-
7/24/2019 Screening Elders for Risk of Hospital Admission (1993)
1/7
Screening Elders
for
Risk of
Hospital Admission
Chad Boult, MD, MPH,* Bryan Dowd, PhD,t David McCaffrey, BA, Lisa Boult, MD, MS,*
Rafael Hernandez, MD,* and Harry Krulewitch, MDS
Objective:To define a set of screening criteria that identifies
elders who are at high risk for repeated hospital admission
in the future.
Design:Longitudinal cohort study. Logistic regression analy-
sis of data from half of the subjects was used to identify risk
factors for repeated hospital admission. The ability of these
risk factors to identify elders who are at high risk for repeated
hospitalization in the future was then tested using data from
the other half of the subjects.
Setting:United States.
Participants: A subsample
n = 5876)
of a multistage prob-
ability sample of all non-institutionalized U.S. civilians who
were
70
years or older in
1984.
Measurements: At baseline (1984), elderly subjects were
asked about their demographic, socioeconomic, medical, and
functional characteristics and about their recent use of health
services. Their subsequent hospital admissions and mortality
were then monitored through the records of the Medicare
program and the National Death Index
(1985-88).
Results:
Among the subjects in the first half of the sample,
eight factors emerged as risk factors for repeated admission:
older age, male sex, poor self-rated general health, availability
of an informal caregiver, having ever had coronary artery
disease, and having had, during the previous year, a hospital
admission, more than six doctor visits, or diabetes. Based on
the presence or absence of these factors in
1984, 7.2%
of the
subjects in the second half of the sample were estimated to
have a high probability of repeated admission
(Pra 0.5)
during 1985-1988. In comparison with subjects estimated to
have a low risk
(Pra14 Bed days in past year
Less
active
in past
year
Health
worse
in
past year
Coronary artery
disease
(CAD)*
Cerebrovascular disease CVD)**
Diabetes in past
year
Hypertension
(ever)
Cancer
(ever)
Arthritis
or
rheumatism
(ever)
Hospital admission in past year
>6 Doctor visits in past year
1
Years
since last doctor
visit
Used visiting nurse
in
past year
No informal caregiver available
Functional limitation***
>1 Fall in past year
Visual impairment
Incontinence (urinary or fecal)
Cognitive impairment#
Hearing
impairment
Predisposing variables
Age
70-74
75-79
80-84
85+
Male
sex
Non-white
race
0-6 Years of education
Lives alone
Has
little control
of
health
< 10,000 Annual income
Enabling
variable
15.8
20.7
30.6
21.1
11.1
13.3
18.6
14.9
16.5
17.7
10.2
44.7
12.5
54.1
21.2
21.4
15.7
2.8
7.0
10.7
10.6
7.9
13.3
26.2
6.5
43.0
30.8
16.5
9.7
42.5
9.3
20.3
36.7
13.6
38.0
~
0.5
1 . 1
9.0
9.0
1.6
2.8
0.7
0.7
0.5
1.4
0 . 0
0.4
0.4
0.3
1.6
1.4
0.8
0.4
0.4
10.0
1.3
0 . 0
0.0
0.0
1.2
0.0
13.6
14.6
* Ever had angina, coronary heart disease, myocardial infarc-
tion or other heart attack.
**
Ever
had arteriosclerosis,hardening of the arteries,
stroke,
or cerebrovascular
accident.
***
Unable to do one or more of five ADLs (eating, using a
toilet, transferring between
a
bed and chair, dressing, bath-
ing) and four IADLs (cooking, shopping, using
a
telephone,
light housecleaning)
without
the assistance of another person
or
special device.
Frequent
or
increasing trouble remembering things or get-
ting confused.
relation
rho)
between:
1)
he error term of a model of
the probability of having repeated admission (inde-
pendent variables = the eight significant risk factors)
and
(2)
the error term of a model of the probability of
having unavailable Medicare records (independent
variables = age, educational level, cognitive impair-
ment, and the use of medical services in the previous
year).22
-
7/24/2019 Screening Elders for Risk of Hospital Admission (1993)
4/7
_ _
14
BOULT ET AL
RESULTS
The baseline (1984) characteristics of the 5876 sub-
jects for whom 1985-1988 hospitalization data were
available are shown in Table
1.
During the 4-year follow-up interval (1985-1988),
1670 subjects (28.4%) were admitted more than once,
and 646
(11.0 )
died.
The Breslow-Day analyses suggested that four two-
way interactions had significant effects on the proba-
bility of future admission P
0.10),
so the original model (without
interaction terms) was retained as the final model.
The results
of
the logistic regression of the probabil-
ity of repeated admission on the 28 independent vari-
ables are shown in Table
2.
Of the significant predic-
tors, the need variables tend to make a larger collective
contribution to the probability of future admission than
do the predisposing variables (age and sex), confirming
a pattern that has been observed before.
After the logistic model was constructed using data
from the first half of the sample,
it
was validated using
data from the second half. Second half subjects were
classified on the basis of their 1984 characteristics as
either high-risk
(PraL
0.5,
n
=
204, 7.2%) or low-risk
(Pra< 0.5,
n
=
2623, 92.8%). Table 3 contrasts the
subsequent admission and mortality experiences of
these two groups during 1985-1988. In order to adjust
for the two groups unequal mortality rates, hospital
days and charges are reported per person-year sur-
vived.
Compared with the low-risk subjects, the high-risk
subjects experienced substantially worse outcomes in
every category: repeated admission (41.8% vs 26.1%),
mortality (44.2% vs 19.0%), hospital days per person-
year
(5 .2
vs 2.6), and hospital charges per person-year
($3731 vs $1841). About 70% (69.6) of the identified
high-risk subjects either died or had repeated admis-
sion during the 4 years of follow-up. By comparison,
only 37.4% of the low-risk subjects died or had re-
peated admission.
In order to test the sensitivity of this studys findings
to our definition of repeated admission, we repeated
the studys regression analyses and validity studies
using three alternate definitions of repeated admission:
one admission in
2
years, two admissions in
3
years,
and three admissions in 4 years. In Table 4, the alter-
nate results are compared with those derived using the
original definition of repeated admission.
All four logistic models included similar variables as
significant risk factors for repeated admission (left col-
umn). In the validation studies, the model based on
the original definition of repeated admission identified
the highest proportion of LSOA subjects as being high-
risk (center column). Furthermore, these identified
high-risk subjects subsequently used more hospital
days (right column) than did the high-risk subjects
identified by any of the models based on alternate
definitions. This successful use of a 4-year interval
supports and extends earlier observations that individ-
ual elders concentrated use of hospitals often occurs
during time periods of at least
3
year^.^,^
IAGS-AUGUST 1993-VOL. 42, NO. 8
TABLE 2. LOGISTIC MODEL OF PREDICT ORS OF
REPEATED ADM ISSION WITHIN FOUR YEARS
Adjusted
(95%
C.I.)
Regression Odd s Ratio
Coefficient
Need
variables
General
health
Excellent
Very good
Good
Fair
Poor
>14
Bed days in past year
Less active in past year
Health worse in past year
Coronary
artery disease
Cerebrovascular disease
Diabetes
in
past
year
Hypertension
Cancer
Arthritis
or
rheumatism
Hospital admission in past
year
>6
Doctor visits
in
past
year
1+
Years since last doctor
visit
Used visiting nurse
in
past
year
No informal caregiver
available
Functional
limitation
>1
Fall
in
past year
Visual
impairment
Incontinence (urinary or
Cognitive impairment
Hearing impairment
Predisposing variables
fecal)
Age
70-74
75-79
80-84
85+
Male sex
Non-white race
0-6
Years
of
education
Lives alone
Has
little control of health
-
7/24/2019 Screening Elders for Risk of Hospital Admission (1993)
5/7
SCREENING FOR RISK OF ADMISSION
815
AGS AUGUST 1993 VOL. 1
NO.
8
biased predictions when applied to a general popula-
tion of elderly persons.
DISCUSSION
In conducting these analyses, we were attempting to
find screening criteria that could be used to identify
groups of frail elders that, without intervention, would
experience frequent health-related crises and consume
a disproportionately high level of hospital resources.
The early identification of such elders would allow the
testing of interventions designed to prevent some of
the medical crises and to reduce the use of expensive
hospital services.
Eight screening criteria were identified. When tested,
these criteria classified
7 .2%
of a national sample of
non-institutionalized elders as being at high risk for
repeated hospital admission within 4 years. This high-
risk designation proved to be appropriate; during 4
years of follow-up, the high-risk group experienced
significantly higher rates of repeated admission and
mortality than did the low-risk group. The high-risk
group's hospital days and charges per person-year
survived were twice those
of
the low-risk group.
Our results suggest that these eight screening criteria
could be used to identify members of non-insti tution-
alized elderly populations who are likely to use hospital
resources frequently in the near future, many of whom
might benefit from preventive interventions. Those
identified could, therefore, be considered possible sub-
jects for future studies of the cost-effectiveness of
preventive interventions, such as geriatric evaluation
and management. Because this set of criteria was de-
rived from data collected after the introduction of
prospective payment for hospital care, it is applicable
to today's elderly community-dwelling population. It
offers a current, comprehensive, quantitative method
for identifying
US
elders at high risk of future hospital
admission.
~~
The results we report here do not appear to have
been biased by the exclusion of subjects whose Medi-
care hospitalization records were not available for
analysis. In most cases, the reason that the hospitali-
zation data were missing was because the subjects'
Social Security numbers, through which Medicare rec-
ords were to be traced, either were not provided or
were provided inaccurately (often by a proxy) during
the LSOA interviews (personal communication, M.
G .
Kovar, director of the LSOA). The excluded subjects
tended to be older, sicker, more functionally limited,
and more likely to be female than the subjects whose
hospitalization data were available. Nevertheless, the
relationships
between the eight risk factors and the
probability of repeated admission appear to be similar
among both the included and the excluded subjects.
Predictor Variables
Other investigators have pre-
viously identified as predictors of hospital admission
seven of the eight factors that we report here (all except
Having
he availability of an informal caregiver).',
11-14
an informal caregiver may predict the heavy use of
hospitals because sick people may be more likely to
have identified caregivers.
One factor that has been reported to predict overall
Medicare but is not included among our screen-
ing criteria, is limitation in the ability to perform the
activities of daily living (ADL) and the instrumental
activities of daily living (IADL).We evaluated the effect
of functional limitation on future hospital admission
by constructing a composite variable that reflected the
inability to perform without help one or more of nine
ADL and IADL (see list in footnote to Table 1). In a
univariate analysis, this variable was strongly associ-
ated with subsequent admission. In the multivariate
analyses, however, it was not significant (95%
CI =
0.8-1.8). Presumably, its univariate association with
repeated admission is explained by its collinearity with
other variables that were significant in the regression
model.
TABLE
3.
OBSERVED EXPERIENCES OF HIGH-RISK AND LOW-RISK GROUPS, 1985-88
Predicted to Have Repeated Admission
No (Low-Risk Yes (High-Risk
Group, n
=
204)
roup,
n
= 2623)
Observed Experiences
n
(%)
n
(%)
Repeated admission
584 (26.1) 85 (41.8)*
Mortality
498 (19.0) 90 44.2)*
Hospital charges per person-year survived
$1841 $3731
Hospital
days
per person-year survived
2.6 5 . 2
*
P < 0 0001
TABLE 4. EFFECTS OF USING ALTERNATE DEFINITIONS OF REPEATED ADMISSION
Hospital Days per
Predictors in Elders Classified as 1984-1988, High-
No. of Significant Proportion of Screened Person-Years
Definition Logistic Model High Risk Risk Elders
1+
Admission
in
2 years
8 3 .8 2 .3
2+
Admissions in
3
years
8 2.6 3.4
2+
Admissions in
4
years
(original definition)
3+
Admissions
in
4
vears
8
6
7.2
0
5.2
-
7/24/2019 Screening Elders for Risk of Hospital Admission (1993)
6/7
816
BOULT ET AL IAGS-AUGUST
1993 VOL.
41, NO 8
Test of Validity In order to test the predictive
validity of the eight risk factors for repeated admission
(two or more admissions in 4 years), we used 1984
interview data to compute, for subjects in the second
half of the sample, the probability of repeated admis-
sion during 1985-1988. The group identified as being
at high risk (Pra2 0.5) did go on to experience consid-
erably higher rates of repeated admission, mortality,
hospital days, and hospital charges than did the low-
risk group (see Table
3).
About half (48.1%)of the false
positives, ie, subjects for whom repeated admission
was predicted but not observed, were subjects who
died during 1985-1988. Of all the subjects who were
predicted to have repeated admission, 69.6% either
died or had repeated admission during the following
4
years. Thus, these screening criteria appear to be es-
pecially accurate in identifying elders at high risk for
one or the other of these undesirable fates. Its predic-
tive validity should now be re-tested on a different
population of non-institutionalized elders.
Use of the Screening Criteria The high-risk seg-
ment of the elderly population identified by our screen-
ing criteria went on to use hospital resources at twice
the rate of their peers. Many such high-risk elders,
except for those who have terminal diseases or unstable
medical conditions, are believed to benefit from pre-
ventive programs like geriatric evaluation and manage-
ment (GEM).24*5 They would, therefore, be appropri-
ate subjects for future research about the cost-effec-
tiveness of GEM and other preventive interventions.
The data required for computing elders probability
of repeated admission (Pra) in the manner described
above are minimal: simple responses to a short self-
administered questionnaire. (Survey materials are
available from the authors upon request.) Alternately,
such data could be extracted from clinical or adminis-
trative records, where available.
An investigator using these screening criteria to iden-
tify high-risk subjects for a study of an intervention
designed to reduce hospital admissions would need to
Spec i f ic i t y (%)
r
. . . . . . .
15
0 2 40 6 8 1
1
-
Spec i f ic i t y
(%)
False Posi t ive Rate)
FIGURE . ROC curve for Pra.The area under the ROC curve (SE)
= 61.0 (1.2) .26
TABLE 5. RELATIONSHIP BETWEEN THRESHOLD FOR HIGH RISK A N D HIGH-RISK ELD ER S OUTCOM ES
High Risk Elders Outcomes, 1985-1988
Hosp. Ch gs. perosp. Days per
Elders at
%
with 2+ Person-Year Person-Year Mor tality
P,, High Risk Adm . Survive d Survive d Rate
0.25 56.1 34.5 3.6 $2515 28.4
0.30 39.2 37.0 4.0 2799 32.3
0.35 27.9 37.9 4.2 2918 34.3
0.40 19.2 40.0 4.6
3127 36.4
0.45 12.7
41.0
4.8 3330 41.0
0.50 7.2 41.8 5.2 3731 44.2
0.55 4.3 39.6 5.7
3820 48.1
0.60 2.7 40.7 5.4
3889 48.3
-
7/24/2019 Screening Elders for Risk of Hospital Admission (1993)
7/7
SCREENING FOR RISK OF ADMISSION
817
AGS-AUGUST
1993 VOL. 41, NO.
8
establish a threshold value for
PI,
bove which elders
would be considered high-risk. Choosing a low thresh-
old would increase the sensitivity of the
PI,,
causing
more elders to be identified as high-risk, whereas
choosing a higher threshold would increase the speci-
ficity of the
PI,,
selecting fewer, but frailer, elders (see
Figure
2).
The threshold value must be chosen with care. As
illustrated in Table
5,
if
Pra= 0.3
were selected as the
threshold,
39.2%
of the population would be classified
as high-risk. The substantial costs of treating such a
large heterogeneous group would be difficult to recover
through reductions in the groups post-treatment use
of hospitals because, even without treatment, its hos-
pital use would be relatively low
( 2799
per person-
year).
At the other extreme, if
Pra =
0.60 were selected,
only 2.7% of the population would be classified as
high-risk. With such a small treatment group, a study
might lack the statistical power necessary to discern
differences in hospital use between the experimental
and control groups. Furthermore, the interventions per
capita costs could not be trimmed through the economy
of scale. Even if significant differences were observed,
the very small market for the intervention, ie, less than
3%
of elderly persons, would limit its potential use in
resolving the larger problem of frequent hospital use
by chronically ill elderly persons.
Future versions of the model that we have described
may be improved by the addition of other variables
(not measured in the LSOA) that reflect life style, use
of medications, psychological conditions, chronic use
of medical services (ie, longer than
1
year), and other
risk factors as yet unrecognized. The predictive validity
of such additional screening criteria should be tested
in the future. Nevertheless, the demonstrated predic-
tive validity of the eight criteria that we report suggests
that they could be used now to estimate elders risk of
repeated hospital admission.
REFERENCES
I.
Gruenberg
L,
Tompkins C, Porell
F.
The health status and utilization
patterns of the elderly: Implications for setting Medicare payments to
HMOs. Adv Health Econ Health Serv Res
1989;10:41-73.
2.
Zook CJ, Moore FD. High-cost users of medical care. N Engl J Med
1980;302:996-1002.
3. Anderson
G,
Knickman JR. Patterns of expenditures among high utilizers
of medical services. Med Care 1984;22:143-149.
4 . Freeborn DK, Pope CR, Mullooly JP et al. Consistently high users
of
medical care among the elderly. Med Care
1990;28:527-540.
5. Riley G, Lubitz 1 Prihoda R et al. Changes in distribution
of
Medicare
expenditures among aged enrollees,
1969-82.
Health Care Fin Rev
6.
Roos
NP, Shapiro E, Tate R. Does a small minority of elderly account for
a majority of health care expenditures?A sixteen year perspective. Milbank
Q
1990;67:347-369.
1986;753-63.
7.
Roos N, Shapiro E. The Manitoba longitudinal study of aging. Med Care
8.
McCall N, Wai HS. An analysis of the use
of
Medicare services by the
continuously enrolled aged. Med Care
1983;21:567-581.
9 . Anderson GF, Steinberg
El
Hospital readmissions in the Medicare pop-
ulation. N Engl
J
Med 1984;31:349-353.
10.
Aday LA, Andersen
R.
A framework
for
the study
of
access to medical
care. Health Serv Res 1974;9:208-220.
11. Branch L, Jette A, Evashwick C et al. Toward understanding elders health
service utilization.
J
Community Health 1981;780-92.
12.
Wolinsky FD, Coe RM. Physician and hospital utilization among non-
institutionalized elderly adults: An analysis of the health interview survey.
J
Gerontol
1984;39:334-341.
13. Beebe
1
Lubitz J, Eggers P. Using prior utilization to determine payments
of Medicare enrollees in health maintenance organizations. Health Care
Finan Rev
1985;6:27-38.
14. Roos NP, Roos LL, Mossey
J
et al. Using administrative data to predict
important health outcomes. Med Care 1988;26:221-239.
15.
Smith DM, Norton JA, McDonald CJ. Nonelective readmissions of medical
patients.
J
Chron Dis 1985;38:213-224.
16.
Anderson GF, Steinberg EP. Predicting hospital readmissions in the Med-
icare population. Inquiry
1985;22:251-258.
17. Gooding
J,
Jette A. Hospital admissions among the elderly. J Am Geriatr
SOC 985;33:595-601.
18. Fethke CC, Smith IM, Johnson N. Risk factors affecting readmission of
the elderly into the health care system. Med Care
1986;24:429-437.
19. Fitti JE, Kovar MG. The supplement on aging to the 1984 national health
interview survey. Vital and Health Statistics 1987, l (21) DHHS Pub. No.
(PHS) 87-1323. Washington DC: U S. Government Printing Office.
20.
SPSS
Inc. SPSS/PC+ Update for
V3.0
and
V3.1
for the IBM PC/XT/AT
and
PS/2.
Chicago: SPSS Inc,
1989.
21. Schlesselman JJ. Case Control Studies. New York: Oxford University Press,
1982.
22. Heckman J. Sample selection bias as a specification error. Econometrica
1979;47:153-161,.
23.
Thomas JW, Lichtenstein RL. Functional measures for adjusting health
maintenance organization capitation rates. Health Care Finan Rev
1985;785.
24. Rubenstein LZ,Goodwin M, Hadley E et al. Working group recommen-
dations: Targeting criteria for geriatric evaluation and management re-
search. J Am Geriatr SOC
991;39S:37S-41S.
25. NIH consensus development conference: Geriatric assessment methods
for
clinical decision-making. Am Geriatr SOC
988;36:342-347.
26. Centor RM. Signal detectability:The use of ROC curves and their analyses.
Med Decis Making 1991;11:102-106.
1981;19:644-657.
APPENDIX
Logistic Formula for Estimating the Probability of
Repeated Admission within Four Years
eBx
PI, =
1 + eBx
where:
P,
=
the probability of repeated admission within
four years
e
=
the natural logarithm
BX = Bo + BiXl + B2X2 + . . . . . .
.
+ B13X13
Bo = a constant
X = 1 or 0, the presence or absence of each risk
B
= the logistic regression coefficient of each risk
factor (including five dummy variables)
factor (including five dummy variables)