RESUMO
BACKGROUND: Severe maternal morbidity is a composite measure of serious obstetric complications that is often identified in administrative data using the International Classification of Diseases (ICD) diagnosis and procedure codes for a set of 21 indicators. Prior studies of screen-positive cases have demonstrated low predictive value for ICD codes relative to the medical record. To our knowledge, the validity of ICD-10 codes for identifying severe maternal morbidity has not been fully described. METHODS: We estimated the sensitivity, specificity, positive predictive value, and negative predictive value of ICD-10 codes for severe maternal morbidity occurring at delivery, compared with medical record abstraction (gold standard), for 1,000 deliveries that took place during 2016-2018 at a large, public hospital. RESULTS: We identified a total of 67 cases of severe maternal morbidity using the ICD-10 definition and 74 cases in the medical record. The sensitivity was 26% (95% confidence interval [CI] = 16%, 37%), the positive predictive value was 28% (95% CI = 18%, 41%), the specificity was 95% (95% CI = 93%, 96%), and the negative predictive value was 94% (95% CI = 92%, 96%). CONCLUSIONS: The validity of ICD-10 codes for severe maternal morbidity in our high-burden population was poor, suggesting considerable potential for bias.