The Effects of a Health Promotion-Health Protection Intervention on

Behavior Change: The WellWorks Study

Glorian Sorensen, PhD, MPH, Anne Stoddard, ScD, Mary Kay Hunt, MPH, RD,James R. Hebert, MSPH, ScD, Judith K. Ockene, PhD, Jill Spitz Avrunin, MS,Jay Himmelstein, MD, MPH, and S. Katharine Hammond, PhD

Mounting evidence has underlined theimportance to cancer prevention efforts ofreducing lifestyle and environmental expo-sures.1-7 Work sites are an important venuefor efforts to reduce cancer risk throughhealth promotion and health protection ini-tiatives--'0: through work sites, it is possibleto influence the health-related behaviors oflarge numbers of people."1-14 Of particularimportance is the potential effect that worksite-based cancer control strategies mayhave on risk reduction among less educatedworkers and those in low-status jobs, amongwhom behavioral risk factors are particularlyhigh. Blue-collar workers are more likely tosmoke than are workers in white-collarjobs.15-17 Blue-collar workers are also morelikely than other workers to be exposed tohazards on the job.'8 Similarly, unhealthydietary habits are more prevalent amongthose with low education'9 or low income.20

Blue-collar workers are twice as likelyto have 2 or more lifestyle risk factors (e.g.,smoking, high-fat diets, and sedentarylifestyles),2' and workers reporting expo-sures to occupational hazards have highersmoking rates than workers without suchexposures.22 In addition, blue-collar workersare less likely to participate in work-sitehealth promotion programs than are white-collar workers.2327 When blue-collar work-ers do participate, these programs are lesslikely to result in health behavior change.27

This report focuses on an innovation inwork-site cancer prevention initiatives of par-ticular relevance for blue-collar workers: anintegrated program targeting both behavioralrisk factors and exposures to hazards on thejob.'0"8 For blue-collar workers, the tophealth priorities may not be the individualbehaviors addressed by work-site health pro-motion programs but rather those risks thatare involuntary, outside personal control, and

undetectable and that seem unfair.283' Indi-vidual health behaviors may fall within a"zone of nonacceptability" for managementactions (meaning that it would not be accept-able to workers for management to takeaction in these areas), while job-related healthand safety issues may be considered a too-often ignored responsibility ofmanagement.32Reduction of job risks may be required inorder to gain credibility and trust with theseworkers and to increase their receptivity tohealth education messages regarding theirown individual health behaviors.33'34

The purpose of this report is 2-fold.First, we present analyses designed to test theeffect of the integrated program on changesin 2 targeted behaviors related to cancer risk:dietary habits, including consumption of fat,fiber, and fruits and vegetables; and cigarettesmoking. Second, analyses are presented toassess whether the intervention effect differedby job category or exposure to occupationalhazards. We hypothesized that this integratedhealth promotion-health protection interven-tion would be more effective in producinghealth behavior changes among blue-collarworkers and workers exposed to occupational

Glorian Sorensen and Mary Kay Hunt are with theDana-Farber Cancer Institute and Harvard School ofPublic Health, Boston, Mass. Anne Stoddard and JillSpitz Avrunin are with the University of Massa-chusetts School of Public Health and HealthSciences, Amherst. James R. Hebert, Judith K.Ockene, and Jay Himmelstein are with the Universityof Massachusetts Medical School, Worcester. S.Katharine Hammond is with the University ofCalifornia School of Public Health, Berkeley.

Requests for reprints should be sent to GlorianSorensen, PhD, Population Sciences, Dana-FarberCancer Institute, 44 Binney Street, Boston, MA02115 (e-mail: glorian_sorensen@dfci.harvard.edu).

This paper was accepted April 22, 1998.

American Journal of Public Health 1685

Sorensen et al.

hazards than among other workers, keepingin mind that traditional health promotionprograms have been less effective with thesegroups.

Methods

Design

The WellWorks Study was conducted at1 of4 intervention research centers participat-ing in the Working Well Cooperative Agree-ment (See Acknowledgments).3s The Work-ing Well Trial was a randomized work-siteintervention trial testing the effectiveness ofhealth promotion interventions in 57 matchedpairs of work sites, using common elementsof study design, intervention methods, datacollection, and statistical analysis. All 4 studycenters targeted the nutritional outcomes, and3 of the 4 targeted smoking. Other targetedrisk factors varied across the 4 study centers,each of which had sufficient power to detectintervention effects independently.

Data for the study reported here werecollected as part of the WellWorks Study,conducted in 24 work sites located in eastemand central Massachusetts.34 After baselineassessments, work sites were matched into 12pairs on the basis of the presence of a cafete-ria, work-site size, type of smoking policy,company type, distribution by sex, distribu-tion of blue-collar and white-collar jobs, andresponse rate to the baseline survey in orderto assure comparability between groupsregarding these factors.2235 One work site ineach pair was then randomly assigned to theintervention condition and the other to thecontrol condition. Among the Working Wellintervention research centers, only the Well-Works Study assessed the effectiveness of amodel integrating health promotion andhealth protection. As in the other study sites,the primary outcomes for WellWorks weresmoking cessation, increased fiber consump-tion, decreased total fat consumption, andincreased fruit and vegetable consumption.35

Sample

The WellWorks Study recruited worksites from a Dun and Bradstreet listing basedon the following criteria: number of workers(250-2500), turnover rate (<20%), non-English-speaking employees (<20%), and useof known or suspected carcinogens in workprocesses. The types of businesses in the finalsample included manufacturers of industrial,chemical, and other products; textile dyeing;firefighting; and newspapers. All participatingwork sites agreed to be randomly assigned, toadminister employee and organizational sur-

veys, and to deliver the intervention ifassigned to the intervention condition.22 3

Intervention Methods

On the basis of a social ecologicalmodel, the WellWorks intervention targetedmultiple levels of influence.'4'36 This inter-vention model has been described previ-ously,34 and it included 3 key elements tar-geting health behavior change: (1) jointworker-management participation in pro-gram planning and implementation, opera-tionalized through an employee advisoryboard and a designated work-site liaison; (2)consultation by project staff with manage-ment on work-site environmental changes,including tobacco control policies, increasedavailability of healthy foods, and reductionin the potential for exposure to occupationalhazards; and (3) health education programstargeting individual behaviors in each of therisk factor areas.

Data Collection

A random sample of workers wasselected at each work site prior to the begin-ning of the intervention (baseline) and againfollowing completion of the intervention(final). Data were collected by means of aself-administered questionnaire distributedthrough work-site channels. Three reminderswere sent to nonrespondents, and incentiveswere provided for participation in the survey.In the 24 WellWorks sites at baseline, 9648surveys were mailed and 5914 completed sur-veys were received (overall completion ratewas 61%; range by work site, 360/o-99%). Atfinal, 8667 surveys were mailed and 5406completed surveys were received (overallcompletion rate was 62%; range, 430/o-92%).

The 2 samples were selected indepen-dently, but 2658 subjects responded to bothsurveys. This report addresses the changes inbehavior reported by the cohort of respon-ders to both the baseline and final surveys.Results of the analysis of the complete cross-sectional data for the Working Well Trial as awhole have been reported elsewhere.37

Measures

Diet was assessed by means of an 88-item semiquantitative food frequency ques-tionnaire that listed portion sizes (176 itemstotal).3' The percentage of energy from fatand grams of fiber per 1000 kilocalories werealso assessed by means of the questionnaire.Following the method used in the WorkingWell Trial as a whole, servings of fruits andvegetables were calculated on the basis of 2questions asking about usual intakes of fruit

(excluding juice) and vegetables (excludingpotatoes and salads), plus responses to itemsabout salad, potato, and fruit juice servings(weighted for serving size).40

The primary outcome for smoking was6-month abstinence, a reasonable approxima-tion of continuous, long-term cessation.41 42 Itwas measured by self-reported abstinence inthe 6 months prior to the final survey forsubjects who reported smoking at the base-line survey.

Self-reported exposure to workplacehazards was based on the response to thequestion, "Some substances used in worksettings may be harmful to your health. Inyour present job are you exposed to any ofthe following substances that may be harm-ful to your health?" Response optionsincluded chemicals (including solvents,cleansers, paints, dyes, oils, etc.); dusts(including metal, wood, etc.); gases, fumes,or vapors; pesticides; herbicides; and other.For the purposes of these analyses, workerswere considered exposed to workplace haz-ards if they reported exposure to any chemi-cals, dusts, gases, fumes, or vapors.

Selected characteristics of the subjectswere assessed, including age, sex, educa-tional level, race, ethnicity, marital status,and job category. Job category was derivedfrom responses to a question asking workersto choose the category that best representedtheir job. Respondents were grouped under 3categories. The first category, skilled andunskilled labor, comprised those whoanswered "skill or craft," "machine opera-tor," "manual labor," or "service work."Another category, office work, comprisedthose who chose scientific technical work orclerical, office, or sales work. The last cate-gory consisted of professional, managerial,and administrative work.

Data Analysis

The work site was the unit of random-ization and intervention, while the employeewas the unit of measurement. Among the2658 subjects in the cohort, 272 wereexcluded from the analysis because they hadout-of-range or missing dietary values oneither the baseline or final survey, leaving2386 subjects for analysis.

To evaluate the representativeness ofthe cohort, we compared characteristics ofsubjects in the cohort with those of subjectssurveyed at baseline but not at the final sur-vey; for this analysis, we used the Cochran-Mantel-Haenszel X2 or mixed-model analy-sis of variance, with work site included as arandom effect.

For testing the intervention effect, weused repeated-measures linear modeling

November 1998, Vol. 88, No. 111686 American Journal of Public Health

Results of the WellWorks Study

techniques where the repeat factor was sur-

vey (baseline vs final) and work site was

included as a random effect. The P value forthe F test ofno survey x intervention interac-tion was used to determine a significantintervention effect.

We then investigated the effectivenessof the intervention, controlling for sex, occu-

pational category, and self-reported exposure

to occupational hazards. To test whether a

characteristic influenced the effectiveness ofthe intervention, we included all 2-way and3-way interactions of the covariable, inter-vention, and survey. To obtain the most par-

simonious model, we removed higher-orderinteractions from the analysis if they were

not statistically significant (P <.05). Maineffects were removed only if the effect itselfand all interactions with that effect were notsignificant. The random work-site effect andthe fixed intervention and survey effectswere always retained because they are ele-ments of the design. The simplest hierarchi-cal models for each outcome are presented.

For continuous outcomes (percentage ofkilocalories from fat, grams of fiber per 1000kilocalories, and number of servings of fruitsand vegetables), we used the repeated-measures mixed-model analysis of vari-ance.43 44For binary outcomes (exposure tohazardous substances, smoking status, and6-month smoking abstinence), we used a

mixed-model logistic regression analysis.45 46

The distributions of the continuous out-comes were examined and scale transforma-tions were performed on several nonnormalcontinuous variables. For grams of fiber per

1000 kilocalories and daily servings of fruitsand vegetables, the natural logarithm of theoriginal value was used in the analyses. Foroutcomes analyzed in the log scale, changeis reported as a percentage increase or

decrease from the baseline value. For out-comes analyzed in the natural scale, changeis reported as a difference from baseline.

less likely to have a college degree (26% vs

30%) but more likely to have some college

(37% vs 32%). Other differences were toosmall to have practical significance.

We also examined the associationbetween self-reported exposures to job haz-ards and the smoking and nutritional out-comes at baseline among those included inthe cohort. Exposure was associated onlywith fiber intake. Unexposed workers hadgreater fiber intake in all job categories (7.86g vs 7.31 g per 1000 kcal, P<.001), but thedifference was greatest for office workers(8.30 g vs 7.13 g,P= .02 for the interaction).

Changes in Risk Factors byIntervention Group

Table 1 presents mean changes in fatintake as a percentage of kilocaloriesbetween baseline and final in the interven-tion and control groups. Those employed inintervention work sites reported significantlygreater reductions in fat consumption thandid those in the control condition. On aver-

age, workers in the intervention sites reducedtheir fat consumption 0.8 percentage pointmore than those in the control work site,reflecting a 2.2% difference in the level ofobserved change. The sex and job categorieswere significantly associated with fat intake,although controlling for these factors did notchange the observed intervention difference.Occupational exposure to harmful sub-stances was not associated with fat intake or

with the effectiveness of the intervention inreducing fat intake.

The increase in fiber consumptionapproached statistical significance whenonly work site was controlled, as shown inTable 2. There was, however, a significantjob category interaction with the interventioneffect. Changes in fiber intake were similarbetween the intervention and control condi-tions for office workers and for professionalsand managers, but these changes were sig-

nificantly greater among skilled andunskilled laborers in the intervention group

than in the control group. Skilled andunskilled laborers in the intervention group

increased their fiber consumption by 7 per-

centage points more than similar workers inthe control group, whereas for both officeworkers and professionals/managers therewas greater change in the control group thanin the intervention group. Sex of theemployee and exposure to hazards on the jobwere associated with fiber intake but did notmoderate the intervention effect.

Workers in the intervention sitesincreased their consumption of fruits andvegetables 0.23 servings per day, comparedwith 0.10 servings among workers in thecontrol sites, reflecting a 6% difference inservings per day, as shown in Table 3. Pro-fessional and managerial workers increasedtheir consumption of fruits and vegetablesmore than other workers did, although thisdifference was apparent for both interventionconditions. Controlling for sex of the workerand job category did not modify theobserved intervention effect.

Six-month smoking abstinence rateswere 15% in the intervention work sitescompared with 9% in the control work sites,controlling for work site (P = .123). Whenwe removed work site from the model, theodds ratio for the intervention effect was1.83 (P= .04). Because ofthe limited samplesize for baseline smokers, we added each ofthe control factors to the model one at a time,controlling for work site. Only job categorywas significantly associated with smoking.Although the intervention-by-job categoryinteraction was not significant (P=.18), thetrend is of interest. For skilled and unskilledlaborers, the 6-month abstinence rate was

twice as high in the intervention as in thecontrol condition (17.9% vs 9.0%). For theother 2 occupational categories, although dif-ferences were small, abstinence rates were

actually higher in the control work sites than

Results

Characteristics ofthe Sample

No significant differences by age, self-reported exposure to hazardous substances,or fiber consumption were observed betweenthose in the cohort and those respondingonly to the baseline survey. Compared withthose responding only at baseline, the cohorthad a higher percentage of men (76% vs

67%), a higher percentage of skilled andunskilled laborers (49% vs 43%), a higherpercentage who were married (76% vs

68%), and a lower smoking prevalence (23%vs 26%). Members of the cohort were also

American Journal of Public Health 1687

TABLE 1-Adjusted Mean Percentage of Kilocalories as Fat by Intervention andSurvey, Controlling for Work Site and for Significant Covariates: TheWellWorks Study, 1989-1994

InterventionCondition Survey Controlling for Work Site Multivariable Modela

Control Baseline 35.72b 35.54cFinal 34.20 33.95Difference -1.52 -1.55

Intervention Baseline 36.14 36.98Final 33.83 33.62Difference -2.31 -3.36

aControlling for work site, gender, and job category.bp= .007 for survey by intervention interaction.cP= .01 for survey x intervention interaction.

November 1998, Vol. 88, No. II

Sorensen et al.

in the intervention sites (for office workers, tion work-site intervention in changing5.1% vs 2.5%; for professionals and man- dietary habits and smoking. These data, fromagers, 18.6% vs 14.2%). a cohort of workers present throughout the

entire intervention period, indicate signifi-cant reductions in the percentage of calories

Discussion consumed as fat and an increase in servingsof fruits and vegetables. The intervention

These analyses assessed the effects of also had a significant effect on daily con-the first randomized, controlled study of an sumption of grams of fiber among skilledintegrated health promotion-health protec- and unskilled laborers. Although the 6% dif-

TABLE 3-Adjusted Geometric Mean Servings of Fruit and Vegetables byIntervention and Survey, Controlling for Work Site and forSignificant Covariates: The WellWorks Study, 1989-1994

Intervention Condition Survey Controlling for Work Site Multivariable Modela

Control Baseline, g 2.26b 2.31CFinal, g 2.36 2.40Change, % +4 +4

Intervention Baseline, g 2.29 2.34Final, g 2.52 2.56Change, % +10 +9

aControlling for work site, gender, and job category.bp= .03 for survey x intervention interaction.cP= .04 for survey x intervention interaction.

ference in 6-month smoking abstinence ratesbetween intervention and control work siteswas not statistically significant, the magni-tude of the effect was relatively large, and itappeared to be largest among skilled andunskilled workers.

Work-site health promotion programs inthe United States have generally emphasizedindividual behavior changes as a means ofreducing disease risk.'434 Noting that healthbehaviors of individual workers are only onepart of the equation of worker health, recentdiscussions have highlighted the importanceof the integration of health protection andhealth promotion efforts.'0'13'14'183447 Despitediffering historical roots and philosophicaltraditions, separate training, and competitionfor scarce resources, an array of opportunitiesexist for synergism between the 2 disciplinesof occupational health and health educa-tion.'013'18 By addressing concems within thebroader work-site environment, integratedhealth promotion-health protection programsare likely to contribute to an environmentmore supportive of general worker health,including health behavior changes.48-50

Prior reports have indicated that tradi-tional health promotion programs are mosteffective among white-collar workers.27 Toestimate the effectiveness of the addition ofthe health protection component, we hypoth-esized that this integrated intervention wouldbe particularly effective for blue-collar work-ers or for workers exposed to occupationalhazards. The intervention had differentialeffects by job category for 1 of the 3 dietaryoutcomes examined: for fiber, the interven-tion was more effective in producingchanges among skilled and unskilled labor-ers than among workers in other types ofjobs. Smoking abstinence rates increasedsubstantially among skilled and unskilledlaborers in the intervention group comparedwith the control group. Although the differ-ences by job category were not statisticallysignificant, the smoking abstinence ratesamong blue-collar workers in the interven-tion group were comparable to abstinencerates among professional and managerialworkers. In addition, we found no differencein intervention effectiveness by job categoryfor the other 2 outcomes, suggesting that thisintervention was at least successful inremoving the diminished intervention effectoften observed among blue-collar workers.

This report does not address the effec-tiveness of the health protection interventionin reducing potential exposures to hazards onthe job. Occupational hazards exposure ismore effectively controlled by changes in thework-site environment than by changes inindividual worker behaviors.5' Therefore,reductions in exposure are more appropriately

November 1998, Vol. 88, No. 11

1688 American Journal of Public Health

TABLE 2-Adjusted Geometric Mean Grams of Fiber per 1000 Kilocalories byIntervention Condition, Survey, and Job Category, Controlling forWork Site and for Significant Covariates: The WellWorks Study,1989-1 994

Intervention Condition Survey Controlling for Work Site Multivariable Modela

Control Baseline,g 4.48bFinal, g 7.87Change, % +5

Intervention Baseline, g 7.43Final, g 8.01Change, % +8

Interaction of job category x intervention condition x surveycSkilled/unskilled labor

Control Baseline, g 7.67Final, g 8.03Change, % +5

Intervention Baseline, g 7.42Final, g 8.31Change, % +12

Office workControl Baseline, g 7.55

Final, g 7.84Change, % +4

Intervention Baseline, g 7.77Final, g 7.88Change, % +1

Professional/managerialControl Baseline, g 8.01

Final, g 8.58Change, % +7%

Intervention Baseline, g 8.09Final, g 8.56Change, % +6

aControlling for work site, gender, self-reported exposure, and job category.bp= .08 for survey x intervention interaction.cp = .012.

assessed at the work-site level than throughindividual self-reports. Self-reported exposuremay provide a useful indicator of perceivedjob risk-an important construct for thehypotheses examined here-but it may not bea useful measure of the effectiveness of theintervention. If effective, the interventionmight either increase workers' awareness ofexposures in their work environment orreduce reported exposures as a consequenceof the introduction of environmental controls.Nonetheless, we tested for a differentialchange in self-reported exposure and foundno change in either group (data not shown).Effective measures to assess work site-levelchanges in the potential for exposure as wellas measures of individual workers' percep-tions of exposure potential are needed to trackchange in future studies.

Cohort analyses provide several advan-tages over cross-sectional surveys, includingassessment of change at the individual levelamong subjects who are exposed for the fullduration of the intervention period. In-migra-tion patterns may dilute the effect observedin the cross-sectional data because theemployees who were hired by the work siteafter initiation of the study may have experi-enced little or no exposure to the interven-tion. Data collected from repeated cross-sectional surveys also are subject to greatersampling variability than the estimates ofbehavioral change obtained from followingcohorts of individuals.52'53 In this study, sta-tistical power may have been reduced owingto the smaller sample size in the cohort.Therefore, despite the potential increasedprecision in the cohort design, the smallersample size was a barrier to detecting statisti-cally significant reductions in smoking, asfound in the cross-sectional analyses.35

This study had numerous strengths. Acommon intervention protocol was followedacross the 12 intervention work sites. Analy-ses focused on changes observed in a cohortof workers present for the full 2-year inter-vention period. However, the interpretationsof these results also must take into accountthe study's limitations. Although work siteswere randomized to the intervention or con-trol condition, work sites agreeing to partici-pate were not randomly selected for thisstudy. These work sites were selected on thebasis of specified eligibility criteria, includ-ing the use of known or suspected occupa-tional carcinogens, and on their willingnessto participate. Thus, the results can be gener-alized only to similar work sites that mayhave high readiness to provide programspromoting worker health. Members of thiscohort of workers differed significantly onseveral important variables from respondentsto the baseline survey only. Of particular

concern for these analyses was the fact thatthose in the cohort included a differentiallylow proportion of smokers and office work-ers. Our ability to assess the actual impact ofthe intervention on smoking cessation mayhave been hampered by these differences. Aswith other large trials, this study had no fea-sible alternative but to rely on self-reports forestimates ofthe intervention effect on behav-iors. Although contexts may differ from thatof this trial, the methods used here have beenvalidated in prior studies.54 60

Despite these limitations, this study rep-resents the first randomized controlled work-site intervention study to assess the effective-ness of an integrated health promotion-health protection intervention. Although thesize of the observed effects is modest, thepopulationwide impact of these effects mustbe considered.61-" For example, recent dataindicate that educational interventions toreduce serum cholesterol are reasonably costeffective if serum cholesterol is reduced byonly 2% or more.65 In this study, the levels ofbehavioral risk factor change among blue-collar workers ranged from 2% for fat con-sumption to 7% for fiber consumption. Weresuch changes to persist on a populationwidebasis, they would be likely to have a mean-ingful effect in terms of cancer-related out-comes as well as for coronary heart diseaseand other diseases.66

Increasingly, others have noted that theconceptualization of health promotion mustbe broadened from its current focus on indi-vidual behavior to consideration of theimpact of environmental influences and thesocial contexts of people's lives.50'6768 A par-ticipatory work-site program aimed at bothhealth promotion and health protectionmoves beyond an exclusive focus on indi-vidual health behavior change to address thecomplexities and politics of worker health asa concern for both employers and workersalike. The data presented here provide pre-liminary evidence from a randomized, con-trolled trial that such a program may beeffective in producing meaningful popula-tion changes in behavioral risk factors, par-ticularly among blue-collar workers. [ii

AcknowledgmentsThis study was supported by a cooperative agree-ment from the National Cancer Institute (grant UOICA5 1686) and from Liberty Mutual Group

The authors are indebted to other investigatorsand staff who participated in this project, includingElizabeth Fulgoni, Lynda Graham-Meho, ElizabethHarden, Jean Hsieh, Patricia Lavin, Ruth Palombo,Judy Phillips, Steven Potter, J. Ellen Thompson, andRichard Youngstrom. The authors also appreciate thecomments ofKaren Emmons from the Working WellTrial. This work would not have been possible with-out the commitment of the participating work sites.

Results of the WellWorks Study

The participating study centers and the respec-tive principal investigators are as follows: BrownUniversity School ofMedicine/The Miriam Hospital,David B. Abrams, PhD; University of Florida, JillVames, EdD; Dana-Farber Cancer Institute and TheUniversity of Massachusetts Medical School,Glorian Sorensen, PhD, MPH; University of TexasM.D. Anderson Cancer Center, Michael Eriksen,ScD (1989-1991), Bryant Boutwell, DrPH(1992-1993), Ellen Gritz, PhD (1993-onward); andFred Hutchinson Cancer Research Center(Coordinating Center), James Grizzle, PhD. TheProgram Director is Jerianne Heimendinger, ScD, ofthe Division of Cancer Prevention and Control,National Cancer Institute.

References1. Doll R, Peto R. The causes of cancer: quantita-

tive estimates of avoidable risk of cancer in theUnited States today. J Natl Cancer Inst. 1981;66:1191-1308.

2. US Dept of Health and Human Services.Strategies to Control Tobacco Use in theUnited States: A Blueprint for Public HealthAction in the 1990s. Bethesda, Md: NationalCancer Institute; 1991.

3. National Research Council. Diet and Health.Washington, DC: National Academy Press;1989.

4. Willett W. Vitamin A and lung cancer. NutrRev. 1990;48:201-211.

5. Ziegler R. Vegetables, fruits, and carotenoidsand the risk of cancer. Am J Clin Nutr. 1991;53(Suppl):251S-259S.

6. Steinmetz K, Potter J. Vegetables, fruit andcancer. I. Epidemiology. Cancer Causes andControl. 1991;2:325-357.

7. Willet W. Diet and health: what should we eat?Science. 1994;264:532-537.

8. Sorensen G, Glasgow R, Corbett K. Promotingsmoking control through worksites in the Com-munity Intervention Trial for Smoking Cessa-tion (COMMIT). Int Q Community HealthEduc. 1991;11:239-258.

9. Glasgow RE, Terborg JR, Hollis JF, SeversonHH, Boles SM. Take heart: results from the ini-tial phase of a work-site wellness program. AmJPublic Health. 1995;85:209-216.

10. Heaney C, Goldenhar L. Worksite health pro-grams: working together to advance employeehealth. Health Educ Q. 1996;23:133-136.

11. Klesges RC, Cigrant JA. Worksite SmokingCessation Programs: Clinical and Method-ological Issues. Newbury Park, Calif: SagePublications; 1988.

12. Abrams DB, Emmons KM, Linnan L, BienerL. Smoking cessation at the workplace: con-ceptual and practical considerations. In: Rich-mond R, ed. International Perspective onSmoking. Baltimore, Md; Williams andWilkins; 1994:137-170.

13. Baker E, Israel B, Schurman S. The integratedmodel: implications for worksite health promo-tion and occupational health and safety prac-tice. Health Educ Q. 1996;23:175-188.

14. Stokols D, Pelletier KR, Fielding JE. The ecol-ogy of work and health: research and policydirections for the promotion of employeehealth. Health Educ Q. 1996;23:137-158.

15. Covey LS, Zang SA, Wydner EL. Cigarettesmoking and occupational status: 1977 to 1990.Am JPublic Health. 1992;82:1230-1234.

November 1998, Vol. 88, No. 11 American Journal of Public Health 1689

Sorensen et al.

16. Sorensen G, Pechacek T. Occupational and sexdifferences in smoking and smoking cessation.J Occup Med. 1986;28:360-364.

17. Sterling T, Weinkam J. Smoking patterns byoccupation, industry, sex and race. Arch Envi-ron Health. 1978;33:313-317.

18. Walsh D, Jennings S, Mangione T, Merrigan D.Health promotion versus health protection?Employees' perceptions and concerns. J PublicHealth Policy. 1991; 12:148-164.

19. Kant A, Block G, Schatzkin A, Ziegler R, Nes-tle M. Dietary diversity in the US population,NHANES II, 1976-1980. J Am Diet Assoc.1991;91:1526-1531.

20. Patterson TL, Sallis JF, Nader PR, et al. Directobservation of physical activity and dietarybehaviors in a structured environment: effectsof a family-based health promotion program.JBehav Med. 1988;11:447-458.

21. Emmons K, Marcus B, Linnan L. Mechanismsin multiple risk factor interventions: smoking,physical activity, and dietary fat intake amongmanufacturing workers. Prev Med. 1994;23:481-489.

22. Sorensen G, Stoddard A, Hammond S. Doublejeopardy: job and personal risks for craftsper-sons and laborers. Am J Health Promotion.1996; 10:355-363.

23. Conrad P. Wellness in the work place: poten-tials and pitfalls of work-site health promotion.Milbank Q. 1987;65:255-275.

24. Glasgow R, McCaul K, Fisher K. Participationin worksite health promotion: a critique of thesource. Health Educ Q. 1993;20:391-408.

25. Sorensen G, Stoddard A, Ockene JK, HuntMK, Youngstrom R. Worker participation in anintegrated health promotion/health protectionprogram: results from the WellWorks Project.Health Educ Q. 1996;23:191-203.

26. Stunkard A, Cohen R, Felix M. Weight losscompetitions at the worksite: how they workand how well. Prev Med. 1989; 18:460-474.

27. Niknian M, Linnan L, Lasater T, Carleton R.Use of population-based data to assess risk fac-tor profiles of blue and white collar workers.JOccupMed. 1991 ;33:29-36.

28. Baker F. Risk Communication about environ-mental hazards. J Public Health Policy. 1990;11:341-359.

29. Fischhoff B, Bostrom A, Quadrel M. Risk per-ception and communication. Annu Rev Health.1993; 14:183-200.

30. Bradbury J. The policy implications of differ-ing concepts of risk. Sci Technol Hum Values.1989; 14:381-396.

31. Weinstein N. Precaution adoption process.Health Psychol. 1988;7:355-368.

32. Barnard C. The Functions of the Executive(30th Anniversary Edition). Cambridge, Mass:Harvard University Press; 1968.

33. Green KL. Issues of control and responsibilityin worker's health. Health Educ Q. 1988;15:473-486.

34. Sorensen G, Himmelstein J, Hunt M. A modelfor worksite cancer prevention: integration ofhealth protection and health promotion in theWellWorks Project. Am J Health Promotion.1995;10:55-62.

35. Abrams D, Boutwell W, Grizzle J,Heimendinger J, Sorensen G, Varnes J. Cancercontrol at the workplace: the Working WellTrial. Prev Med. 1994;23:1-13.

36. McLeroy KR, Bibeau D, Steckler A, Glanz K.An ecological perspective on health promotionprograms. Health Educ Q. 1988;15:351-377.

37. Sorensen G, Thompson B, Glanz K, et al. Worksite-based cancer prevention: primary resultsfrom the Working Well Trial. Am J PublicHealth. 1996;86:939-947.

38. Kristal AR, Shattuck AL, William AE. FoodFrequency Questionnaires for Diet Interven-tion Research. Washington, DC: InternationalLife Sciences Institute; 1994:110-125.

39. Glanz K, Patterson RE, Kristal AR, et al.Stages of change in adopting healthy diets: fat,fiber, and correlates of nutrient intake. HealthEduc Q. 1994;21:499-519.

40. Block G. Health habits and history question-naire: diet history and other risk factors.Bethesda, Md: National Cancer Institute; 1988.

41. COMMIT. Community Intervention Trial forSmoking Cessation (COMMIT): summary ofdesign and intervention. J Nat Cancer Inst.1991;83:1620-1628.

42. Cohen S, Lichtenstein E, Prochaska J. Debunk-ing myths about self-quitting. Am Psychol.1989;44:1135-1365.

43. Jennrich R, Schuchter M. Unbalancedrepeated-measures models with structuredcovariance matrices. Biometrics. 1986;42:805-820.

44. Schall R. Estimation in generalized linear mod-els with random effects. Biometrika. 1991;78:719-727.

45. Liang K, Zeger S. Longitudinal data analysisusing generalized linear models. Biometrika.1986;73: 13-22.

46. Wolfinger R. Laplace's approximation for non-linear mixed models. Biometrika. 1993;80:791-795.

47. Sorensen G, Himmelstein J. Worksite interven-tions promoting reduction of coronary heartdisease risk. In: Ockene I, Ockene J, eds. ThePrevention of Coronary Heart Disease: ASkills-Based Approach. Boston, Mass: LittleBrown & Co; 1992.

48. Green L, Richard L, Potrin L. Ecological foun-dations of health promotion. Am J Health Pro-motion. 1996;10:270-281.

49. Stokols D. Translating social ecological theoryinto guidelines for community health promotion.Am JHealth Promotion. 1996;10:282-295.

50. Breslow L. Social ecological strategies for pro-moting healthy lifestyles. Am J Health Promo-tion. 1996;10:253-256.

51. Preventing Illness and Injury in the Workplace.Washington, DC: Office of Technology Assess-ment; 1985.

52. Gail M, Byar D, Pechacek T, Corle D. Aspectsof statistical design for the Community Inter-vention Trial for Smoking Cessation (COM-MIT). Control Clin Trials. 1992; 13:6-21.

53. Gail M, Byar D, Pechacek T, Corle D. Aspectsof statistical design for the Community Inter-vention Trial for Smoking Cessation (COM-MIT). Control Clin Trials. 1993; 14:253-254.

54. Block G, Hartman A. Dietary Assessment Meth-ods. New York, NY: Marcel Dekker; 1989.

55. Block G, Woods M, Potssky A, Clifford C. Val-idation of a self-administered diet history ques-tionnaire using multiple diet records. J ClinEpidemiol. 1990;41:1327-1335.

56. Sobell J, Block G, Koslowe P, Tobin J, AndresR. Validation of a retrospective questionnaireassessing diet 10- 15 years ago. Am J Epi-demiol. 1989; 130:173-187.

57. Block G, Thompson F, Hartman A, Larkin F,Guire K. Comparison of two dietary question-naires validated against multiple dietaryrecords collected during a 1 year period.JAMA. 1992;92:686-693.

58. Coates R, Eley J, Block G. An evaluation of afood frequency questionnaire for assessingdietary intake of specific carotenoids and vita-min E among low-income black women. Am JEpidemiol. 1991;134:658-671.

59. Mares-Perlman J, Klein R. A diet history ques-tionnaire ranks nutrient intakes in middle-agedand older men and women similarly to multiplefood record. JNutr Ed. 1993; 123:489-501.

60. Velicer W, Prochaska J, Rossi J, Snow M.Assessing outcome in smoking cessation stud-ies. Psychol Bull. 1992;l 11 :23-4 1.

61. Susser M. Editorial: the tribulations of trials-interventions in communities. Am J PublicHealth. 1995;85:156-158.

62. Fisher EBJ. Editorial: the results of the COM-MIT trial. Am J Public Health. 1995;85:159-160.

63. Rose G. The Strategy of Preventive Medicine.Oxford, England: Oxford University Press;1995.

64. McKinley JB. The promotion of health throughplanned sociopolitical change: challenges forresearch and policy. Soc Sci Med. 1992;36:109-117.

65. Tosteson ANA, Weinstein MC, Hunink MGM,et al. Cost-effectiveness of populationwideeducational approaches to reduce serum cho-lesterol levels. Circulation. 1997;95:24-30.

66. Ockene I, Ockene J. Prevention of CoronaryHeart Disease. Boston, Mass: Little, Brown &Co; 1992.

67. Ottawa Charter for Health Promotion. FirstInternational Conference on Health Promotion:New York; Oxford University Press; 1986.

68. Terris M. Concepts of health promotion: duali-ties in public health theory. J Public HealthPolicy. 1991;12:267-275.

November 1998, Vol. 88, No. 111690 American Journal of Public Health