Improving US maternal mortality reporting by analyzing literal text on death certificates, United States, 2016-2017.

Changes in data collection and processing of US maternal mortality data across states over time have led to inconsistencies in maternal death reporting. Our purpose was to identify possible misclassification of maternal deaths and to apply alternative coding methods to improve specificity of maternal causes. We analyzed 2016-2017 US vital statistics mortality data with cause-of-death literals (actual words written on the death certificate) added. We developed an alternative coding strategy to code the "primary cause of death" defined as the most likely cause that led to death. We recoded deaths with or without literal pregnancy mentions to maternal and non-maternal causes, respectively. Originally coded and recoded data were compared for overall maternal deaths and for a subset of deaths originally coded to ill-defined causes. Among 1691 originally coded maternal deaths, 597 (35.3%) remained a maternal death upon recoding and 1094 (64.7%) were recoded to non-maternal causes. The most common maternal causes were eclampsia and preeclampsia, obstetric embolism, postpartum cardiomyopathy, and obstetric hemorrhage. The most common non-maternal causes were diseases of the circulatory system and cancer, similar to the leading causes of death among all reproductive-age women (excluding injuries). Among 735 records originally coded to ill-defined causes, 94% were recoded to more specific, informative causes from literal text. Eighteen deaths originally coded as non-maternal mentioned pregnancy in the literals and were recoded as maternal deaths. Literal text provides more detailed information on cause of death which is often lost during coding. We found evidence of both underreporting and overreporting of maternal deaths, with possible overreporting predominant. Accurate data is essential for measuring the effectiveness of maternal mortality reduction programs