A Meta-Analysis of Interventions to Improve Chronic Illness Care Alexander Tsai 1 S.C. Morton 2, C.M. Mangione 3, E.B. Keeler 2 1 Case School of Medicine; 2 RAND Health; 3 David Geffen School of Medicine at UCLA AcademyHealth Annual Research Meeting, June 7, 2004
The Chronic Care Model
Objective Lack of controlled studies of the CCM –But there have been controlled studies of interventions that incorporate one or more CCM elements Using meta analysis, we sought to: –Determine the extent to which CCM-style interventions improve chronic illness care –Determine whether any specific CCM elements were essential to improved outcomes
Table 1. Outcomes of Interest Clinical OutcomesQuality of LifeProcesses (Continuous)(Dichotomous)(Continuous)(Dichotomous) Asthma# ED visits% with at least one ED visit Quality of life% with long- acting meds CHF# hospital readmissions % with at least 1 readmission Quality of life% with ACE inhibitor DepressionDepression Scale % depressed /symptomatic Quality of life or SF-36 MCS % with antidepressant DiabetesHbA1c% with HbA1c > 7% Quality of life% tested for HbA1c level
Data Sources 1.Bibliographies of 23 recently published systematic reviews and meta-analyses: asthma (5), CHF (6), diabetes (7), depression (2), general chronic care (2), information systems (1) 2.MEDLINE Chronic Care Bibliography
Inclusion/Exclusion Criteria Inclusion criteria – –Asthma, CHF, depression, diabetes –Controlled (randomized or non-randomized) –Outcomes of Interest Exclusion criteria –Not written in English –Non-adult patient population –Insufficient statistics
Data Abstraction Data obtained from all relevant associated articles and attributed to the primary citation Only 12-month follow-up data recorded if multiple follow-up times assessed If missing data, SD conservatively assumed to be 1/4 of the theoretical range for that measure
Statistical Analysis Comparisons at follow-up Pooled analysis by condition –Hedges’ g (continuous), risk ratio (binary) Relative effectiveness of CCM elements –Random-effects meta-regression model Funnel plots to detect publication bias Cochran’s Q to assess heterogeneity Sensitivity analysis for Jadad score ≥3
Table 2. Summary Statistics (N=112) Element Type DSDSMSDSCISCRHCO N # ElementsOneTwoThreeFourFiveSix N Jadad scoreZeroOneTwoThreeFourFive N
Table 3. By Condition Clinical OutcomesQuality of lifeProcesses [continuous] (lower=better) [dichotomous] (lower=better) [continuous] (higher=better) [dichotomous] (higher=better) Effect SizeRREffect SizeRR OVERALL *0.84 *0.11 *1.19 * Asthma 0.82 * CHF 0.81 *0.28 *1.13 * Depression *0.83 *0.18 *1.28 * Diabetes * * * P<0.05
Table 4. By CCM Element Clinical OutcomesQuality of lifeProcesses [continuous] (lower=better) [dichotomous] (lower=better) [continuous] (higher=better) [dichotomous] (higher=better) Effect SizeRREffect SizeRR DSD-0.21 *0.77 * * SMS-0.22 *0.81 * * DS * CIS * P<0.05
Conclusions 1.Interventions that contained one or more CCM elements improved clinical outcomes and processes of care for four chronic illnesses 2.Effect on quality of life was mixed 3.The specific CCM elements most responsible for the beneficial effects could not be determined
Limitations Testing the CCM vs. testing CCM elements –Unable to assess intensity of implementation Unexplained heterogeneity in aggregating across conditions and types of interventions Conclusions limited to selected outcomes and selected conditions
For additional information:
Fig 1. Clinical Outcomes (Continuous) Pooled Effect Size = (-0.31, -0.15) favoring intervention Q=230, df=51, P<0.001 Depression Diabetes
Fig 2. Clinical Outcomes (Binary) Pooled RR = 0.84 (0.78, 0.90) favoring intervention Q=135, df=45, P<0.001 Asthma CHF Depression Diabetes
Fig 3. Quality of Life Pooled Effect Size = 0.11 (0.02, 0.21) favoring intervention Q=93, df=23, P<0.001 Asthma CHF Depression Diabetes
Fig 4. Processes of Care Pooled RR = 1.19 (1.10, 1.28) favoring intervention Q=312, df=31, P<0.001 Asthma CHF Depression Diabetes