We have some concerns about the validity of the study by Gilbody, et al.1 Patients were recruited for a randomised controlled trial of collaborative care for depression in primary care. Thirty-six of 96 (93 in the abstract) patients (37.5 %) were diagnosed with major depressive disorder according to SCID. Such high prevalence indicates that patients were not randomly chosen from practices. Receiver-operating curve statistics was applied on this obviously highly-selected group of patients. This is misleading; any depression screening instrument may demonstrate excellent performance in such groups. For instruments to prove useful in general practice, statistics should be based on representative practice population samples.2–4
展开▼
