Abstract
Previous research has relied on International Classification of Diseases (ICD) codes to define maternal injuries. However, the validity of these codes remains unclear. We aimed to validate ICD-10 codes used to ascertain maternal injuries using medical chart reviews as the gold standard. A retrospective cohort study of all births occurring at Atrium Health Wake Forest Baptist Medical Center in 2022-2023. We randomly selected 100 subjects with ICD-10-indicated injury and 100 subjects without indication of injury. Two independent reviewers, blinded to the ICD-10-based classification, conducted the chart review. We examined the validity of relevant injury-related codes (V00-Y38; S00-T79; O9A.2-O9A.4) and calculated positive predictive values (PPVs) for different algorithms defined by varying the encounter type and the list of codes used. The algorithm that included all injury-related ICD-10 codes without encounter-type restrictions showed moderate PPV (71% [95% CI, 61%-79%]) and high negative predictive value (96% [90%-98%]). PPV was maximized when including codes V00-Y38 and restricting encounter type to inpatient or emergency department encounters (PPV 100% [93%-100%]). This study characterizes the accuracy of ICD-10-based algorithms for ascertaining maternal injuries during pregnancy. These findings can help improve inference by providing bias parameters for future research.