A Meta-analysis of Interventions to Improve Care for Chronic Illnesses

August 1, 2005
Alexander C. Tsai, PhD

,
Sally C. Morton, PhD

,
Carol M. Mangione, MD, MSPH

,
Emmett B. Keeler, PhD

Volume 11, Issue 8

Objective: To use empirical data from previously published literatureto address 2 research questions: (1) Do interventions thatincorporate at least 1 element of the Chronic Care Model (CCM)result in improved outcomes for specific chronic illnesses? (2) Areany elements essential for improved outcomes?

Study Design: Meta-analysis.

Methods: Articles were identified from narrative literaturereviews and quantitative meta-analyses, each of which coveredmultiple bibliographic databases from inception to March 2003.We supplemented this strategy by searching the MEDLINE database(1998-2003) and by consulting experts. We included randomizedand nonrandomized controlled trials of interventions thatcontained 1 or more elements of the CCM for asthma, congestiveheart failure (CHF), depression, and diabetes. We extracted data onclinical outcomes, quality of life, and processes of care. We thenused random-effects modeling to compute pooled standardizedeffect sizes and risk ratios.

Results: Of 1345 abstracts screened, 112 studies contributeddata to the meta-analysis: asthma, 27 studies; CHF, 21 studies;depression, 33 studies; and diabetes, 31 studies. Interventions withat least 1 CCM element had consistently beneficial effects on clinicaloutcomes and processes of care across all conditions studied.The effects on quality of life were mixed, with only the CHF anddepression studies showing benefit. Publication bias was noted forthe CHF studies and a subset of the asthma studies.

Conclusions: Interventions that contain at least 1 CCM elementimprove clinical outcomes and processes of care—and to a lesserextent, quality of life—for patients with chronic illnesses.

(Am J Manag Care. 2005;11:478-488)

Approximately 90 million persons in the UnitedStates live with 1 or more chronic illnesses.1 Themanagement of healthcare delivery for personswith chronic illnesses has advanced substantially inrecent decades, yet considerable deficiencies in thequality of chronic-illness care remain.2-6 Critics arguethat systems of care designed to deal with acuteepisodes do not serve the needs of patients with chronicillness well.7 Chronic illnesses rank among thenation's costliest conditions8; and 5 chronic illnesses—asthma, diabetes, heart disease, hypertension, andmood disorders—account for nearly one half of UShealthcare expenditures.9 Managed care organizationsare increasingly becoming the main source of healthcareservices for persons with chronic illnesses such asdiabetes10 and may be particularly well suited to adoptingreforms to optimize care for such patients.

The Chronic Care Model (CCM) is a primarycare-based framework aimed at improving the care ofpatients with chronic illnesses.11-14 The model integratesa number of elements into a plausible packagedesigned to foster more productive interactionsbetween prepared, proactive teams and well-informed,motivated patients. Details on the concept, implementation,and evidence base of the CCM are available athttp://www.improvingchroniccare.org.

Although individual components of the CCM havebeen rigorously studied, accounts of the CCM's beneficialeffects as a whole on processes, outcomes, and/or costscome largely from self-reported, uncontrolled studies.11,14,15 Therefore, we sought in this meta-analysis toclassify previously published studies according to theCCM component(s) implemented, in order to address 2related research questions: (1) To what extent do interventionsthat incorporate 1 or more elements of theCCM result in improved outcomes of interest for specificchronic illnesses? (2) Are some elements of the CCMmore effective than others? Although the CCM isintended to be generic, applicable across all types ofchronic illnesses,11,16 we focused on evaluating theeffects of diverse interventions on clinical outcomes,quality of life, and processes of care for 4 chronic illnessesthat would be of particular interest to managedcare organizations: asthma, congestive heart failure,depression, and diabetes.

METHODS

Conceptual Model

The CCM identifies 6 elements deemed to be essentialfor providing high-quality care to patients withchronic illnesses: delivery system design, self-managementsupport, decision support, clinical informationsystems, community resources, and healthcare organization.11,13,14 Because of the limited information availablein published descriptions of interventions, andbecause the CCM elements are broadly defined, we categorizedinterventions according to the CCM elementsthey incorporated (Box).17 Any given intervention couldcontain more than 1 element, up to a maximum of 6.

Article Selection and Data Abstraction

One author (ACT) was responsible for all aspects ofstudy selection and data abstraction. First, we usedMEDLINE and the Cochrane Library to identify 20recently published systematic reviews and meta-analysesof the 4 chronic illnesses of interest: asthma,18-22congestive heart failure,23-28 non-insulin dependent diabetesmellitus,29-35 and depression.36,37 These 20 reviews,in addition to 3 other reviews that were not conditionspecific,38-40 formed the primary substrate for identifyingstudies for potential inclusion in our study sample.These previously published reviews covered multipledatabases—including the Cumulative Index to Nursingand Allied Health Literature (CINAHL), the CochraneLibrary, the Dissertation Index, Embase, Healthstar, theHealth Management Information Service (HELMIS),MEDLINE, PsycINFO, PsychLit, and the System forInformation on Grey Literature in Europe (SIGLE)—from inception to March 2003 (depending on the datethe review was published). We also performed aMEDLINE search for more recent studies (January1998-June 2003) that may have been missed. Finally,additional studies were identified by consulting expertsand searching independently maintained bibliographies(eg, the Chronic Care Bibliography41).

In identifying abstracts and articles for potentialinclusion in our sample, the authors and their affiliationswere not masked. We screened all identifiedabstracts for potentially relevant studies that assessedthe effects of interventions containing 1 or more of theCCM elements. Randomized and nonrandomized controlledstudies were eligible for inclusion. Uncontrolledstudies were excluded, as were studies written in a languageother than English. Because we sought to includeonly the most recent decade of published evidence inour report, we excluded studies published before 1993.

These studies used a wide range of outcome variables.We identified a clinical outcome, a quality-of-lifeoutcome, and a key process of care for each condition(Table 1). To make the analysis more tractable and tomaximize our ability to pool data, we opted to collectdata on the variables that were reported by the greatestnumber of studies (eg, for the diabetes studies we selectedglycosylated hemoglobin [HbA1c] as the clinical outcomerather than cholesterol level, and HbA1cmonitoring as the key process of care rather than funduscopictesting). We also accepted a variety of similarmeasures for some domains. For example, we acceptedany of the most commonly used measures for the clinicaloutcome among the depression studies, includingthe Hopkins Symptom Checklist-20 (9 studies), theHamilton Depression Rating scale (6 studies), and theCenter for Epidemiological Studies-Depression scale (6studies). The most commonly used measures of disease-specificquality of life were the St. George's RespiratoryQuestionnaire (4 studies), the Juniper Asthma Qualityof Life Questionnaire (4 studies), and the MinnesotaLiving with Heart Failure Questionnaire (3 studies).

Among the selected studies with data on the clinicaloutcomes, some reported data as count variables orcontinuous variables, whereas others reported data asdichotomous variables. We extracted data in both formswhere possible. Too few studies on asthma and congestiveheart failure reported the clinical outcome data ascount variables (number of emergency department visitsor readmissions during the study period), so forthose 2 condition-specific analyses we used only thedata that were reported as dichotomous variables (numberwith any emergency department visit during studyperiod, number readmitted during study period).

Using a standardized data collection form, weabstracted data on study characteristics, including CCMelements incorporated in the interventions and effectson clinical outcomes, quality of life, and processes ofcare. If the results of a study were reported in multiplearticles, data were abstracted from all articles andattributed to the primary citation. If a study reportedcomparisons at multiple follow-up times, we abstracteddata on outcomes closest to 12 months of follow-up. Foreach study that reported data as continuous variables,the mean and standard deviation values were extracted(if available). For studies that did not report a standarddeviation value at follow-up, the standard deviation wasassumed to be equal to one quarter of the theoreticalrange for that measure. This imputation approach isbased on the assumption that the underlying distributionswere approximately normal, and it is a very conservativeassumption in that the multiplication factorfor truly normal distributions should be closer to onesixth. For the clinical outcomes, a higher score representeda worse outcome, whereas for quality of life andprocesses of care, a higher score represented a betteroutcome; some studies required recoding of quality-of-life measures for consistency. For example, if a studyused the St. George's Respiratory Questionnaire as ameasure of quality of life (0-100, with higher scoresindicating poorer quality of life), we recoded the variablesso that higher scores represented better quality oflife. We excluded studies that did not provide sufficientstatistics for data abstraction if our efforts to contact thestudy's authors for additional data were unsuccessful.

Statistical Analysis

g,

We conducted all analyses using the Stata statisticalsoftware package (version 8.0, Stata Corporation,College Station, Tex).42-47 For each study, effect sizeswere calculated for the intervention group relative to thecomparison group at follow-up. If outcomes were measuredon a continuous scale, we computed Hedges thestandardized effect size, with an adjustment to correctfor small-sample bias.48 For each intervention-controlcomparison of a dichotomous outcome, we calculatedthe risk ratio and its standard deviation.49 Effect sizeswere estimated for each type of outcome variable usingrandom-effects models (incorporating both between-studyand within-study variance50), first pooled acrossconditions and then stratified by condition. A priori, wedeemed pooled estimates based on fewer than 5 studiesto be unreliable for statistical hypothesis testing, asnoted in the results. To check for publication bias(which may result from the nonpublication of small negativestudies), we visually assessed funnel plots for asymmetryand used the regression asymmetry test.51

We used a multivariate approach to independentlyassess the effect of each CCM element on the estimatedpooled effect size, after adjusting for the presence of theother elements if the study's intervention contained morethan 1 element. To do this, we fit random-effects metaregressionmodels52 for each of the 4 types of outcomes.The only covariates included in these regressions were aconstant term and 6 indicator variables equal to unity ifthe intervention included that particular CCM element,zero otherwise.53 Some of the CCM elements were implemented in too few studies for a pooled estimate to becomputed, so we labeled those situations as "notestimable" in the results. All statistical hypothesis testswere carried out at the 2-sided .05 level of confidence.

Q

I2

The degree of between-study heterogeneity wasassessed by the chi-square test for heterogeneity based onthe Cochran test.54 We also calculated the heterogeneitystatistic , which is independent of the number ofstudies and the effect-size metric, and can be interpretedas the proportion of total variation in the estimated treatmenteffect that is due to between-study heterogeneityrather than chance.55 We fit random-effects meta-regressionmodels52 to determine whether study-level variablesexplained the heterogeneity in the treatment effects,including in these models the following study-level variables:indicators for CCM element, indicators for typeof chronic illness, duration of follow-up, inpatient oroutpatient setting, and Jadad quality score.

In post hoc meta-regression analyses, we includedvariables to determine whether synergies were associatedwith the use of more CCM elements. The first specificationincluded an ordinal variable for the number ofCCM elements in the intervention. The second specificationexamined only the subset of studies that includeddelivery system design (the most frequently studiedCCM component) in the intervention. Studies thatincorporated delivery system design plus any additionalCCM elements were compared with studies that incorporateddelivery system design only. The third specificationwas similar but examined only the subset ofstudies that included self-management support (the secondmost frequently studied CCM component).

Sensitivity Analysis

We used the Jadad scale to assess the quality of studiesin our sample.56 Each study received from 0 to 5points (with higher scores representing higher methodologicalquality), depending on whether it was describedas randomized or double blind, if the randomizationsequence or blinding procedure was appropriate, and ifit provided detailed information on withdrawals anddropouts. Because studies of low methodological qualityhave been found to overestimate treatment benefit,57 weconducted a sensitivity analysis by re-estimating pooledeffects and refitting meta-regression models using onlythe 36 studies with a Jadad quality score of 3 or higher.

RESULTS

Identification, Distribution, and Quality of Evidence

Figure 1 illustrates the flow of literature from the originalsources to final acceptance for our review. We identified1345 abstracts of studies published between January1993 and June 2003. On the basis of screening theabstracts, we excluded 1041 abstracts and requested 304articles for detailed review. Of these, we excluded an additional192 articles. The Evidence Table, available athttp://www.rand.org/health/icice, provides a list of referencesfor the 112 studies included, as well as further detailson each of these studies. Any given study could contributedata on 1 or more outcomes of interest, so we report samplesizes for the stratified analyses in the tables below.

Summary statistics for the sample of studies are providedin Table 2. Of the 112 studies in our analysis, 93studies (83%) were described as randomized. Only 36studies (32%) scored a 3 on the Jadad scale, and nonescored higher than 3. The primary limitation of thesestudies was the lack of double blinding, but double blindingis rarely possible in studies of organizational interventions.Almost half of the studies (46%) contained only1 CCM element in the intervention. The most commonCCM elements contained in the interventions were self-managementsupport (n = 80), delivery system design(n = 60), and decision support (n = 38). The 2 most commonelements, self-management support and deliverysystem design, were frequently bundled together with atleast 1 other element (64% [51/80] and 83% [50/60],respectively). Only 8 studies included 4 CCM elements inthe intervention. Most interventions were carried out inthe outpatient setting, but 4 congestive heart failure studiesand 1 depression study tested inpatient interventions.

Overall Results

P

P

P

P

We first examined the overall effectiveness of interventionswith 1 or more CCM elements. Overall, the interventionsled to statistically significant improvementsin each of the 4 outcomes of interest. In particular, forthe 52 studies that reported clinical outcomes as continuousvariables (depression and diabetes only), there wasa statistically significant pooled effect size of -0.23 infavor of the intervention (95% confidence interval[CI] = -0.31, -0.15; < .001 for whether the reductionwas less than 0; see Figure 2). The 46 studies thatreported clinical outcomes as dichotomous variablesyielded a statistically significant pooled relative risk of0.84 in favor of the intervention (95% CI = 0.78, 0.90;< .001 for whether the relative risk was less than 1; seeFigure 3). The overall effect on quality-of-life outcomes,based on data from 24 studies, was statistically significantand favored the intervention (0.11; 95% CI =0.02, 0.21; = .023; see Figure 4). Finally, 32 studiescontributed data on process outcomes of interestfor a pooled relative risk of 1.19 (95% CI = 1.10, 1.28;< .001; see Figure 5).

The overall effects were pooled across conditions. Toenhance the clinical interpretation of our findings, wereport the condition-specific pooled effect sizes in Table3. Across conditions, the interventions had consistenteffects on clinical outcomes and processes. With respectto quality of life, however, the asthma studies showedequivocal results, despite an adequate number of studies.

The pooled effect size (-0.19) for the diabetes studiesreporting continuous data on HbA1c indicates alower HbA1c value in the intervention group at followupcompared with the control group. Assuming arange of standard deviation values from 1.56 to 2.47(the interquartile range for the studies in our sample),this effect size is equivalent to a reduction in HbA1c of0.30% to 0.47%. The positive effect size reported forthe congestive heart failure studies (0.28) indicates ahigher quality-of-life scale score in the interventiongroup compared with the control group and is equivalentto an increase of 5.6-6.7 points on the ChronicHeart Failure Questionnaire.

We assessed publication bias along 2 dimensions: outcomemeasure and condition. Based on funnel plotinspection and the regression asymmetry test, evidenceof publication bias was apparent for the congestiveheart failure studies (all outcomes) and the asthmastudies with dichotomous data on the clinical outcome.

Relative Effectiveness of Chronic CareModel Elements

In the meta-regression analyses, we found that 4 elementsof the CCM (delivery system design, self-managementsupport, decision support, and clinicalinformation systems) were associated with better outcomesand processes, after adjusting for the presenceof other elements if the intervention contained morethan 1 element (Table 4). Our sample had too fewstudies that implemented the community resourcesand health care organization elements to judge the relativeeffectiveness of those 2 elements. The statisticallysignificant effects observed for delivery systemdesign and self-management support could be attributedto both the larger number of studies and thelarger estimated effects. Decision support improvedprocess significantly, but not outcomes. We observedno statistically significant effects for clinical informationsystems, perhaps due to the smaller number ofstudies. We also noted that no single element of theCCM was essential to improved outcomes (Appendix, availableat http://www.rand.org/health/icice).

I

Q

df

P

I

Q

df

P

I

Q

df

P

I

Q

df

P

I

Evidence of substantial between-study heterogeneitywas noted for the analyses of continuous clinical outcomes(2 = 78%; Cochran's &#967;2 test for homogeneity =230.4, = 51, < .001); dichotomous clinical outcomes(2 = 67%; =135.19, = 45, < .001); qualityof life (2 = 75%, = 92.81; = 23; < .001); andprocesses (2 = 90%; = 311.59, = 31, < .001).These 2 values indicate that more than two thirds of thevariation in the estimated treatment effects may beattributable to between-study heterogeneity. We fit random-effects meta-regression models to identify potentialexplanations for this variation, but the study-level variableswe considered—indicators for type of chronic illness,indicators for CCM element, duration of follow-up,inpatient or outpatient setting, and Jadad quality score—did not appreciably reduce the unexplained variance.

Synergistic Effects

P

P

In our random-effects meta-regression analyses,the number ofCCM elementsincorporated inthe study interventionwas notassociated withbetter outcomes,with values rangingfrom 0.38 to0.81. Only 2CCM domains,delivery systemdesign and self-managementsupport, wererepresented inenough studiesfor a meaningfulcomparison ofinterventionsconsisting of thatelement aloneversus that elementin conjunctionwith otherelements. In ourdata, these comparisonswere never statistically significant, with valuesranging from 0.13 to 0.61.

Sensitivity Analysis

We re-fit the meta-regression models to the subset ofstudies (n = 36) that scored a 3 on the Jadad scale.57This sensitivity analysis yielded results that were qualitativelysimilar to the main analysis (Appendix Tables6 and 7, available at http://www.rand.org/health/icice).The pooled effect sizes were generally larger (favoringthe intervention), with larger confidence intervals.

DISCUSSION

Interventions that incorporated 1 or more elementsof the CCM had beneficial effects on clinical outcomesand processes of care for patients, and the results wereconsistent across a variety of chronic illnesses. Althoughour estimates of pooled effect size were small tomoderate,58 they also are broadly consistent with thosereported in prior meta-analyses.22,28,35,40 Interventionsdirected at diabetes care, for example, led to a 0.30%-0.47% reduction in HbA1c. Managed care organizationsmay realize benefits from even smaller reductions inmean population values for continuous risk factors suchas lipid levels and HbA1c. For example, the EuropeanProspective Investigation of Cancer and Nutrition(EPIC-Norfolk) estimated that a population reductionof 0.2% in HbA1c could reduce the prevalence of menwith high HbA1c levels (5%-6.9%) from 79% to 57% andreduce excess mortality by 10%.59 We found that interventionsdirected at congestive heart failure led to a5.6- to 6.7-point improvement in the Chronic HeartFailure Questionnaire, slightly less than the 7- to 9-point difference that is regarded as a minimal clinicallyimportant difference on that scale.60

The evidence was mixed for quality of life, with theasthma and diabetes studies showing no benefit. It hasbeen well established that condition-specific quality-of-lifescales are more sensitive to changes in clinical statusthan are generic measures of quality of life. Most ofthe studies included in our meta-analyses used condition-specific quality of life scales (see the EvidenceTable, available at http://www.rand.org/health/icice).For some conditions, one might reasonably expect disease-specific interventions to have a more direct effect onclinical outcomes and processes of care than on qualityof life. Improvements in diabetes care, for example, arefocused on preventing long-term microvascular complicationsbeyond the end point of the studies we examined,61 with less focus on improving short-term qualityof life. We speculate that our meta-analyses mighthave yielded different results, for example, had weused a quality-of-life measure that is more sensitive tothe short-term benefits of improved glycemic control.62,63

The CCM elements most responsible for these benefitscould not be determined from the data. Effects appearedto be somewhat stronger for delivery system design andself-management support, although decision support hadsignificant beneficial effects on processes. The other elementsof the CCM may be critical infrastructure for providinghigh-quality chronic care but are more difficult totest scientifically. For example, leadership support maybe necessary to promote and sustain higher quality, butrandomizing managed care organizations to receivechanges in leadership support is clearly infeasible. It isno wonder that these elements have not had many scientifictrials. The fact that such linkages are hard tostudy scientifically does not mean they are unimportant.Instead, they are supported by common sense andreports from successful organizations.

The CCM has been promoted as a unified package.Evidence that interventions with multiple componentsdo better than interventions with single components64has been interpreted as supporting synergistic effects inwhich the whole is bigger than the sum of the parts.Some components of the model, such as building anelectronic patient registry, may facilitate other componentsand reduce their costs. We found that single interventionswere quite successful. In post hoc analyses, weattempted to identify whether there may be some advantageto having more components, but that advantage wasnever statistically significant and does not appear to bemore than additive.

One limitation of our work is that the studies in our sampleonly incorporated elements of the CCM and were notdesigned to test the entire CCM package.11,13,14 TheRAND/University of California-Berkeley Improving ChronicIllness Care Evaluation (ICICE) is nearing completion, and itis the first independent and controlled evaluation of theeffects of implementing the CCM as a whole. Organizationssigned up for the Institute of Healthcare Improvement'sCollaboratives to improve care for specific conditions65 andworked together to learn about the CCM and about how tomake organizational changes to improve quality of care. Thedesign of the ICICE has been published,66 and results fromthe evaluation are posted at http://www.rand.org/health/ICICE/ as they become available. Despite the large scale ofthe evaluation—24 organizations with both intervention andcontrol sites, and 12 organizations with intervention sitesonly—the number of participating organizations was toosmall to determine which components of the CCM weremost critical to success. The organizations' characteristicsand what they did differed in many ways, many timesmore than the number that could be studied statistically.

A second limitation is that the use of meta-analyticmethods necessarily forces what are likely complex, multivariateinterventions into a narrow linear framework. Inthis meta-analysis we aggregated results across conditionsand across interventions. We attempted to investigate thesources of variation between studies, but we were unableto explain much of it. We also were unable to assess interactionsbetween CCM element and type of chronic illness.For example, a clinical information systems interventionfeaturing physician reminders may be particularly effectivefor improving care for one type of chronic illness but notfor other types, and a pooled analysis would not identifythe interaction. A related limitation is that we were unableto assess the intensity of implementation in the studyinterventions.17,67 Perhaps the interventions we studiedwere successful because doing trials requires energy andcommitment to the intervention concept. This energy maybe an important component of initial success that is hardto transfer. If there is significant variation in the intensityof implementation of these elements across studies, simplifiedcomparisons based on the presence or absence ofthese elements may mask important between-study differences.In addition, we focused our data collection onselected outcomes. We needed to do so in order to aggregateacross studies, recognizing that interventions mayhave had different effects on other outcomes and processesof care. However, the outcomes we selected werereported in a large number of studies and likely reflectoutcomes of interest to managed care organizations. Afinal limitation is that we used an unconventional searchstrategy by relying on prior meta-analyses as the primarysubstrate for identifying our sample of studies. Doing somay have introduced unpredictable biases, but we alsosystematically searched the MEDLINE database and theChronic Care Bibliography41 to identify more recentlypublished studies.

Despite these limitations, our meta-analysis showsthat interventions that contain 1 or more elements ofthe CCM can improve outcomes and processes for severalchronic illnesses of interest to managed care organizations.How to transfer the gains from these efficacystudies into the chaotic real world of healthcare is a differentbut equally important issue.

From the Department of Epidemiology and Biostatistics, Case Western Reserve UniversitySchool of Medicine, Cleveland, Ohio (ACT); RAND Health, Santa Monica, Calif (SCM, EBK);and the Division of General Internal Medicine and Health Services Research, Departmentof Medicine, David Geffen School of Medicine at UCLA, Los Angeles, Calif (CMM).

This research was supported by grants 034984 and 035678 from the Robert Wood JohnsonFoundation. At the time the research was conducted, Dr Tsai was a Graduate Student SummerAssociate at RAND Health and a National Research Service Award Trainee supported by USAgency for Healthcare Research and Quality Institutional Training Award T32 HS 00059-06,Case Western Reserve University, and AHRQ Dissertation Research Grant R36 HS 014151-01.Dr Mangione received support from the UCLA Center for Health Improvement in MinorityElders/Resource Centers for Minority Aging Research, NIH/NIA, under Grant AG-02-004.An earlier version of this manuscript was presented at the AcademyHealth AnnualResearch Meeting, San Diego, Calif, June 7, 2004.

Address correspondence to: Alexander C. Tsai, PhD, Case Western Reserve UniversitySchool of Medicine, WG-57 10900 Euclid Avenue, Cleveland, OH 44106-4945. E-mail:act2@case.edu.

JAMA.

1. Hoffman C, Rice D, Sung HY. Persons with chronic conditions. Their prevalenceand costs. 1996;276:1473-1479.

Am J Manag Care.

2. Legorreta AP, Liu X, Zaher CA, Jatulis DE. Variation in managing asthma: experienceat the medical group level in California. 2000;6:445-453.

Arch Intern Med.

3. Ni H, Nauman DJ, Hershberger RE. Managed care and outcomes of hospitalizationamong elderly patients with congestive heart failure. 1998;158:1231-1236.

Arch Gen Psychiatry.

4. Young AS, Klap R, Sherbourne CD, Wells KB. The quality of care for depressiveand anxiety disorders in the United States. 2001;58:55-61.

JAMA.

5. Clark CM, Fradkin JE, Hiss RG, et al. Promoting early diagnosis and treatment oftype 2 diabetes: the National Diabetes Education Program. 2000;284:363-365.

N Engl J Med.

6. McGlynn EA, Asch SM, Adams J, et al. The quality of health care delivered toadults in the United States. 2003;348:2635-2645.

Crossing the Quality Chasm: A New Health System for

the Twenty-first Century.

7. Institute of Medicine. Washington, DC: National Academy Press; 2001.

Health Aff (Millwood).

8. Cohen JW, Krauss NA. Spending and service use among people with the fifteenmost costly medical conditions, 1997. 2003;22:129-138.

Health Aff (Millwood).

9. Druss BG, Marcus SC, Olfson M et al. Comparing the national economic burdenof five chronic conditions. 2001;20:233-241.

Ann Intern Med.

10. Quickel KE Jr. Diabetes in a managed care system. 1996;124(1 pt 2):160-163.

Milbank Q.

11. Wagner EH, Austin BT, Von Korff M. Organizing care for patients with chronicillness. 1996;74:511-544.

Manag Care Q.

12. Wagner EH, Davis C, Schaefer J, Von Korff M, Austin B. A survey of leadingchronic disease management programs: are they consistent with the literature?1999;7(3):56-66.

Health Aff (Millwood).

13. Wagner EH, Austin BT, Davis C, et al. Improving chronic illness care: translatingevidence into action. 2001;20:64-78.

JAMA.

14. Bodenheimer T, Wagner EH, Grumbach K. Improving primary care forpatients with chronic illness. 2002;288:1775-1779.

Ann Intern Med.

15. Von Korff M, Gruman J, Schaefer J, Curry SJ, Wagner EH. Collaborativemanagement of chronic illness. 1997;127:1097-1102.

Milbank Q.

16. Glasgow RE, Orleans CT, Wagner EH. Does the chronic care model servealso as a template for improving prevention? 2001;79:579-612, iv-v.

Health Serv Res.

17. Pearson ML, Wu SY, Schaefer J, et al. Assessing the implementation of theChronic Care Model in quality improvement collaboratives. 2005;40(4):978-996.

Qual Health Care.

18. Eastwood AJ, Sheldon TA. Organisation of asthma care: what difference doesit make? A systematic review of the literature. 1996;5:134-143.

Thorax.

19. Sudre P, Jacquemet S, Uldry C, Perneger TV. Objectives, methods and contentof patient education programmes for adults with asthma: systematic review ofstudies published between 1979 and 1998. 1999;54:681-687.

Cochrane Database Syst Rev.

20. Gibson PG, Powell H, Coughlan J, et al. Limited (information only) patienteducation programs for adults with asthma. 2002;(2):CD001005.

J Fam Pract.

21. Lefevre F, Piper M, Weiss K, et al. Do written action plans improve patientoutcomes in asthma? An evidence-based analysis. 2002;51:842-848.

Cochrane Database Syst Rev.

22. Gibson PG, Powell H, Coughlan J, et al. Self-management education and regularpractitioner review for adults with asthma. 2003;(1):CD001117.

J Gen Intern Med.

23. Philbin EF. Comprehensive multidisciplinary programs for the management ofpatients with congestive heart failure. 1999;14:130-135.

J Card Fail.

24. Rich MW. Heart failure disease management: a critical review. 1999;5:64-75.

Circulation.

25. Grady KL, Dracup K, Kennedy G, et al. Team management of patients withheart failure: A statement for healthcare professionals from The Cardiovascular NursingCouncil of the American Heart Association. 2000;102:2443-2456.

Annu Rev Nurs Res.

26. Moser DK. Heart failure management: optimal health care delivery programs.2000;18:91-126.

Prog Cardiovasc Dis.

27. Quaglietti SE, Atwood JE, Ackerman L, Froelicher V. Management of thepatient with congestive heart failure using outpatient, home, and palliative care.2000;43:259-274.

Am J Med.

28. McAlister FA, Lawson FM, Teo KK, Armstrong PW. A systematic review ofrandomized trials of disease management programs in heart failure. 2001;110:378-384.

BMJ.

29. Griffin S. Diabetes care in general practice: meta-analysis of randomised controltrials. 1998;317(7155):390-396.

Cochrane Database Syst Rev.

30. Valk GD, Kriegsman DM, Assendelft WJ. Patient education for preventingdiabetic foot ulceration. 2001;(4):CD001488.

Diabetes Care.

31. Renders CM, Valk GD, Griffin SJ, et al. Interventions to improve the managementof diabetes in primary care, outpatient, and community settings: a systematicreview. 2001;24:1821-1833.

Am J Prev Med.

32. Norris SL, Nichols PJ, Caspersen CJ, et al. Increasing diabetes self-managementeducation in community settings. A systematic review. 2002;22(4 suppl):39-66.

Am J Prev Med.

33. Norris SL, Nichols PJ, Caspersen CJ, et al. The effectiveness of disease andcase management for people with diabetes. A systematic review. 2002;22(4 suppl):15-38.

Cochrane Database Syst Rev.

34. Loveman E, Royle P, Waugh N. Specialist nurses in diabetes mellitus.2003;(2):CD003286.

35. Shekelle PG, Morton SC, Chodosh J, et al. Chronic disease self-managementfor diabetes, osteoarthritis, post-myocardial infarction care, and hypertension.Prepared by the Southern California Evidence-based Practice Center for the USDepartment of Health and Human Services under contract 500-98-0281. 2003.Evidence Report/Technology Assessment.

JAMA.

36. Gilbody SM, Whitty PM, Grimshaw JM, Thomas RE. Educational and organizationalinterventions to improve the management of depression in primary care: asystematic review. 2003;289:3145-3151.

Qual Saf Health Care.

37. Gilbody SM, Whitty PM, Grimshaw JM, Thomas RE. Improving the detectionand management of depression in primary care. 2003;12:149-155.

J

Gen Intern Med.

38. Ferguson JA, Weinberger M. Case management programs in primary care. 1998;13:123-126.

JAMA.

39. Hunt DL, Haynes RB, Hanna SE, Smith K. Effects of computer-based clinicaldecision support systems on physician performance and patient outcomes: a systematicreview. 1998;280:1339-1346.

BMJ.

40. Weingarten SR, Henning JM, Badamgarav E, et al. Interventions used in diseasemanagement programmes for patients with chronic illness—which oneswork? Meta-analysis of published reports. 2002;325(7370):925.

41. Improving Chronic Illness Care. A national program of The Robert WoodJohnson Foundation. Chronic care bibliography. Available at: http://www.improvingchroniccare.org/resources/bibliography/index.html. Accessed June 12, 2003.

Stata Tech Bull.

42. Sharp S, Sterne J. sbe16: meta-analysis. 1997;38:9-14.

Stata Tech Bull.

43. Sharp S, Sterne J. sbe16.1: new syntax and output for the meta-analysis command.1998;42:6-8.

Stata Tech Bull.

44. Steichen T. sbe19: tests for publication bias in meta-analysis. 1998;41:9-15.

Stata Tech Bull.

45. Steichen T, Egger M, Sterne J. sbe19.1: tests for publication bias in metaanalysis.1998;44:3-4.

Stata Tech Bull.

46. Sharp S. sbe23: meta-analysis regression. 1998;42:16-24.

Stata Tech Bull.

47. Bradburn MJ, Deeks JJ, Altman DG. sbe24: metan—an alternative metaanalysiscommand. 1998;44:4-15.

Psychol Bull.

48. Hedges LV. Estimation of effect size from a series of independent experiments.1982;92:490-499.

Systematic Reviews in Health Care: Meta-Analysis in

Context.

49. Deeks JJ, Altman DG, Bradburn MJ. Statistical methods for examining heterogeneityand combining results from several studies in meta-analysis. In: Egger M,Smith GD, Altman DG, eds. London, UK: BMJ Publishing Group; 2001:285-312.

Control Clin Trials.

50. DerSimonian R, Laird N. Meta-analysis in clinical trials. 1986;7:177-188.

BMJ.

51. Egger M, Davey SG, Schneider M, Minder C. Bias in meta-analysis detectedby a simple, graphical test. 1997;315(7109):629-634.

Stat Med.

52. Berkey CS, Hoaglin DC, Mosteller F, Colditz GA. A random-effects regressionmodel for meta-analysis. 1995;14:395-411.

Ann Intern

Med.

53. Stone EG, Morton SC, Hulscher ME, et al. Interventions that increase use ofadult immunization and cancer screening services: a meta-analysis. 2002;136:641-651.

Biometrics.

54. Cochran WG. The combination of estimates from different experiments.1954;10:101-129.

Stat

Med.

55. Higgins JP, Thompson SG. Quantifying heterogeneity in a meta-analysis. 2002;21:1539-1558.

Control Clin Trials.

56. Jadad AR, Moore RA, Carroll D, et al. Assessing the quality of reports of randomizedclinical trials: is blinding necessary? 1996;17:1-12.

Lancet.

57. Moher D, Pham B, Jones A, et al. Does quality of reports of randomised trialsaffect estimates of intervention efficacy reported in meta-analyses? 1998;352(9128):609-613.

Statistical Power Analysis for the Behavioral Sciences.

58. Cohen J. New York,NY: Academic Press; 1977.

BMJ.

59. Khaw KT, Wareham N, Luben R, et al. Glycated haemoglobin, diabetes, andmortality in men in Norfolk cohort of European prospective investigation of cancerand nutrition (EPIC-Norfolk). 2001;322(7277):15-18.

J Gen Intern Med.

60. Guyatt GH, Nogradi S, Halcrow S, et al. Development and testing of a newmeasure of health status for clinical trials in heart failure. 1989;4:101-107.

N Engl J Med.

61. Retinopathy and nephropathy in patients with type 1 diabetes four years aftera trial of intensive therapy. The Diabetes Control and ComplicationsTrial/Epidemiology of Diabetes Interventions and Complications Research Group.2000;342:381-389.

JAMA.

62. Testa MA, Simonson DC. Health economic benefits and quality of life duringimproved glycemic control in patients with type 2 diabetes mellitus: a randomized,controlled, double-blind trial. 1998;280:1490-1496.

Diabetes Care.

63. Testa MA, Simonson DC, Turner RR. Valuing quality of life and improvementsin glycemic control in people with type 2 diabetes. 1998;21(suppl 3):C44-C52.

JAMA.

64. Bodenheimer T, Wagner EH, Grumbach K. Improving primary care forpatients with chronic illness: the chronic care model, part 2. 2002;288:1909-1914.

IHI's Collaborative Model for

Achieving Breakthrough Improvement.

The Breakthrough Series.

65. Institute for Healthcare Improvement. Boston, Mass: Institute for HealthcareImprovement; 2003.

Eval Rev.

66. Cretin S, Shortell SM, Keeler EB. An evaluation of collaborative interventions toimprove chronic illness care. Framework and study design. 2004;28:28-51.

67. Wu SY, Pearson ML, Keeler EB, et al. Sustainability and spread of chronic illnesscare improvement. Paper presented at: AcademyHealth Annual ResearchMeeting; June 6-8, 2004; San Diego, Calif.