When we examined fidelity to the collaborative care model,9 all studies had a case manager, but several studies deviated from the model in that they did not have access to specialist input. These studies with lower fidelity showed a lower pooled effect size and were more heterogeneous (SMDlow fidelity, 0.187; I2low fidelity, 73.3%; SMDhigh fidelity, 0.30; I2high fidelity, 4.6%), although this difference was not significant (meta-regression β, 0.09; 95% CI, −0.08 to 0.25; P = .29; I2 = 50.7%). Two study-level variables, regular supervision and the mental health background of case managers, were significantly related to study effect size. The use of regular and planned supervision of the case manager, usually by a psychiatrist, was related to a more positive clinical outcome (SMDusual supervision, 0.29; SMDunplanned and ad hoc supervision, 0.14; meta-regression β, 0.15; 95% CI, −0.02 to 0.31; P = .07; I2 = 49.3%). Case managers with a specific mental health background also achieved better outcomes (SMDCM mental health background, 0.34; SMDCM non–mental health background, 0.164; meta-regression β, 0.18; 95% CI, 0.04-0.32; P = .02; I2 = 42.4%). However, the addition of a specific form of psychotherapy to medication management in collaborative care was not associated with any significantly increased effect size (SMDpsychotherapy + medication management, 0.30; SMDmedication management only, 0.21; meta-regression β, 0.10; 95% CI, −0.05 to 0.25; P = .20; I2 = 49.3%). Similarly, studies in which antidepressant medication was prescribed at entry to the trial were no more effective (SMDantidepressants at entry, 0.21; SMDantidepressants not consistently prescribed, 0.30; meta-regression β, −0.09; 95% CI, −0.24 to 0.06; P = .23; I2 = 50.7%). The number of case management sessions ranged from 2 to 14, but the number of sessions was not related to outcome (meta-regression β, 0.02; 95% CI, −0.008 to 0.04, P = .19; I2 = 50.9%) (Figure 5).