Using Administrative Data to Identify Mental Illness: What Approach Is Best?

TLDR

The study evaluated the validity of administrative data algorithms for detecting mental health conditions among 133,068 diabetic Veterans using VHA records and a 1999 survey. They compared multiple ICD‑9 based algorithms against self‑reported depression, PTSD, and schizophrenia from the survey. The algorithms achieved PPVs of 0.65–0.86 and NPVs of 0.68–0.77, with PPV improved by requiring ≥2 ICD‑9 instances or mental‑health‑visit codes, and NPV enhanced by adding Medicare data; these results guide algorithm choice in mental‑health quality improvement and research.

Abstract

The authors estimated the validity of algorithms for identification of mental health conditions (MHCs) in administrative data for the 133 068 diabetic patients who used Veterans Health Administration (VHA) nationally in 1998 and responded to the 1999 Large Health Survey of Veteran Enrollees. They compared various algorithms for identification of MHCs from International Classification of Diseases, 9th Revision (ICD-9) codes with self-reported depression, posttraumatic stress disorder, or schizophrenia from the survey. Positive predictive value (PPV) and negative predictive value (NPV) for identification of MHC varied by algorithm (0.65-0.86, 0.68-0.77, respectively). PPV was optimized by requiring ≥2 instances of MHC ICD-9 codes or by only accepting codes from mental health visits. NPV was optimized by supplementing VHA data with Medicare data. Findings inform efforts to identify MHC in quality improvement programs that assess health care disparities. When using administrative data in mental health studies, researchers should consider the nature of their research question in choosing algorithms for MHC identification.