Our main research question was whether there are specific design features of collaborative care interventions that are consistently associated with greater impact on depression symptoms compared to a usual care control group. We also aimed to explore additional outcomes including patient satisfaction and functioning. In addition, we asked whether there were specific design features of randomized trial evaluations of collaborative care that were associated with consistently greater effects. Secondarily, we aimed to assess whether any patient characteristics, such as comorbidities, were associated with differential collaborative care effects, and the degree to which model effects persisted over time. We investigated these goals based on the following research questions. (1) Primary Research Question: What is the core set of intervention features that characterize collaborative care interventions, and which additional features are most linked to enhanced outcome effects? (2) Secondary Research Question: Are there specific evaluation features among randomized trials of collaborative care that are associated with effect size differences, independently of intervention features? (3) Secondary Research Question: To what extent is collaborative care more effective than usual care for decreasing depressive symptoms among patients with comorbid mental health conditions (PTSD, dementia, anxiety, dysthymia, substance abuse) or medical conditions? Methods: We used a set of articles identified and preliminarily reviewed as part of an earlier, nonquantitative literature review on depression care models (Williams, Gerrity et al. 2007) to carry out quantitative meta-regression analysis of collaborative care features. Studies were high quality randomized trials of depression collaborative care interventions compared to usual care that incorporated at least two features of the chronic illness care model. At least one of these features had to directly support patients in completing depression treatment. We did not review studies that only sought to change primary care clinician behavior (e.g., using reminders), without an additional patient-directed component, such as care management. We contacted authors extensively to identify, clarify, or verify study variables such as chronic illness care features or patient population characteristics. We began our analyses by assessing correlations between features. For study outcomes, we evaluated the effect size across studies for changes in depression symptoms, and relative risk across studies for changes in rates of resolution of depression. For these analyses we used study effect sizes comparing intervention to usual care arms as the unit of analysis. The effect size analyses treated short (six weeks to four months), medium (five to eight months), and long (nine to twelve months) outcomes separately. We also measured intervention impact (high, medium, low and little or none) for each study based on reviewer ratings of de-identified sets of study outcomes, including adherence, patient satisfaction, and functioning. We eliminated variables with inadequate distributions for meaningful quantitative analysis, using a rule of thumb of at least three studies per variable category. We carried out univariate and multivariate regression to determine relationships between intervention and evaluation features and effectiveness. Finally, we conducted cross-case qualitative analysis (Miles and Huberman 1994) of intervention and evaluation features, including comorbidities, against intervention impact.