Reliability of post-mortem chart diagnoses of schizophrenia and dementia

The reliability of psychiatric diagnosis has a direct effect on the validity of post-mortem analyses of neuropathological data, yet little is known about the reliability of retrospective diagnostic procedures which rely on review of medical records. In this paper, we report on the reliability of DSM-III-R psychiatric diagnoses assigned by a pool of 8 raters to a set of 106 state hospital charts of elderly, chronic patients who had died while institutionalized and were autopsied. Diagnoses were grouped by general diagnostic class, and Kappa coefficients computed for agreement among raters, as well as for agreement between ultimate consensus diagnoses and those made while subjects were living. Interrater agreement for those diagnoses that occurred most frequently in this sample (e.g. Schizophrenia and Dementia) was excellent, and comparable to the the agreement observed for ratings of live patients. Interrater agreement for less frequently occurring diagnoses (e.g. Mental Retardation, Mood Disorders, other non-Schizophrenic Psychoses) ranged from excellent to poor. We found high agreement between our rates diagnoses and those assigned by state hospital personnel while patients were living, although post-mortem review produced lower rates of diagnosis of both schizophrenia and Alzheimer-type dementias. Overall, results suggest that the reliability of chart review diagnosis is comparable to that obtained from interviews of live patients when experienced raters are used and diagnostic base rates are high enough to produce stable estimates of reliability.