Interventions to increase physical activity and healthy eating among overweight and obese children...
Transcript of Interventions to increase physical activity and healthy eating among overweight and obese children...
Interventions to Increase Physical Activity among HealthyAdults: Meta-Analysis of Outcomes
Vicki S. Conn, PhD, RN, FAAN, Adam R. Hafdahl, PhD, and David R. Mehr, MD, MS
AbstractObjective—This meta-analysis summarized effects of interventions designed to increase physicalactivity among healthy adults.
Methods—Comprehensive searching located 358 reports eligible for inclusion. Random-effectsanalyses were used to synthesize data. Potential moderator variables were examined with meta-analytic analogues of ANOVA and regression. Moderator variable robustness and publication biaswere explored.
Results—Meta-analytic results were computed from studies including 99,011 subjects. Theoverall mean effect size (d) for treatment vs. control groups comparisons was 0.19 (higher meanfor treatment than control subjects). This 0.19 is consistent with a mean difference of 496 steps/day between treatment and control subjects. Exploratory moderator analyses suggested that thecharacteristics of the most effective interventions included behavioral interventions instead ofcognitive interventions, face-to-face delivery versus mediated interventions (e.g. via telephone,mail, etc.), and targeted individuals instead of communities. Participant characteristics wereunrelated to physical activity effect sizes. Substantial between-studies heterogeneity remainedbeyond individual moderators.
Conclusions—These findings suggest that interventions designed to increase activity aremodestly effective. Findings suggest interventions should emphasize behavioral strategies overcognitive strategies to increase physical activity.
Adequate physical activity (PA) is linked with important health outcomes, including reducedcardiovascular disease,1 type 2 diabetes,2–3 some cancers,4–5 falls,6 osteoporotic fractures,7and depression,8 as well as improved physical function,9–11 weight management,12–15
cognitive function,16–17 and quality of life.18 Despite this compelling evidence, healthyadults commonly get inadequate PA.19
Extensive primary research has tested interventions to increase PA. Although many meta-analyses have addressed health outcomes of PA, few have examined PA behavior outcomes.The seminal 1996 meta-analysis of interventions to increase PA behavior reported amoderate effect size across 127 studies of healthy and chronically ill adults and children.20
Their moderator analyses documented larger effect sizes when interventions used behaviormodification, had face-to-face delivery vs. mediated delivery (e.g. telephone), focused on
Corresponding author: Vicki S. Conn, PhD, RN, FAAN, S317 School of Nursing, University of Missouri, Columbia, MO 65211,[email protected], 573 882 0231 (office), 573 884 4544 (fax).Human participant protection: The Institutional Review Board for the Protection of Human Subjects categorized the project asexempt from requiring individual subject consent.Author contributions: Dr. Conn conceived the project, developed methods, supervised study conduct, led interpretation of findings,wrote the first draft of the manuscript, and is the principal investigator of the grant which funded the work. Dr. Hafdahl participated indeveloping the research protocol, analyzed data, facilitated interpretation of findings, and actively participated in manuscriptdevelopment. Dr. Mehr participated in conceptual discussions of project development and design, he also actively participated inmanuscript development and revision.
NIH Public AccessAuthor ManuscriptAm J Public Health. Author manuscript; available in PMC 2012 April 1.
Published in final edited form as:Am J Public Health. 2011 April ; 101(4): 751–758. doi:10.2105/AJPH.2010.194381.
NIH
-PA Author Manuscript
NIH
-PA Author Manuscript
NIH
-PA Author Manuscript
healthy people, measured active leisure vs. structured exercise, measured low-intensityactivity, encouraged unsupervised PA vs. supervised PA, targeted subjects of diverse ages,and targeted groups vs. individuals. A recent comprehensive meta-analysis of worksiteprograms for healthy adults documented a d effect size (standardized mean difference) of0.21 but did not conduct moderator analyses to determine the intervention characteristicslinked with the largest PA increases.21 Other meta-analyses have integrated acrosschronically ill adults22 or focused on small specific interventions or populations, such asprimary-care-based referrals to PA programs,23 older adults,24 computer-basedinterventions,25 or environmental interventions.26 Many meta-analyses have been plaguedby small samples that hinder moderator analyses.27–28 For example, only 19 studies wereincluded in the most recent Cochrane review which aggregated randomized controlled trialswith at least six-month follow-up data following interventions to increase PA amongsedentary adults.27 Given the importance of PA and the proliferation of studies testinginterventions to increase PA, we sought to move this area of science forward by conductinga comprehensive meta-analysis to estimate the overall effect of interventions and moreimportantly to conduct moderator analyses to identify intervention characteristics associatedwith the best outcomes. The project addressed these questions: (1) What are the overalleffects of interventions designed to increase PA on PA behavior after completion ofinterventions? (2) Do interventions’ effects on PA behavior vary depending on intervention,methodology, or sample characteristics?
METHODSSearch Strategies
We used multiple comprehensive search strategies to avoid the bias resulting from narrowsearches.29–30 An expert reference librarian conducted searches in 13 databases (e.g.MEDLINE, Dissertation Abstracts, SCOPUS). The National Institutes of Health ComputerRetrieval of Information on Scientific Projects was examined for potential studies. Thirty-six research registers were searched.31 Broad terms ensured comprehensive searches.Ancestry searches were completed for review papers and eligible studies. Computerizeddatabase searching was conducted for senior authors and principal investigators of alleligible studies. Our staff hand-searched 82 journals from 1960 through 2007.31 Thisextensive searching yielded 54,642 papers to consider for inclusion.
Inclusion CriteriaWe included English-language reports of interventions to increase PA among healthy adults.PA was defined as any bodily movement that increased energy expenditure beyond basallevels. Diverse PA behavior change interventions were eligible (e.g. education sessions,supervised exercise practice sessions) if PA was measured separately from the intervention.To reduce biases, we included both published and unpublished studies.32–33 We alsoincluded small-sample studies and pre-experimental studies.
Data ExtractionA coding frame was developed from related meta-analyses, review articles, and extensiveexamination of primary studies.34 This captured studies’ results as well as characteristics ofthe source, participants, method, and intervention. Participant characteristics coded includedage, gender, overweight status, previous exercise, and minority status. Methodologicalfeatures coded were method of assigning subjects, attrition, PA measure, and intervalbetween intervention and PA assessment.
We coded 74 intervention characteristics. These included intervention social context(individual, group); social structure target (individual, community); theoretical framework
Conn et al. Page 2
Am J Public Health. Author manuscript; available in PMC 2012 April 1.
NIH
-PA Author Manuscript
NIH
-PA Author Manuscript
NIH
-PA Author Manuscript
for intervention (Social Cognitive Theory, Transtheoretical Model, other theories werereported too infrequently for analyses); behavioral target (PA only vs. PA plus other healthbehaviors); recommended PA (form, intensity, duration/session, frequency/week); andexercise session characteristics among studies with supervised PA (form, intensity, duration/session, frequency/week, total number of sessions). Specific intervention content wasrecorded: access enhancement, barriers management, competition, contracting,consequences/rewards, cues/stimulus control, decision making, education about the healthbenefits of PA, exercise prescription, feedback, goal setting, modeling, monitoring PAbehavior by research staff, motivational interviewing, problem solving, relapse preventioneducation, and self-monitoring. The presence of individually tailored interventions—intervention with specific content matched to individual subjects’ attributes such as itemsidentified as personal barriers to PA—was coded. Interventions that used a ‘train-the-trainer’(i.e. teach PA behavior change interventions to local community members or health careproviders so they can deliver the interventions to individuals) approach were noted. Wecoded special intervention targets including entire communities, worksites, and ambulatoryhealth care settings and we coded the mode of delivery (e.g. face-to-face, mass media,mediated by telephone, mail, email). Thirty studies were pilot coded.
Data were coded at a micro-level to enhance validity.35 To enhance reliability, twoextensively trained coders independently extracted all data from each report. All data werecompared between coders to achieve 100% agreement. A third coder verified effect-sizedata. Discrepancies were resolved by consulting the first author or another member of theresearch team. We extracted data on 564 pairwise comparisons from 358 reports.
Data AnalysisFour types of effect-size (ES) comparisons were calculated.36 Treatment vs. control post-intervention ESs refer to treatment group results compared to control group results afterinterventions. Treatment vs. control pre-post ESs were calculated as a comparison betweentreatment group pre-post ES and control group pre-post ES. Treatment pre-post ESs arewithin-group ESs calculated for studies which provided treatment group baseline andoutcome data. Control pre-post comparisons are the same but for control subjects. Astandardized mean difference (d) ES was calculated for each primary study comparison.Positive d reflects more favorable scores for the treatment group or at post-test. The fourtypes of comparisons were analyzed separately.
Main Analyses of Treatment vs. Control Data—Treatment vs. control post-intervention ESs were calculated as the treatment post-test mean minus the control grouppost-test mean, divided by the pooled post-test SD. A second two-group ES, treatment vs.control pre-post ES, was generated to address possible treatment changes from time-relatedeffects in addition to the intervention effect (e.g., maturation, testing, regression). These ESswere calculated as treatment group pre-post ES minus control group pre-post ES, whereeach pre-post ES was computed as post-test mean minus pre-test mean, divided by pre-testSD. ESs were adjusted for bias.37 Larger samples were given more influence in the analysisby weighting each ES by the inverse of its sampling variance (i.e., precision). Homogeneitywas assessed using a conventional heterogeneity statistic (Q) and I2, which is an index ofbetween-studies heterogeneity relative to within-study sampling error used to assess theimpact of (in)consistency among trials on meta-analytic results. Studies with two or moretreatment groups compared to a single group were included in the meta-analysis byaccounting for the dependence caused by the shared control group: A two-stage approachwas used where each study’s dependent ESs were combined into a single independent ES38
then submitted to standard univariate random-effects analysis. Estimates of mean PA ESswere converted to the original metrics of steps/day and minutes/week. To detect possible
Conn et al. Page 3
Am J Public Health. Author manuscript; available in PMC 2012 April 1.
NIH
-PA Author Manuscript
NIH
-PA Author Manuscript
NIH
-PA Author Manuscript
publication bias, we used multiple statistical procedures since all strategies havelimitations.39–43 Outliers were detected statistically, omitting each ES one at a time andchecking for large externally standardized residuals or substantially reduced measures ofheterogeneity.
Estimated ESs we report here are based on the random-effects model with the betweenstudies variance component, , estimated by weighted method of moments, unlessotherwise designated. The random-effects model assumes individual ESs vary as a result ofboth subject-level sampling error and sources of study-level error.44 The random-effectsmodel is appropriate when study implementation is heterogeneous. Inclusion criteriavariations, intervention differences, dose variations, and study execution differencescontribute to heterogeneity.45–47 We expect heterogeneity in behavior change research andused four strategies to manage it. First, findings of the random-effects model, whichassumes heterogeneity, are presented. Second, both a location parameter and variabilityparameter are reported. Third, potential study-level moderators were explored to understandsources of heterogeneity. Fourth, findings were interpreted in light of heterogeneity. Thesestrategies are important because they help us interpret the extent to which heterogeneityaffects meta-analysis conclusions.45
Analyses of Alternative Designs—Treatment group pre-post comparisons includedstudies designed as single-group projects, those with multiple treatment groups and nocontrol group, and studies designed as treatment vs. control comparisons that also providedpre-intervention data so treatment pre-post comparisons were possible. Control pre-postcomparisons were calculated from baseline and outcome control group scores. Each single-group ES was calculated as the post-minus pre-intervention mean divided by the baselineSD. Correlations between pre- and post-intervention scores were solicited from primarystudy authors to calculate sampling variances. Adjustment for bias, sample size weighting,detection of outliers, random-effects models, estimation of publication bias, and assessmentof heterogeneity as described above were applied to these single-group ESs. Findings fromsingle-group comparisons are presented as ancillary evidence to the more internally validtwo-group comparisons. Within-group control subject findings are presented as empiricalevidence to address the common concern that control subjects may experience some benefitfrom participation in study procedures.
Moderator Analyses—Exploratory moderator analyses were conducted with treatmentvs. control post-intervention comparisons. A mixed-effects meta-analytic analogue ofregression was used for moderator analyses. This incorporates between-studiesheterogeneity into the estimate and test of the moderator’s relationship with mean ES. Forcontinuous moderators this estimates and tests the (unstandardized) regression slope, β. Fordichotomous moderators, a special case of regression estimates and tests the differencebetween two mean ESs. The moderator’s effect is tested with a heterogeneity statistic, eitherfor the model (QModel) or between groups (QBetween). These moderator analyses should beinterpreted as hypothesis generating, given the lack of consistent previous findings to form afirm basis for hypothesis testing. Moderator pairs were analyzed to address a givenmoderator’s robustness and generalizability in the presence of other moderators bydetermining how much its effect changed when each other moderator was controlled andhow much it interacted with each other moderator. Robustness was summarized as excellent,good, mixed, mediocre, and poor, based on the extent of agreement in rankings ofinteraction significance, interaction size, and size of change in main effect.
Conn et al. Page 4
Am J Public Health. Author manuscript; available in PMC 2012 April 1.
NIH
-PA Author Manuscript
NIH
-PA Author Manuscript
NIH
-PA Author Manuscript
RESULTSPrimary Study Characteristics
We calculated ESs from about 99,011 subjects’ data. Treatment vs. control post-interventionanalyses included 74,852 subjects (206 comparisons). Treatment pre-post comparisons useddata from 43,701 subjects (498 comparisons). Study characteristics are described in Table 1.The median of mean age was 44 years. Median sample size was 72 participants (range 5 to17,579). Women were well-represented with a median of 74% females, but the median forminority participants was only 14% among studies that reported such data. Interventionsranged from a single motivational education session to extensive supervised exercisesessions occurring over many weeks. The median duration of supervised exercise was 45minutes. The median number of sessions was 27 supervised exercise encounters.Motivational interventions’ median duration was 60 minutes delivered in a median of 5sessions.
Effect of InterventionsTable 2 shows the effects of interventions on PA behavior outcomes. We found a mean ES(d) estimate of 0.19 for both treatment vs. control post-intervention comparisons and fortreatment vs. control pre-post comparisons. A mean ES (d) of 0.33 was documented fortreatment pre-post comparisons. These ESs indicate that, on average, interventions didincrease the overall PA after completion of the intervention. In contrast, control subjects didnot experience increased PA by participating in studies, as evidenced by a mean ES of .00(d). Findings from heterogeneity analyses (Q and I2) suggest substantial variation in true ESamong studies. The two-group comparison mean ES of 0.19 is consistent with a meandifference of 14.7 minutes/week of PA or 496 steps/day between the treatment and controlgroups. If we assume true ESs are normally distributed with mean 0.19 and SD 0.17 (seeTable 2), then the middle 95% of true ESs fall between −0.14 and 0.53. Expressing thisinterval in an original metric gives (−11.0, 40.3) min/wk or (−371, 1363) steps/day. That is,for instance, a randomly selected study’s true mean difference could plausibly range from 11min/week less for treatment subjects to 40 min/week more for treatment subjects. Evidencesuggested possible publication bias among ESs for treatment vs. control post-intervention,treatment vs. control pre-post, and treatment pre-post ESs. No publication bias was apparentfor control pre-post ESs.
Moderator AnalysesTables 3 and 4 present the results of dichotomous and continuous moderator analyses oftreatment vs. control post-intervention ESs. Tabled results from the analyses of multiple-dfcategorical moderators and moderator pairs are available from the first author. Analyseswith fewer studies providing information on a characteristic (smaller k) should beinterpreted more cautiously than comparisons with more comparisons (larger k). Moderatoranalyses should be considered exploratory.
Report, Sample, and Methodological Moderators—Neither publication nor fundingstatus was related to PA ESs (see QB in Table 3). Studies published more recently had largermean ESs (Qmodel in Table 4). Although the dichotomous moderator analyses suggestedstudies of samples who exercised prior to the intervention reported lower ES (0.14) thanstudies of sedentary samples (0.27), these findings were not robust in joint moderatoranalyses. ESs were unrelated to sample characteristics (e.g. age), to random versusnonrandom assignment, or to method of measuring PA. The attrition difference betweentreatment and control subjects was related to ES in both the individual moderator analysesand the joint moderator analyses to assess robustness of findings. Specifically, studies withsmaller treatment-group attrition rates as compared to control-group attrition rates reported
Conn et al. Page 5
Am J Public Health. Author manuscript; available in PMC 2012 April 1.
NIH
-PA Author Manuscript
NIH
-PA Author Manuscript
NIH
-PA Author Manuscript
larger ESs. The days between intervention and outcome measurement was unrelated to ES inthe continuous moderator analyses (Table 4) or in the joint moderator analyses.
Intervention Moderators—When considered individually, 13 of the dichotomousmoderators tested were associated with differences in PA outcomes (Table 3). Studies thatdid not use Social Cognitive Theory reported significantly larger ES (0.20) than studies thatdid (0.12). Studies without the Transtheoretical Model reported larger ES (0.21) than studieswith the model (0.15). These findings regarding better outcomes among studies withoutSocial Cognitive Theory or without the Transtheoretical model were robust in the jointmoderator analyses. A comparison of ES between studies using the two models did notreveal a statistically significant difference. Multiple-df analyses documented the largest ESfor studies using neither model (0.23). The pattern of findings suggested Social CognitiveTheory was more detrimental to ES values than the Transtheoretical Model.
Although the dichotomous moderator analyses suggested studies including exerciseprescription reported larger ESs for PA (0.30) than studies without prescription (0.17), thesefindings were not robust. The presence of supervised exercise in the intervention wasassociated with better PA outcomes (0.29 vs. 0.17) in dichotomous moderator analyses, butthis was not supported in joint analyses. The joint moderator analyses revealed mixedsupport for the finding that studies where research staff modeled exercise behavior wereassociated with larger ESs (0.38) than studies without modeling (0.17).
The joint analyses supported the finding that interventions which included a train-the-trainerapproach were less effective (0.09) than interventions with research staff providinginterventions directly to subjects (0.21). The finding that standardized interventions (0.20)were more effective than individually tailored interventions (0.04) received mixed support inthe joint moderator analyses. The dichotomous moderator finding that interventionsincluding relapse-prevention strategies (0.34) were more effective than interventions without(0.17) was not confirmed in the joint analyses.
Both the dichotomous and joint moderator analyses confirmed that interventions whichtargeted entire communities (0.09) were less effective than interventions aimed atindividuals (0.19). The finding that studies with mass media approaches (0.08) were lesseffective than studies using other strategies to increase PA (0.19) was confirmed in the jointanalyses. Interventions with mediated delivery of interventions (e.g. mail, telephone), hadsmaller ESs (0.15) than interventions that were delivered face-to-face (0.29) in the singlevariable analysis. The joint moderator analyses did not confirm the better ES for face-to-faceinterventions. Worksite- and primary-care-based interventions (0.21) did not report differentESs compared to interventions without these characteristics (0.18).
We grouped interventions into approaches that were either behavioral (e.g. goal setting,contracting, self-monitoring, cues, rewards) or cognitive (e.g. decision making, healtheducation, providing information). Interventions that exclusively used behavioral strategies(0.25) were more effective than other interventions (0.17). Multiple-df analyses confirmedthat the largest ESs were for interventions that focused entirely on behavioral interventions.The joint moderator analyses confirmed that the superiority of behavioral approaches was arobust finding.
Interventions often recommended PA form, intensity, or duration following interventions.None of these were significantly linked with ESs for PA. Neither the number of interventionstrategies nor the total minutes of intervention content (including total minutes of supervisedPA) was associated with PA outcomes.
Conn et al. Page 6
Am J Public Health. Author manuscript; available in PMC 2012 April 1.
NIH
-PA Author Manuscript
NIH
-PA Author Manuscript
NIH
-PA Author Manuscript
DISCUSSIONThis comprehensive meta-analysis found a moderate mean ES (d=0.19) across diversestudies designed to increase PA among healthy adults. Moderator analyses identified severalrobust and moderately robust ES moderators associated with larger PA ES: behavioralinterventions (vs. cognitive interventions which target knowledge, attitudes, beliefs),interventions delivered directly to individuals (vs. mass media interventions andinterventions targeting entire communities), interventions delivered by project staff (vs.train-the-trainer models), modeling PA behavior by research staff, standardizedinterventions (vs. individually tailored interventions), and absence of interventions based onSocial Cognitive Theory or the Transtheoretical Model.
The ES from these studies of healthy adults is smaller than the ES reported for chronicallyill adults22 (d=0.45) and the ES for chronically ill and healthy adults and children(d=0.72).20 These results are similar to the ES reported for older adults (d=0.26)24 in whichinterventions targeting specific disease patient populations elicited larger PA behaviorchanges than interventions not targeting such groups. The presence of chronic illness maycause patients to be more responsive to interventions. Our smaller ES than that reported byDishman20 could have resulted from our greatly expanded search strategies locating moreobscure studies with smaller ESs.
The magnitude of PA behavior change is modest. The achieved steps per day do not meetpublic health goals of 10,000 steps per day.48–49 It is unclear if entirely sedentary peoplegain incremental health benefits when they add even small amounts of PA. Futureintervention research should report outcomes in terms of understandable amounts of PAincreases such as steps per day or minutes per week.
The moderator analysis finding that behavioral strategies are superior to cognitive strategiesis consistent with meta-analytic findings for chronically ill adults22 and older adults.24
Behavioral strategies include goal setting, self-monitoring, PA behavior feedback,consequences, exercise prescription, and cues. Health care providers and public healthprograms often emphasize PA’s health benefits, but we found that health education did notincrease ES. Perhaps the public already is convinced of PA’s health benefits, so programsusing behavioral strategies to change PA behavior may be most effective. Future researchcomparing behavioral interventions to cognitive interventions in large randomizedcontrolled trials would help confirm these findings. Further primary research comparingspecific types of behavioral interventions (e.g. contracting, self-monitoring, cues, rewards)could identify the most effective behavioral intervention components. Public health workersdesigning interventions should emphasize behavioral strategies over cognitive approaches.
The pattern of findings across mediated delivery (e.g. delivered via email, telephone), massmedia (e.g. delivered via television or newspaper), social structure target (individual vs.community), and train-the-trainer (i.e. teach PA behavior change interventions to localcommunity members or health care providers so they can deliver the interventions toindividuals) approaches suggest that delivering interventions face-to-face to individuals ismost effective. This finding moves beyond trends suggested in previous meta-analyses thatdid not achieve statistical significance.22,24 Audience attention to the message may behigher in individually delivered face-to-face interventions and may seem more important torecipients. Moreover, behavioral interventions may be easier to deliver in face-to-faceindividual encounters.
This meta-analysis was limited by the studies retrieved and the available information instudy reports. Primary study quality varies widely. Many quality aspects, such as treatmentfidelity and allocation concealment, are poorly reported and could not be examined in
Conn et al. Page 7
Am J Public Health. Author manuscript; available in PMC 2012 April 1.
NIH
-PA Author Manuscript
NIH
-PA Author Manuscript
NIH
-PA Author Manuscript
moderator analyses. Publication bias suggests studies with negative or low ESs remaininaccessible. Meta-analyses are unable to assess publication bias favoring studies withparticular characteristics. The comprehensive nature of this meta-analysis, and resultingheterogeneity among studies, is both a strength and limitation. The overall effect size shouldbe interpreted in light of discovered heterogeneity; not all interventions are equally effective.The results of the moderator analyses should be used to interpret findings: These effect sizecomparisons may be more important than the overall ES. The moderator analyses findingsshould be interpreted in the context of associations among moderators, substantial residualbetween-studies heterogeneity, and lack of hypotheses from the literature.
In conclusion, our comprehensive meta-analysis found that PA interventions producemoderate statistically significant increases in PA behavior and that behavioral interventionsappear more effective than cognitive interventions. These findings suggest interventions toincrease PA should emphasize behavioral components such as self-monitoring, stimuli toincrease PA, rewards, behavioral goal setting, and modeling PA behavior in standardizedinterventions delivered to individuals. Future research should explore which components ofbehavioral interventions are most effective.
AcknowledgmentsFinancial support provided by a grant from the National Institutes of Health (R01NR009656) to Vicki Conn,principal investigator. The content is solely the responsibility of the authors and does not necessarily represent theofficial views of the National Institutes of Health.
References1. Hamer M, Chida Y. Walking and primary prevention: A meta-analysis of prospective cohort studies.
Br J Sports Med. 2008; 42:238–243. [PubMed: 18048441]2. Roumen C, Blaak EE, Corpeleijn E. Lifestyle intervention for prevention of diabetes: Determinants
of success for future implementation. Nutr Rev. 2009; 67:132–146. [PubMed: 19239628]3. Orozco LJ, et al. Exercise or exercise and diet for preventing type 2 diabetes mellitus. Cochrane
Database of Systematic Reviews. 2008:CD003054.4. Friedenreich CM, Cust AE. Physical activity and breast cancer risk: Impact of timing, type and dose
of activity and population subgroup effects. Br J Sports Med. 2008; 42:636–647. [PubMed:18487249]
5. Tardon A, et al. Leisure-time physical activity and lung cancer: A meta-analysis. Cancer CausesControl. 2005; 16:389–397. [PubMed: 15953981]
6. Sherrington C, et al. Effective exercise for the prevention of falls: A systematic review and meta-analysis. J Am Geriatr Soc. 2008; 56:2234–2243. [PubMed: 19093923]
7. Moayyeri A. The association between physical activity and osteoporotic fractures: A review of theevidence and implications for future research. Ann Epidemiol. 2008; 18:827–835. [PubMed:18809340]
8. Teychenne M, Ball K, Salmon J. Physical activity and likelihood of depression in adults: A review.Prev Med. 2008; 46:397–411. [PubMed: 18289655]
9. Manini TM, Pahor M. Physical activity and maintaining physical function in older adults. Br JSports Med. 2009; 43:28–31. [PubMed: 18927164]
10. Mian OS, et al. The impact of physical training on locomotor function in older people. Sports Med.2007; 37:683–701. [PubMed: 17645371]
11. Chin APMJ, et al. The functional effects of physical exercise training in frail older people: Asystematic review. Sports Med. 2008; 38:781–793. [PubMed: 18712944]
12. Keller C, et al. Interventions for weight management in postpartum women. J Obstet GynecolNeonatal Nurs. 2008; 37:71–79.
13. Seo DC, et al. A meta-analysis of psycho-behavioral obesity interventions among us multiethnicand minority adults. Prev Med. 2008; 47:573–582. [PubMed: 18201758]
Conn et al. Page 8
Am J Public Health. Author manuscript; available in PMC 2012 April 1.
NIH
-PA Author Manuscript
NIH
-PA Author Manuscript
NIH
-PA Author Manuscript
14. Shaw K, et al. Exercise for overweight or obesity. Cochrane Database Syst Rev. 2006:CD003817.[PubMed: 17054187]
15. Ohkawara K, et al. A dose-response relation between aerobic exercise and visceral fat reduction:Systematic review of clinical trials. Int J Obes (Lond). 2007; 31:1786–1797. [PubMed: 17637702]
16. Erickson KI, Kramer AF. Aerobic exercise effects on cognitive and neural plasticity in olderadults. Br J Sports Med. 2009; 43:22–24. [PubMed: 18927158]
17. Liu-Ambrose T, Donaldson MG. Exercise and cognition in older adults: Is there a role forresistance training programmes? Br J Sports Med. 2009; 43:25–27. [PubMed: 19019904]
18. Bize R, Johnson JA, Plotnikoff RC. Physical activity level and health-related quality of life in thegeneral adult population: A systematic review. Prev Med. 2007; 45:401–415. [PubMed:17707498]
19. Centers for Disease Control and Prevention. US Physical activity statistics. 2010. 2010 [cited 2010January 5, 2010]; Available from: http://apps.nccd.cdc.gov/PASurveillance/StateSumV.asp
20. Dishman RK, Buckworth J. Increasing physical activity: A quantitative synthesis. Med Sci SportsExerc. 1996; 28:706–719. [PubMed: 8784759]
21. Conn V, et al. Meta-analysis of workplace physical activity interventions. Am J Prev Med. 2009;37:330–339. [PubMed: 19765506]
22. Conn VS, et al. Meta-analysis of patient education interventions to increase physical activityamong chronically ill adults. Patient Educ Couns. 2008; 70:157–172. [PubMed: 18023128]
23. Williams NH, et al. Effectiveness of exercise-referral schemes to promote physical activity inadults: Systematic review. Br J Gen Pract. 2007; 57:979–986. [PubMed: 18252074]
24. Conn VS, Valentine JC, Cooper HM. Interventions to increase physical activity among agingadults: A meta-analysis. Ann Behav Med. 2002; 24:190–200. [PubMed: 12173676]
25. Kroeze W, et al. A systematic review of randomized trials on the effectiveness of computer-tailored education on physical activity and dietary behaviors. Ann Behav Med. 2006; 31:205–223.[PubMed: 16700634]
26. Matson-Koffman DM, et al. A site-specific literature review of policy and environmentalinterventions that promote physical activity and nutrition for cardiovascular health: What works?Am J Health Promot. 2005; 19:167–193. [PubMed: 15693346]
27. Foster C, Hillsdon M, Thorogood M. Interventions for promoting physical activity. CochraneDatabase of Systematic Reviews. 2005:CD003180.
28. Kuoppala J, et al. Work health promotion, job well-being, and sickness absences--a systematicreview and meta-analysis. J Occup Environ Med. 2008; 50:1216–1227. [PubMed: 19001948]
29. Cooper, H.; Hedges, L.; Valentine, J., editors. The handbook of research synthesis and meta-analysis. 2. Russell Sage Foundation; New York: 2009. p. 615
30. White, H. Scientific communication and literature retrieval. In: Cooper, H.; Hedges, L.; Valentine,J., editors. The handbook of research synthesis and meta-analysis. Russell Sage Foundation; NewYork: 2009. p. 51-71.
31. Reed, J.; Baxter, P. Using reference databases. In: Cooper, H.; Hedges, L.; Valentine, J., editors.The handbook of research synthesis and meta-analysis. Russell Sage Foundation; New York:2009. p. 73-101.
32. Cook DJ, et al. Should unpublished data be included in meta-analyses? Current convictions andcontroversies. JAMA. 1993; 269:2749–2753. [PubMed: 8492400]
33. Rothstein, HR.; Hopewell, S. Grey literature. In: Cooper, H.; Hedges, L.; Valentine, J., editors. Thehandbook of research synthesis and meta-analysis. Russell Sage Foundation; New York: 2009. p.103-125.
34. Lipsey, M. Identifying interesting variables and analysis opportunities. In: Cooper, H.; Hedges, L.;Valentine, J., editors. The handbook of research synthesis and meta-analysis. Russell SageFoundation; New York: 2009. p. 147-158.
35. Orwin, R.; Vevea, J. Evaluating coding decisions. In: Cooper, H.; Hedges, L.; Valentine, J.,editors. The handbook of research synthesis and meta-analysis. Russell Sage Foundation; NewYork: 2009. p. 177-203.
Conn et al. Page 9
Am J Public Health. Author manuscript; available in PMC 2012 April 1.
NIH
-PA Author Manuscript
NIH
-PA Author Manuscript
NIH
-PA Author Manuscript
36. Morris SB, DeShon RP. Combining effect size estimates in meta-analysis with repeated measuresand independent-groups designs. Psychological Methods. 2002; 7:105–125. [PubMed: 11928886]
37. Hedges, L.; Olkin, I. Statistical methods for meta-analysis. Orlando, FL: Academic Press; 1985.38. Gleser, LJ.; Olkin, I. Stochastically dependent effect sizes. In: Cooper, H.; Hedges, L.; Valentine,
J., editors. The handbook of research synthesis and meta-analysis. Russell Sage Foundation; NewYork: 2009. p. 357-376.
39. Sutton, AJ. Publicaton bias. In: Cooper, H.; Hedges, L.; Valentine, J., editors. The handbook ofresearch synthesis and meta-analysis. Russell Sage Foundation; New York: 2009. p. 435-452.
40. Gleser LJ, Olkin Ipb. Modelsfor estimating the number of unpublished studies. Stat Med. 1996;15:2493–2507. [PubMed: 8961459]
41. Rosenthal R. The file drawer problem and tolerance for null results. Psychol Bull. 1979; 86:638–641.
42. Sterne JA, Egger Mpb. Funnel plots for detecting bias in meta-analysis: Guidelines on choice ofaxis. J Clin Epidemiol. 2001; 54:1046–1055. [PubMed: 11576817]
43. Vevea JL, Hedges LV. A general linear model for estimating effect size in the presence ofpublication bias. Psychometrika. 1995; 60:419–435.
44. Raudenbush, S. Analyzing effect sizes: Random-effect models. In: Cooper, H.; Hedges, L.;Valentine, J., editors. The handbook of research synthesis and meta-analysis. Russell SageFoundation; New York: 2009. p. 295-315.
45. Higgins JP, et al. Measuring inconsistency in meta-analyses. BMJ. 2003; 327:557–560. [PubMed:12958120]
46. Colditz GA, Burdick E, Mosteller F. Heterogeneity in meta-analysis of data from epidemiologicstudies: A commentary. Am J Epidemiol. 1995; 142:371–382. [PubMed: 7625401]
47. Berlin JA. Invited commentary: Benefits of heterogeneity in meta-analysis of data fromepidemiologic studies. Am J Epidemiol. 1995; 142:383–387. [PubMed: 7625402]
48. Le Masurier GC, et al. Accumulating 10,000 steps: Does this meet current physical activityguidelines? Res Q Exerc Sport. 2003; 74:389–394. [PubMed: 14768840]
49. Tudor-Locke C, et al. How many steps/day are enough? Preliminary pedometer indices for publichealth. Sports Med. 2004; 34:1–8. [PubMed: 14715035]
Conn et al. Page 10
Am J Public Health. Author manuscript; available in PMC 2012 April 1.
NIH
-PA Author Manuscript
NIH
-PA Author Manuscript
NIH
-PA Author Manuscript
NIH
-PA Author Manuscript
NIH
-PA Author Manuscript
NIH
-PA Author Manuscript
Conn et al. Page 11
Tabl
e 1
Cha
ract
eris
tics o
f Prim
ary
Stud
ies I
nclu
ded
in P
A M
eta-
Ana
lyse
s
Cha
ract
eris
tick
Min
Q1
Mdn
Q3
Max
Mea
n ag
e (y
ears
)16
418
4044
5288
Sam
ple
size
per
stud
y35
85
3372
207
1757
9
Perc
enta
ge a
ttriti
on fr
om c
ompa
rison
gro
up82
05
1222
77
Perc
enta
ge a
ttriti
on fr
om tr
eatm
ent g
roup
153
06
1629
73
Perc
enta
ge a
ttriti
on fr
om to
tal s
ampl
ea85
05
1322
62
Perc
enta
ge fe
mal
e20
70
5674
100
100
Perc
enta
ge m
inor
ity88
07
1490
100
Min
utes
of s
uper
vise
d ex
erci
se p
er se
ssio
n25
1830
4560
60
Tota
l num
ber o
f sup
ervi
sed
exer
cise
sess
ions
286
1627
4815
6
Min
utes
of e
duca
tion/
mot
ivat
ion
per s
essi
on60
324
6069
120
Tota
l num
ber o
f edu
catio
nal/m
otiv
atio
nal s
essi
ons
165
11
512
59
Num
ber o
f wee
ks in
terv
entio
n w
as d
eliv
ered
205
01
1026
313
Not
e. In
clud
es a
ll pr
imar
y st
udie
s tha
t con
tribu
ted
one
effe
ct si
ze fo
r any
type
of c
ompa
rison
. Ind
epen
dent
sam
ples
with
in st
udie
s agg
rega
ted
by su
mm
ing
sam
ple
size
s and
usi
ng w
eigh
ted
mea
n of
oth
erch
arac
teris
tics (
wei
ghte
d by
sam
ple
size
). k
= nu
mbe
r of s
tudi
es p
rovi
ding
dat
a on
cha
ract
eris
tic; Q
1 =
first
qua
rtile
, Q3
= th
ird q
uarti
le.
a Incl
udes
onl
y st
udie
s with
bot
h tre
atm
ent a
nd c
ompa
rison
sam
ples
.
Am J Public Health. Author manuscript; available in PMC 2012 April 1.
NIH
-PA Author Manuscript
NIH
-PA Author Manuscript
NIH
-PA Author Manuscript
Conn et al. Page 12
TAB
LE 2
Ran
dom
-Eff
ects
Phy
sica
l Act
ivity
Beh
avio
r Out
com
e Es
timat
es a
nd T
ests
ES
type
kμ̂ δ
p (μ̂
δ)SE
(μ̂δ)
μ δ 9
5% C
Iσ̂ δ
Qp
(Q)
I2
Trea
tmen
t vs.
cont
rol p
ost-i
nter
vent
ion
com
paris
ona
206
0.19
<.00
10.
019
(0.1
5, 0
.23)
0.17
554.
4<.
001
.67
Trea
tmen
t vs.
cont
rol p
re-p
ost c
ompa
rison
b14
60.
19<.
001
0.02
1(0
.15,
0.2
3)0.
1529
9.7
<.00
1.5
2
Trea
tmen
t pre
-pos
t com
paris
on49
80.
33<.
001
0.01
4(0
.30,
0.3
5)0.
2629
45.3
<.00
1.8
3
Con
trol p
re-p
ost c
ompa
rison
115
0.00
.792
0.01
7(−
0.04
, 0.0
3)0.
1228
1.2
<.00
1.5
9
Not
e. T
reat
men
t vs.
cont
rol p
ost-i
nter
vent
ion
com
paris
on re
fers
to tr
eatm
ent p
ost-t
est m
ean
min
us c
ontro
l pos
t-tes
t mea
n. T
reat
men
t vs.
cont
rol p
re-p
ost c
ompa
rison
refe
rs to
trea
tmen
t gro
up p
re-p
ost E
Sm
inus
con
trol g
roup
pre
-pos
t ES.
Tre
atm
ent p
re-p
ost c
ompa
rison
ESs
refe
rs to
trea
tmen
t gro
up p
ost-t
est m
ean
min
us b
asel
ine
mea
n. C
ontro
l pre
-pos
t com
paris
on E
S re
fers
to c
ontro
l gro
up p
ost-t
est m
ean
min
us b
asel
ine
mea
n. k
=num
ber o
f com
paris
ons. μ̂ δ
= e
stim
ated
mea
n ES
. . Q
= h
eter
ogen
eity
stat
istic
. Und
er h
omog
enei
ty (H
0: δ
i = δ
) Q is
dist
ribut
ed a
s chi
-squ
are
with
df =
k −
1, w
here
k is
the
num
ber o
f (po
ssib
ly d
epen
dent
) obs
erve
d ef
fect
size
s; th
is a
lso
test
s H0:
. I
2 =
quan
tific
atio
n of
impa
ct o
f het
erog
enei
ty. P
oten
tial o
utlie
rsex
clud
ed b
ased
on
stan
dard
ized
rand
om-e
ffec
ts re
sidu
als f
or tr
eatm
ent v
s. co
ntro
l pos
t-int
erve
ntio
n (1
4 [6
% o
f com
paris
ons)
, tre
atm
ent v
s. co
ntro
l pre
-pos
t (7
[5%
of c
ompa
rison
s]),
treat
men
t pre
-pos
t (43
[8%
of c
ompa
rison
s]),
and
cont
rol p
re-p
ost (
13 [1
0% o
f com
paris
ons]
).
a Acc
omm
odat
ing
mul
tiple
-trea
tmen
t dep
ende
nce
due
to 2
3, 6
, and
1 m
ultip
le-tr
eatm
ent p
airs
, trip
lets
, and
qua
drup
lets
, res
pect
ivel
y.
b Igno
ring
depe
nden
ce d
ue to
18
and
3 m
ultip
le-tr
eatm
ent p
airs
and
trip
lets
resp
ectiv
ely.
Am J Public Health. Author manuscript; available in PMC 2012 April 1.
NIH
-PA Author Manuscript
NIH
-PA Author Manuscript
NIH
-PA Author Manuscript
Conn et al. Page 13
TAB
LE 3
Trea
tmen
t vs.
Con
trol P
ost-I
nter
vent
ion
Com
paris
ons:
Dic
hoto
mou
s Mod
erat
ors,
Mix
ed-E
ffec
ts A
naly
ses
Mod
erat
ork 0
k 1μ̂ δ 0
μ̂ δ1
SEdi
fQ
Bσ̂ δ
aI2
Rob
ustn
ess
Sour
ce c
hara
cter
istic
s
Publ
icat
ion
stat
us25
179
0.17
0.19
0.06
80.
10.
17.6
4
Fund
ing
7712
90.
210.
170.
038
1.1
0.17
.63
Parti
cipa
nt c
hara
cter
istic
s
Prev
ious
exe
rcis
ers
8811
80.
270.
140.
036
13.3
***
0.16
.60
poor
Mos
tly o
verw
eigh
t22
580.
210.
180.
062
0.2
0.15
.51
Res
earc
h m
etho
ds c
hara
cter
istic
s
Ran
dom
ass
ignm
ent
8512
10.
160.
210.
036
1.9
0.17
.63
Obj
ectiv
e vs
. sel
f-re
porte
d PA
mea
sure
4416
20.
230.
180.
050
1.1
0.17
.63
Epis
odic
vs.
over
all P
A m
easu
reb
1581
0.08
0.18
0.05
92.
50.
13.4
6
Post
-inte
rven
tion
lag
befo
re m
easu
re15
104
0.39
0.18
0.08
36.
3*0.
14.4
5po
or
Inte
rven
tion
char
acte
ristic
s
Soci
al c
ogni
tive
theo
ry (S
CT)
169
370.
200.
120.
045
3.5†
0.17
.63
exce
llent
Tran
sthe
oret
ical
mod
el (T
M)
146
600.
210.
150.
038
2.8†
0.17
.64
good
SCT
vs. T
M41
180.
140.
080.
065
0.9
0.14
.54
poor
Acc
ess e
nhan
cem
ent
193
130.
190.
100.
070
1.6
0.17
.64
Bar
riers
man
agem
ent
157
490.
190.
160.
042
0.7
0.17
.64
Com
petit
ion/
cont
ests
198
80.
180.
240.
089
0.5
0.17
.63
Con
tract
ing
197
90.
190.
080.
082
1.7
0.17
.64
Con
sequ
ence
165
410.
180.
200.
046
0.2
0.17
.63
Dec
isio
n m
akin
g19
511
0.19
0.15
0.08
00.
20.
17.6
4
Exer
cise
pre
scrip
tion
176
300.
170.
300.
053
6.2*
0.17
.63
poor
Feed
back
165
410.
190.
180.
043
0.0
0.17
.63
Fitn
ess t
estin
g19
97
0.18
0.30
0.10
61.
30.
17.6
3
Goa
l set
ting
151
550.
170.
210.
039
1.2
0.16
.61
Hea
lth e
duca
tion
144
620.
190.
170.
039
0.6
0.17
.63
Am J Public Health. Author manuscript; available in PMC 2012 April 1.
NIH
-PA Author Manuscript
NIH
-PA Author Manuscript
NIH
-PA Author Manuscript
Conn et al. Page 14
Mod
erat
ork 0
k 1μ̂ δ 0
μ̂ δ1
SEdi
fQ
Bσ̂ δ
aI2
Rob
ustn
ess
Mas
s med
ia19
511
0.19
0.08
0.05
83.
8†0.
16.5
8go
od
Mod
elin
g18
521
0.17
0.38
0.06
411
.4**
*0.
16.5
9m
ixed
Mon
itorin
g17
036
0.17
0.24
0.04
91.
90.
17.6
3
Mot
ivat
iona
l int
ervi
ewin
g19
79
0.18
0.20
0.08
40.
00.
17.6
3
Prob
lem
solv
ing
184
220.
180.
200.
057
0.1
0.17
.63
Rel
apse
pre
vent
ion
190
160.
170.
340.
066
6.4*
0.17
.62
poor
Self-
mon
itorin
g13
769
0.18
0.20
0.03
90.
30.
17.6
4
Stim
ulus
con
trol
191
150.
190.
140.
067
0.6
0.17
.64
Supe
rvis
ed e
xerc
ise
172
340.
170.
290.
055
4.8*
0.17
.62
poor
Stan
dard
ized
vs.
indi
vidu
ally
tailo
red
196
100.
200.
040.
071
4.6*
0.17
.63
mix
ed
Targ
eted
174
320.
190.
170.
046
0.1
0.17
.64
Trai
n-th
e-tra
iner
app
roac
h17
333
0.21
0.09
0.04
56.
9**
0.17
.63
good
Any
beh
avio
ral i
nter
vent
ion
6514
10.
180.
190.
038
0.1
0.17
.62
exce
llent
Beh
avio
ral i
nter
vent
ions
onl
y15
155
0.17
0.25
0.04
34.
0*0.
17.6
3m
ixed
Beh
avio
ral o
nly
vs. w
/cog
nitiv
e86
550.
160.
250.
045
4.3*
0.16
.53
Any
cog
nitiv
e in
terv
entio
n10
010
60.
220.
160.
036
2.5
0.17
.63
Cog
nitiv
e in
terv
entio
ns o
nly
186
200.
190.
170.
059
0.1
0.17
.64
Cog
nitiv
e on
ly v
s. w
/beh
avio
ral
8620
0.15
0.17
0.05
70.
00.
15.5
3m
ixed
Mul
tiple
beh
avio
rs v
s. PA
onl
y ta
rget
115
870.
200.
170.
037
1.0
0.17
.63
Indi
vidu
al v
s. gr
oup
soci
al c
onte
xt84
122
0.21
0.17
0.03
71.
70.
17.6
2
Targ
ets i
ndiv
idua
ls v
s. co
mm
uniti
es19
412
0.19
0.09
0.05
73.
2†0.
16.5
8go
od
Med
iate
d de
liver
y (e
.g. t
elep
hone
)71
135
0.29
0.15
0.04
111
.7**
*0.
16.6
1po
or
Wor
ksite
pro
gram
157
490.
180.
210.
043
0.8
0.17
.63
Link
ed to
prim
ary
care
173
330.
190.
160.
048
0.3
0.17
.64
Rec
omm
end
spec
ific
PA12
185
0.18
0.20
0.03
70.
40.
17.6
3
Rec
omm
end
wal
king
for P
A14
838
0.17
0.23
0.04
81.
20.
18.6
6
Rec
omm
end
PA in
tens
ity15
490.
220.
180.
076
0.2
0.17
.61
Am J Public Health. Author manuscript; available in PMC 2012 April 1.
NIH
-PA Author Manuscript
NIH
-PA Author Manuscript
NIH
-PA Author Manuscript
Conn et al. Page 15N
ote.
k0
= nu
mbe
r of E
S es
timat
es w
ith th
e m
oder
ator
var
iabl
e ab
sent
. k1
= nu
mbe
r of E
S es
timat
es w
ith th
e m
oder
ator
var
iabl
e pr
esen
t. μ̂ δ
0=mea
n ES
for s
tudi
es w
ithou
t int
erve
ntio
n co
mpo
nent
. μ̂δ 1
=mea
n ES
for s
tudi
es w
ith in
terv
entio
n co
mpo
nent
pre
sent
. QB
= b
etw
een
grou
ps h
eter
ogen
eity
stat
istic
(dis
tribu
ted
as c
hi-s
quar
e on
df =
1 u
nder
H0:
μδ 0
=μδ 1
).
. I2
= qu
antif
icat
ion
of im
pact
of h
eter
ogen
eity
. Ana
lysi
s rep
orte
d if
k 0 ≥
3 a
nd k
1 ≥
3. R
obus
tnes
s ass
esse
d in
join
t mod
erat
or a
naly
ses.
† p <
.10,
* p <
.05,
**p
< .0
1,
*** p
< .0
01 (f
or Q
B).
a For a
ll ta
bled
mod
erat
ors a
test
of H
0:
yie
lded
p <
.000
01.
b Excl
udes
com
paris
ons b
ased
on
fitne
ss m
easu
res (≥
6 m
onth
s afte
r sup
ervi
sed
PA tr
eate
d as
PA
beh
avio
r out
com
e).
Am J Public Health. Author manuscript; available in PMC 2012 April 1.
NIH
-PA Author Manuscript
NIH
-PA Author Manuscript
NIH
-PA Author Manuscript
Conn et al. Page 16
TAB
LE 4
Line
ar a
nd C
ubic
Mix
ed-E
ffec
ts M
oder
ator
Ana
lyse
s for
PA
Beh
avio
r Out
com
es fo
r Tre
atm
ent v
s. C
ontro
l Pos
t-Int
erve
ntio
n C
ompa
rison
s
Mod
erat
ork
MSD
Lin
ear β̂
Lin
ear
I2
Qm
odel
σ̂ δ
LC
LC
Publ
icat
ion
year
(−19
60)
206
39.4
16.
26−0.008
.63
9.9*
*10
.8*
0.17
0.17
Parti
cipa
nt c
hara
cter
istic
s
Mea
n sa
mpl
e ag
e13
647
.31
10.5
50.
002
.48
1.2
3.0
0.15
0.15
Prop
ortio
n fe
mal
e16
30.
600.
250.
027
.57
0.2
1.4
0.16
0.16
Prop
ortio
n et
hnic
min
ority
520.
250.
310.
030
.58
0.1
1.2
0.18
0.18
Mea
n ye
ars o
f edu
catio
n13
12.8
91.
490.
117
.54
2.5
3.3
0.25
0.30
Prop
ortio
n at
tritio
n, C
ontro
l13
00.
170.
150.
138
.47
1.0
2.9
0.14
0.14
Prop
ortio
n at
tritio
n, T
reat
men
t13
10.
180.
14−0.131
.46
0.9
1.2
0.13
0.14
Prop
ortio
n at
tritio
n, a
ll12
60.
170.
140.
041
.47
0.1
1.6
0.14
0.14
Diff
eren
ce in
pro
p. a
ttriti
on12
60.
000.
09−0.583
.46
6.6*
*8.
9*0.
140.
14
Log
odds
ratio
of a
ttriti
on96
0.01
0.72
−0.071
.49
4.5*
6.2
0.14
0.14
Log 1
0 day
s sin
ce in
terv
entio
n10
42.
090.
61−0.016
.44
0.1
4.6
0.14
0.14
Inte
rven
tion
char
acte
ristic
s
Num
ber b
ehav
iora
l stra
tegi
es20
61.
041.
340.
012
.62
0.8
3.1
0.17
0.17
Num
ber c
ogni
tive
stra
tegi
es20
60.
700.
93−0.004
.63
0.0
4.6
0.17
0.17
Num
ber s
trate
gies
206
3.17
3.07
0.00
5.6
30.
81.
90.
170.
17
Log 1
0 rec
. min
utes
/wk
462.
180.
13−0.183
.60
0.4
1.3
0.19
0.20
Log 1
0 rec
. sup
ervi
sed
PA27
3.25
0.51
0.13
3.5
40.
72.
00.
280.
28
Log 1
0 rec
. mot
ivat
e/ed
ucat
e50
1.97
0.72
−0.002
.57
0.0
4.3
0.18
0.19
Log 1
0 rec
. tot
al m
inut
es59
2.01
0.76
0.04
3.5
50.
84.
70.
190.
19
Not
e. k
= n
umbe
r of (
poss
ibly
dep
ende
nt) E
S es
timat
es. E
ach
mod
erat
or’s
wei
ghte
d m
ean
(M) a
nd st
anda
rd d
evia
tion
(SD
) com
pute
d fr
om a
ll av
aila
ble
case
s. M
oder
ator
cen
tere
d at
M. P
olyn
omia
l mod
els
with
deg
ree
m =
1 o
r 3: L
= li
near
, C =
cub
ic. H
eter
ogen
eity
stat
istic
s: Q
tota
l = to
tal,
igno
ring
mod
erat
or; Q
mod
el =
due
to a
ll po
lyno
mia
l ter
ms o
f mod
erat
or (x
) in
linea
r (β 1
x) o
r cub
ic (β
1x +
β2x
2 +
β 3x3
) mod
el, d
istri
bute
d as
chi
-squ
are
on d
f = m
und
er H
0: β
= 0
, whe
re β
is β
1, o
r [β 1β 2β 3
]T, r
espe
ctiv
ely,
for l
inea
r or c
ubic
mod
el. I
2 =
quan
tific
atio
n of
deg
ree
of h
eter
ogen
eity
. Ana
lysi
s rep
orte
d if
k≥
m +
5.
† p <
.10,
Am J Public Health. Author manuscript; available in PMC 2012 April 1.