Introduction/Objective: To examine overviews of psychological rehabilitation trials for evidence of publication bias. Publication bias in medical research, particularly of new therapies, is recognised but the possible magnitude of its effect on statistical overviews of trials is not always appreciated.

Methods: To overview the published trials, seek missing information from authors and compare findings with a subsequent large multi-centre trial.

Results: Eleven controlled trials of myocardial infarction compared psychological rehabilitation with 'usual care' (2269 patients). Eight reported total mortality, one to five years after trial entry; 123 deaths among 1954 patients. An initial statistical overview of these trials suggests a 'relative risk' (RR) of mortality of 0.65 with 95% confidence interval (CI) of 0.46-0.91, a greater mortality reduction than aspirin, beta blocker or thrombolysis. A 'funnel plot' suggests some publication bias. Correspondence with principal investigators of those trials, that did not report total mortality provides a revised estimate of RR = 073 (CI 0.53-1.00) corroborating the suspicion. Individual trials were mostly small and many were conducted by therapists. A multi-centre trial, with sufficient power to detect a 20% reduction in one year mortality and larger numbers than all previous trials combined, found no difference in mortality RR = 1 01 (CI 0.75 - 1 37).

Discussion: Published reports of small trials may misrepresent the effects of health care interventions and incomplete overviews can give a very misleading impression of effect size.