How good are the data? Reliability of one health care data bank.

This study investigates the reliability of the Manitoba Health Services Commission data bank from a variety of perspectives. Emphasizing diagnostic and surgical procedures, the research focuses on those areas in which problems exist and in which the data can be relied upon. Computerized comparisons are stressed, since they can provide cost-effective checks on data quality. One key to performing reliability studies inexpensively is finding information recorded independently: by separate individuals or organizations, at two different times, or in two or more data files. When a particular event has certain logical implications vis-à-vis another, inconsistencies can be located. Face sheet information and data on the performance of major surgical procedures were found to be reliably recorded in the Manitoba data bank. Collapsing ICD-8 diagnosis from medical claims into several categories proved much better than relying upon individual diagnoses. Problems in working with the data included difficulty in distinguishing between closely related surgical procedures and the underreporting of inhospital consultations and nonsurgical procedures.