Release bias in accessing medical records in clinical trials: a STAR*D report.

Clinical trials often require subjects to sign medical record releases to allow investigators to measure treatment fidelity, off-protocol care use, and care costs. Little is known, however, if limiting samples to those willing to sign releases impacts external validity. Data came from outpatients with non-psychotic major depressive disorder who enrolled in the multisite Sequenced Treatment Alternatives to Relieve Depression (STAR*D) study. Differences between those who signed (n = 3116) and who did not sign (n = 925) releases were assessed using logistic regression and two-part, three-level log-transformed regression models, corrected for site clustering and repeated measures. Patients who released records tended to believe care was helpful, were younger, and married. However, release status had little material or consistent associations with patient health outcomes or use of care. With appropriate adjustments to data, requiring patient medical records may pose only minimal challenges to external validity in cost-outcome studies.