). 2011;30(3):490–9.
13. Elixhauser A, Steiner C, Harris DR, Coffey RM. Comorbidity measures for use with administrative data. Med Care. 1998;36(1):8–27.
14. Bureau of Labor Statistics. Consumer Price Index. Washington (DC): BLS; 2013 Mar. 15. Bureau of Labor Statistics. Employer costs for employee compensation summary. Washington
(DC): BLS; 2013 Mar 12. 16. To access the Appendix, click on the Appendix link to the right of the article online. 17. Gowrisankaran G, Norberg K, Kymes S, Chernew ME, Stwalley D, Kemper L, et al. A hospital
system’s wellness program linked to health plan enrollment cut hospitalizations but not overall costs. Health Aff (Millwood). 2013;32(3):477–85.
18. Aldana SG, Jacobson BH, Harris CJ, Kelley PL, Stone WJ. Influence of a mobile worksite health promotion program on health care costs. Am J Prev Med. 1993;9(6):378–83.
19. Bly JL, Jones RC, Richardson JE. Impact of worksite health promotion on health care costs and utilization. Evaluation of Johnson & Johnson’s Live for Life program. JAMA. 1986; 256(23):3235–40.
20. Fries JF, Harrington H, Edwards R, Kent LA, Richardson N. Randomized controlled trial of cost reductions from a health education program: the California Public Employees’ Retirement System (PERS) study. Am J Health Promot. 1994;8(3): 216–23.
21. Leigh JP, Richardson N, Beck R, Kerr C, Harrington H, Parcell CL, et al. Randomized controlled study of a retiree health promotion program: the Bank of America study. Arch Intern Med. 1992;152(6):1201.
22. Shi L. Health promotion, medical care use, and costs in a sample of worksite employees. Eval Rev. 1993;17(5):475–87.
23. The first state restriction on smoking in the workplace was enacted by California in 1994 as Assembly Bill 13 and became law in 1995 (Labor Code 6404.5).
24. The first statin (lovastatin) for the lowering of cholesterol was approved by the Food and Drug Administration in 1987.
25. Ozminkowski RJ, Dunn RL, Goetzel RZ, Cantor RI, Murnane J, Harrison M. A return on investment evaluation of the Citibank, N.A., health management program. Am J Health Promot. 1999;14(1):31–43.
26. Naydeck BL, Pearson JA, Ozminkowski RJ, Day BT, Goetzel RZ. The impact of the Highmark employee wellness programs on 4‐year healthcare costs. J Occup Environ Med. 2008;50(2):146–56.
27. Bolnick H, Millard F, Dugas JP. Medical care savings from workplace wellness programs: what is a realistic savings potential? J Occup Environ Med. 2013;55(1):4–9.
28. The minimum medical loss ratio allowed by the Affordable Care Act, section 158.210, is 0.85.
15
29. Kaiser Family Foundation. Employer health benefits: 2012 annual survey [Internet]. Menlo Park (CA): KFF; 2012 [cited 2013 Dec 11]. Available from:http://kaiserfamilyfoundation.files.wordpress.com/2013/03/8345‐employer‐health-benefits‐annual‐survey‐full‐report‐0912.pdf
30. Obtained by applying the 18.4 percent savings estimate and the minimum medical loss ratio of85 percent permitted by the Affordable Care Act for the large‐group market to the averageannual premium for single coverage in employer‐sponsored health insurance ($5,615).
31. Osilla KC, Van Busum K, Schnyer C, Larkin JW, Eibner C, Mattke S. Systematic review of theimpact of worksite wellness programs. Am J Manag Care. 2012;18(2):e68–81.
32. Cassels A. Seeking sickness: medical screening and the misguided hunt for disease. Vancouver(BC): Greystone Books; 2012.
33. Krogsbøll LT, Jørgensen KJ, Gøtzsche PC. General health checks in adults for reducing morbidityand mortality from disease. JAMA. 2013;309(23):2489–90.
34. Sox HC. The health checkup: was it ever effective? Could it be effective? JAMA.2013;309(23):2496–7.
35. Asch DA, Muller RW, Volpp KG. Conflicts and compromises in not hiring smokers. N Engl J Med.2013;368(15):1371–3.
36. Schmidt H, Voigt K, Emanuel EJ. The ethics of not hiring smokers. N Engl J Med.2013;368(15):1369–71.
37. Centers for Disease Control and Prevention.Workplace health promotion: workplace healthmodel [Internet]. Atlanta (GA): CDC; [cited 2013 Dec 11]. Available from:http://www.cdc.gov/workplacehealthpromotion/model/index.html
38. Department of Health and Human Services. Incentives for nondiscriminatory wellness programsin group health plans; final rule. Federal Register [serial on the Internet]. 2013;78(106):33157–92. Available from: http://www.gpo.gov/fdsys/pkg/FR‐2013‐06‐03/pdf/2013‐12916.pdf
39. Health‐contingent wellness programs may use incentives up to 30 percent of the cost ofemployee only coverage or up to 50 percent if the program is designed to prevent or reducetobacco use. Using the average annual premium for single coverage, employer‐sponsored healthinsurance ($5,615), the maximum average incentive allowed under the Affordable Care Act isabout $1,680, or $2,800 if also targeting tobacco use.
40. Horwitz JR, Kelly BD, DiNardo JE. Wellness incentives in the workplace: cost savings through costshifting to unhealthy workers. Health Aff (Millwood). 2013;32(3):468–76.
16
EXHIBITS
EXHIBIT 1. Aggregate impact of lifestyle management and disease management on per member per month health care costs at PepsiCo, 2004–11.
SOURCE: Authors’ analysis of PepsiCo health plan and Healthy Living program data.
NOTES: Cost estimates are adjusted by demographics, comorbidities, and calendar years based on propensity score matching and regression analyses. This exhibit assumes that members participated continuously during 2004–11; 2003 is the baseline year.
0
50
100
150
200
250
2003 2004 2005 2006 2007 2008 2009 2010 2011
Total PMPM Cost (2012 $)
Year
Participation in 2004‐2011 No participation
17
EXHIBIT 2. Per member per month cost savings at PepsiCo, by Healthy Living Program component,
2004–11.
SOURCE: Authors’ analysis of PepsiCo health plan and Healthy Living program data.
NOTES: All savings are difference‐in‐differences estimates. “Lifestyle management” is the lifestyle management component; “disease management” is the disease management component; “any program” represents our measure of aggregate participation (participation in either lifestyle management or disease management, or both). Intervals for each estimate represent 95 percent confidence intervals. *p < 0:05.
‐30
‐3
‐136
‐250
‐200
‐150
‐100
‐50
0
50
Any Program* LM DM*
Total per‐member per‐month cost, 2012 USD
18
EXHIBIT 3. Return on investment for PepsiCo’s Healthy Living Program, by program component, 2011.
SOURCE: Authors’ analysis of PepsiCo health plan and Healthy Living program data.
NOTES: Program effects are difference‐in‐differences estimates based on 2004–11 participation data; program costs represent those incurred in 2011. Return on investment denotes savings for each dollar spent. For descriptions of program components, see Exhibit 2 notes.
0.24
3.83
1.30
0.24
‐0.04
0.160.48
3.78
1.46
‐1.0
0.0
1.0
2.0
3.0
4.0
5.0
Lifestyle management Disease management Any program
Return on investment ($)
Healthcare Productivity Overall
19
SUPPLEMENTARYMATERIALS
TABLE OF CONTENTS
Table S1. Member characteristics before and after propensity score matching: cost and utilization
Table S2. Member characteristics before and after propensity score matching: absenteeism
Table S3. Comparison between matched participants and unmatched participants: cost and utilization
Table S4. Comparison between matched participants and unmatched participants: absenteeism
Table S5. Propensity score regression: cost and utilization
Table S6. Propensity score regression: absenteeism
Table S7. Individual fixed effect regression after matching: cost
Table S8. Individual fixed effect regression after matching: inpatient admission
Table S9. Individual fixed effect regression after matching: emergency room visits
Table S10. Random effect interval regression after matching: absenteeism
20
Table S1. Member characteristics before and after propensity score matching: cost and utilization
Member characteristics
Before matching After matching Program
participants (n=46,759)
Non‐participants (n=20,782)
Program participants (n=22,204)
Non‐participants (n=22,204)
Age, No. (%) 18‐34 years 13,385(28.6) 8,217(39.5)** 7,525(33.9) 7,340(33.1) 35‐44 years 16,734(35.8) 7,438(35.8) 8,531(38.4) 8,699(39.2) 45‐54 years 13,343(28.5) 4,345(20.9) 5,436(24.5) 5,382(24.2) 55‐64 years 3,297(7.1) 782(3.8) 712(3.2) 783(3.5)
Male, No. (%) 27,941(59.8) 12,366(59.5) 14,190(63.9) 14,016(63.1) Comorbidities, No. (%) Myocardial infarction 77(0.2) 11(<0.1)** 6(<0.1) 5(<0.1) Congestive heart failure 129(0.3) 26(0.1)** 22(0.1) 25(0.1) Peripheral vascular disease 145(0.3) 39(0.2)** 29(0.1) 31(0.1) Cerebrovascular disease 292(0.6) 63(0.3)** 58(0.3) 49(0.2) Chronic pulmonary disease 2,315(5.0) 769(3.7)** 690(3.1) 662(3.0) Rheumatic disease 325(0.7) 80(0.4)** 67(0.3) 68(0.3) Peptic ulcer disease 101(0.2) 37(0.2) 37(0.2) 30(0.1) Hemiplegia or paraplegia 16(<0.1) 5(<0.1) 1(<0.1) 9(<0.1)* Renal disease 92(0.2) 21(0.1)** 18(<0.1) 14(<0.1) Malignancy 478(1.0) 156(0.8)** 164(0.7) 178(0.8) Metastatic solid tumor 26(<0.1) 8(<0.1) 6(<0.1) 10(<0.1) AIDs 46(0.1) 16(<0.1) 14(<0.1) 16(<0.1) Diabetes 266(0.6) 93(0.5)* 69(0.3) 83(0.4) Liver disease 189(0.4) 23(0.1)** 14(<0.1) 10(<0.1)
Region, No. (%) North East 4,400(9.4) 2,604(12.5)** 2,245(10.1) 2,145(9.7)** North Central 8,662(18.5) 4,203(20.2) 4,547(20.5) 4,901(22.1) West 29,999(64.2) 11,741(56.5) 13,479(60.7) 13,300(59.9) South 3,698(7.9) 2,234(10.8) 1,933(8.7) 1,858(8.4)
Baseline year, No. (%) 2003 18,743(40.1) 7,673(36.9)** 11,184(50.4) 11,184(50.4) 2004 13,301(28.5) 3,037(14.6) 4,010(18.1) 4,010(18.1) 2005 2,944(6.3) 1,371(6.6) 933(4.2) 933(4.2) 2006 2,975(6.4) 1,886(9.1) 1,147(5.2) 1,147(5.2) 2007 2,439(5.2) 1,898(9.1) 1,317(5.9) 1,317(5.9) 2008 2,740(5.9) 2,175(10.5) 1,556(7.0) 1,556(7.0) 2009 2,392(5.1) 1,731(8.3) 1,399(6.3) 1,399(6.3) 2010 1,225(2.6) 1,011(4.9) 658(3.0) 658(3.0)
Number of years in data, mean (standard deviation) 6.1(2.3) 5.5(2.4)** 6.4(2.3) 6.4(2.3)
PMPM Cost in 2012 US Dollars, mean (standard deviation) 279.1(830.9) 210.7(684.1)** 177.9(549.0) 169.2(510.4)
Utilization per 1,000 employee years, mean (standard deviation)
Emergency room visits 203.2(663.4) 196.8(633.8) 159.8(579.2) 160.5(498.4)
21
Member characteristics
Before matching After matching Program
participants (n=46,759)
Non‐participants (n=20,782)
Program participants (n=22,204)
Non‐participants (n=22,204)
Inpatient admissions 31.9(229.7) 24.4(193.6)** 18.9(172.9) 18.0(151.4)
NOTES: For comparisons between participants and non‐participants: * p<0.05; ** p<0.01.
22
Table S2. Member characteristics before and after propensity score matching: absenteeism
Member characteristics
Before matching After matching
Program participants (n=11,235)
Non‐participants (n=2,764)
Program participants (n=3,542)
Non‐participants (n=3,542)
Age, No. (%)
18‐34 years 3,507(31.2) 1,258(45.5)** 978(27.6) 970(27.4)** 35‐44 years 3,805(33.9) 863(31.2) 1,282(36.2) 1,143(32.3) 45‐54 years 3,246(28.9) 542(19.6) 1,074(30.3) 1,229(34.7)
55‐64 years 677(6.0) 101(3.7) 208(5.9) 200(5.7) Male, No. (%) 9,064(80.7) 2,303(83.3)** 2,825(79.8) 2,796(78.9) Comorbidities, No. (%)
Myocardial infarction 5(<0.1) 1(<0.1) 0(0) 3(<0.1) Congestive heart failure 18(0.2) 0(0)* 2(<0.1) 0(0) Peripheral vascular disease 24(0.2) 2(<0.1) 4(0.1) 1(<0.1) Cerebrovascular disease 44(0.4) 7(0.3) 8(0.2) 13(0.4) Chronic pulmonary disease 393(3.5) 56(2.0)** 67(1.9) 59(1.7) Rheumatic disease 46(0.4) 5(0.2) 5(0.1) 3(<0.1) Peptic ulcer disease 19(0.2) 9(0.3) 1(<0.1) 1(<0.1) Hemiplegia or paraplegia 0(0) 0(0) 0(0) 0(0) Renal disease 5(<0.1) 1(<0.1) 0(0) 0(0) Malignancy 77(0.7) 9(0.3)* 14(0.4) 8(0.2) Metastatic solid tumor 1(<0.1) 0(0) 0(0) 0(0) AIDs 6(<0.1) 2(<0.1) 1(<0.1) 0(0) Diabetes 40(0.4) 6(0.2) 8(0.2) 6(0.2)
Liver disease 16(0.1) 2(<0.1) 0(0) 0(0) Region, No. (%)
North East 1,937(17.2) 514(18.6)** 444(12.5) 387(10.9) North Central 2,368(21.1) 509(18.4) 1,001(28.3) 986(27.8) West 5,454(48.5) 1,331(48.2) 1,767(49.9) 1,800(50.8)
South 1,476(13.1) 410(14.8) 330(9.3) 369(10.4) Baseline year, No. (%)
2004 205(1.8) 41(1.5)** 4(0.1) 4(0.1) 2005 3,422(30.5) 950(34.4) 395(11.2) 395(11.2) 2006 1,060(9.4) 301(10.9) 89(2.5) 89(2.5) 2007 712(6.3) 243(8.8) 58(1.6) 58(1.6) 2008 498(4.4) 263(9.5) 56(1.6) 56(1.6) 2009 5,167(46.0) 921(33.3) 2,930(82.7) 2,930(82.7) 2010 171(1.5) 45(1.6) 10(0.3) 10(0.3)
Number of years in data, mean (standard deviation) 3.3(1.5) 3.2(1.4)* 2.5(1.1) 2.5(1.1)
PMPM Cost in 2012 US Dollars, mean (standard deviation) 186.2(445.7) 116.9(288.9)** 146.7(382.1) 135.9(303.4)
Utilization per 1,000 employee years, mean (standard deviation)
Emergency room visits 132.9(412.4) 125.5(413.9) 84.4(317.0) 95.7(342.2)
Inpatient admissions 12.9(120.5) 6.5(80.5)** 5.1(71.1) 5.4(73.1)
23
Member characteristics
Before matching After matching
Program participants (n=11,235)
Non‐participants (n=2,764)
Program participants (n=3,542)
Non‐participants (n=3,542)
Absenteeism, No. (%)
0 days 6,656(59.2) 1,575(57.0) 2,402(67.8) 2,442(68.9) 1‐2 days 2,659(23.7) 697(25.2) 750(21.2) 689(19.5) 3‐5 days 1,375(12.2) 361(13.1) 298(8.4) 315(8.9) 6 days or more 545(4.9) 131(4.7) 92(2.6) 96(2.7)
NOTES: For comparisons between participants and non‐participants: * p<0.05; ** p<0.01.
24
Table S3. Comparison between matched participants and unmatched participants: cost and utilization
Member characteristics Matched participants (n=22,204) Non‐matched participants (n=24,555)
Age, No. (%) 18‐34 years 7,525(33.9) 5,860(23.9)** 35‐44 years 8,531(38.4) 8,203(33.4) 45‐54 years 5,436(24.5) 7,907(32.2) 55‐64 years 712(3.2) 2,585(10.5)
Male, No. (%) 14,190(63.9) 13,751(56.0)** Comorbidities, No. (%) Myocardial infarction 6(<0.1) 71(0.3)** Congestive heart failure 22(0.1) 107(0.4)** Peripheral vascular disease 29(0.1) 116(0.5)** Cerebrovascular disease 58(0.3) 234(1.0)** Chronic pulmonary disease 690(3.1) 1,625(6.6)** Rheumatic disease 67(0.3) 258(1.1)** Peptic ulcer disease 37(0.2) 64(0.3)* Hemiplegia or paraplegia 1(<0.1) 15(<0.1)** Renal disease 18(<0.1) 74(0.3)** Malignancy 164(0.7) 314(1.3)** Metastatic solid tumor 6(<0.1) 20(<0.1)* AIDs 14(<0.1) 32(0.1)* Diabetes 69(0.3) 197(0.8)** Liver disease 14(<0.1) 175(0.7)**
Region, No. (%) North East 2,245(10.1) 2,155(8.8)** North Central 4,547(20.5) 4,115(16.8) West 13,479(60.7) 16,520(67.3) South 1,933(8.7) 1,765(7.2)
Baseline year, No. (%) 2003 11,184(50.4) 7,559(30.8)** 2004 4,010(18.1) 9,291(37.8) 2005 933(4.2) 2,011(8.2) 2006 1,147(5.2) 1,828(7.4) 2007 1,317(5.9) 1,122(4.6) 2008 1,556(7.0) 1,184(4.8) 2009 1,399(6.3) 993(4.0) 2010 658(3.0) 567(2.3)
Number of years in data, mean (standard deviation) 6.4(2.3) 5.9(2.2)**
PMPM Cost in 2012 US Dollars, mean (standard deviation) 177.9(549.0) 370.6(1,012.2)**
Utilization per 1,000 employee years, mean (standard deviation)
Emergency room visits 159.8(579.2) 242.5(729.0)** Inpatient admissions 18.9(172.9) 43.6(270.5)**
NOTES: For comparisons between matched participants and non‐matched participants: * p<0.05; ** p<0.01.
25
Table S4. Comparison between matched participants and unmatched participants: absenteeism
Member characteristics Matched participants (n=3,542) Non‐matched participants (n=7,695)
Age, No. (%) 18‐34 years 978(27.6) 2,529(32.9)** 35‐44 years 1,282(36.2) 2,524(32.8) 45‐54 years 1,074(30.3) 2,173(28.2) 55‐64 years 208(5.9) 469(6.1)
Male, No. (%) 2,825(79.8) 6,241(81.1) Comorbidities, No. (%) Myocardial infarction 0(0) 5(<0.1) Congestive heart failure 2(<0.1) 16(0.2) Peripheral vascular disease 4(0.1) 20(0.3) Cerebrovascular disease 8(0.2) 36(0.5) Chronic pulmonary disease 67(1.9) 326(4.2)** Rheumatic disease 5(0.1) 41(0.5)** Peptic ulcer disease 1(<0.1) 18(0.2)* Hemiplegia or paraplegia 0(0) 0(0) Renal disease 0(0) 5(<0.1) Malignancy 14(0.4) 63(0.8)* Metastatic solid tumor 0(0) 1(<0.1) AIDs 1(<0.1) 5(<0.1) Diabetes 8(0.2) 32(0.4) Liver disease 0(0) 16(0.2)**
Region, No. (%) North East 444(12.5) 1,493(19.4)** North Central 1,001(28.3) 1,367(17.8) West 1,767(49.9) 3,689(47.9) South 330(9.3) 1,146(14.9)
Baseline year, No. (%) 2004 4(0.1) 203(2.6)** 2005 395(11.2) 3,027(39.3) 2006 89(2.5) 971(12.6) 2007 58(1.6) 654(8.5) 2008 56(1.6) 442(5.7) 2009 2,930(82.7) 2,237(29.1) 2010 10(0.3) 161(2.1)
Number of years in data, mean (standard deviation) 2.5(1.1) 3.7(1.6)**
PMPM Cost in 2012 US Dollars, mean (standard deviation) 146.7(382.1) 204.3(470.9)**
Utilization per 1,000 employee years, mean (standard deviation)
Emergency room visits 84.4(317.0) 155.2(447.8)** Inpatient admissions 5.1(71.1) 16.5(137.2)**
Absenteeism, No. (%) 0 days 2,402(67.8) 4,256(55.3)**
26
Member characteristics Matched participants (n=3,542) Non‐matched participants (n=7,695)
1‐2 days 750(21.2) 1,909(24.8) 3‐5 days 298(8.4) 1,077(14.0) 6 days or more 92(2.6) 453(5.9)
NOTES: For comparisons between matched participants and non‐matched participants: * p<0.05; ** p<0.01.
27
Table S5. Propensity score regression: cost and utilization
Dependent variable = program participation, LM, DM, or both Model form = multinomial probit
Coefficient Standard error
Outcome = LM participation
Age
18‐34 years (reference)
35‐44 years 0.04 0.02 45‐54 years 0.09** 0.02 55‐64 years 0.12* 0.05
Employee 0.17** 0.02 Male ‐0.20** 0.02 Region
South (reference)
North East <0.01 0.04 North Central 0.25** 0.03 West 0.07* 0.03
Program year
2003 (reference)
2004 0.12** 0.03 2005 0.05 0.04 2006 ‐0.12** 0.04 2007 ‐0.05 0.04 2008 0.19** 0.04 2009 0.53** 0.05 2010 0.58** 0.06
Number of data years available 0.10** 0.01 Emergency room visits ‐0.04** 0.02 Inpatient admissions ‐0.06 0.05 Office visits <0.01 <0.01 PMPM cost, logged 0.03** <0.01 Comorbidities
Myocardial infarction ‐0.21 0.42 Congestive heart failure 0.11 0.27 Peripheral vascular disease 0.03 0.23 Cerebrovascular disease 0.10 0.16 Chronic pulmonary disease 0.06 0.05 Rheumatic disease 0.28* 0.14 Peptic ulcer disease 0.01 0.21 Hemiplegia or paraplegia 0.18 0.55 Renal disease ‐0.11 0.33 Malignancy 0.03 0.10 Metastatic solid tumor 0.68 0.49 AIDs 0.05 0.31 Diabetes 0.17 0.13 Liver disease <0.01 0.27
28
Dependent variable = program participation, LM, DM, or both Model form = multinomial probit
Coefficient Standard error
Outcome = DM participation
Age
18‐34 years (reference)
35‐44 years 0.01 0.03 45‐54 years 0.28** 0.03 55‐64 years 0.72** 0.05
Employee ‐0.01 0.03 Male 0.33** 0.03 Region
South (reference)
North East ‐0.02 0.05 North Central ‐0.21** 0.05 West 0.07 0.04
Program year
2003 (reference)
2004 0.33** 0.05 2005 ‐0.01 0.05 2006 ‐0.51** 0.05 2007 ‐0.98** 0.05 2008 ‐1.54** 0.07 2009 ‐2.58** 0.12 2010 ‐0.10** 0.01
Number of data years available <0.01 0.02 Emergency room visits ‐0.02 0.04 Inpatient admissions 0.01 <0.01 Office visits 0.02** <0.01 PMPM cost, logged <0.01** <0.01 Comorbidities
Myocardial infarction 0.54 0.28 Congestive heart failure 0.17 0.19 Peripheral vascular disease 0.02 0.18 Cerebrovascular disease 0.30* 0.13 Chronic pulmonary disease 0.15** 0.05 Rheumatic disease 0.38** 0.13 Peptic ulcer disease ‐0.10 0.20 Hemiplegia or paraplegia 0.51 0.54 Renal disease 0.10 0.22 Malignancy ‐0.05 0.10 Metastatic solid tumor ‐0.05 0.39 AIDs 0.11 0.31 Diabetes ‐0.23 0.13 Liver disease 0.74** 0.18
Outcome = Both LM and DM participation
29
Dependent variable = program participation, LM, DM, or both Model form = multinomial probit
Coefficient Standard error
Age
18‐34 years (reference)
35‐44 years 0.02 0.03 45‐54 years 0.33** 0.04 55‐64 years 0.74** 0.06
Employee 0.09* 0.03 Male 0.18** 0.03 Region
South (reference)
North East ‐0.12* 0.06 North Central ‐0.06 0.05 West 0.04 0.05
Program year
2003 (reference)
2004 1.49** 0.04 2005 0.26** 0.06 2006 <0.01 0.06 2007 ‐0.32** 0.07 2008 ‐0.64** 0.07 2009 ‐1.07** 0.09 2010 ‐1.76** 0.15
Number of data years available 0.03** 0.01 Emergency room visits ‐0.04 0.02 Inpatient admissions ‐0.02 0.06 Office visits 0.02** <0.01 PMPM cost, logged 0.05** 0.01 Comorbidities
Myocardial infarction 0.75* 0.37 Congestive heart failure 0.19 0.26 Peripheral vascular disease 0.18 0.25 Cerebrovascular disease 0.32 0.17 Chronic pulmonary disease 0.26** 0.06 Rheumatic disease 0.48** 0.16 Peptic ulcer disease ‐0.43 0.25 Hemiplegia or paraplegia ‐0.78 0.84 Renal disease ‐0.04 0.31 Malignancy ‐0.17 0.12 Metastatic solid tumor 0.30 0.53 AIDs ‐0.12 0.38 Diabetes ‐0.19 0.16 Liver disease 0.91** 0.23
NOTES: * p<0.05; ** p<0.01.
30
Table S6. Propensity score regression: absenteeism
Dependent variable = program participation, LM, DM, or both Model form = multinomial probit
Coefficient Standard error
Outcome = LM participation
Age
18‐34 years (reference)
35‐44 years 0.14** 0.05 45‐54 years 0.16** 0.06 55‐64 years 0.15 0.10
Male ‐0.10 0.05 Region
South (reference)
North East 0.03 0.08 North Central 0.11 0.07 West 0.08 0.07
Program year
2010 (reference)
2004 ‐1.57** 0.24 2005 ‐2.30** 0.17 2006 ‐2.03** 0.17 2007 ‐1.49** 0.17 2008 ‐1.44** 0.16 2009 0.27 0.14
Number of data years available 0.32** 0.03 Emergency room visits ‐0.05 0.06 Inpatient admissions 0.17 0.25 Office visits 0.01 0.01 PMPM cost, logged 0.02 0.01 Absenteeism
0 days (reference)
1‐2 days ‐0.02 0.05 3‐5 days ‐0.16* 0.07 6 days or more ‐0.23* 0.11
Comorbidities
Myocardial infarction ‐0.43 0.96 Congestive heart failure 4.15 45.73 Peripheral vascular disease 0.19 0.58 Cerebrovascular disease ‐0.03 0.40 Chronic pulmonary disease 0.09 0.14 Rheumatic disease 0.28 0.40 Peptic ulcer disease ‐0.69 0.46 Renal disease ‐1.40 1.23 Malignancy 0.24 0.31 Metastatic solid tumor 6.40 7939 AIDs ‐0.67 0.78
31
Dependent variable = program participation, LM, DM, or both Model form = multinomial probit
Coefficient Standard error
Diabetes 0.21 0.42 Liver disease ‐0.49 0.77
Outcome = DM participation
Age
18‐34 years (reference)
35‐44 years ‐0.24* 0.10 45‐54 years ‐0.26* 0.11 55‐64 years 0.03 0.17
Male 0.40** 0.10 Region
South (reference)
North East ‐0.02 0.14 North Central ‐0.21 0.14 West ‐0.20 0.12
Program year
2010 (reference)
2004 3.43** 0.68 2005 3.09** 0.55 2006 3.34** 0.56 2007 3.11** 0.56 2008 2.87** 0.56 2009 0.87 0.54
Number of data years available 0.20** 0.04 Emergency room visits ‐0.06 0.08 Inpatient admissions 0.66* 0.33 Office visits <0.01 0.02 PMPM cost, logged ‐0.03 0.02 Absenteeism
0 days (reference)
1‐2 days 0.21* 0.10 3‐5 days 0.05 0.12 6 days or more 0.09 0.16
Comorbidities
Myocardial infarction ‐1.72 1.42 Congestive heart failure 4.27 45.72 Peripheral vascular disease 1.02 0.75 Cerebrovascular disease ‐0.30 0.53 Chronic pulmonary disease 0.43* 0.19 Rheumatic disease 0.73 0.57 Peptic ulcer disease ‐0.77 0.59 Renal disease ‐0.61 1.16 Malignancy 0.54 0.43 Metastatic solid tumor 0.12 9331 AIDs ‐0.66 1.73
32
Dependent variable = program participation, LM, DM, or both Model form = multinomial probit
Coefficient Standard error
Diabetes ‐0.42 0.55 Liver disease ‐0.28 0.79
Outcome = Both LM and DM participation
Age
18‐34 years (reference)
35‐44 years 0.06 0.11 45‐54 years <0.01 0.12 55‐64 years 0.49** 0.17
Male 0.41** 0.10 Region
South (reference)
North East ‐0.32* 0.15 North Central ‐0.36* 0.15 West ‐0.23 0.13
Program year
2010 (reference)
2004 2.35** 0.58 2005 1.61** 0.41 2006 2.07** 0.41 2007 2.27** 0.41 2008 2.32** 0.41 2009 0.65 0.37
Number of data years available 0.44** 0.06 Emergency room visits ‐0.03 0.09 Inpatient admissions 0.70* 0.34 Office visits 0.01 0.02 PMPM cost, logged ‐0.04 0.02 Absenteeism
0 days (reference)
1‐2 days 0.30** 0.10 3‐5 days 0.11 0.12 6 days or more ‐0.09 0.17
Comorbidities
Myocardial infarction ‐1.15 1.28 Congestive heart failure 5.05 45.73 Peripheral vascular disease 0.45 0.72 Cerebrovascular disease 0.36 0.52 Chronic pulmonary disease 0.49* 0.20 Rheumatic disease 1.02 0.54 Peptic ulcer disease ‐1.79* 0.75 Renal disease ‐1.23 1.30 Malignancy 0.66 0.42 Metastatic solid tumor ‐0.74 10358
33
Dependent variable = program participation, LM, DM, or both Model form = multinomial probit
Coefficient Standard error
AIDs 0.20 1.03 Diabetes ‐0.14 0.59 Liver disease 0.98 0.75
NOTES: * p<0.05; ** p<0.01. Employee status and hemiplegia dropped because of collinearity.
34
Table S7. Individual fixed effect regression after matching: cost
Dependent variable = Difference in PMPM cost Coefficient Standard
error
Program participation
LM current year ‐0.26 2.74 DM current year 20.39* 10.26 LM one year ago ‐4.59 4.34 LM two years ago ‐4.71 3.39 LM three years ago 1.58 3.20 LM four years ago ‐4.41 3.85 LM five years ago 10.26* 4.13 LM six years ago 18.62* 7.35 DM one year ago ‐35.43** 11.29 DM two years ago ‐33.46** 11.98 DM three years ago ‐60.39** 11.80 DM four years ago ‐49.83** 13.02 DM five years ago ‐53.99** 14.27 DM six years ago ‐35.56 24.77 DM seven years ago ‐64.17** 21.60
Age
18‐34 years (reference)
35‐44 years ‐3.16 4.62 45‐54 years ‐3.28 6.96 55‐64 years ‐3.66 9.94
Region
South (reference)
North East 82.79 42.53 North Central 71.64 47.18 West 75.80 41.82
Program year
2004 (reference)
2005 20.24** 5.01 2006 17.10** 4.98 2007 16.48** 4.99 2008 15.18** 5.09 2009 19.88** 5.25 2010 19.68** 5.35 2011 14.04* 5.69
Comorbidities
Myocardial infarction 921.97** 300.15 Congestive heart failure 215.84* 94.20 Peripheral vascular disease 192.32** 53.12 Cerebrovascular disease 280.34** 51.50 Chronic pulmonary disease 63.27** 7.59 Rheumatic disease 59.77** 19.33 Peptic ulcer disease 225.79** 56.11 Hemiplegia or paraplegia 13.50 176.41 Renal disease ‐19.02 98.35
35
Dependent variable = Difference in PMPM cost Coefficient Standard
error
Malignancy 296.55** 48.55 Metastatic solid tumor 978.45** 361.80 AIDs 202.03** 67.26 Diabetes 192.96** 39.54 Liver disease 173.45** 56.64
NOTES: * p<0.05; ** p<0.01.
36
Table S8. Individual fixed effect regression after matching: inpatient admission
Dependent variable = Difference in number of IP admissions Coefficient Standard
error
Program participation
LM current year 0.00019 0.00088 DM current year 0.00509 0.00281 LM one year ago 0.00063 0.00104 LM two years ago 0.00024 0.00102 LM three years ago 0.00067 0.00099 LM four years ago ‐0.00032 0.00105 LM five years ago 0.00178 0.00133 LM six years ago 0.00260 0.00185 DM one year ago ‐0.00910** 0.00298 DM two years ago ‐0.00764* 0.00302 DM three years ago ‐0.01067** 0.00306 DM four years ago ‐0.01098** 0.00327 DM five years ago ‐0.01044** 0.00362 DM six years ago ‐0.01035* 0.00456 DM seven years ago ‐0.01348* 0.00584
Age
18‐34 years (reference)
35‐44 years 0.00216 0.00172 45‐54 years 0.00129 0.00228 55‐64 years 0.00085 0.00321
Region
South (reference)
North East 0.02233* 0.00940 North Central 0.02533** 0.00872 West 0.02060* 0.00889
Program year
2004 (reference)
2005 0.00058 0.00187 2006 0.00155 0.00193 2007 ‐0.00032 0.00194 2008 0.00053 0.00197 2009 ‐0.00042 0.00199 2010 0.00082 0.00202 2011 ‐0.00017 0.00210
Comorbidities
Myocardial infarction 0.21729** 0.08015 Congestive heart failure 0.05147* 0.02611 Peripheral vascular disease 0.03833** 0.01326 Cerebrovascular disease 0.06318** 0.01071 Chronic pulmonary disease 0.01072** 0.00271 Rheumatic disease ‐0.00621 0.00663 Peptic ulcer disease 0.04944** 0.01426 Hemiplegia or paraplegia ‐0.10435 0.07375 Renal disease 0.05074 0.03149
37
Dependent variable = Difference in number of IP admissions Coefficient Standard
error
Malignancy 0.01937* 0.00925 Metastatic solid tumor 0.09157** 0.03240 AIDs 0.11656 0.08764 Diabetes 0.03060** 0.00767 Liver disease 0.01460* 0.00630
NOTES: * p<0.05; ** p<0.01.
38
Table S9. Individual fixed effect regression after matching: emergency room visits
Dependent variable = Difference in number of ER visits Coefficient Standard
error
Program participation
LM current year 0.00533* 0.00245 DM current year 0.01364* 0.00683 LM one year ago ‐0.00548 0.00297 LM two years ago ‐0.00160 0.00288 LM three years ago ‐0.00060 0.00287 LM four years ago ‐0.00835* 0.00356 LM five years ago 0.00043 0.00411 LM six years ago ‐0.00466 0.00610 DM one year ago ‐0.01620* 0.00669 DM two years ago ‐0.00500 0.00606 DM three years ago ‐0.00421 0.00669 DM four years ago ‐0.00598 0.00780 DM five years ago ‐0.01071 0.00762 DM six years ago 0.01580 0.00931 DM seven years ago ‐0.10371** 0.01467
Age
18‐34 years (reference)
35‐44 years ‐0.01462** 0.00546 45‐54 years ‐0.02129** 0.00718 55‐64 years ‐0.01460 0.00925
Region
South (reference)
North East ‐0.07390 0.05272 North Central ‐0.01387 0.04160 West ‐0.07798 0.04223
Program year
2004 (reference)
2005 0.01547** 0.00500 2006 0.01959** 0.00510 2007 0.02587** 0.00528 2008 0.02209** 0.00538 2009 0.02610** 0.00549 2010 0.02522** 0.00564 2011 0.02626** 0.00589
Comorbidities
Myocardial infarction 0.01226 0.09251 Congestive heart failure 0.06119 0.03765 Peripheral vascular disease 0.06502* 0.02507 Cerebrovascular disease 0.13578** 0.02058 Chronic pulmonary disease 0.06175** 0.00674 Rheumatic disease 0.01720 0.01562 Peptic ulcer disease 0.05086* 0.02529 Hemiplegia or paraplegia 0.07860 0.11145 Renal disease ‐0.12958 0.15779
39
Dependent variable = Difference in number of ER visits Coefficient Standard
error
Malignancy 0.02505 0.01797 Metastatic solid tumor 0.05803 0.06586 AIDs 0.08837 0.06145 Diabetes 0.05931** 0.01465 Liver disease 0.01251 0.02368
NOTES: * p<0.05; ** p<0.01.
40
Table S10. Random effect interval regression after matching: absenteeism
Dependent variable = Difference in number days missed Coefficient Standard
error
Program participation
LM current year ‐0.12950** 0.01927 DM current year 0.06593 0.03841
Age
18‐34 years (reference)
35‐44 years 0.03552 0.02527 45‐54 years ‐0.07597** 0.02731 55‐64 years ‐0.14277** 0.03843
Region
South (reference)
North East ‐0.02398 0.04156 North Central ‐0.01369 0.03712 West 0.10857** 0.03396
Program year
2005 (reference)
2006 0.39970 0.27655 2007 0.22501 0.27635 2008 0.12205 0.27620 2009 0.05255 0.27594 2010 0.11703 0.27609 2011 0.13892 0.27632
Comorbidities
Myocardial infarction ‐0.29251 0.86262 Congestive heart failure 0.15025 0.30132 Peripheral vascular disease ‐0.28645 0.15577 Cerebrovascular disease 0.04610 0.12348 Chronic pulmonary disease 0.37585** 0.05751 Rheumatic disease 0.36007 0.22792 Peptic ulcer disease ‐0.34711 0.29649 Hemiplegia or paraplegia 0.27188 0.75424 Renal disease ‐0.14250 0.65812 Malignancy 0.25001* 0.09790 Metastatic solid tumor ‐0.04644 0.61670 AIDs ‐0.59425 0.87849 Diabetes 0.19026 0.16691 Liver disease ‐0.31401 0.20448
NOTES: * p<0.05; ** p<0.01.
41
PAPER2:MONETARYPENALTIESMARGINALLYINCREASEPARTICIPATIONINWORKPLACETOBACCOCESSATIONPROGRAMSBUTALSOSHIFTCOSTSTOTHOSEWITHLOWERSOCIOECONOMICSTATUS
Caloyeras JP1,2 MPhil
1. Pardee RAND Graduate School, Santa Monica, CA
2. RAND Corporation, Santa Monica, CA
42
MAINTEXT
INTRODUCTION
Wellness programs broadly aim to improve health by identifying health risks, promoting healthy
behaviors, and helping those with manifest diseases manage their conditions. In 2016, 83% of employers
(with 200 or more workers) providing health benefits offered behavioral lifestyle management
intervention(s) and 59% offered health risk assessments.1
To improve employee health, wellness programs must first get employees to participate. Frustrated with
low participation rates,2 employers and wellness vendors have tried a variety of approaches to increase
program uptake, such as “nudging” employees to participate with modest financial incentives.3
Although financial incentives are common—48% of large employers reported using them in 20161,4—
some firms are now deploying incentive schemes that go well beyond nudges and instead represent real
hits to take‐home pay. Employer survey data from 2016 found 16% of firms with a wellness program
offered incentives exceeding $1,000 per employee per year, and 7% of firms had incentives surpassing
the $2,000 mark. To frame the magnitude of these incentives, consider the median US worker with
annual earnings of about $45,000:5 a $2,000 wellness incentive equates to over 4% of income, or about
the same amount the average American spends on gasoline each year.6
Concerns regarding the impact of large incentives on employees have led the EEOC to sue several
employers7‐9 and the AARP to sue the EEOC.10 These suits, in part, question the degree to which program
participation remains voluntary given not participating comes with a real hit to take‐home pay.
Despite concerns regarding incentive magnitudes, the Preserving Employee Wellness Programs Act (H.R.
1313) introduced in March of 2017 proposes to increase incentives even further by allowing the 30% cap
(50% if tied to tobacco use) on incentives to be applied to the cost of family coverage rather than
individual coverage.11 Using 2016 averages for the cost of employer sponsored individual ($6,435) and
family ($18,142) coverage and assuming the employee is a tobacco user, means the proposed law would
allow incentives of over $9,100 per employee per year to be used.
Against this backdrop of legal and policy activities regarding wellness incentives, it is unfortunate that
the study of wellness incentives remains insufficient, particularly from the employee perspective.12 To
that end, this study aimed to generate evidence on wellness incentive benefits and risks, taking the lens
of both employees and the employer, by exploring the distributional and behavioral effects of a $600
43
non‐participation penalty tied to tobacco use in a Fortune‐100 employer’s wellness program.
Distributional effects of the penalty are examined by highlighting differences between employees
eligible for the penalty with those who are not. An accounting exercise is conducted to show the net
transfers of penalty dollars between employees, the employer and the wellness vendor. Behavioral
effects of the penalty are estimated as the degree to which the penalty increased program participation,
and if so, whether this effect is modified by employee characteristics such as income or health status.
STUDY DATA AND METHODS
THE TOBACCO CESSATION AND LIFESTYLE MANAGEMENT INTERVENTIONS
The analysis centers on the Fortune‐100 firm’s tobacco cessation intervention, which is one of five
telephonic‐based, lifestyle management (LM) interventions offered in the firm’s wellness program (the
others are weight management, nutrition management, physical activity, and pre‐diabetes
management). Telephonic sessions are conducted by wellness coaches, who provide education, advice,
and motivational support tailored to the needs of each participant, typically over a six to twelve‐month
period. LM intervention eligibility is determined by health‐risk assessments. As described below,
eligibility for the tobacco cessation intervention expanded after restructuring of the wellness program in
2009 to pull in employee tobacco status as captured during mandatory reporting by employees during
the annual health insurance enrollment period.
THE TOBACCO CESSATION SURCHARGE
The firm’s wellness program began in 2003, with all telephonic LM interventions implemented in year 3
(2005) of the program. In 2009, the firm restructured components of their wellness program, which
included the introduction of a $600 non‐participation penalty (or “surcharge”) for members identified as
eligible for the tobacco cessation intervention, determined from mandatory tobacco‐use reporting by
employees during annual health insurance enrollment. The $600 surcharge is deducted on a bi‐weekly
basis from employee paychecks in $25 increments. To avoid the surcharge eligible members are
required to participate in the cessation program as determined by proprietary vendor rules.
DATA AND STUDY SAMPLE
Years of member health and pharmacy plan claims from 2005 to 2011 were merged with eligibility and
participation data from the wellness program. Wellness program data from 2009, the year the program
was restructured and the tobacco surcharge was introduced, were excluded because during the
44
program restructuring year newly eligibles were not identified (participants in 2009 were made up of
only those identified in 2008 that chose to continue participating).
The analytic sample was limited to employees aged 18‐64 years, who were not enrolled in a HMO and
not eligible for the complex‐care management component of the wellness program (designed for those
with very serious illnesses like cancer). Member‐years with evidence of pregnancies were excluded, and
members were required to have at least two complete years of health plan enrollment. The resulting
sample, containing 64,966 unique members and 383,395 member‐years of data, is used for analyses
focused on comparing the characteristics of tobacco users to non‐users. Analyses of the effect of the
tobacco‐use surcharge on telephonic LM participation employs a sample restricted to member‐years
with telephonic LM eligibility, which reflects 49,515 unique members and 90,570 member‐years of data.
ANALYTIC APPROACH AND MEASURES
A difference‐in‐differences (DD) analytic design was used to estimate the incremental effect of the $600
tobacco cessation surcharge on the tobacco cessation participation rate. The DD approach allows for
isolation of the effect of the surcharge on the tobacco cessation participation rate by accounting for
changes that occurred in the four (similar) telephonic LM interventions in which no surcharges were
introduced.
To implement the DD approach the group of tobacco users (identified above) were compared over time
to a clean “comparison group” made up of individuals eligible for any of the other four telephonic LM
interventions offered in the wellness program. The comparison group, by not being exposed to a
surcharge for telephonic LM non‐participation, thereby served as a control group for what changes
would have been expected to the participation rate in the tobacco group had the surcharge not been
introduced.
As is formally shown in the Appendix, the outcome measure is a dichotomous variable equal to one if a
telephonic LM eligible employee in a given year participated and zero, otherwise. For the tobacco users
the participation outcome variable was defined using only tobacco cessation participation. By contrast,
the participation outcome was defined for the comparison group using any participation in any of the
four other telephonic LM interventions.
The “treatment” variable is equal to one if the employee was eligible to participate (i.e., subject to the
surcharge) in the tobacco cessation intervention and zero, otherwise. The “treatment” variable is
interacted with a “post” variable equal to one for the years in which the non‐participation penalty was
45
in effect and zero in years prior. Thus, the effect of the non‐participation penalty is identified by the
coefficient on this interaction term (treatment*post) in a standard DD framework.
The model is estimated using a logistic regression where additional covariates are included such as
categorical age, gender, pay type (salaried or hourly), median household income in the employee’s zip
code of residence, lagged health care costs and lagged health status (Charlson comorbidity index). To
explore whether patient characteristics may modify the effect of the surcharge on the participate rate, a
series of interaction terms were added to each regression in three separate analyses (regressions were
repeated with the addition of these triple interaction terms between employee characteristics and the
DD estimator).
The main analysis was conducted using only data on member characteristics from medical and
pharmacy eligibility and claims data. As a robustness check, two additional specifications were
estimated restricting to employees who completed the HRA: first, the same model as described above
was estimated by simply restricting to the 52,875 employees for whom HRA data are available. Second,
this restricted “HRA” sample was used and included additional covariates obtained from the HRA
(categorical variables for education level, ethnicity, job classification and BMI).
For all regressions predicted probabilities of participation were generated for the tobacco user group
and for the comparison group in both the pre and post surcharge periods. All predictions were
calculated as average marginal effects using the margins command in Stata v9.
Employees who use tobacco are compared to those who do not using independent samples t‐tests for
continuous variables and Chi‐square tests for categorical variables. Logistic regression is used to
examine the impact of individual characteristics on the likelihood of tobacco use; regression results are
translated into predicted probabilities of tobacco use using average marginal effects. The proportion of
tobacco users who chose to pay the surcharge (rather than participate) is estimated using tobacco
cessation program participation data; a $150 per‐participant per‐year fee is assumed for the cost paid by
the employer to the wellness vendor. Participation intensity is examined by constructing a boxplot of
participation sessions per‐tobacco user per‐year, restricted to those who had at least one telephonic
session.
LIMITATIONS
46
The analysis centers on data from a large employer with an established wellness program; as such,
caution is warranted in generalizing the results to smaller employers or to those with newer wellness
programs.
The DD estimator will produce biased results if unobserved factors that influence the decision to
participate changed between the tobacco group and comparison group before and after the
introduction of the surcharge. For example, I cannot rule out that as part of the wellness program
restructuring factors like how individuals were invited to participate (e.g., personalized phone calls vs.
generic emails) did not change between the tobacco cessation intervention and the other lifestyle
management telephonic coaching interventions. The estimated effect may also be biased downwards if
tobacco users are opting to participate in a cessation program outside of the employer’s wellness
program to avoid the penalty, which is allowed in the firm’s program.
The results are limited by the fact that employees may falsely report their tobacco use when enrolling
annually for health benefits. However, this is a limitation inherent in any wellness program that relies
upon employee self‐reports to determine program eligibility.
To avoid the non‐participation surcharge employees were required to participate in the cessation
program. However, as the rules that define “active participation” are proprietary to the wellness vendor,
the definition used in the study may differ. A disconnect between the true definition to avoid the
surcharge in the program and the definition used in this study may bias results. Future analyses could
explore the impact of different definitions of participation required to avoid the surcharge, including
altering the outcome variable to an indicator variable for program completion.
RESULTS
In the four years prior to the implementation of the tobacco cessation surcharge, the participation rate
in the tobacco cessation intervention averaged 22.4 percent (Exhibit 1). Following the introduction of
the surcharge, the participation rate increased to 26.9 percent—an (unadjusted) increase of 4.5
percentage points.
The average number of telephonic coaching sessions among tobacco users with at least one session
increased during the initial years of the program to a high of 3.4 sessions per‐participant per‐year in
2008 (Exhibit 2). In the two years after the introduction of the surcharge the average sessions per‐
participant per‐year was 2.8 in 2010, increasing to 3.4 in 2011.
47
The introduction of the tobacco surcharge is estimated to have increased participation in the tobacco
cessation intervention by between 5.5 and 12.7 percentage points (all estimates significantly different
from zero at p<0.001) (Exhibit 3). Estimates were higher for analyses using the full analytic sample
versus those using the restricted sample to employees that completed at least one HRA. The adjusted
participation rates for the cases (tobacco cessation intervention eligibles) and controls (lifestyle
management intervention eligibles except for tobacco cessation) in the pre and post surcharge periods
are provided for each analysis in Appendix Exhibit A5.
Employee characteristics found to positively modify the effect of the surcharge on the participation rate
included being paid hourly versus salaried (4.2 percentage point difference), being a non‐unionized
versus unionized employee (28.7 percentage point difference), and living in the top (vs. bottom) decile
of employee income zip‐codes (4.7 percentage point difference; only statistically significant in the
analysis using the full analytic sample).
Poorer health status, as measured by being in the top (vs. bottom) decile of per‐member per‐month
health care costs was estimated to increase the effect of the surcharge on participation by 3.0
percentage points. Being obese or overweight versus normal weight similarly had positive effects (2.9
and 0.9 percentage point differences, respectively) on the participation rate. Appendix Exhibit A6
provides all incentive effect estimates for each employee characteristic examined.
The accounting exercise of the surcharge and program participation fees indicates the net transfer of
dollars is primarily from employees to the employer, and to a lesser extent from the employer to the
wellness vendor (Exhibit 4). The estimated net “cost” of the tobacco cessation intervention per tobacco
using employee was ‐$398 (i.e., a cost savings) during the two years the surcharge was in place.
Tobacco using versus non‐using employees were consistently found to be of lower socioeconomic status
in both unadjusted (Appendix Exhibit A7) and adjusted analyses (Appendix Exhibits A8 and A9). To
illustrate, being a unionized (vs. non‐unionized) employee increases the likelihood of using tobacco by
2.0 to 3.5 percentage points (p<0.01 for both); similar trends were seen for being paid hourly (vs.
salaried). Earning less money (as measured using zip‐code median earnings) increased the likelihood of
being a tobacco user by 2.9 to 4.6 percentage points (bottom vs. top decile earners; p<0.001 for both).
Higher (vs. lower) paying job classifications tended to have lower likelihoods of smoking: for example
being an “executive” lowered the likelihood of smoking by 2.6 percentage points vs. being a “laborer”
(p<0.05). Employees with college degrees and those with some college had likelihoods of smoking that
48
were 3.6 and 1.4 percentage points lower vs. those with high school or less (both p<0.01). Interestingly,
employees with the highest health care costs (top decile) were the least likely to use tobacco—a
difference of up to 2.5 to 2.9 percentage points vs. employees in the bottom decile (p<0.001 for both), a
trend that held for comparisons of the 75th and 25th health care cost quartiles.
DISCUSSION
This study found that within a Fortune‐100 employer’s established wellness program, only one in four
tobacco users—even when faced with a $600 non‐participation surcharge—chose to participate in the
offered tobacco cessation intervention. Although both the unadjusted and adjusted analyses point
towards the introduction of the surcharge having a large relative increase in participation (between 20
and 89 percent), the absolute increase falls more modestly between 4.5 and 12.7 percentage points.
Monetary incentives for tobacco cessation activities as part of health benefits have been used as far
back as 1988.13 More recently, an estimated 40% of large employers in a 2014 Towers Watson with a
tobacco cessation program used incentives to boost program uptake.14 However, among employers in
the Towers Watson survey offering the largest tobacco‐cessation incentives (more than $250 per‐
member per‐year) only 10% achieved a “high” participation rate defined as more than 11 percent,
underscoring low program uptake even among incentivized tobacco cessation interventions. A search of
the literature for studies isolating the effect of incentives on tobacco cessation program participation
returned few results; the most applicable found incentives to raise participation by about 10 percentage
points to 15.4 percent15,16‐‐ roughly in line with this study both in terms of absolute and relative
increases.
The ultimate goal of tobacco cessation interventions is not program participation but true cessation.
Each year in the US an estimated 3 to 6 percent of prevalent tobacco users are able to quit,17‐20 but only
50% of the 91 million persons who have smoked at least 100 cigarettes during their lifetimes have been
able to quit smoking.21 Roughly 70 percent of smokers wish they could quit,22 and 70 percent report
having made at least one quit attempt.23 This disconnect between wanting to quit and actually quitting
underscores how hard quitting can be. To illustrate, while measuring the number of quit attempts
required on average before successful has methodologic issues (e.g., recall bias), the most reliable
estimates suggest between 6 and 15 attempts are required before true cessation.24‐27
It thus may not be surprising that participating in tobacco cessation programs more often than not does
not lead to quitting. While the current analysis could not examine quitting, a recent literature review of
49
incentives for quitting found only one study with positive results.16,28 A more recent study of CVS
Caremark employees found only two of four intervention schemes to result in higher (with statistical
significance) quit rates measured at 1 year, with an increase from about 4 percent in usual care (no
incentives) to 7.5 and 8.5 percent in the two incentive groups— a large relative, but small absolute,
increase.15 An analysis of Georgia’s state employees’ health benefit plan, which implemented monthly
tobacco use surcharges similar to the employer in this analysis, found 45% of enrollees reported
themselves to be tobacco‐free a year after surcharges were introduced, implying either a highly
effective program, dishonest self‐reporting, or a combination of the two.18 In another recent and related
study, the authors exploited variation in state tobacco surcharges to estimate whether there was an
association between tobacco cessation and surcharges; while this analysis is of surcharges alone (there
were not specific tobacco cessation interventions offered as in this Fortune‐100 employer analysis), the
mechanism (premium surcharge) is similar; the study found individuals faced with higher surcharges
were not more likely to stop using tobacco.17
Stepping back, it is important to recognize the many monetary and non‐monetary incentives that
already exist for tobacco cessation: the harmful health effects are broadly known and disseminated on
product packaging and the average cigarette smoker (13.8 cigarettes per day29) at an average retail price
per pack ($6.0530) spends over $2,500 on cigarettes alone each year. In addition there are other financial
costs of smoking such as higher life and home insurance premiums, lower car and home resale values,
and perhaps the most significant the opportunity cost of dollars spent on tobacco instead of being
saved; even with modest return on investment assumptions, the opportunity cost of spending $2,500
per year on cigarettes alone for years, let alone decades, is tremendous.
The impact of changes in the cost of tobacco products on consumption has received substantial
attention (usually in the form of changes in state and federal taxes, or price changes following legal
settlements). This body of literature has been summarized in CBO analyses, which have concluded the
long‐run price elasticity to fall between ‐0.3 and ‐0.7, meaning a 1 percent rise in price causes overall
consumption to decline by between 0.3 and 0.7 percent.31 In long‐term tobacco tax revenue forecasts,
the CBO has assumed half of elasticities to result from reductions in the number of people who smoke
(quitting), and half to reductions in cigarettes consumed per smoker (still smoking, but smoking less).31
Applying such assumptions to the $600 surcharge in this study to the aforementioned average cigarette
prices and cigarettes smoked per day means a $600 surcharge increases the cost of smoking ($2,500) by
23.6 percent to roughly $3,100 per year. Using an elasticity of demand of ‐0.3 percent (about half of the
50
overall estimate) indicates the surcharge should yield a 7 percent decline in the number of tobacco
users. But this approximation comes with important caveats such as the premium surcharges in the
current study being a form of “conditioning,” which is not thought to be an effective way in behavioral
economics of changing behavior.32
Interestingly, in a large review of literature on why tobacco users quit,33 the primary motivator (among
75 percent of smokers) were health concerns, a finding consistent with more recent studies.34‐36 By
contrast, only 10‐15 percent cited economic reasons as a primary motivator.33 To support this, studies of
health care costs before and after quitting, have found costs to spike near the time of quitting, which
has been interpreted by the CBO to be that quitters often do so in response to a serious illness.31,37
Taken together, such evidence suggests tobacco cessation interventions would best motivate cessation
attempts by focusing on the negative health consequences (rather than economic costs) of tobacco use.
Although the evidence points to monetary incentives having limited effectiveness both in motivating
participation and in achieving cessation, at least in absolute terms, a fairness argument is commonly
made that tobacco users should be forced to pay for their bad behavior. While the evidence points
towards current or former smokers having health care costs vs. never smokers that are between 11 and
16 percent higher across all age categories,31 a more recent analysis found smokers to cost slightly less
overall, particularly at younger ages.38 Given current law allows surcharges to be up to 50% of the cost of
coverage, the amount tobacco users could be charged would appear to be well above any excess cost
they impose because of their tobacco use.
Tobacco users are more likely to be of lower socioeconomic status: for example, 26.3 percent of
individuals below the poverty line smoke, whereas only 5.4 at or above poverty level do so.29 Similarly,
over 40 percent of individuals with a GED smoke, whereas only 5 percent of those with a graduate
degree do so.29 While I did not find such extreme differences in the analyses between tobacco users and
non‐users (employees of the same firm are likely much more homogenous in characteristics), the
findings do point towards tobacco users being of lower socioeconomic status on average. Among the
Fortune‐100 employer studied, I thus conclude from the results that given low program uptake, the
primary effect of the $600 surcharge is a regressive tax on tobacco use (as has been written
elsewhere32), shifting health care costs to those who are least able to pay. Such concerns were validated
in an analysis which found tobacco users to be more likely to face unaffordable plans vs. non‐users using
8% of household income as a threshold for affordability.38
51
Better policy making might begin with the Fortune‐100 employer in the analysis adopting more lessons
from behavioral economics, such as the notion that “small but tangible and frequent positive feedback
or rewards” work best.39 Applying such logic and tying incentives to behaviors that are easy to change
should produce the greatest effects possible. Thus, if the Fortune‐100 employer wants to keep their
incentive in the form of a $600 annual penalty, then allowing the penalty to be earned back through
individual participation events could potentially yield greater effects. Such a program tweak could
provide paid time during the work‐day for participation and provide $25 in cash at the completion of
each class.
A final policy consideration is that if incentives are to continue to be used, then more evaluation
research is needed, as called for by DHHS guidelines,40 with a focus on incentive effectiveness and any
unintended consequences. A prior study by Horwitz and colleagues41 raised the concern that wellness
incentives may primarily be resulting in a shifting of costs from employers to employees. The current
study validates such concerns, underscoring the need for additional evidence to identify when and
under which circumstances incentives are effective so that employers and policy makers can make
informed decisions to achieve the ultimate policy goal of improved health and lower health care
spending.
CONCLUSIONS
This study analyzed the effect of a $600 tobacco cessation surcharge on participation in a tobacco
cessation intervention, finding that even in the presence of the surcharge only 1 in 4 tobacco users
opted to participate. As such, the main effect, in light of low participation and assumed quit rates, is that
tobacco surcharges are shifting health care costs to tobacco using employees who are more likely than
their non‐tobacco using counterparts to be of lower socioeconomic status.
NOTES
1. Kaiser Family Foundation/Health Research and Educational Trust. Employer Health Benefits: 2016 Annual Survey (Publication #8775). Copyright © 2016 Henry J. Kaiser Family Foundation, Menlo Park, California, and Health Research & Educational Trust, Chicago, Illinois.
2. Mattke S, Liu H, Caloyeras JP, Huang CY, Van Busum KR, Khodyakov D, et al. Workplace Wellness Programs Study. Santa Monica (CA): RAND Corporation; 2013. (Pub. No. RR‐254‐DOL).
3. Mattke, Soeren, Kandice A. Kapinos, John P. Caloyeras, Erin Audrey Taylor, Benjamin Saul Batorsky, Hangsheng Liu, Kristin R. Van Busum and Sydne Newberry. Workplace Wellness Programs: Services Offered, Participation, and Incentives. Santa Monica, CA: RAND Corporation, 2014. http://www.rand.org/pubs/research_reports/RR724.html.
52
4. Wellness programs defined in the KFF/HRET survey as those offering either "Programs to HelpEmployees Stop Smoking", "Programs to Help Employees Lose Weight", or "Other Lifestyle orBehavioral Coaching.".
5. https://www.bls.gov/news.release/pdf/wkyeng.pdf.6. Bureau of Labor Statistics. Economic news release: consumer expensitures ‐‐ 2015. Tues, August
30, 2016. Available at: https://www.bls.gov/news.release/cesan.nr0.htm. Accessed 31 July2017.
7. EEOC v. Orion Energy Systems, Inc., No. 1:14‐01019 (E.D. Wis. 2014).8. EEOC v. Flambeau, Inc., No. 3:14‐00638 (W.D. Wis. 2014).9. EEOC v. Honeywell, No. 0:14‐04517 (D.MN 2014).10. AARP v. Equal Employment Opportunity Commission, U.S. District Court for the District of
Columbia, No. 1:16‐cv‐02113.11. H.R.1313 — 115th Congress (2017‐2018). Available at: https://www.congress.gov/bill/115th‐
congress/house‐bill/1313.12. Madison KM. The risks of using workplace wellness programs to foster a culture of health.
Health Affairs. 2016;35(11):2068‐2074.13. Penner M. Economic incentives to reduce employee smoking: A health insurance surcharge for
tobacco using state of Kansas employees. American Journal of Health Promotion. 1989;4(1):5‐11.
14. Towers Watson, July 2010. Boosting wellness participation without breaking the bank. Availableat: https://www.towerswatson.com/en‐US/Insights/Newsletters/Americas/insider/2010/boosting‐wellness‐participation‐without‐breaking‐the‐bank.
15. Halpern SD, French B, Small DS, et al. Randomized trial of four financial‐incentive programs forsmoking cessation. New England Journal of Medicine. 2015.
16. Volpp KG, Troxel AB, Pauly MV, et al. A randomized, controlled trial of financial incentives forsmoking cessation. New England Journal of Medicine. 2009;360(7):699‐709.
17. Friedman AS, Schpero WL, Busch SH. Evidence Suggests That The ACA’s Tobacco SurchargesReduced Insurance Take‐Up And Did Not Increase Smoking Cessation. Health Affairs.2016;35(7):1176‐1183.
18. Liber AC, Hockenberry JM, Gaydos LM, Lipscomb J. The Potential and Peril of Health InsuranceTobacco Surcharge Programs: Evidence From Georgia’s State Employees Health Benefit Plan.nicotine & tobacco research. 2013:ntt216.
19. Mendez D, Warner KE. Setting a challenging yet realistic smoking prevalence target for HealthyPeople 2020: learning from the California experience. American journal of public health.2008;98(3):556‐559.
20. Mendez D, Warner KE, Courant PN. Has smoking cessation ceased? Expected trends in theprevalence of smoking in the United States. American Journal of Epidemiology. 1998;148(3):249‐258.
21. MMWR. Cigarette smoking among adults ‐‐‐ Unites States, 2006. Available at:https://www.cdc.gov/mmwr/preview/mmwrhtml/mm5644a2.htm.
53
22. Hymowitz N, Cummings KM, Hyland A, Lynn WR, Pechacek TF, Hartwell TD. Predictors ofsmoking cessation in a cohort of adult smokers followed for five years. Tobacco control.1997;6(suppl 2):S57.
23. US Public Health Service (2000, June). Treating tobacco use and dependence. Fact Sheet.Available at; http://www.surgeongeneral.gov/tobacco/smokfact.htm.
24. Chaiton M, Diemert L, Cohen JE, et al. Estimating the number of quit attempts it takes to quitsmoking successfully in a longitudinal cohort of smokers. BMJ open. 2016;6(6):e011045.
25. American Cancer Society. Guide to Quitting Smoking.http://www.cancer.org/healthy/stayawayfromtobacco/guidetoquittingsmoking/guide‐to‐quitting‐smoking‐pdf (accessed 26 Jul 2013).
26. US Department of Health Human Services. Women and smoking: a report of the surgeongeneral. Rockville, MD: US Department of Health and Human Services, Public Health Service,Centers for Disease Control, Center for Chronic Disease Prevention and Health Promotion, Officeon Smoking and Health, 2001.
27. Sharecare.com. How often do people try to quit smoking?http://www.sharecare.com/health/smoking‐treatment/how‐often‐do-people‐try‐to‐quit (accessed 26 Jul 2013).
28. Cahill K, Perera R. Competitions and incentives for smoking cessation. The Cochrane Library.2011.
29. MMWR. Current cigarette smoking among adults ‐‐‐ United States, 2005‐2014. November 13,2015/64(44);1233‐1240.
30. http://www.tobaccofreekids.org/research/factsheets/pdf/0097.pdf. Campaign for Tobacco‐FreeKids. 2016. Accessed 4 Dec 2016.
31. CBO. Raising the excise tax on cigarettes: effects on health and the federal budget. June 2012.Available at: https://www.cbo.gov/publication/43319.
32. Loewenstein G, Asch DA, Volpp KG. Behavioral economics holds potential to deliver betterresults for patients, insurers, and employers. Health Affairs. 2013;32(7):1244‐1250.
33. McCaul KD, Hockemeyer JR, Johnson RJ, Zetocha K, Quinlan K, Glasgow RE. Motivation to quitusing cigarettes: a review. Addictive behaviors. 2006;31(1):42‐56.
34. Baha M, Le Faou A‐L. Smokers’ reasons for quitting in an anti‐smoking social context. Publichealth. 2010;124(4):225‐231.
35. Gallus S, Muttarak R, Franchi M, et al. Why do smokers quit? European Journal of CancerPrevention. 2013;22(1):96‐101.
36. Pisinger C, Aadahl M, Toft U, Jørgensen T. Motives to quit smoking and reasons to relapse differby socioeconomic status. Preventive medicine. 2011;52(1):48‐52.
37. Hockenberry JM, Curry SJ, Fishman PA, et al. Healthcare costs around the time of smokingcessation. American journal of preventive medicine. 2012;42(6):596‐601.
38. Kaplan CM, Graetz I, Waters TM. Most exchange plans charge lower tobacco surcharges thanallowed, but many tobacco users lack affordable coverage. Health Affairs. 2014;33(8):1466‐1473.
39. Volpp KG, Asch DA, Galvin R, Loewenstein G. Redesigning employee health incentives—lessonsfrom behavioral economics. New England Journal of Medicine. 2011;365(5):388‐390.
54
40. Fiore M, Jaen CR, Baker T, et al. Treating tobacco use and dependence: 2008 update. Rockville, MD: US Department of Health and Human Services. 2008.
41. Horwitz JR, Kelly BD, DiNardo JE. Wellness incentives in the workplace: cost savings through cost shifting to unhealthy workers. Health Affairs. 2013;32(3):468‐476.
ACKNOWLEDGEMENTS
This study was funded in part by the US Department of Labor.
Analyses of data from this Fortune‐100 employer were previously conducted and published, as part of work towards this dissertation, in the RAND Corporation study report cited below. The difference‐in‐differences analyses presented in the current study overlap in methods and approach to this prior publication. This current study builds upon those analyses, with slightly modified samples, and provides a deeper (multivariable) exploration of the characteristics of tobacco users vs. non‐users. This current study thus leverages prior work but should be viewed as a unique stand‐alone research study that will be submitted to a peer‐reviewed journal for publication.
Prior presentations:
Mattke, Soeren, Kandice A. Kapinos, John P. Caloyeras, Erin Audrey Taylor, Benjamin Saul Batorsky, Hangsheng Liu, Kristin R. Van Busum and Sydne Newberry. Workplace Wellness Programs: Services Offered, Participation, and Incentives. Santa Monica, CA: RAND Corporation, 2014. http://www.rand.org/pubs/research_reports/RR724.html.
55
EXHIBITS
EXHIBIT 1. Eligibility and participation in the tobacco cessation intervention, by program year.
SOURCE: Author’s analysis of Fortune‐100 employer data.
NOTES: Participation defined as at least one telephonic encounter with a tobacco cessation coach.
0
20
40
60
80
100
0
5,000
10,000
15,000
20,000
25,000
2005 2006 2007 2008 2009 2010 2011
Participation rate (%)Person‐years (N)
Year
Non‐Participant Participant Participation rate
Program restructure and surcharge introduction
56
EXHIBIT 2. Participation intensity in the tobacco cessation intervention, by program year.
SOURCE: Author’s analysis of Fortune‐100 employer data.
NOTES: Boxplot represents minimum, 25th percentile, median, 75th percentile and maximum sessions per‐employee per‐year; means are represented by the black circles and numeric data labels.
1.42.1 2.2
3.42.8
3.3
0
1
2
3
4
5
6
7
8
9
10
11
12
2005 2006 2007 2008 2009 2010 2011
Telephonic coaching sessions
Program year
57
EXHIBIT 3. Surcharge effects for tobacco cessation on participation in lifestyle management programs: difference in differences estimate by model and sample specification.
SOURCE: Author’s analysis of Fortune‐100 employer data.
NOTES: Estimates represent average marginal effects; error bars represent 95 percent confidence intervals. All estimates significantly different from zero at p<0.001.
5.5
6.0
12.5
7.2
6.6
12.7
0 5 10 15
HRA w/ HRA vars w/interactions
HRA w/out HRA vars w/interactions
Full sample w/ interactions
HRA w/ HRA vars
HRA w/out HRA vars
Full sample
Effect on participation rate (%)
58
EXHIBIT 3. Employer and employee direct costs of the tobacco cessation intervention, by year.
Program year Program cost
Tobacco use surcharges
Participation fees*
Total program cost
2005 $0 $8,400 $8,400
2006 $0 $69,300 $69,300
2007 $0 $88,650 $88,650
2008 $0 $166,050 $166,050
2009 n/a n/a n/a
2010 ($1,458,000) $135,300 ($1,322,700)
2011 ($1,830,600) $167,550 ($1,663,050)
Total (all years) ($3,288,600) $635,250 ($2,653,350)
SOURCE: Author’s analysis of Fortune‐100 employer data.
NOTES: Wellness program data from 2009 not included in the analysis due to program restructuring during that year. Analysis assumes a modest $150 per‐participant per‐year fee as actual vendor fees are confidential. Accounting exercise is limited to direct costs of the tobacco components of the Fortune‐100 employer’s wellness program.
59
SUPPLEMENTARYMATERIALS
TABLE OF CONTENTS
EXHIBIT A1. Creation of analytic sample
EXHIBIT A2. Pooled eligibility and participation in all lifestyle management interventions except for tobacco cessation, by program year
EXHIBIT A3. Characteristics of lifestyle management eligibles, overall and by tobacco use status
EXHIBIT A4. Regression coefficients from difference‐in‐differences lifestyle management participation regressions
EXHIBIT A5. Predicted probability of lifestyle management participation (except tobacco cessation) and tobacco cessation participation pre and post introduction of the tobacco use surcharge, by model and sample specification
EXHIBIT A6. Incentive effects for smoking cessation on participation in lifestyle management programs
EXHIBIT A7. Characteristics of Fortune‐100 employees, overall and by tobacco use status
EXHIBIT A8. Regression coefficients from predictors of tobacco use regressions
EXHIBIT A9. Predicted probability of tobacco use by select patient characteristics
60
EXHIBIT A1. Creation of analytic sample.
Inclusion Criteria Employee Years (N) Unique Employees (N) 1. Ages 18–64 683,206 166,842 2. Not enrolled in HMO 677,609 164,204 3. Not pregnant during data year 651,277 160,336 4. One or more full‐years of enrollment 610,517 119,576 5. Two or more full‐years of enrollment 565,883 97,259 6. Employees only 383,395 64,966 Additional criteria for analytic subsamples
Completed the HRA (at least once) 103,268 52,875 Eligible for any lifestyle management intervention
90,570 49,515
Eligible for tobacco cessation intervention (tobacco user)
18,643 9,988
SOURCE: Author’s analysis of Fortune‐100 employer data.
NOTES: Analytic samples are reduced further in analyses as variables such as lagged medical costs imposes the requirement of two consecutive full‐years of enrollment.
61
EXHIBIT A2. Pooled eligibility and participation in all lifestyle management interventions except for tobacco cessation, by program year.
SOURCE: Author’s analysis of Fortune‐100 employer data.
0
20
40
60
80
100
0
5,000
10,000
15,000
20,000
25,000
2005 2006 2007 2008 2009 2010 2011
Participation rate (%)
Person‐years (N)
Year
Non‐Participant Participant Participation rate
Program restructure and surcharge introduction
62
EXHIBIT A3. Characteristics of lifestyle management eligibles, overall and by tobacco use status.
Characteristic Overall Non‐tobacco users Tobacco users p‐value
Sex (n, %)
Female 7,555 20.7% 5,580 21.0% 1,958 20.0% 0.04
Male 28,932 79.3% 21,024 79.0% 7,843 80.0%
Age (years) (n, %)
18‐34 8,760 24.0% 6,471 24.3% 2,225 22.7% <.01
35‐44 12,205 33.5% 8,937 33.6% 3,235 33.0%
45‐54 11,901 32.6% 8,440 31.7% 3,460 35.3%
55‐64 3,621 9.9% 2,756 10.4% 881 9.0%
Age (years) (mean, SD) 42.06 9.4 42.03 9.50 42.30 9.20 0.01
Region (n, %)
East 5,811 15.9% 4,440 16.7% 1,356 13.8% <.01
Midwest 8,399 23.0% 5,820 21.9% 2,558 26.1%
South 17,947 49.2% 13,034 49.0% 4,879 49.8%
West 4,330 11.9% 3,310 12.4% 1,008 10.3%
Pay type (n, %)
Hourly 22,217 60.9% 15,203 57.1% 6,984 71.3% <.01
Salaried 14,270 39.1% 11,401 42.9% 2,817 28.7%
Union status (n, %)
Non‐union 27,977 76.7% 20,348 76.5% 7,350 75.0% <.01
Union 8,510 23.3% 6,256 23.5% 2,451 25.0%
Zip code median income, 2008 (10Ks, 2013 USD) (n, %)
<35K 1,546 4.2% 1,063 4.0% 474 4.8% <.01
35K ‐ 49K 11,943 32.7% 8,106 30.5% 3,794 38.7%
50K ‐ 64K 10,515 28.8% 7,532 28.3% 2,969 30.3%
65K ‐ 79K 5,921 16.2% 4,472 16.8% 1,441 14.7%
80K ‐ 99K 4,004 11.0% 3,226 12.1% 767 7.8%
>=100K 2,558 7.0% 2,205 8.3% 356 3.6%
Zip code median income, 2008 (10Ks, 2013 USD) (mean, SD)
6.16 2.3 6.32 2.40 5.71 1.89 <.01
Years of LM eligibility (mean, SD) 1.84 1.0 1.72 0.96 2.18 1.02 <.01
63
Characteristic Overall Non‐tobacco users Tobacco users p‐value
PMPM cost (2013 USD) (mean, SD) 190.15 325.7 197.69 330.55 170.95 315.14 <.01
Utilization per 1,000 member years (mean, SD)
Emergency room visits 120.95 376.6 115.96 369.46 134.37 394.35 <.01
Inpatient admissions 13.95 125.9 13.27 124.19 15.81 134.22 0.09
Office visits for primary care eval. and man.
2,522.7 2,905.2 2,650.3 2,946.9 2,178.3 2,763.8 <.01
Charlson comorbidity (weighted) index (mean, SD)
0.13 0.45 0.13 0.46 0.11 0.41 <.01
Completed HRA (n, %)
Non‐participant 8,189 22.4% 5,099 19.2% 3,073 31.4% <.01
Participant 28,298 77.6% 21,505 80.8% 6,728 68.6%
Ethnicity (n, %)
African‐American 2,236 10.7% 1,872 11.9% 363 7.1% <.01
Asian or Pacific Islander 552 2.6% 480 3.0% 71 1.4%
Caucasian / Non‐Hispanic 15,982 76.3% 11,627 73.7% 4,330 84.4%
Hispanic 1,631 7.8% 1,419 9.0% 210 4.1%
Native American / Alaskan Native 380 1.8% 244 1.5% 136 2.7%
Multiracial 71 0.3% 63 0.4% 7 0.1%
Other 82 0.4% 69 0.4% 13 0.3%
Education (n, %)
High school or less 6,486 30.8% 4,599 28.1% 1,860 40.6% <.01
Some college 7,927 37.7% 6,079 37.1% 1,823 39.8%
College graduate 6,617 31.5% 5,708 34.8% 899 19.6%
Job classification (n, %)
Executive 1,669 8.0% 1,437 8.8% 230 5.1% <.01
Professional 2,569 12.29% 2,236 13.7% 330 7.3%
Technical support 671 3.21% 505 3.1% 162 3.6%
Sales 7,939 37.97% 6,187 38.0% 1,731 38.1%
Clerical 1,598 7.64% 1,241 7.6% 353 7.8%
Service 452 2.16% 346 2.1% 106 2.3%
Production 998 4.77% 699 4.3% 294 6.5%
64
Characteristic Overall Non‐tobacco users Tobacco users p‐value
Laborer 5,013 23.98% 3,649 22.4% 1,341 29.5%
BMI category (n, %)
Normal weight 6,196 22.70% 4,314 20.9% 1,865 28.1% <.01
Overweight 11,899 43.59% 9,061 44.0% 2,822 42.5%
Obese 9,205 33.72% 7,231 35.1% 1,954 29.4%
BMI (mean, SD) 28.66 5.17 28.87 5.14 28.00 5.20 <.01
Blood pressure category (n, %)
Normal 3,456 24.84% 2,654 24.6% 796 25.9% 0.18
Pre‐hypertensive 8,486 61.00% 6,642 61.5% 1,829 59.4%
Stage 1 hypertension 1,579 11.35% 1,220 11.3% 356 11.6%
Stage 2 hypertension 390 2.80% 292 2.7% 96 3.1%
Total cholesterol category (n, %)
Desirable 5,233 67.43% 4,192 67.2% 1,036 68.2% 0.11
Borderline high 2,007 25.86% 1,638 26.3% 367 24.2%
High 521 6.71% 404 6.5% 115 7.6%
Days of work missed due to illness (past 12 mos) (mean, SD)
1.29 3.50 1.23 2.79 1.52 5.58 0.18
Hours missed from work b/c of health problems (past 4 wks) (mean, SD)
1.69 6.80 1.60 6.42 1.96 8.08 0.35
Hours actually worked (past 4 wks) (mean, SD)
176.08 54.58 176.89 56.06 174.82 51.18 0.51
SOURCE: Author’s analysis of Fortune‐100 employer data.
NOTES: The following variables were obtained from completed HRAs and thus were only available for a subset of employees: ethnicity, education, job classification, BMI, blood pressure (systolic and diastolic), total cholesterol, days of work missed due to illness, hours missed from work because of health problems and hours actually worked. P‐values generated using Chi‐squared tests for categorical variables and t‐tests for numerical variables.
65
EXHIBIT A4. Regression coefficients from difference‐in‐differences lifestyle management participation regressions.
Model: (#1) Full sample
(#2) HRA w/out
HRA variables
(#3) HRA w/ HRA variables
(#4) Full sample,
w/ interactions
(#5) HRA w/out
HRA variables, w/
interactions
(#6) HRA w/ HRA variables, w/ interactions
Eligible for smoking cessation (0/1) ‐0.748***(0.0307)
‐0.287***(0.0483)
‐0.252*** (0.0487)
‐1.124***(0.156)
‐0.459*(0.257)
‐0.527(0.418)
Eligible for smoking cessation (0/1) ‐0.748*** (0.0307)
‐0.287*** (0.0483)
‐0.252*** (0.0487)
‐1.124*** (0.156)
‐0.459* (0.257)
‐0.527 (0.418)
Eligible for smoking non‐participation surcharge (0/1)
‐0.362*** (0.0210)
0.190*** (0.0317)
0.204*** (0.0319)
‐0.710*** (0.103)
0.122 (0.163) ‐0.606** (0.263)
Interaction: Eligible for smoking cessation (0/1) XXX Eligible for smoking non‐participation surcharge (0/1)
0.611*** (0.0430)
0.359*** (0.0785)
0.381*** (0.0789)
0.748*** (0.229)
0.752* (0.453)
1.233 (0.790)
Male (0/1) 0.137*** (0.0229)
0.0568* (0.0335)
0.0243 (0.0384)
0.157*** (0.0378)
‐0.0392 (0.0586)
‐0.0101 (0.0684)
Interaction: Eligible for smoking non‐participation surcharge (0/1) XXX Male (0/1)
‐0.0119 (0.0518)
0.179** (0.0756)
0.0568 (0.0877)
Interaction: Eligible for smoking cessation (0/1) XXX Male (0/1)
‐0.0392 (0.0751)
‐0.0567 (0.111)
‐0.128 (0.131)
Interaction: Eligible for smoking cessation (0/1) XXX Eligible for smoking non‐participation surcharge (0/1) XXX Male (0/1)
‐0.0492 (0.0882)
0.219 (0.156) 0.204 (0.174)
Age 35 to 44 (0/1) ‐0.100*** (0.0258)
0.153*** (0.0456)
0.138*** (0.0460)
‐0.133*** (0.0420)
0.144* (0.0741)
0.156** (0.0747)
Age 45 to 54 (0/1) ‐0.00954 (0.0255)
0.417*** (0.0443)
0.416*** (0.0451)
‐0.0402 (0.0428)
0.351*** (0.0737)
0.364*** (0.0750)
Age 55 to 64 (0/1) 0.0803** (0.0333)
0.671*** (0.0543)
0.674*** (0.0552)
‐0.0979 (0.0601)
0.484*** (0.100)
0.491*** (0.102)
Interaction: Eligible for smoking non‐participation surcharge (0/1) XXX Age 35 to 44 (0/1)
0.0670 (0.0589)
0.0271 (0.100)
‐0.0341 (0.101)
Interaction: Eligible for smoking non‐participation surcharge (0/1) XXX Age 45 to 54 (0/1)
0.116** (0.0587)
0.148 (0.0982)
0.108 (0.100)
Interaction: Eligible for smoking non‐participation surcharge (0/1) XXX Age 55 to 64 (0/1)
0.377*** (0.0776)
0.316** (0.125)
0.292** (0.128)
66
Model: (#1) Full sample
(#2) HRA w/out
HRA variables
(#3) HRA w/ HRA variables
(#4) Full sample,
w/ interactions
(#5) HRA w/out
HRA variables, w/
interactions
(#6) HRA w/ HRA variables, w/ interactions
Interaction: Eligible for smoking cessation (0/1) XXX Age 35 to 44 (0/1)
0.0780 (0.0877)
0.00724 (0.152)
0.0106 (0.154)
Interaction: Eligible for smoking cessation (0/1) XXX Age 45 to 54 (0/1)
‐0.134 (0.0877)
‐0.105 (0.149)
‐0.0524 (0.152)
Interaction: Eligible for smoking cessation (0/1) XXX Age 55 to 64 (0/1)
‐0.0688 (0.125)
‐0.106 (0.211)
‐0.0648 (0.214)
Interaction: Eligible for smoking cessation (0/1) XXX Eligible for smoking non‐participation surcharge (0/1) XXX Age 35 to 44 (0/1)
‐0.0243 (0.0904)
‐0.163 (0.226)
‐0.133 (0.230)
Interaction: Eligible for smoking cessation (0/1) XXX Eligible for smoking non‐participation surcharge (0/1) XXX Age 45 to 54 (0/1)
‐0.0501 (0.0877)
0.0775 (0.213)
0.0799 (0.218)
Interaction: Eligible for smoking cessation (0/1) XXX Eligible for smoking non‐participation surcharge (0/1) XXX Age 55 to 64 (0/1)
0.0102 (0.118)
0.172 (0.264) 0.148 (0.269)
Region = Central (0/1) ‐0.352*** (0.0289)
‐0.138*** (0.0451)
‐0.154*** (0.0455)
‐0.709*** (0.0490)
‐0.0741 (0.0798)
‐0.0675 (0.0802)
Region = South (0/1) ‐0.250*** (0.0256)
‐0.237*** (0.0416)
‐0.261*** (0.0422)
‐0.384*** (0.0442)
‐0.183** (0.0756)
‐0.178** (0.0762)
Region = West (0/1) ‐0.0240 (0.0345)
‐0.231*** (0.0616)
‐0.274*** (0.0623)
‐0.0483 (0.0589)
‐0.150 (0.108)
‐0.156 (0.109)
Interaction: Eligible for smoking non‐participation surcharge (0/1) XXX Region = Central (0/1)
0.616*** (0.0662)
‐0.128 (0.101)
‐0.183* (0.102)
Interaction: Eligible for smoking non‐participation surcharge (0/1) XXX Region = South (0/1)
0.259*** (0.0583)
‐0.103 (0.0939)
‐0.170* (0.0954)
Interaction: Eligible for smoking non‐participation surcharge (0/1) XXX Region = West (0/1)
0.134* (0.0785)
‐0.160 (0.138)
‐0.243* (0.140)
Interaction: Eligible for smoking cessation (0/1) XXX Region = Central (0/1)
0.231** (0.101)
0.0706 (0.166)
0.0522 (0.167)
Interaction: Eligible for smoking cessation (0/1) XXX Region = South (0/1)
0.0155 (0.0928)
0.0544 (0.158)
0.0194 (0.161)
Interaction: Eligible for smoking cessation (0/1) XXX Region = West (0/1)
‐0.102 (0.127)
0.257 (0.217) 0.207 (0.221)
67
Model: (#1) Full sample
(#2) HRA w/out
HRA variables
(#3) HRA w/ HRA variables
(#4) Full sample,
w/ interactions
(#5) HRA w/out
HRA variables, w/
interactions
(#6) HRA w/ HRA variables, w/ interactions
Interaction: Eligible for smoking cessation (0/1) XXX Eligible for smoking non‐participation surcharge (0/1) XXX Region = Central (0/1)
0.443*** (0.103)
‐0.157 (0.229)
‐0.183 (0.234)
Interaction: Eligible for smoking cessation (0/1) XXX Eligible for smoking non‐participation surcharge (0/1) XXX Region = South (0/1)
0.213** (0.0941)
‐0.224 (0.219)
‐0.231 (0.224)
Interaction: Eligible for smoking cessation (0/1) XXX Eligible for smoking non‐participation surcharge (0/1) XXX Region = West (0/1)
‐0.159 (0.126)
‐0.632** (0.314)
‐0.574* (0.319)
Salaried (0/1) 0.479*** (0.0204)
0.228*** (0.0335)
0.263*** (0.0441)
0.431*** (0.0354)
0.164*** (0.0582)
0.166** (0.0744)
Interaction: Eligible for smoking non‐participation surcharge (0/1) XXX Salaried (0/1)
0.0592 (0.0472)
0.0577 (0.0754)
0.162 (0.0993)
Interaction: Eligible for smoking cessation (0/1) XXX Salaried (0/1)
0.177** (0.0710)
0.154 (0.112) 0.0598 (0.146)
Interaction: Eligible for smoking cessation (0/1) XXX Eligible for smoking non‐participation surcharge (0/1) XXX Salaried (0/1)
0.376*** (0.0749)
0.240 (0.166) 0.346 (0.219)
Union Member (0/1) 0.0362 (0.0233)
0.0806 (0.0493)
0.0885* (0.0503)
0.606*** (0.0557)
0.0121 (0.0968)
0.0139 (0.0983)
Interaction: Eligible for smoking non‐participation surcharge (0/1) XXX Union member (0/1)
‐0.941*** (0.0640)
0.0318 (0.118)
0.0429 (0.121)
Interaction: Eligible for smoking cessation (0/1) XXX Union member (0/1)
‐0.305*** (0.106)
4.54e‐05 (0.178)
‐0.0503 (0.181)
Interaction: Eligible for smoking cessation (0/1) XXX Eligible for smoking non‐participation surcharge (0/1) XXX Union member (0/1)
0.0286 (0.0799)
0.505*** (0.192)
0.464** (0.198)
Household income (zip code median) ‐0.0724*** (0.00427)
‐0.0283***(0.00639)
‐0.0175*** (0.00675)
‐0.0791*** (0.00681)
‐0.00287 (0.0105)
‐0.000109 (0.0112)
Interaction: Eligible for smoking non‐participation surcharge (0/1) XXX Household income (zip code median)
0.00110 (0.00942)
‐0.0485*** (0.0139)
‐0.0287* (0.0147)
Interaction: Eligible for smoking cessation (0/1) 0.0426*** 0.0204 0.00254
68
Model: (#1) Full sample
(#2) HRA w/out
HRA variables
(#3) HRA w/ HRA variables
(#4) Full sample,
w/ interactions
(#5) HRA w/out
HRA variables, w/
interactions
(#6) HRA w/ HRA variables, w/ interactions
XXX Household income (zip code median) (0.0157) (0.0235) (0.0250)Interaction: Eligible for smoking cessation (0/1) XXX Eligible for smoking non‐participation surcharge (0/1) XXX Household income (zip code median)
0.00964 (0.0174)
‐0.0413 (0.0385)
‐0.0479 (0.0402)
Charlson Comorbidity index (prior year) 0.0660*** (0.0196)
0.0393 (0.0291)
0.0246 (0.0294)
0.0384 (0.0356)
‐0.00687 (0.0566)
‐0.00732 (0.0568)
Interaction: Eligible for smoking non‐participation surcharge (0/1) XXX Charlson comorbidity index (prior year)
0.0545 (0.0448)
0.0673 (0.0678)
0.0429 (0.0684)
Interaction: Eligible for smoking cessation (0/1) XXX Charlson comorbidity index (prior year)
‐0.127* (0.0765)
‐0.0846 (0.122)
‐0.0939 (0.123)
Interaction: Eligible for smoking cessation (0/1) XXX Eligible for smoking non‐participation surcharge (0/1) XXX Charlson comorbidity index (prior year)
0.112 (0.0812)
0.165 (0.181) 0.249 (0.186)
Log(Total PMPM costs) (prior year) 0.0399*** (0.00334)
0.0559*** (0.00552)
0.0529*** (0.00556)
0.0234*** (0.00572)
0.0329*** (0.00962)
0.0336*** (0.00965)
Interaction: Eligible for smoking non‐participation surcharge (0/1) XXX Log(Total PMPM costs) (prior year)
0.0213*** (0.00787)
0.0503*** (0.0126)
0.0416*** (0.0127)
Interaction: Eligible for smoking cessation (0/1) XXX Log(Total PMPM costs) (prior year)
0.0299*** (0.0109)
0.00670 (0.0177)
0.00444 (0.0179)
Interaction: Eligible for smoking cessation (0/1) XXX Eligible for smoking non‐participation surcharge (0/1) XXX Log(Total PMPM costs) (prior year)
0.0279** (0.0108)
‐0.0152 (0.0231)
‐0.0101 (0.0234)
Ethnicity = Asian or Pacific Islander (0/1) 0.309*** (0.0994)
0.201(0.171)
Ethnicity = Caucasian / Non‐Hispanic (0/1) ‐0.0760 (0.0498)
‐0.0166 (0.0893)
Ethnicity = Hispanic (0/1) ‐0.185** (0.0735)
‐0.00169 (0.128)
69
Model: (#1) Full sample
(#2) HRA w/out
HRA variables
(#3) HRA w/ HRA variables
(#4) Full sample,
w/ interactions
(#5) HRA w/out
HRA variables, w/
interactions
(#6) HRA w/ HRA variables, w/ interactions
Ethnicity = Native American / Alaskan Native or Multiracial or Other (0/1)
‐0.101 (0.109)
‐0.0140 (0.190)
Interaction: Eligible for smoking non‐participation surcharge (0/1) XXX Ethnicity = Asian or Pacific Islander (0/1)
0.146(0.219)
Interaction: Eligible for smoking non‐participation surcharge (0/1) XXX Ethnicity = Caucasian / Non‐Hispanic (0/1)
‐0.0972(0.112)
Interaction: Eligible for smoking non‐participation surcharge (0/1) XXX Ethnicity = Hispanic (0/1)
‐0.305* (0.162)
Interaction: Eligible for smoking non‐participation surcharge (0/1) XXX Ethnicity = Native American / Alaskan Native or Multiracial or Other (0/1)
‐0.0627 (0.246)
Interaction: Eligible for smoking cessation (0/1) XXX Ethnicity = Asian or Pacific Islander (0/1)
0.771** (0.385)
Interaction: Eligible for smoking cessation (0/1) XXX Ethnicity = Caucasian / Non‐Hispanic (0/1)
0.0615 (0.208)
Interaction: Eligible for smoking cessation (0/1) XXX Ethnicity = Hispanic (0/1)
0.0422 (0.314)
Interaction: Eligible for smoking cessation (0/1) XXX Ethnicity = Native American / Alaskan Native or Multiracial or Other (0/1)
‐0.434 (0.409)
Interaction: Eligible for smoking cessation (0/1) XXX Eligible for smoking non‐participation surcharge (0/1) XXX Ethnicity = Asian or Pacific Islander (0/1)
‐0.400 (0.593)
Interaction: Eligible for smoking cessation (0/1) XXX Eligible for smoking non‐participation surcharge (0/1) XXX Ethnicity = Caucasian / Non‐Hispanic (0/1)
‐0.363 (0.257)
Interaction: Eligible for smoking cessation (0/1) XXX Eligible for smoking non‐participation surcharge (0/1) XXX Ethnicity = Hispanic (0/1)
‐0.331 (0.421)
70
Model: (#1) Full sample
(#2) HRA w/out
HRA variables
(#3) HRA w/ HRA variables
(#4) Full sample,
w/ interactions
(#5) HRA w/out
HRA variables, w/
interactions
(#6) HRA w/ HRA variables, w/ interactions
Interaction: Eligible for smoking cessation (0/1) XXX Eligible for smoking non‐participation surcharge (0/1) XXX Ethnicity = Native American / Alaskan Native or Multiracial or Other (0/1)
0.247(0.503)
Job = Professional (0/1) 0.0985 (0.0620)
0.123 (0.0977)
Job = Tech. support (0/1) 0.277*** (0.0919)
0.223(0.158)
Job = Sales (0/1) 0.355*** (0.0628)
0.189* (0.101)
Job = Clerical (0/1) 0.319*** (0.0736)
0.247** (0.125)
Job = Service (0/1) 0.396*** (0.123)
0.115(0.249)
Job = Production (0/1) 0.108 (0.0955)
0.0517 (0.174)
Job = Laborer (0/1) 0.271*** (0.0748)
0.153(0.129)
Interaction: Eligible for smoking non‐participation surcharge (0/1) XXX Job = Professional (0/1)
‐0.00396 (0.133)
Interaction: Eligible for smoking non‐participation surcharge (0/1) XXX Job = Tech. support (0/1)
0.198(0.206)
Interaction: Eligible for smoking non‐participation surcharge (0/1) XXX Job = Sales (0/1)
0.371*** (0.137)
Interaction: Eligible for smoking non‐participation surcharge (0/1) XXX Job = Clerical (0/1)
0.167(0.164)
Interaction: Eligible for smoking non‐participation surcharge (0/1) XXX Job = Service (0/1)
0.529* (0.302)
Interaction: Eligible for smoking non‐participation surcharge (0/1) XXX Job = Production (0/1)
0.192(0.222)
Interaction: Eligible for smoking non‐participation surcharge (0/1) XXX Job = Laborer (0/1)
0.233(0.168)
Interaction: Eligible for smoking cessation (0/1) ‐0.286
71
Model: (#1) Full sample
(#2) HRA w/out
HRA variables
(#3) HRA w/ HRA variables
(#4) Full sample,
w/ interactions
(#5) HRA w/out
HRA variables, w/
interactions
(#6) HRA w/ HRA variables, w/ interactions
XXX Job = Professional (0/1) (0.224)Interaction: Eligible for smoking cessation (0/1) XXX Job = Tech. support (0/1)
‐0.469 (0.318)
Interaction: Eligible for smoking cessation (0/1) XXX Job = Sales (0/1)
‐0.139 (0.211)
Interaction: Eligible for smoking cessation (0/1) XXX Job = Clerical (0/1)
‐0.0829 (0.243)
Interaction: Eligible for smoking cessation (0/1) XXX Job = Service (0/1)
‐0.0164 (0.427)
Interaction: Eligible for smoking cessation (0/1) XXX Job = Production (0/1)
‐0.201 (0.319)
Interaction: Eligible for smoking cessation (0/1) XXX Job = Laborer (0/1)
‐0.0265 (0.252)
Interaction: Eligible for smoking cessation (0/1) XXX Eligible for smoking non‐participation surcharge (0/1) XXX Job = Professional (0/1)
0.853* (0.475)
Interaction: Eligible for smoking cessation (0/1) XXX Eligible for smoking non‐participation surcharge (0/1) XXX Job = Tech. support (0/1)
0.551(0.536)
Interaction: Eligible for smoking cessation (0/1) XXX Eligible for smoking non‐participation surcharge (0/1) XXX Job = Sales (0/1)
0.739(0.459)
Interaction: Eligible for smoking cessation (0/1) XXX Eligible for smoking non‐participation surcharge (0/1) XXX Job = Clerical (0/1)
0.462(0.485)
Interaction: Eligible for smoking cessation (0/1) XXX Eligible for smoking non‐participation surcharge (0/1) XXX Job = Service (0/1)
0.652(0.622)
Interaction: Eligible for smoking cessation (0/1) XXX Eligible for smoking non‐participation surcharge (0/1) XXX Job = Production (0/1)
0.680(0.533)
Interaction: Eligible for smoking cessation (0/1) XXX Eligible for smoking non‐participation
0.806* (0.486)
72
Model: (#1) Full sample
(#2) HRA w/out
HRA variables
(#3) HRA w/ HRA variables
(#4) Full sample,
w/ interactions
(#5) HRA w/out
HRA variables, w/
interactions
(#6) HRA w/ HRA variables, w/ interactions
surcharge (0/1) XXX Job = Laborer (0/1) Education = Some college (0/1) 0.0692*
(0.0364) 0.00133 (0.0679)
Education = College graduate (0/1) 0.131*** (0.0433)
0.0445 (0.0773)
Interaction: Eligible for smoking non‐participation surcharge (0/1) XXX Education = Some college (0/1)
0.0750 (0.0853)
Interaction: Eligible for smoking non‐participation surcharge (0/1) XXX Education = College graduate (0/1)
0.0882 (0.0987)
Interaction: Eligible for smoking cessation (0/1) XXX Education = Some college (0/1)
0.296** (0.123)
Interaction: Eligible for smoking cessation (0/1) XXX Education = College graduate (0/1)
0.352** (0.149)
Interaction: Eligible for smoking cessation (0/1) XXX Eligible for smoking non‐participation surcharge (0/1) XXX Education = Some college (0/1)
‐0.0259 (0.156)
Interaction: Eligible for smoking cessation (0/1) XXX Eligible for smoking non‐participation surcharge (0/1) XXX Education = College graduate (0/1)
0.126(0.208)
BMI = Overweight (0/1) 0.238*** (0.0387)
‐0.0417 (0.0682)
BMI = Obese (0/1) 0.431*** (0.0402)
‐0.0650 (0.0714)
Interaction: Eligible for smoking non‐participation surcharge (0/1) XXX BMI = Overweight (0/1)
0.526*** (0.0891)
Interaction: Eligible for smoking non‐participation surcharge (0/1) XXX BMI = Obese (0/1)
0.955*** (0.0926)
Interaction: Eligible for smoking cessation (0/1) XXX BMI = Overweight (0/1)
0.154(0.127)
73
Model: (#1) Full sample
(#2) HRA w/out
HRA variables
(#3) HRA w/ HRA variables
(#4) Full sample,
w/ interactions
(#5) HRA w/out
HRA variables, w/
interactions
(#6) HRA w/ HRA variables, w/ interactions
Interaction: Eligible for smoking cessation (0/1) XXX BMI = Obese (0/1)
0.204(0.134)
Interaction: Eligible for smoking cessation (0/1) XXX Eligible for smoking non‐participation surcharge (0/1) XXX BMI = Overweight (0/1)
0.0203 (0.157)
Interaction: Eligible for smoking cessation (0/1) XXX Eligible for smoking non‐participation surcharge (0/1) XXX BMI = Obese (0/1)
‐0.370** (0.186)
Constant ‐0.213*** (0.0473)
‐1.341*** (0.0765)
‐1.901*** (0.120)
0.0421 (0.0754)
‐1.311*** (0.127)
‐1.491*** (0.203)
Observations 59,962 26,571 26,571 59,962 26,571 26,571SOURCE: Author’s analysis of Fortune‐100 employer data.
NOTES: Standard errors in parentheses; *** p<0.01; ** p<0.05; * p<0.1.
74
EXHIBIT A5. Predicted probability of lifestyle management participation (except tobacco cessation) and tobacco cessation participation pre and post introduction of the tobacco use surcharge, by model and sample specification.
SOURCE: Author’s analysis of Fortune‐100 employer data.
NOTES: All within group pre surcharge vs. post surcharge differences are significant with p<0.001. Error bars represent 95 percent confidence intervals. Estimates represent average marginal effects.
29.9
31.0
29.9
31.1
31.7
27.8
30.1
32.8
30.1
31.6
31.1
28.3
26.1
21.7
26.1
21.3
41.1
24.6
26.1
21.6
26.4
21.3
39.1
23.6
0 10 20 30 40 50
LM (except tob. cessation)
Tob. cessation
LM (except tob. cessation)
Tob. cessation
LM (except tob. cessation)
Tob. cessation
LM (except tob. cessation)
Tob. cessation
LM (except tob. cessation)
Tob. cessation
LM (except tob. cessation)
Tob. cessation
Participation rate (%)Pre surcharge Post surcharge
Full sample, w/out interactions
HRA sample, w/out HRA vars., w/out interactions
HRA sample, w/ HRA vars.,
w/out interactions
Full sample, w/ interactions
HRA sample, w/out HRA vars., w/ interactions
HRA sample, w/ HRA vars., w/ interactions
75
EXHIBIT A6. Incentive effects for smoking cessation on participation in lifestyle management programs.
Characteristic Full sample HRA, w/out HRA variables HRA w/ HRA variables
Sex
Female 0.005 (‐0.029,0.038)
0.013 (‐0.040,0.065)
0.037 (‐0.021,0.096)
Male ‐0.001 (‐0.010,0.009)
‐0.004 (‐0.023,0.014)
‐0.013 (‐0.034,0.007)
Age (years)
18‐34 ‐0.025 (‐0.060,0.011)
‐0.005 (‐0.078,0.068)
0.000 (‐0.072,0.072)
35‐44 0.013 (‐0.011, 0.037)
0.028 (‐0.015,0.072)
0.016 (‐0.028,0.060)
45‐54 ‐0.013 (‐0.035, 0.009)
‐0.021 (‐0.059,0.017)
‐0.016 (‐0.054,0.023)
55‐64 0.046* (‐0.002, 0.094)
‐0.009 (‐0.101,0.083)
0.004 (‐0.088,0.096)
Pay type
Hourly 0.014* (‐0.001, 0.030)
0.006 (‐0.026,0.037)
0.012 (‐0.027,0.050)
Salaried ‐0.028** (‐0.054,‐0.002)
‐0.011 (‐0.058,0.035)
‐0.022 (‐0.083,0.040)
Union status
Non‐union 0.061*** (0.051,0.072)
0.011** (0.001,0.021)
0.010** (0.000,0.020)
Union ‐0.226*** (‐0.266,‐0.186)
‐0.095** (‐0.185,‐0.004)
‐0.090* (‐0.180,0.001)
Zip code median income, 2008 (10Ks, 2013 USD)
39K (10th percentile) ‐0.020* (‐0.041,0.001)
‐0.006 (‐0.051,0.039)
‐0.012 (‐0.058,0.035)
46K (25th percentile) ‐0.014* (‐0.029,0.001)
‐0.004 (‐0.037,0.028)
‐0.008 (‐0.042,0.025)
73K (75th percentile) 0.010** (0.000,0.019)
0.001 (‐0.013,0.016)
0.004 (‐0.012,0.019)
93K (90th percentile) 0.027** (0.002,0.051)
0.006 (‐0.041,0.053)
0.012 (‐0.037,0.061)
Charlson comorbidity (weighted) index
0 (10th through 90th percentiles) 0.005 (‐0.001,0.010)
0.005 (‐0.007,0.017)
0.008 (‐0.004,0.020)
1 (95th percentile) ‐0.030 (‐0.066,0.006)
‐0.030 (‐0.102,0.043)
‐0.051 (‐0.125,0.024)
2 (99th percentile) ‐0.065 (‐0.144,0.015)
‐0.065 (‐0.226,0.096)
‐0.111 (‐0.277,0.054)
PMPM cost (2013 USD)
76
0 (10th percentile) ‐0.020 (‐0.052,0.012)
‐0.075** (‐0.138,‐0.011)
‐0.056* (‐0.119,0.006)
6 (25th percentile) ‐0.007 (‐0.017,0.004)
‐0.028** (‐0.050,‐0.006)
‐0.021* (‐0.043,0.001)
195 (75th percentile) 0.007 (‐0.004,0.018)
0.021** (0.003,0.038)
0.015* (‐0.002,0.032)
467 (90th percentile) 0.010 (‐0.006,0.026)
0.033** (0.006,0.060)
0.024* (‐0.003,0.052)
Ethnicity
African‐American ‐0.059 (‐0.175,0.056)
Asian or Pacific Islander 0.215 (‐0.045,0.474)
Caucasian / Non‐Hispanic 0.008 (‐0.010,0.026)
Hispanic ‐0.041 (‐0.191,0.108)
Native American / Alaskan Native / Multiracial / Other
‐0.188* (‐0.402,0.026)
Education
High school or less ‐0.041 (‐0.092,0.010)
Some college 0.025 (‐0.018,0.068)
College graduate 0.006 (‐0.050,0.061)
Job classification
Executive 0.095 (‐0.027,0.218)
Professional ‐0.104* (‐0.216,0.008)
Technical support ‐0.037 (‐0.180,0.107)
Sales 0.013 (‐0.034,0.060)
Clerical 0.036 (‐0.077,0.148)
Service 0.086 (‐0.114,0.286)
Production ‐0.016 (‐0.141,0.109)
Laborer ‐0.009 (‐0.085,0.067)
BMI category
Normal weight ‐0.151*** (‐0.205,‐0.097)
Overweight ‐0.034* (‐0.070,0.001)
Obese 0.142*** (0.097,0.187)
77
SOURCE: Author’s analysis of Fortune‐100 employer data.
NOTES: Point estimates (and corresponding 95 percent confidence intervals shown with error bars) are the DDD estimates, representing the modifying effect (if any) of individual characteristics on participation in the smoking cessation program. For example, a coefficient of 0.1 for a given characteristic would mean the presence of that characteristics increase the effect of the tobacco cessation surcharge on the participation rate by 10 percentage points.
78
EXHIBIT A7. Characteristics of Fortune‐100 employees, overall and by tobacco use status.
Characteristic Overall Non‐tobacco users Tobacco users p‐value
Sex (n, %)
Female 8,908 17.9% 8,162 18.2% 746 15.4% <.01
Male 40,847 82.1% 36,757 81.8% 4,090 84.6%
Age (years) (n, %)
18‐34 12,026 24.2% 11,080 24.7% 946 19.6% <.01
35‐44 15,426 31.0% 14,026 31.2% 1,400 28.9%
45‐54 16,263 32.7% 14,415 32.1% 1,848 38.2%
55‐64 6,040 12.1% 5,398 12.0% 642 13.3%
Age (years) (mean, SD) 42.56 9.7% 42.41 9.77 43.86 9.36 <.01
Region (n, %)
East 8,280 16.6% 7,591 16.9% 689 14.2% <.01
Midwest 10,587 21.3% 9,407 20.9% 1,180 24.4%
South 24,038 48.3% 21,624 48.1% 2,414 49.9%
West 6,850 13.8% 6,297 14.0% 553 11.4%
Pay type (n, %)
Hourly 33,798 67.9% 29,989 66.8% 3,809 78.8% <.01
Salaried 15,957 32.1% 14,930 33.2% 1,027 21.2%
Union status (n, %)
Non‐union 30,399 61.1% 27,880 62.1% 2,519 52.1% <.01
Union 19,356 38.9% 17,039 37.9% 2,317 47.9%
Zip code median income, 2008 (10Ks, 2013 USD) (n, %)
<35K 2,143 4.3% 1,897 4.2% 246 5.1% <.01
35K ‐ 49K 15,678 31.5% 13,782 30.7% 1,896 39.2%
50K ‐ 64K 14,580 29.3% 13,120 29.2% 1,460 30.2%
65K ‐ 79K 8,262 16.6% 7,537 16.8% 725 15.0%
80K ‐ 99K 5,519 11.1% 5,162 11.5% 357 7.4%
>=100K 3,573 7.2% 3,421 7.6% 152 3.1%
Zip code median income, 2008 (10Ks, 2013 USD) (mean, SD)
6.19 2.31 6.25 2.34 5.63 1.82 <.01
PMPM cost (2013 USD) (mean, SD) 174.9 323.3 176.1 323.6 163.6 320.1 0.01
79
Characteristic Overall Non‐tobacco users Tobacco users p‐value
Utilization per 1,000 member years (mean, SD)
Emergency room visits 109.4 359.03 108.15 357.49 120.97 372.87 0.02
Inpatient admissions 9.9 105.63 9.37 102.84 15.10 128.55 <.01
Office visits for primary care eval. and man.
2,331.4 2,869.4 2,359.3 2,885.0 2,072.0 2,706.8 <.01
Charlson comorbidity (weighted) index (mean, SD)
0.12 0.45 0.12 0.45 0.12 0.44 0.97
Completed HRA (n, %)
Non‐participant 17,613 35.4% 15,540 34.6% 2,073 42.9% <.01
Participant 32,142 64.6% 29,379 65.4% 2,763 57.1%
Ethnicity (n, %)
African‐American 2,334 11.0% 2,232 11.3% 102 6.7% <.01
Asian or Pacific Islander 581 2.7% 552 2.8% 29 1.9%
Caucasian / Non‐Hispanic 16,100 75.7% 14,812 75.0% 1,288 84.7%
Hispanic 1,770 8.3% 1,713 8.7% 57 3.7%
Native American / Alaskan Native 266 1.3% 235 1.2% 31 2.0%
Multiracial 99 0.5% 94 0.5% 5 0.3%
Other 116 0.6% 107 0.5% 9 0.6%
Education (n, %)
High school or less 7,866 30.7% 6,944 29.5% 922 44.2% <.01
Some college 9,606 37.5% 8,752 37.2% 854 40.9%
College graduate 8,123 31.7% 7,811 33.2% 312 14.9%
Job classification (n, %)
Executive 2,080 8.2% 2,002 8.6% 78 3.8% <.01
Professional 3,122 12.3% 2,993 12.8% 129 6.2%
Technical support 807 3.2% 727 3.1% 80 3.8%
Sales 9,530 37.5% 8,791 37.6% 739 35.6%
Clerical 1,789 7.0% 1,647 7.0% 142 6.8%
Service 578 2.3% 525 2.2% 53 2.6%
Production 1,140 4.5% 987 4.2% 153 7.4%
Laborer 6,398 25.2% 5,694 24.4% 704 33.9%
80
Characteristic Overall Non‐tobacco users Tobacco users p‐value
BMI category (n, %)
Normal weight 8,015 25.8% 7,177 25.3% 838 31.0% <.01
Overweight 13,545 43.5% 12,417 43.7% 1,128 41.7%
Obese 9,546 30.7% 8,810 31.0% 736 27.2%
BMI (mean, SD) 28.3 5.1 28.3 5.1 27.7 5.6 <.01
Blood pressure category (n, %)
Normal 3,898 26.1% 3,622 26.2% 276 24.7% 0.15
Pre‐hypertensive 9,076 60.8% 8,402 60.8% 674 60.3%
Stage 1 hypertension 1,560 10.5% 1,430 10.4% 130 11.6%
Stage 2 hypertension 393 2.6% 355 2.6% 38 3.4%
Total cholesterol category (n, %)
Desirable 5,535 69.2% 5,243 69.4% 292 66.5% 0.04
Borderline high 1,946 24.3% 1,840 24.3% 106 24.1%
High 516 6.5% 475 6.3% 41 9.3%
Days of work missed due to illness (past 12 mos) (mean, SD)
1.4 4.2 1.4 4.1 2.3 6.7 0.04
Hours missed from work b/c of health problems (past 4 wks) (mean, SD)
2.1 9.0 2.0 8.8 3.7 12.9 0.06
Hours actually worked (past 4 wks) (mean, SD)
171.4 56.8 171.3 56.9 174.4 54.5 0.59
SOURCE: Author’s analysis of Fortune‐100 employer data.
NOTES: Most recent year of data for tobacco users identified during 2010 and 2011 used in the analysis.
81
EXHIBIT A8. Regression coefficients from predictors of tobacco use regressions.
Model: Full sample HRA w/out HRA variables
HRA w/ HRA variables
Male (0/1) ‐0.0226 ‐0.230** ‐0.0526 (0.0466) (0.0925) (0.105)
Age 35 to 44 (0/1) 0.257*** 0.305** 0.317** (0.0459) (0.130) (0.132)
Age 45 to 54 (0/1) 0.525*** 0.609*** 0.500*** (0.0446) (0.125) (0.128)
Age 55 to 64 (0/1) 0.452*** 0.378** 0.192 (0.0578) (0.152) (0.155)
Region = Central (0/1) 0.213*** 0.199 0.215 (0.0531) (0.141) (0.143)
Region = South (0/1) 0.140*** 0.154 0.291** (0.0482) (0.135) (0.137)
Region = West (0/1) ‐0.107* 0.0766 0.244 (0.0624) (0.177) (0.180)
Salaried (0/1) ‐0.351*** ‐0.504*** ‐0.156 (0.0415) (0.102) (0.143)
Union Member (0/1) 0.392*** 0.344*** 0.384*** (0.0326) (0.105) (0.110)
Household income (zip code median) ‐0.105*** ‐0.123*** ‐0.108*** (0.00896) (0.0219) (0.0230)
Charlson Comorbidity index (prior year) ‐0.106** 0.0237 0.0553 (0.0431) (0.0894) (0.0900)
Log(Total PMPM costs) (prior year) ‐0.0330*** ‐0.0569*** ‐0.0549*** (0.00521) (0.0131) (0.0133)
Ethnicity = Asian or Pacific Islander (0/1) 0.551 (0.339)
Ethnicity = Caucasian / Non‐Hispanic (0/1) 0.881*** (0.161)
Ethnicity = Hispanic (0/1) ‐0.307 (0.252)
Ethnicity = Native American / Alaskan Native (0/1) 0.740
(0.546) Ethnicity = Multiracial (0/1) 0.655
(0.543) Ethnicity = Other (0/1) 1.096***
(0.405) Job = Professional (0/1) 0.232
(0.254) Job = Tech. support (0/1) 0.856***
(0.296) Job = Sales (0/1) 0.265
(0.250)
82
Model: Full sample HRA w/out HRA variables
HRA w/ HRA variables
Job = Clerical (0/1) 0.676*** (0.262)
Job = Service (0/1) 0.518 (0.360)
Job = Production (0/1) 0.833*** (0.289)
Job = Laborer (0/1) 0.548** (0.265)
Education = Some college (0/1) ‐0.234*** (0.0874)
Education = College graduate (0/1) ‐0.714*** (0.126)
BMI = Overweight (0/1) ‐0.351*** (0.0912)
BMI = Obese (0/1) ‐0.534*** (0.104)
Constant ‐1.979*** ‐2.117*** ‐3.048*** (0.0878) (0.221) (0.387)
Observations 46,455 13,080 13,080 SOURCE: Author’s analysis of Fortune‐100 employer data.
NOTES: Standard errors in parentheses; *** p<0.01, ** p<0.05, * p<0.1.
83
EXHIBIT A9. Predicted probability of tobacco use by select patient characteristics.
Characteristic Full sample HRA, w/out HRA variables HRA w/ HRA variables
Sex
Female 0.099*** (0.091,0.106)
0.068*** (0.058,0.078)
0.060*** (0.051,0.069)
Male 0.097*** (0.094, 0.100)
0.055*** (0.051, 0.060)
0.057*** (0.053, 0.062)
Age (years)
18‐34 0.073*** (0.068, 0.078)
0.040*** (0.032, 0.048)
0.043*** (0.034, 0.051)
35‐44 0.092*** (0.088, 0.097)
0.053*** (0.046, 0.060)
0.057*** (0.050, 0.064)
45‐54 0.117*** (0.112, 0.122)
0.070*** (0.063, 0.078)
0.068*** (0.061, 0.074)
55‐64 0.110*** (0.101, 0.118)
0.057*** (0.046, 0.067)
0.051*** (0.041, 0.060)
Region
East 0.088*** (0.082, 0.095)
0.051*** (0.039, 0.062)
0.047*** (0.036, 0.057)
Midwest 0.107*** (0.101, 0.113)
0.061*** (0.053, 0.068)
0.057*** (0.050, 0.064)
South 0.100*** (0.096, 0.104)
0.058*** (0.053, 0.064)
0.061*** (0.055, 0.067)
West 0.080*** (0.074, 0.087)
0.054*** (0.042, 0.067)
0.059*** (0.045, 0.072)
Pay type
Hourly 0.105*** (0.102, 0.108)
0.066*** (0.061, 0.072)
0.060*** (0.054, 0.066)
Salaried 0.077*** (0.072, 0.081)
0.041*** (0.035, 0.048)
0.052*** (0.041, 0.063)
Union status
Non‐union 0.083*** (0.080, 0.087)
0.055*** (0.051, 0.059)
0.055*** (0.050, 0.059)
Union 0.118*** (0.113, 0.123)
0.075*** (0.063, 0.088)
0.077*** (0.064, 0.091)
Zip code median income, 2008 (10Ks,
84
2013 USD)
39K (10th percentile) 0.116*** (0.112, 0.121)
0.073*** (0.066, 0.080)
0.070*** (0.063, 0.078)
46K (25th percentile) 0.109*** (0.105, 0.113)
0.067*** (0.062, 0.073)
0.066*** (0.060, 0.071)
73K (75th percentile) 0.085*** (0.081, 0.088)
0.049*** (0.045, 0.054)
0.050*** (0.046, 0.055)
93K (90th percentile) 0.070*** (0.065, 0.074)
0.039*** (0.033, 0.045)
0.041*** (0.034, 0.048)
Charlson comorbidity (weighted) index
0 (10th through 90th percentiles) 0.098*** (0.095, 0.101)
0.058*** (0.054, 0.062)
0.058*** (0.053, 0.062)
1 (95th percentile) 0.089*** (0.083, 0.096)
0.059*** (0.050, 0.068)
0.061*** (0.051, 0.070)
2 (99th percentile) 0.081*** (0.069, 0.093)
0.060*** (0.041, 0.079)
0.064*** (0.044, 0.083)
PMPM cost (2013 USD)
0 (10th percentile) 0.113*** (0.107, 0.119)
0.078*** (0.067, 0.090)
0.077*** (0.066, 0.089)
6 (25th percentile) 0.100*** (0.097, 0.103)
0.063*** (0.058, 0.068)
0.063*** (0.058, 0.067)
195 (75th percentile) 0.090*** (0.087, 0.093)
0.052*** (0.048, 0.057)
0.052*** (0.048, 0.057)
467 (90th percentile) 0.088*** (0.084, 0.091)
0.050*** (0.045, 0.055)
0.050*** (0.045, 0.055)
Ethnicity
African‐American 0.029*** (0.021, 0.038)
Asian or Pacific Islander 0.049*** (0.022, 0.076)
Caucasian / Non‐Hispanic 0.067*** (0.062, 0.072)
Hispanic 0.022*** (0.013, 0.030)
Native American / Alaskan Native 0.059** (0.003, 0.114)
85
Multiracial 0.054** (0.003, 0.106)
Other 0.081*** (0.028, 0.134)
Education
High school or less 0.074*** (0.065, 0.082)
Some college 0.060*** (0.053, 0.066)
College graduate 0.038*** (0.031, 0.045)
Job classification
Executive 0.039*** (0.022, 0.056)
Professional 0.049*** (0.033, 0.065)
Technical support 0.086*** (0.058, 0.114)
Sales 0.050*** (0.044, 0.056)
Clerical 0.073*** (0.053, 0.094)
Service 0.063*** (0.034, 0.093)
Production 0.084*** (0.063, 0.106)
Laborer 0.065*** (0.055, 0.075)
BMI category
Normal weight 0.076*** (0.067, 0.085)
Overweight 0.055*** (0.049, 0.061)
Obese 0.047*** (0.040, 0.053)
SOURCE: Author’s analysis of Fortune‐100 employer data.
86
NOTES: *** p<0.01, ** p<0.05, * p<0.1. Prevalence of tobacco use: 0.097(SD 0.296) for full sample and 0.058(SD 0.234) for HRA sample. Predictions generated using average marginal effects.
87
PAPER3:ASURVEYOFPHYSICIANPERCEPTIONSONUSEOFTIMEANDTHEAPPROPRIATENESSOFCAREPROVIDED
Caloyeras JP1,2,3 MPhil, Kanter MH4,5 MD, Ives NR4 MHA, Kim CY4 PhD, Kanzaria HK2,6,7 MD MSHPM, Berry
SH1,2 MA, Brook RH1,2,8,9 MD ScD
1. Pardee RAND Graduate School, Santa Monica, CA
2. RAND Corporation, Santa Monica, CA
3. Amgen, Inc., Thousand Oaks, CA
4. Southern California Permanente Medical Group, Pasadena, CA
5. The Permanente Federation
6. University of California, San Francisco, San Francisco, CA
7. San Francisco General Hospital, San Francisco, CA
8. David Geffen School of Medicine, University of California, Los Angeles, Los Angeles, CA
9. Fielding School of Public Health, University of California, Los Angeles, Los Angeles, CA
88
ABSTRACT
Roughly 30 percent—or $530 billion per year—of total United States health care spending is thought to
be on activities like unnecessary and inefficiently delivered services that do not improve patient health.
The study objective was to measure, using a cross‐sectional, online survey, the perceptions of frontline
physicians regarding the efficiency with which they use their time and the appropriateness of care they
and others provide. Participants were clinic‐based physicians (N=1,034) practicing within Southern
California Permanente Medical Group. The main study measures were the efficiency with which
physicians use their time and the appropriateness of care they provide. The average perception among
the 636 respondents (61.5% response rate) was that 15% of their time spent on direct patient care could
be shifted to non‐physicians or automated or computerized systems. Reasons for not delegating direct
patient care tasks were of two kinds: organizational (e.g., availability of support staff) and personal
beliefs and preferences. Between 10% and 16% of care provided was perceived to be equivocal or
inappropriate. The most preferred strategies for reducing equivocal or inappropriate care were
increased use of evidence‐based clinical decision rules, patient or family education and tort reform. The
low proportion of equivocal and inappropriate care may either indicate that there is little room for
improvement or it may indicate physicians have difficulty in assessing care appropriateness. The latter
case suggests attempts to reduce or eliminate inappropriate care may be unsuccessful until physician
beliefs, knowledge or behaviors are better understood and addressed. Based upon our results we
conclude that within Southern California Permanente Medical Group the opportunity to increase value
through shifting tasks and avoiding inappropriate care is small and less than commonly accepted
national wisdom.
89
MAINTEXT
INTRODUCTION
Roughly 30 percent—or $530 billion per year— of total United States health care spending is thought by
some to be spent on activities that do not improve patient health1‐3 including the provision of
unnecessary and inefficiently delivered services.1 Reviews of studies evaluating the appropriateness of
care have additionally indicated that 30 percent or more of performed procedures, tests and
medications may be of questionable benefit.4,5 Nearly three quarters of physicians in a recent survey say
unnecessary tests and procedures are a very or somewhat serious problem for our health care system.6
Against this backdrop of evidence, a variety of policy levers have been pulled to help reduce non‐value
added activities in health care. For example, Accountable Care Organizations (ACOs),7 medical
malpractice reforms,8 value‐based insurance design9 and the “Choosing Wisely” campaign,10 have all
been implemented, in part, to reduce unnecessary services. Process improvement strategies such as
Lean and Six Sigma have been widely adopted to accomplish objectives including increasing the
efficiency, reliability and quality of delivered services.1,11
Somewhat surprising, given all these policy activities, is how little is known about what frontline
physicians think and feel regarding health care value.12 For example, do frontline physicians believe
they provide unnecessary services or recognize inefficient delivery of care in their practice? If yes, what
are the magnitudes of these non‐value added activities? What do they perceive as the primary reasons
driving these activities and the best solutions for addressing them?
To that end, we developed a survey to engage physicians along the two domains of health care value in
which they play a direct role on a day‐to‐day basis: the efficiency with which they use their time on
direct patient care tasks and the appropriateness of care perceived to be provided by other physicians
with whom they are familiar and by themselves. Our study thus directly contributes much needed
evidence that should be leveraged both to better understand the available opportunities for improving
health care value and to inform the selection of policy levers with the best chances of succeeding.
METHODS
STUDY DESIGN
We conducted a cross‐sectional study of 1,034 Southern California Permanente Medical Group (SCPMG)
physicians from four Kaiser Permanente Southern California (KPSC) medical centers using an online
90
survey focused on two domains of health care value: how physicians use their time and the
appropriateness of care they provide. The survey instrument was developed by the authors at the RAND
Corporation (RAND). Dissemination of the survey among SCPMG physicians, confidential presentation of
results within SCPMG, and drafting of this manuscript were conducted as a partnership between the
RAND and SCPMG leadership. All analyses were performed by the authors at the RAND with advice from
SCPMG leadership. Approvals from the RAND IRB were obtained for all phases of work; approval from
the KPSC IRB was also obtained for survey dissemination and all subsequent activities.
INSTRUMENT DEVELOPMENT
The two survey concepts—how physicians use their time and the appropriateness of care they provide—
were chosen as the study focus to reflect areas frontline physicians can immediately impact.13 Draft
items to explore these concepts were developed, which were then incorporated into a focus group
discussion guide. Multispecialty physicians from the greater Los Angeles area (none were part of
SCPMG) who were engaged with clinic‐based care were recruited for two focus groups. The first was
composed of nine specialists including an anesthesiologist, a neurologist, surgeons, an emergency
physician, a radiologist, and internal medicine subspecialists, while the second was made up of nine
generalists including family physicians, pediatricians, general internists, and obstetrician/gynecologists.
Qualitative analyses of focus group data were used to refine survey concept descriptions, items, and
item responses so that they best aligned with the perceptions and experiences of practicing physicians.
Our final survey instrument implemented within SCPMG is available for download (and may be re‐used
for free) from the RAND Health Surveys webpage.14
SURVEY SAMPLE
All clinic‐based physicians from four geographically and operationally distinct KPSC medical centers
(ranging in size from 248‐279 physicians) were invited by email to complete the online survey. The sites
represented a convenience sample, selected by SCPMG regional leadership using internal metrics on
back office support; two sites with below average and two sites with above average scores were
selected for participation. Full‐time SCPMG associate and partner physicians were invited to complete
the survey.
SURVEY ADMINISTRATION
SCPMG regional staff and medical center leadership distributed a memorandum to eligible physicians to
introduce the survey and the partnership with the RAND Corporation. Email invitations were
91
subsequently sent with individualized links to the online survey. To maximize the response rate up to
four rounds of reminder emails to complete the survey were sent. Physicians were given $25 Amazon
gift cards for survey completion.
STUDY MEASURES AND VARIABLES
Our first study concept, the perceived efficiency with which physicians use their time, was measured by
asking respondents to estimate the percent of direct patient care time spent on tasks that (a) “require
MY clinical / specialty training as a physician (or another physician who has similar years of clinical
training)”; (b) “could be performed by physicians who have fewer years of clinical training”; (c) “could be
performed by non‐physicians”; and (d) “could be performed primarily by an automated or computerized
system.” A question on total time working and total time spent on direct patient care activities on
average per week over the past month were asked so as to allow quantification of potential “freed up”
time if tasks could be shifted.
Our second study concept, the appropriateness of care provided, was measured by asking physicians to
estimate the proportion of care provided by others (physicians with whom they are familiar that have
the same specialty, excluding themselves) across eight clinical activity, test or procedure categories that
are perceived to be appropriate, equivocal and inappropriate. After completing the items on others
physicians were then asked to answer each question based upon care they personally provide, skipping
categories where they reported not ordering, performing or reviewing in the past month. Respondents
were given established definitions for appropriate (potential health benefit exceeds potential health
risk), equivocal (potential health benefit equal to potential health risk) and inappropriate (potential
health benefit less than potential health risk).15 Physicians were instructed to make these judgments
considering only the potential health benefit and risk to the individual patient, without assessment of
cost.
Physicians reporting that five percent or more of their time was spent on tasks that could be performed
by others were asked to indicate the type(s) of personnel or system that would be needed, followed by
the perceived reasons others do not perform these tasks currently. Physicians reporting any equivocal or
inappropriate care were asked to evaluate potential reasons for such care; all respondents (regardless of
whether they reported equivocal or inappropriate care) were asked to evaluate the helpfulness of
strategies for reducing levels of equivocal and inappropriate care.
92
Participant descriptors not available in the SCPMG administrative data were gathered from respondents;
these included area of clinical practice (primary care, medical specialty, general surgery or surgical
subspecialty, or other), years of post‐graduate training, average hours worked per week as a KPSC
physician and average hours per week spent on direct patient care.
STATISTICAL ANALYSIS
Descriptive statistics (counts, means, standard deviations and 95% confidence intervals) were used to
summarize survey results. Potential differences between survey respondents and non‐respondents were
explored using independent samples T‐tests for numeric variables and Pearson chi‐square tests for
categorical variables. Data management and statistical analyses were performed using SAS 9.4 (SAS
Institute, Cary, NC).
LIMITATIONS
We studied four sites within a single integrated delivery system; repeating this study in fee‐for‐service
settings or non‐integrated delivery systems could help determine the degree to which our results are
generalizable to other settings. However, we believe the strength of our study is to establish a lower
bound estimate on the opportunity to improve value through reducing equivocal and inappropriate care
and increasing the efficiency with which physician use their time. This is because the characteristics of
SCPMG (i.e., integrated, non‐FFS delivery system) suggest appropriateness of care should be higher vs.
traditional FFS settings, for example. Recall bias and response bias could influence the physician
responses regarding their use of time and appropriateness of care provided; the latter we attempted to
explore by having physicians first share their perceptions on appropriateness of care provided by others,
before asking them to share perceptions of care personally provided.
Readers may point out that we did not attempt to benchmark physician appropriateness of care,
meaning we do not know to what degree physicians responding to our survey can correctly identify care
that is equivocal or inappropriate. However, we were not concerned with whether physicians are
correct in their assessment of appropriateness because it is their assessment itself that underlies their
response to attempts to reduce or eliminate inappropriate care; we elaborate upon this point in our
discussion section.
RESULTS
SAMPLE CHARACTERISTICS
93
Characteristics of the 1,034 physicians invited to participate in our survey are shown in Table 1, as are
the characteristics of respondents (N=636; 61.5% response rate) and non‐respondents (N=398).
Respondents were on average 1.2 years younger, had 1.4 fewer years since medical school and were
more likely to be SCPMG “associates” (i.e., partner‐track) vs. SCPMG partners. Response rates were
similar across the four participating KPSC sites.
USE OF TIME
Physicians reported that of their time spent on direct patient care tasks 70.4% (SD 22.0%) was spent on
tasks that require their clinical / specialty training as a physician (or another physician who has similar
years of clinical training), 14.2% (SD 16.5%) on tasks that could be performed by physicians who have
fewer years of clinical training, 11.6% (SD 9.6%) on tasks that could be performed by non‐physicians and
3.8% (SD 5.2%) on tasks that could be performed primarily by an automated or computerized system.
The proportion of direct patient care time spent on tasks that requires the respondent’s clinical /
specialty training as a physician was lowest among PCPs (65.8%; SD 23.4%) followed by general surgeons
or surgical subspecialists (69.8% SD 21.3%), physicians working in an “other” discipline (73.5% SD 19.6%)
and was highest among medical specialists (76.8%; SD 19.2%) (Appendix).
SHIFTING OF DIRECT PATIENT CARE TASKS
Nearly all physicians (86.2%) responding to the survey indicated at least 5% or more of their direct
patient care time was spent on tasks that could be performed by someone other than themselves or a
physician like them with similar years of training. The staff types cited as needed by over 50% of
respondents were nurse practitioners (68.4%), physician assistants (67.7%), registered nurses (54.4%)
and PCPs (52.7%) (PCP was offered as a choice only to physicians not reporting their area of clinical
practice as primary care).
The reasons physicians gave for not delegating direct patient care tasks to alternative providers
(including non‐clinical staff and automated or computerized systems) were of two kinds: organizational
such as “type of practice organization I’m in doesn’t include them” (35.2%), “can’t find and / or retain
qualified staff” (21.9%), and personal beliefs and preferences such as “patients prefer for me to do these
tasks personally” (36.5%), and “don’t like to delegate, prefer to take care of patients myself” (15.3%).
Multiplying the 15.4% (11.6%+3.8%) of time spent on direct patient care tasks perceived to be possible
for non‐physicians to perform by the mean hours (43.5; Table 1) spent by respondents per week on
direct patient care tasks indicates that 6.7 hours per week could theoretically be repurposed for other
94
activities. For additional statistics pertaining to time‐related questions in the survey please refer to the
Appendix.
APPROPRIATENESS OF CARE
Physicians reported that across all services provided by “others” (i.e., physicians with whom they are
familiar that have the same specialty, excluding themselves) 11.1% (SD 11.2%) is perceived to be
equivocal and 5.1% (SD 7.8%) inappropriate (Figure 1). By contrast, physicians reported that across all
services they personally provide 7.4% (SD 7.8%) is perceived to be equivocal and 2.8% (SD 8.1%)
inappropriate (Figure 1). The category with the greatest perceived equivocal or inappropriate care was
“order, perform, or review non‐invasive diagnostic studies (such as x‐rays)” for both the assessment of
others (22.0%) and self (12.6%). “Recommend or perform surgeries or procedures” and “Provide
counseling or education” had the least equivocal and inappropriate care for the assessment of others
(12.3%) and self (6.4%). Stratifications of appropriateness of care by area of clinical practice for others
and self are provided in the Appendix.
REASONS FOR EQUIVOCAL OR INAPPROPRIATE CARE
Nearly 70% of physicians perceived “patient or family concerns or expectations” to be “often a reason”
for equivocal or inappropriate care—the highest of the 17 reasons offered. The four other reasons
marked as “often a reason” by at least 25% of respondents can be seen in Figure 2. The Appendix
provides statistics for all reasons included in the survey both overall and by area of clinical practice.
STRATEGIES FOR REDUCING EQUIVOCAL OR INAPPROPRIATE CARE
The two changes perceived by more than 80% of respondents as “extremely or very helpful” for
reducing the overall level of equivocal or inappropriate care were related to increased use of evidence‐
based clinical decision rules and patient or family education (Figure 3). “Change malpractice laws” was
cited by 75.5% of physicians as perceived to be “extremely or very helpful.” The other 10 strategies
reported by more than 50% of physicians as “extremely or very helpful” are shown in Figure 3; the full
list, including stratifications by area of clinical practice, can be found in the Appendix.
WILLINGNESS TO WORK TO IMPROVE VALUE
The vast majority of physicians indicated they were “very willing” or “somewhat willing” to work with
administrators, staff, and colleagues to change the way they practice to facilitate better use of time
spent on direct patient care (67.5 percent and 22.7 percent, respectively). Reported willingness was
95
similarly high to change the way they practice to minimize equivocal or inappropriate care (66.8 percent
and 24.6 percent, respectively).
DISCUSSION
We engaged a sample of frontline physicians practicing within SCPMG for their perceptions along two
domains of health care value—the use of time on direct patient care tasks and the appropriateness of
care provided—using a newly developed survey. The average perception was that 15 percent of their
time (or 6.7 hours per week) spent on direct patient care could be shifted to non‐physicians or
automated or computerized systems. Between 10 percent and 16 percent of care provided was
perceived to be equivocal (7.4 percent to 11.1 percent) or inappropriate (2.8 percent to 5.1 percent).
We conclude from our results that within SCPMG physicians perceive the opportunity to increase health
care value through the shifting of tasks and avoidance of inappropriate care to be small and less than
commonly accepted wisdom would suggest.1,4 We additionally conclude that to improve value along the
two domains in our study, policy activities should broadly focus on providing physicians (and their
patient care teams) with training and resources for discussing, communicating, identifying and managing
the preferences and expectations of patients and their families. It may also be useful to test the
perceptions of physicians about patients’ preferences for who provides care.
Because many theoretically “shiftable” tasks likely occur at irregular or inconsistent time intervals, it
may be neither feasible nor necessarily efficient for physicians to try and shift every possible task that
could be performed by others. However, despite the relatively limited opportunity identified, SCPMG
leadership is evaluating existing16,17 and new interventions focused on educating and communicating
with patients on having non‐physician staff perform some tasks within their scope of practice.
For example, ways to increase the use of the online personal action plan, expand the use of pharmacists
and nurses to assist in the management of chronic conditions, and have health educators take a greater
lead for weight management and diabetes education are being examined. In evaluating these options,
existing evidence can be leveraged: prior studies have shown how to identify specific tasks appropriate
for shifting,1,11 assess the feasibility of shifting a given task,1,11 and then monitor outcomes (including
patient satisfaction) after shifting a task.1,11,18,19
We believe the apparent disconnect between the levels of inappropriate care perceived by SCPMG
physicians and current national wisdom point to three possibilities. First, national wisdom surrounding
96
the level of inappropriate care provided may simply be overstated. If so, expectations for lowering costs
through avoiding inappropriate care, such as through the Choosing Wisely campaign,10 should be
adjusted downwards. This scenario seems unlikely given the evidence that does exist on the topic of
overuse, even though the problem is grossly understudied.5
A second possibility is that integrated, pre‐paid delivery systems like KPSC may have already eliminated
the vast majority of inappropriate care. If true, then as other health care providers create more
integrated delivery systems we might expect declines in the provision of inappropriate care.
Lastly, it is possible the surveyed physicians may be unable to recognize that some of the care they
provide is inappropriate, as mentioned in our limitations section. Under this scenario, it may be
necessary to address the beliefs and behaviors of physicians regarding what is appropriate, equivocal
and inappropriate care. To accomplish this leaders of medicine and institutions may want to consider
investments to improve the evidence base. These activities should be complemented with physician
education so physicians can better develop the skills needed to recognize in real‐time when the
expected risks of a given treatment option are equal to or less than the expected benefits, working with
their patients to find the best path forward.
The perception among more than 80% of physicians that patient and family education would be an
extremely or very helpful change to minimize equivocal and inappropriate care, has led SCPMG
leadership to take a number of steps. Physicians are being encouraged to more proactively discuss the
risks and benefits of treatment options with patients and their families. Ways to educate physicians
around how to communicate with patients who may desire services not medically appropriate are being
explored. Physicians are being reminded of shared‐decision making programs currently available, and
leadership is examining how these programs may be refined or expanded to help address issues around
equivocal care—a situation in which it is particularly important to identify the course of action best
aligned with a patient’s values and preferences, including cost.
During discussions of survey results with SCPMG physicians, a common concern reported by physicians
was ensuring patients are “satisfied” in instances when what a patient wants is withheld because it is
medically inappropriate. In a recent ABIM survey, 23 percent of physicians viewed “wanting to keep
patients happy” as a “major reason” for why they sometimes end up ordering an unnecessary test or
procedure.6 These two pieces of evidence suggest that the physicians’ perception of the patient’s care
experience—which is about the quality of the care experience in meeting the patient’s health care
97
needs, not about keeping patients happy— would need to be addressed to implement significant
practice changes aimed at reducing inappropriate care.
Although it is not surprising malpractice fears emerged as a concern with respect to equivocal care,6,20,21
it is important to note the available evidence points to changes in malpractice laws having little effect on
intensity of practice measures like imaging and hospital admission rates.22 What is clear from the
literature is that open and honest communication between physicians and patients has a protective
effect against malpractice claims.23‐25 Evidence is beginning to form supporting an inverse association
between patient experience scores and patient complaints26 and malpractice costs,27 though findings
are mixed and more research is needed.28 While studies have shown malpractice fears can be a barrier
to the use of shared‐decision making,29,30 (though the PPACA may help in this regard),31 future research
is warranted to determine the best ways (e.g., development of better treatment protocols or
algorithms) to reduce malpractice fears caused by the joint decision between a doctor and their patient
to not proceed with equivocal care or to withhold inappropriate care desired by the patient.
CONCLUSIONS
Physicians within SCPMG perceived the opportunity to increase value through shifting tasks and
avoiding inappropriate care to be small and less than commonly accepted national wisdom. Policy
activities identified as most helpful in increasing value along these two domains centered on providing
physicians (and their teams) with training and resources for discussing, communicating, identifying and
managing the preferences and expectations of patients and their families. If actual levels of equivocal
and inappropriate care are higher within SCPMG than perceived by SCPMG physicians, then our results
underscore the barriers policies aimed at reducing such care will face until steps are taken to address
physician beliefs or behaviors regarding what is equivocal and inappropriate care, and how it can be
recognized in day‐to‐day clinical practice.
REFERENCES
1. Young PL, Olsen L. The healthcare imperative: lowering costs and improving outcomes: workshop series summary. National Academies Press; 2010.
2. Chapurlat RD, Bauer DC, Nevitt M, Stone K, Cummings SR. Incidence and risk factors for a second hip fracture in elderly women. The Study of Osteoporotic Fractures. Osteoporosis International. 2003;14(2):130‐136.
3. Berwick DM, Hackbarth AD. Eliminating waste in US health care. Jama. 2012;307(14):1513‐1516. 4. McGlynn EA. Assessing the Appropriateness of Care: How Much Is Too Much?. Santa Monica,
CA: RAND Corporation, 1998. http://www.rand.org/pubs/research_briefs/RB4522.html.
98
5. Korenstein D, Falk R, Howell EA, Bishop T, Keyhani S. Overuse of health care services in the United States: an understudied problem. Archives of internal medicine. 2012;172(2):171‐178.
6. Kanis JA, Oden A, Johnell O, De Laet C, Jonsson B. Excess mortality after hospitalisation for vertebral fracture. Osteoporosis international : a journal established as result of cooperation between the European Foundation for Osteoporosis and the National Osteoporosis Foundation of the USA. Feb 2004;15(2):108‐112.
7. Devers K, Berenson R. Can accountable care organizations improve the value of health care by solving the cost and quality quandaries? Washington, DC: Robert Wood Johnson Foundation and The Urban Institute. 2009.
8. Mello MM, Chandra A, Gawande AA, Studdert DM. National costs of the medical liability system. Health affairs. 2010;29(9):1569‐1577.
9. Chernew ME, Rosen AB, Fendrick AM. Value‐based insurance design. Health Affairs. 2007;26(2):w195‐w203.
10. Prince R, Sipos A, Hossain A, et al. Sustained Nonvertebral Fragility Fracture Risk Reduction After Discontinuation of Teriparatide Treatment. Journal of Bone and Mineral Research. 2005;20(9):1507‐1513.
11. Eibner C, Hussey PS, Ridgely MS, McGlynn EA. Controlling Health Care Spending in Massachusetts: An Analysis of Options. Santa Monica, CA: RAND Corporation, 2009. http://www.rand.org/pubs/technical_reports/TR733.html.
12. Kanzaria HK, Brook RH. The silent physician. Journal of general internal medicine. 2013;28(11):1389.
13. Bentley TG, Effros RM, Palar K, Keeler EB. Waste in the US health care system: a conceptual framework. Milbank Quarterly. 2008;86(4):629‐659.
14. Stevenson M, Jones ML, De Nigris E, Brewer N, Davis S, Oakley J. A systematic review and economic evaluation of alendronate, etidronate, risedronate, raloxifene and teriparatide for the prevention and treatment of postmenopausal osteoporosis. 2005.
15. Brook RH, Chassin MR, Fink A, Solomon DH, Kosecoff J, Park RE. A method for the detailed assessment of the appropriateness of medical technologies. International journal of technology assessment in health care. 1986;2(01):53‐63.
16. Kanter MH, Lindsay G, Bellows J, Chase A. Complete care at Kaiser Permanente: transforming chronic and preventive care. The Joint Commission Journal on Quality and Patient Safety. 2013;39(11):484‐494.
17. Henry SL, Shen E, Ahuja A, Gould MK, Kanter MH. The Online Personal Action Plan: A Tool to Transform Patient‐Enabled Preventive and Chronic Care. American journal of preventive medicine. 2016.
18. Martínez‐González NA, Tandjung R, Djalali S, Rosemann T. The impact of physician–nurse task shifting in primary care on the course of disease: a systematic review. Human resources for health. 2015;13(1):1.
19. Laurant M, Reeves D, Hermens R, Braspenning J, Grol R, Sibbald B. Substitution of doctors by nurses in primary care. Cochrane Database Syst Rev. 2005;2(2).
20. Studdert DM, Mello MM, Sage WM, et al. Defensive medicine among high‐risk specialist physicians in a volatile malpractice environment. Jama. 2005;293(21):2609‐2617.
99
21. Bishop TF, Federman AD, Keyhani S. Physicians’ views on defensive medicine: a national survey. Archives of Internal Medicine. 2010;170(12):1081‐1083.
22. Waxman DA, Greenberg MD, Ridgely MS, Kellermann AL, Heaton P. The effect of malpractice reform on emergency department care. New England Journal of Medicine. 2014;371(16):1518‐1525.
23. Roter D. Patient‐Physician Relationship and Its Implications for Malpractice Litigation, The. J. Health Care L. & Pol'y. 2006;9:304.
24. Huntington B, Kuhn N. Communication gaffes: a root cause of malpractice claims. Paper presented at: Baylor University Medical Center. Proceedings 2003.
25. Carroll AE. To Be Sued Less, Doctors Should Consider Talking to Patients More. The New York Times. June 1, 2015, 2015.
26. Stelfox HT, Gandhi TK, Orav EJ, Gustafson ML. The relation of patient satisfaction with complaints against physicians and malpractice lawsuits. The American journal of medicine. 2005;118(10):1126‐1133.
27. Patient satisfaction and physician communication: Drivers of medical malpractice costs. Available at: http://www.beckershospitalreview.com/finance/patient‐satisfaction‐and‐physician‐communication‐drivers‐of‐medical‐malpractice‐costs.html.
28. Rodriguez HP, Rodday AMC, Marshall RE, Nelson KL, Rogers WH, Safran DG. Relation of patients' experiences with individual physicians to malpractice risk. International Journal for Quality in Health Care. 2008;20(1):5‐12.
29. Lewis MH, Gohagan JK, Merenstein DJ. The locality rule and the physician's dilemma: local medical practices vs the national standard of care. JAMA. 2007;297(23):2633‐2637.
30. Merenstein D. Winners and losers. Jama. 2004;291(1):15‐16. 31. Shkolnikov V, Barbieri M, Wilmoth J. The Human Mortality Database. www.mortality.org/.
Accessed January 1, 2016.
ACKNOWLEDGEMENTS
Contributors: The authors thank Dina Chau, RAND Survey Research Group, for administrative help conducting the online survey.
Funders: Development and proof‐of‐concept work for the survey used in this study was funded through a donation to the RAND Corporation by David Richards and by SCPMG. The implementation of the survey within SCPMG, including data analysis and manuscript preparation, was funded through a donation to the RAND Corporation by David Richards.
Prior presentations:
Caloyeras J, Kanter M, Cowell N, Kim C, Kanzaria H, Berry S, Brook R. A survey of physician perceptions on improving health care value. Podium Presentation. AcademyHealth 2014 Annual Research Meeting. San Diego, CA. June 8‐10, 2014.
100
EXHIBITS
TABLES
TABLE 1. Characteristics of SCPMG physicians invited to complete a survey, stratified by response.
Physician characteristic or response
Overall sample
(N=1,034)
Respondents
(N=636)
Non‐respondents
(N=398) p‐value
Sex, No. (%)
Women 431 (41.7) 268 (42.1) 163 (41.0) 0.71
Men 603 (58.3) 368 (57.9) 235 (59.0) ‐
Age (years), No. (%)
30 – 39 295 (28.5) 199 (31.3) 96 (24.1) 0.01
40 – 49 382 (36.9) 225 (35.4) 157 (39.4) ‐
50 – 59 231 (22.3) 147 (23.1) 84 (21.1) ‐
60 – 69 126 (12.2) 65 (10.2) 61 (15.3) ‐
Age (years), Mean (SD) 46.32 (9.2) 45.85 (9.2) 47.09 (9.2) 0.03
Medical school type, No. (%)
Public 401 (38.8) 263 (41.4) 138 (34.7) 0.08
Private 465 (45.0) 270 (42.5) 195 (49.0) ‐
International 168 (16.2) 103 (16.2) 65 (16.3) ‐
Years since medical school, Mean (SD) 19.03 (9.58) 18.5 (9.6) 19.92 (9.5) 0.02
Years of post‐graduate training, Mean (SD) n/a 4.78 (2.2) n/a n/a
Average total hours per week working as a SCPMG physician, Mean (SD)
n/a 48.80 (10.4) n/a n/a
Average hours per week for direct patient care, Mean (SD)
n/a 43.49 (11.8) n/a n/a
SCPMG partner status, No. (%)
Associate 222 (21.5) 154 (24.2) 68 (17.1) <.01
Partner 812 (78.5) 482 (75.8) 330 (82.9) ‐
KPSC site, No. (%)
Site 1 248 (24.0) 149 (23.4) 99 (24.9) 0.16
Site 2 261 (25.2) 172 (27.0) 89 (22.4) ‐
Site 3 246 (23.8) 139 (21.9) 107 (26.9) ‐
101
Site 4 279 (27.0) 176 (27.7) 103 (25.9) ‐
SOURCE: Authors’ analysis of SCPMG survey data.
NOTES: No. = number; SD = standard deviation; SCPMG = Southern California Permanente Medical Group; KPSC = Kaiser Permanente Southern California. Percentages may not sum to 100 because of rounding. P‐values generated using t‐tests for continuous variables and Chi‐square tests for categorical variables.
102
FIGURES
FIGURE 1. Perceived appropriateness of care provided for assessment of others and self, by clinical activity, test, or procedure category.
SOURCE: Authors’ analysis of SCPMG survey data.
NOTES: Appropriate = potential health benefit exceeds potential health risk; Equivocal = potential health benefit equal to potential health risk; Inappropriate = potential health benefit less than potential health risk. Percent only provided in figure if 5 percent or greater. Others = physicians with whom respondent is familiar that has same specialty, excluding respondent. Self = perception of care personally provided by respondent, restricted to only categories for which respondent ordered, performed or reviewed in past month.
89.8
83.8
93.6
87.6
93.8
86.8
90.4
82.2
91.9
84.0
87.4
78.0
88.0
78.4
88.9
84.2
89.6
84.0
7.4
11.1
5.2
8.9
4.9
10.0
7.8
12.2
6.4
11.1
10.1
14.7
10.0
14.9
8.5
11.3
8.1
11.1
2.8
5.1
1.2
3.4
1.3
3.3
1.8
5.6
1.7
4.9
2.5
7.3
2.1
6.6
2.6
4.5
2.4
4.9
0 25 50 75 100
Self (N=400)
Others (N=463)
Self (N=386)
Others (N=504)
Self (N=524)
Others (N=524)
Self (N=526)
Others (N=527)
Self (N=351)
Others (N=507)
Self (N=495)
Others (N=525)
Self (N=519)
Others (N=529)
Self (N=383)
Others (N=504)
Self (N=526)
Others (N=532)
Percent (%)
Clinical activity, test, or procedure category
Appropriate Equivocal Inappropriate
Answer consult from another physician
Conduct patient visits
Order, perform, or review non‐invasive diagnostic studies (such as x‐rays)
Recommend or perform surgeries or procedures
Order, perform, or review invasive diagnostic studies
Provide counseling or education
Order, prescribe, or administer medications
Order, perform, or review lab tests (such as blood
chemistries)
All services
103
FIGURE 2. Perceived reasons for equivocal or inappropriate care, limited to reasons reported as “often a reason” by at least 25 percent of respondents.
SOURCE: Authors’ analysis of SCPMG survey data.
NOTES: Percent only provided in figure if 5 percent or greater.
26.3
27.8
34.9
37
69.5
46.5
44.3
54
51.4
26.3
27.2
27.8
11.2
11.6
4.2
0 25 50 75 100
Required to justify subsequenttreatment (N=430)
Standard of practice in myspecialty (N=431)
To be sure about diagnosis, even ifno treatment implications (N=430)
To avoid any potential malpracticeissues (N=432)
Patient or family concerns orexpectations (N=433)
Percent (%)
Reason
Often a reason Sometimes a reason Rarely or never a reason
104
FIGURE 3. Perceived helpfulness of strategies to reduce overall level of equivocal or inappropriate care.
SOURCE: Authors’ analysis of SCPMG survey data.
NOTES: App. = application; Alg. = algorithm; Incr. = increase. Percent only provided in figure if 5 percent or greater.
51.7
52.4
56.5
58.7
65.3
65.3
69.3
71.2
74.4
74.9
75.5
80.8
80.7
28.4
30.4
29.7
28.3
21.3
23.5
21.5
21.9
20.3
19.7
18.9
16.5
16.9
13.6
11.6
9.5
11.0
11.1
8.1
7.2
5.0
6.3
5.6
0 25 50 75 100
Make patients and/or families pay more out‐of‐pocket for thecare they request (N=553)
Standardize billing and reimbursement procedures (N=550)
Use more PA, RN, and other non‐physician staff to performroutine care that doesn't require an MD/specialist (N=555)
Incr. public emphasis on patient safety and reducing medicalerrors (N=555)
Decrease the influence of the drug and device industry on thepractice of care (N=549)
Eliminate direct‐to‐patient advertising (N=553)
Make more use of evidence‐based criteria in determiningreimbursement levels (N=553)
Incr. use of advance directives for end‐of‐life care (N=549)
Better reimburse time spent on patient education andcounseling (N=551)
Improve design of electronic health records and prescribingsystems (N=553)
Change malpractice laws (N=555)
Incr. app. of protocols or alg. for clinical problems that have aclear evidence base supporting consistent app. of a clinical
protocol for most patients (N=552)
Educate patients and families about need to minimize care forwhich the potential health benefit is less than the potential
health risk (N=555)
Percent (%)
Strategy
Extremely or very helpful Some‐what helpful Not very helpful Not helpful at all
105
SUPPLEMENTARYMATERIALS
TABLE OF CONTENTS
Table S1. Physician reported use of time, overall and by area of clinical practice
Table S2. Staff perceived as needed for performing direct patient care tasks that could be performed by someone other than a physician with the respondent’s specialty
Table S3. Perceived reasons for not having someone or something else perform “shiftable” direct patient care tasks
Table S4. Perceived appropriateness of care for each clinical activity, test, or procedure category, overall and by area of clinical practice
Table S5. Perceived reasons for equivocal or inappropriate care for each clinical activity, test, or procedure category, overall and by area of clinical practice
Table S6. Perceived helpfulness of strategies to reduce overall level of equivocal or inappropriate care
Table S7. Willingness to work with administrators, staff, and colleagues to improve value
106
Table S1. Physician reported use of time, overall and by area of clinical practice [Mean (SD) for all].
Percent of direct patient care time on tasks that: Overall Sample
(N=596)
Area of Clinical Practice
Primary Care (N=260)
Medical Specialty (N=177)
General Surg. or Surg. Subspec.
(N=117) Other (N=42)
(1) Require MY clinical / specialty training as a physician (or another physician who has similar years of clinical training)
70.4 (22.0) 65.8 (23.4) 76.8 (19.2) 69.8 (21.3) 73.5 (19.6)
(2) Could be performed by physicians who have fewer years of clinical training
14.2 (16.5) 15.8 (18.1) 11.9 (14.8) 13.8 (15.8) 14.8 (14.1)
(3) Could be performed by non‐physicians 11.6 (9.6) 13.5 (10.3) 8.9 (7.7) 12.4 (10.3) 9.0 (7.2)
(4) Could be performed primarily by an automated or computerized system
3.8 (5.2) 4.9 (5.8) 2.4 (3.9) 4.0 (5.5) 2.7 (3.8)
SOURCE: Authors’ analysis of SCPMG survey data.
NOTES: None.
107
Table S2. Staff perceived as needed for performing direct patient care tasks that could be performed by someone other than a physician with the respondent’s specialty [N (%) for all].
Staff type: Overall Sample
(N=548)
Area of Clinical Practice
Primary Care (N=250)
Medical Specialty (N=157)
General Surg. or Surg. Subspec.
(N=106) Other (N=35)
Other kinds of clinical staff
Primary care physician* 157 (52.7) n/a 88 (56.1) 48 (45.3) 21 (60)A specialist / another specialist 108 (19.7) 55 (22.0) 30 (19.1) 18 (17) 5 (14.3)Physician assistant 371 (67.7) 199 (79.6) 75 (47.8) 78 (73.6) 19 (54.3)Nurse practitioner 375 (68.4) 202 (80.8) 90 (57.3) 57 (53.8) 26 (74.3)Nurse anesthetist 3 (0.5) ‐ 1 (0.6) 1 (0.9) 1 (2.9)Nurse / midwife 36 (6.6) 12 (4.8) 7 (4.5) 8 (7.5) 9 (25.7)Registered nurse 298 (54.4) 163 (65.2) 67 (42.7) 53 (50) 15 (42.9)Licensed vocational nurse / certified medical assistant
176 (32.1) 108 (43.2) 29 (18.5) 29 (27.4) 10 (28.6)
Licensed practical nurse 71 (13) 45 (18) 11 (7) 14 (13.2) 1 (2.9)Medical office assistant 147 (26.8) 82 (32.8) 34 (21.7) 26 (24.5) 5 (14.3)
Other kinds of staff
Dietician / nutritionist 195 (35.6) 142 (56.8) 39 (24.8) 9 (8.5) 5 (14.3)Health coach 115 (21) 86 (34.4) 18 (11.5) 5 (4.7) 6 (17.1)Doula ‐ ‐ ‐ ‐ ‐
Social worker 180 (32.8) 120 (48) 45 (28.7) 5 (4.7) 10 (28.6)Health educator 229 (41.8) 149 (59.6) 49 (31.2) 25 (23.6) 6 (17.1)Care coordinator 146 (26.6) 90 (36) 30 (19.1) 13 (12.3) 13 (37.1)Medical records specialist 67 (12.2) 37 (14.8) 14 (8.9) 11 (10.4) 5 (14.3)Insurance or billing specialist 91 (16.6) 39 (15.6) 23 (14.6) 25 (23.6) 4 (11.4)Administrative staff 99 (18.1) 44 (17.6) 26 (16.6) 21 (19.8) 8 (22.9)Other kind of staff – what kind? 48 (8.8) 16 (6.4) 17 (10.8) 11 (10.4) 4 (11.4)
SOURCE: Authors’ analysis of SCPMG survey data.
NOTES: *Respondents indicating area of clinical practice as primary care not allowed to select primary care physician. Question only asked to respondents reporting that 5 percent or more of their time could be performed by someone or something else.
108
Table S3. Perceived reasons for not having someone or something else perform “shiftable” direct patient care tasks [N (%) for all].
Reason: Overall Sample
(N=548)
Area of Clinical Practice
Primary Care (N=250)
Medical Specialty (N=157)
General Surg. or Surg. Subspec.
(N=106) Other (N=35)
Can’t find and / or retain qualified staff 120 (21.9) 62 (24.8) 28 (17.8) 20 (18.9) 10 (28.6)Don’t have someone to supervise them 65 (11.9) 33 (13.2) 21 (13.4) 5 (4.7) 6 (17.1)Can’t get reimbursed to cover them 75 (13.7) 26 (10.4) 24 (15.3) 22 (20.8) 3 (8.6)Couldn’t keep them busy – not enough work
30 (5.5) 6 (2.4) 16 (10.2) 5 (4.7) 3 (8.6)
Type of practice organization I’m in doesn’t include them
193 (35.2) 87 (34.8) 49 (31.2) 41 (38.7) 16 (45.7)
Concerned with malpractice issues 38 (6.9) 20 (8) 9 (5.7) 6 (5.7) 3 (8.6)Don’t like to delegate, prefer to take care of patients myself
84 (15.3) 40 (16) 23 (14.6) 19 (17.9) 2 (5.7)
Don’t trust automated / computerized systems for clinical duties
32 (5.8) 18 (7.2) 7 (4.5) 4 (3.8) 3 (8.6)
Referral barriers make it simpler to do these tasks myself
66 (12) 36 (14.4) 15 (9.6) 9 (8.5) 6 (17.1)
Patients prefer for me to do these tasks personally
200 (36.5) 102 (40.8) 52 (33.1) 32 (30.2) 14 (40)
Legal barriers 26 (4.7) 16 (6.4) 6 (3.8) 2 (1.9) 2 (5.7)Other reason(s) – what are they? 159 (29) 68 (27.2) 50 (31.8) 29 (27.4) 12 (34.3)SOURCE: Authors’ analysis of SCPMG survey data.
NOTES: Question only posed to respondents reporting that 5 percent or more of their time could be performed by someone or something else.
109
Table S4. Perceived appropriateness of care for each clinical activity, test, or procedure category, overall and by area of clinical practice [Mean (SD) for all; Ns refer to overall sample].
Clinical activity, test, or procedure
Overall Sample
Area of Clinical Practice
Primary Care Medical Specialty
General Surg. or Surg. Subspec. Other
App. Eq. In. App. Eq. In. App. Eq. In. App. Eq. In. App. Eq. In.Conduct patient visits (N=526)
89.6 (13.0)
8.1 (9.9)
2.4 (5.8)
88.9 (13.5)
8.4 (9.6)
2.8 (7.0)
91.2 (11.1)
7.1 (9.1)
1.7 (4.2)
88.2 (15.0)
9.4 (12.3)
2.4 (5.2)
91.0 (11.2)
6.6 (7.5)
2.5 (5.1)
Answer consult from
another physician (N=383) 88.9
(16.4) 8.5
(13.3) 2.6
(7.7) 90.1
(20.1) 7.7
(15.6) 2.2
(11.4) 89.1
(14.0) 8.3
(11.1) 2.6
(6.1) 87.5
(16.8) 9.6
(14.4) 2.9
(5.9) 88.4
(12.9) 9.1
(11.2) 2.5
(4.6) Order, perform, or review lab tests (such as blood chemistries) (N=519)
88.0 (13.9)
10.0 (11.5)
2.1 (5.9)
84.9 (14.8)
12.5 (12.8)
2.6 (5.4)
91.7 (12.0)
6.4 (8.2)
1.9 (7.3)
88.1 (13.1)
10.6 (11.9)
1.3 (4.1)
90.5 (13.6)
7.6 (10.1)
1.9 (5.5)
Order, perform, or review non‐invasive diagnostic studies (such as x‐rays) (N=495)
87.4 (15.1)
10.1 (12.5)
2.5 (6.3)
85.7 (13.6)
11.3 (10.8)
3.0 (6.0)
89.4 (16.2)
8.4 (13.1)
2.2 (7.9)
89.3 (14.6)
8.7 (12.4)
2.0 (5.0)
85.5 (20.2)
12.8 (19.4)
1.7 (4.3)
Order, perform, or review invasive diagnostic studies (N=351)
91.9 (13.8)
6.4 (11.1)
1.7 (6.9)
91.7 (11.9)
6.3 (8.6)
2.0 (7.3)
92.6 (14.2)
5.5 (10.2)
1.9 (8.4)
92.4 (13.6)
6.6 (12.2)
1.0 (3.6)
87.8 (21.9)
10.8 (21.4)
1.4 (3.2)
Order, prescribe, or administer medications (N=526)
90.4 (12.1)
7.8 (9.8)
1.8 (4.8)
88.4 (13.0)
9.4 (10.8)
2.3 (5.2)
93.5 (8.8)
5.6 (7.3)
0.9 (2.8)
90.2 (13.5)
7.9 (10.5)
1.9 (5.6)
90.7 (12.3)
6.8 (8.7)
2.4 (5.6)
Provide counseling or education (N=524)
93.8 (12.4)
4.9 (9.7)
1.3 (5.3)
94.2 (11.8)
4.9 (10.1)
0.9 (4.0)
94.4 (13.0)
3.9 (8.7)
1.7 (7.2)
92.2 (12.6)
6.2 (10.1)
1.6 (4.9)
93.3 (12.6)
5.1 (9.6)
1.5 (4.0)
Recommend or perform
surgeries or procedures (N=386)
93.6 (12.1)
5.2 (10.0)
1.2 (5.8)
93.2 (12.5)
5.3 (8.8)
1.5 (8.4)
93.9 (13.4)
5.1 (12.2)
1.0 (3.6)
93.6 (11.0)
5.3 (9.5)
1.1 (3.4)
94.2 (8.4)
4.8 (7.0)
1.0 (2.8)
All services (N=400) 89.8 (12.2)
7.4 (7.8)
2.8 (8.1)
87.5 (14.9)
8.5 (8.3)
4.0 (12.0)
91.5 (9.4)
6.5 (6.9)
2.0 (3.8)
91.3 (10.5)
6.7 (8.4)
2.0 (3.8)
90.4 (9.9)
7.6 (7.1)
2.0 (3.7)
SOURCE: Authors’ analysis of SCPMG survey data.
110
NOTES: For each clinical activity, test, or procedure the responding physician had to indicate that they ordered, performed, or reviewed in the past month (yes / no) to enter the proportion appropriate, equivocal and inappropriate.
111
Table S5. Perceived reasons for equivocal or inappropriate care for each clinical activity, test, or procedure category, overall and by area of clinical practice [N (%) for all].
Likely reasons for equivocal or inappropriate:
Overall Sample
Area of Clinical Practice
Primary Care Medical Specialty
General Surg. or Surg. Subspec. Other
Often Some‐times
Rarely or
never Often Some‐times
Rarely or
never Often Some‐ times
Rarely or
never Often Some‐times
Rarely or
never Often Some‐times
Rarely or
never Patient or family concerns or expectations
301 (69.5)
114 (26.3)
18(4.2)
152 (75.6)
47 (23.4)
2(1)
77 (63.6)
35 (28.9)
9(7.4)
49 (60.5)
26 (32.1)
6(7.4)
23 (76.7)
6(20)
1(3.3)
Required to justify subsequent treatment
113 (26.3)
200 (46.5)
117 (27.2)
55 (27.5)
101 (50.5)
44(22)
28 (23.1)
55 (45.5)
38 (31.4)
21 (26.6)
32 (40.5)
26 (32.9)
9(30)
12(40)
9(30)
To qualify for “pay‐for‐performance” incentives
33 (7.7)
86(20)
311 (72.3)
16(8)
52(26)
132 (66)
9(7.4)
18 (14.9)
94 (77.7)
3(3.8)
14 (17.7)
62 (78.5)
5(16.7)
2(6.7)
23 (76.7)
To be sure about diagnosis, even if no treatment implications
150 (34.9)
232 (54)
48 (11.2)
74 (36.8)
114 (56.7)
13(6.5)
39 (32.5)
63 (52.5)
18(15)
26 (32.9)
42 (53.2)
11 (13.9)
11 (36.7)
13 (43.3)
6(20)
Easier or faster to order full set of tests in electronic systems
58 (13.5)
171 (39.7)
202 (46.9)
31 (15.3)
91(45)
80 (39.6)
13 (10.8)
44 (36.7)
63 (52.5)
10 (12.7)
25 (31.6)
44 (55.7)
4(13.3)
11 (36.7)
15(50)
To avoid any potential malpractice issues
160 (37)
222 (51.4)
50 (11.6)
79 (39.1)
103 (51)
20(9.9)
35 (29.2)
68 (56.7)
17 (14.2)
35 (43.8)
37 (46.3)
8(10)
11 (36.7)
14 (46.7)
5(16.7)
Reordered a test that was not done properly
40 (9.3)
225 (52.2)
166 (38.5)
17(8.4)
107 (53)
78 (38.6)
9(7.5)
59 (49.2)
52 (43.3)
10 (12.7)
45(57)
24 (30.4)
4(13.3)
14 (46.7)
12(40)
Test or imaging result or medical record was not available, or could not find it
22 (5.1)
139 (32.3)
269 (62.6)
9(4.5)
64 (31.8)
128 (63.7)
5(4.2)
37 (30.8)
78(65)
6(7.6)
30(38)
43 (54.4)
2(6.7)
8(26.7)
20 (66.7)
Needed to use test / treatment facilities in order to ensure they stay in business and available for patients
5 (1.2)
18(4.2)
408 (94.7)
1(0.5)
7(3.5)
193 (96)
1(0.8)
6 (5)
114 (94.2)
2(2.5)
3(3.8)
74 (93.7)
1(3.3)
2(6.7)
27(90)
112
Likely reasons for equivocal or inappropriate:
Overall Sample
Area of Clinical Practice
Primary Care Medical Specialty
General Surg. or Surg. Subspec. Other
Often Some‐times
Rarely or
never Often Some‐times
Rarely or
never Often Some‐ times
Rarely or
never Often Some‐times
Rarely or
never Often Some‐times
Rarely or
never To see how well test / therapy works for patients
76 (17.6)
197 (45.7)
158 (36.7)
38 (18.8)
96 (47.5)
68 (33.7)
19 (15.8)
50 (41.7)
51 (42.5)
13 (16.5)
32 (40.5)
34(43)
6(20)
19 (63.3)
5(16.7)
To change the level of billing for an encounter or visit
5 (1.2)
19(4.4)
406 (94.4)
‐ 9(4.5)
191 (95.5)
2(1.7)
5 (4.1)
114 (94.2)
‐ 5(6.3)
74 (93.7)
3(10)
‐ 27(90)
Needed practice revenues to cover costs
3 (0.7)
16(3.7)
412 (95.6)
‐ 6(3)
196 (97)
1(0.8)
4 (3.3)
115 (95.8)
‐ 4(5.1)
75 (94.9)
2(6.7)
2(6.7)
26 (86.7)
Needed to maintain a reasonable level of personal income
5 (1.2)
17(4)
408 (94.9)
‐ 5(2.5)
197 (97.5)
1(0.8)
5 (4.2)
113 (95)
1(1.3)
5(6.3)
73 (92.4)
3(10)
2(6.7)
25 (83.3)
Expected standard of practice in my geographic area
65 (15.2)
182 (42.4)
182 (42.4)
33 (16.4)
89 (44.3)
79 (39.3)
17 (14.2)
47 (39.2)
56 (46.7)
8(10.3)
31 (39.7)
39(50)
7(23.3)
15(50)
8(26.7)
Standard of practice in my medical group, or among closest colleagues
84 (19.5)
195 (45.2)
152 (35.3)
45 (22.4)
94 (46.8)
62 (30.8)
22 (18.2)
52 (43)
47 (38.8)
7(8.9)
34(43)
38 (48.1)
10 (33.3)
15(50)
5(16.7)
Standard of practice in my specialty
120 (27.8)
191 (44.3)
120 (27.8)
53 (26.2)
94 (46.5)
55 (27.2)
33 (27.5)
52 (43.3)
35 (29.2)
21 (26.6)
33 (41.8)
25 (31.6)
13 (43.3)
12(40)
5(16.7)
Influence of the drug and device industry
4 (0.9)
35(8.1)
392 (91)
‐ 13(6.4)
189 (93.6)
1(0.8)
13 (10.7)
107 (88.4)
1(1.3)
5(6.4)
72 (92.3)
2(6.7)
4(13.3)
24(80)
SOURCE: Authors’ analysis of SCPMG survey data.
NOTES: Question only posed to respondents reporting some equivocal or inappropriate care.
113
Table S6. Perceived helpfulness of strategies to reduce overall level of equivocal or inappropriate care [N (%) for all].
Helpfulness of change in reducing
equivocal or inappropriate
care:
Overall Sample
Area of Clinical Practice
Primary Care Medical Specialty
General Surg. or Surg. Subspec. Other
Extrem. or very
Some‐ what
Not very
Not at all
Extrem. or very
Some‐what
Not very
Not at all
Extrem. or very
Some‐what
Not very
Not at all
Extrem. or very
Some‐what
Not very
Not at all
Extrem. or very
Some‐what
Not very
Not at all
Change malpractice laws
419 (75.5)
105 (18.9)
28 (5)
3 (0.5)
180 (73.8)
47 (19.3)
15 (6.1)
2 (0.8)
126 (77.3)
29 (17.8)
7 (4.3)
1 (0.6)
82 (76.6)
21 (19.6)
4 (3.7)
‐ 31 (75.6)
8 (19.5)
2 (4.9)
‐
Change laws or standards to permit delegation of more care activities to persons with different levels of training or other professional background
221 (40)
184 (33.3)
110 (19.9)
37 (6.7)
101 (41.4)
81 (33.2)
46 (18.9)
16 (6.6)
63 (38.7)
58 (35.6)
29 (17.8)
13 (8)
43 (41.3)
32 (30.8)
23 (22.1)
6 (5.8)
14 (34.1)
13 (31.7)
12 (29.3)
2 (4.9)
Eliminate direct‐to‐patient advertising
361 (65.3)
130 (23.5)
45 (8.1)
17 (3.1)
166 (67.8)
56 (22.9)
18 (7.3)
5(2)
101 (61.6)
42 (25.6)
13 (7.9)
8 (4.9)
62 (60.2)
26 (25.2)
13 (12.6)
2 (1.9)
32(78)
6 (14.6)
1 (2.4)
2 (4.9)
Better reimburse time spent on patient education and counseling
410 (74.4)
112 (20.3)
27 (4.9)
2 (0.4)
200 (82.6)
38 (15.7)
4 (1.7)
‐ 118 (72)
35 (21.3)
10 (6.1)
1 (0.6)
60 (57.1)
33 (31.4)
11 (10.5)
1(1)
32(80)
6(15)
2(5)
‐
Better support medical education to reduce physician debt
239 (43.3)
164 (29.7)
112 (20.3)
37 (6.7)
114 (46.9)
69 (28.4)
48 (19.8)
12 (4.9)
73 (44.5)
48 (29.3)
29 (17.7)
14 (8.5)
35 (33.3)
34 (32.4)
28 (26.7)
8 (7.6)
17 (42.5)
13 (32.5)
7 (17.5)
3 (7.5)
Standardize billing and reimbursement
288 (52.4)
167 (30.4)
64 (11.6)
31 (5.6)
123 (50.8)
76 (31.4)
32 (13.2)
11 (4.5)
84 (51.5)
56 (34.4)
12 (7.4)
11 (6.7)
58 (55.2)
26 (24.8)
17 (16.2)
4 (3.8)
23 (57.5)
9 (22.5)
3 (7.5)
5 (12.5)
114
Helpfulness of change in reducing
equivocal or inappropriate
care:
Overall Sample
Area of Clinical Practice
Primary Care Medical Specialty
General Surg. or Surg. Subspec. Other
Extrem. or very
Some‐ what
Not very
Not at all
Extrem. or very
Some‐what
Not very
Not at all
Extrem. or very
Some‐what
Not very
Not at all
Extrem. or very
Some‐what
Not very
Not at all
Extrem. or very
Some‐what
Not very
Not at all
procedures Allow more flexibility in billing and reimbursement procedures
207 (37.9)
190 (34.8)
108 (19.8)
41 (7.5)
89 (37.1)
85 (35.4)
52 (21.7)
14 (5.8)
68 (41.7)
57 (35)
24 (14.7)
14 (8.6)
35 (33.7)
33 (31.7)
28 (26.9)
8 (7.7)
15 (38.5)
15 (38.5)
4 (10.3)
5 (12.8)
Make patients and / or families pay more out‐of‐pocket for the care they request
286 (51.7)
157 (28.4)
75 (13.6)
35 (6.3)
125 (51.2)
69 (28.3)
30 (12.3)
20 (8.2)
82(50)
45 (27.4)
26 (15.9)
11 (6.7)
62 (57.9)
29 (27.1)
13 (12.1)
3 (2.8)
17 (44.7)
14 (36.8)
6 (15.8)
1 (2.6)
Make more use of evidence‐based criteria in determining reimbursement levels
383 (69.3)
119 (21.5)
40 (7.2)
11 (2)
174 (71.3)
51 (20.9)
16 (6.6)
3 (1.2)
108 (65.9)
43 (26.2)
9 (5.5)
4 (2.4)
72 (67.9)
20 (18.9)
12 (11.3)
2 (1.9)
29 (74.4)
5 (12.8)
3 (7.7)
2 (5.1)
Increase use of advance directives for end‐of‐life care
391 (71.2)
120 (21.9)
20 (3.6)
18 (3.3)
168 (69.7)
56 (23.2)
10 (4.1)
7 (2.9)
115 (70.1)
39 (23.8)
6 (3.7)
4 (2.4)
79 (75.2)
19 (18.1)
2 (1.9)
5 (4.8)
29 (74.4)
6 (15.4)
2 (5.1)
2 (5.1)
Use more PA, RN, and other non‐physician staff to perform routine care that doesn’t require an MD / specialist
314 (56.6)
165 (29.7)
53 (9.5)
23 (4.1)
149 (60.8)
64 (26.1)
22(9)
10 (4.1)
77 (46.7)
57 (34.5)
22 (13.3)
9 (5.5)
69 (65.1)
27 (25.5)
6 (5.7)
4 (3.8)
19 (48.7)
17 (43.6)
3 (7.7)
‐
115
Helpfulness of change in reducing
equivocal or inappropriate
care:
Overall Sample
Area of Clinical Practice
Primary Care Medical Specialty
General Surg. or Surg. Subspec. Other
Extrem. or very
Some‐ what
Not very
Not at all
Extrem. or very
Some‐what
Not very
Not at all
Extrem. or very
Some‐what
Not very
Not at all
Extrem. or very
Some‐what
Not very
Not at all
Extrem. or very
Some‐what
Not very
Not at all
Educate patients and families about need to minimize care for which the potential health benefit is less than the potential health risk
448 (80.7)
94 (16.9)
12 (2.2)
1 (0.2)
207 (85.2)
33 (13.6)
3 (1.2)
‐ 128 (77.1)
35 (21.1)
3 (1.8)
‐ 79 (73.8)
22 (20.6)
5 (4.7)
1 (0.9)
34 (87.2)
4 (10.3)
1 (2.6)
‐
Increase public emphasis on patient safety and reducing medical errors
326 (58.7)
157 (28.3)
61 (11)
11 (2)
160 (65.3)
57 (23.3)
26 (10.6)
2 (0.8)
96 (57.8)
53 (31.9)
13 (7.8)
4 (2.4)
47 (44.8)
37 (35.2)
17 (16.2)
4 (3.8)
2 (59)
10 (25.6)
5 (12.8)
1 (2.6)
Decrease the influence of the drug and device industry on the practice of care
359 (65.4)
117 (21.3)
61 (11.1)
12 (2.2)
163 (67.4)
56 (23.1)
20 (8.3)
3 (1.2)
101 (61.6)
35 (21.3)
21 (12.8)
7 (4.3)
66 (63.5)
20 (19.2)
18 (17.3)
‐ 29 (74.4)
6 (15.4)
2 (5.1)
2 (5.1)
Increase application of protocols or algorithms for clinical problems that have clear evidence base supporting consistent
446 (80.8)
91 (16.5)
9 (1.6)
6 (1.1)
210 (86.4)
30 (12.3)
2 (0.8)
1 (0.4)
121 (73.3)
38 (23)
4 (2.4)
2 (1.2)
82 (78.1)
18 (17.1)
3 (2.9)
2 (1.9)
33 (84.6)
5 (12.8)
‐ 1 (2.6)
116
Helpfulness of change in reducing
equivocal or inappropriate
care:
Overall Sample
Area of Clinical Practice
Primary Care Medical Specialty
General Surg. or Surg. Subspec. Other
Extrem. or very
Some‐ what
Not very
Not at all
Extrem. or very
Some‐what
Not very
Not at all
Extrem. or very
Some‐what
Not very
Not at all
Extrem. or very
Some‐what
Not very
Not at all
Extrem. or very
Some‐what
Not very
Not at all
application of a clinical protocol for most patients (e.g., protocol for acute myocardial infarction care in the emergency room) Improve design of electronic health records and prescribing systems
414 (74.9)
109 (19.7)
23 (4.2)
7 (1.3)
187 (76.3)
47 (19.2)
8 (3.3)
3 (1.2)
121 (73.3)
34 (20.6)
8 (4.8)
2 (1.2)
77(72.6)
22 (20.8)
6 (5.7)
1 (0.9)
29 (78.4)
6 (16.2)
1 (2.7)
1 (2.7)
Other – specify: 44 (83)
3 (5.7) 1 (1.9)
5 (9.4)
18 (78.3)
‐ 1 (4.3)
4 (17.4)
14 (93.3)
1 (6.7)
‐ ‐ 5(71.4)
2 (28.6)
‐ ‐ 7(87.5)
‐ ‐ 1 (12.5)
SOURCE: Authors’ analysis of SCPMG survey data.
NOTES: Extrem. = extremely.
117
Table S7. Willingness to work with administrators, staff, and colleagues to improve value [N (%) for all]
Willingness to work with administrators, staff, and colleagues to:
Overall Sample
Area of Clinical Practice
Primary Care
Medical Specialty
General Surg. or Surg.
Subspec. Other Change the way you organize your practice, so that you are making more appropriate or efficient use of your time spent on direct patient care?
Very willing 414 (67.5)
191 (70.2)
114 (64) 78 (66.1) 31 (68.9)
Somewhat willing 139 (22.7)
63 (23.2) 38 (21.3) 29 (24.6) 9 (20)
Neutral 35 (5.7) 12 (4.4) 14 (7.9) 7 (5.9) 2 (4.4)Somewhat unwilling 9 (1.5) 4 (1.5) 3 (1.7) 2 (1.7) ‐
Very unwilling 1 (0.2) ‐ 1 (0.6) ‐ ‐
Not sure / don’t know 10 (1.6) 1 (0.4) 4 (2.2) 2 (1.7) 3 (6.7) 5 (0.8) 1 (0.4) 4 (2.2) ‐ ‐
Change the way you organize your practice to minimize equivocal or inappropriate care?
Very willing 373 (66.8) 169 (69)
104 (62.7) 74 (69.2) 26 (65)
Somewhat willing 137 (24.6) 58 (23.7) 44 (26.5) 24 (22.4) 11 (27.5)
Neutral 27 (4.8) 11 (4.5) 10 (6) 5 (4.7) 1 (2.5)Somewhat unwilling 4 (0.7) 1 (0.4) 2 (1.2) 1 (0.9) ‐
Very unwilling 6 (1.1) 1 (0.4) 4 (2.4) ‐ 1 (2.5)Not sure / don’t know 8 (1.4) 4 (1.6) 1 (0.6) 2 (1.9) 1 (2.5) 3 (0.5) 1 (0.4) 1 (0.6) 1 (0.9) ‐
SOURCE: Authors’ analysis of SCPMG survey data.
NOTES: None.