Diagnosing depression in the medically ill: validity of a lay-administered structured diagnostic interview

Understanding the validity of structured psychiatric diagnostic interviews in medically ill patients will advance the ability to conduct research into the treatment and management of these disorders in general medical settings. We compared the University of Michigan version of the CIDI (Composite International Diagnostic Interview) for major depression to a clinical gold standard, derived through Spitzer's Longitudinal, Expert, All Data (LEAD) criteria based on the SCID-III-R. A convenience sample of medical inpatients was administered the SCID-III-R and the CIDI for major depression in random order. A physician panel reviewed the SCID interview and other pertinent data and determined whether patients had a lifetime or current (past month) diagnosis of major depression. The CIDI was scored with and without hierarchical exclusions for mania, hypomania, substance use, or medical illness. When the UM-CIDI was scored for a lifetime diagnosis of major depression without hierarchical exclusions, agreement above chance (kappa) was very good (kappa = 0.67) between the CIDI and the physician panel and good (kappa = 0.46) when the UM-CIDI was scored with exclusions. Agreement above chance for diagnosis of a recent disorder was better for UM-CIDI scoring with exclusions (kappa = 0.51) compared to scoring without exclusions (kappa = 0.43). Predictive value-positive was excellent in both scoring versions for a lifetime diagnosis (82%) and good to very good for current depression (46% and 62%). In all cases predictive value-negative was very good to excellent (77-93%). Discordant cases were almost uniformly due to difficulties in attribution of symptoms to medical illnesses. We conclude that the CIDI can perform acceptably as a research instrument to diagnose major depression in medically ill patients, potentially supplemented by clinician review of cases identified by the CIDI with current disorder.