Co-morbidity data in outcomes research: are clinical data derived from administrative databases a reliable alternative to chart review?

Evaluation of co-morbidity data is essential in health outcomes research. Co-morbidity data derived from administrative databases has been criticized for lacking the accuracy required for clinical research. We compared co-morbidity data derived from a Canadian provincial hospitalization database with chart review in 817 adults treated with a percutaneous coronary intervention at a single tertiary care hospital between 1994 and 1995. While the administrative database tended to under-estimate the prevalence of some co-morbid conditions, the agreement between chart review and administrative data was good to very good for most conditions. Asymptomatic conditions were noted to have lower levels of agreement. Multivariate risk models for all-cause mortality constructed from both data sources were almost identical, suggesting minimal misclassification. The results indicate that clinical data abstracted from most Canadian hospitalization databases can provide reliable information regarding baseline co-morbid conditions believed to influence survival in a population undergoing percutaneous coronary interventions.