Information Technology in Mental Health Research: Impediments and Implications in One Chronic Pain Study Population

Abstract

Objective. The Department of Veterans Affairs and Veterans' Administration hospitals began using the electronic medical record in 1994, streamlining provider-to-provider communication of vital clinical information, and simultaneously providing investigators access to vast amounts of clinical data for research purposes. Administrative and coded data, including symptoms and diagnoses as derived from chart content, are one of the most ready substrates for analysis of these massive databases to answer various research-related inquiries. We evaluated the capability of this type of coded data to accurately identify patients with schizophrenia from a group of 819 patients taking opioids for chronic pain in the primary care clinic of the Philadelphia Veterans' Administration Medical Center. Methods. Patients were identified for inclusion in this analysis by their sequential enrollment in an Opioid Renewal Clinic as referred by their primary care providers, as well as a nonreferred cohort of chronic pain patients maintained on opioids by their primary care providers (N = 819). The prevalence of schizophrenia obtained by coded diagnostic labeling by computerized chart review was compared with the prevalence of schizophrenia obtained by systematic independent manual chart review of all cases by a research psychiatrist. Results. Based purely on coded diagnostic labeling, the prevalence of schizophrenia in this population was 14%, nearly 14 times the estimated general population prevalence. However, manual independent chart review of the identified cases by a research psychiatrist (N = 119) resulted in the actual prevalence in this group to be estimated at 1.8%. None of the patients with three or less incidents of diagnostic labeling by code had schizophrenia. Of those with four or more incidents, 74% had schizophrenia. Conclusions. Better accuracy is obtained when cases of schizophrenia are identified by multiple coding incidents derived from longitudinal clinical encounters. These findings suggest that the extraction of up to three coding incidents for schizophrenia alone is not an accurate means of identifying this clinical population for research and epidemiological purposes. Further, when designing automated, information technology-based approaches to the identification of subjects for clinical research and extracting their data from electronic medical records using International Classification of Disease, Ninth Revision, Clinical Modification (ICD-9-CM) codes, mental health diagnoses must be uniquely considered on a disease-by-disease basis.