Identifying stigmatizing language in hospital birthing care: The ID-STIGMA study

Despite knowledge of obstetric, medical, and social risk factors that contribute to poor perinatal morbidity and mortality outcomes, significant variations in pregnancy outcomes persist across different patient populations. Healthcare delivery patterns and documentation practices vary across healthcare settings and patient populations. Natural language processing (NLP) is a data science approach that has emerged as a promising tool to analyze documentation patterns in clinical documentation. Documentation patterns can influence the quality of care provided to pregnant women. Previous NLP studies have shown that documentation patterns can vary by patient demographics and clinical characteristics. Correctly identifying documentation patterns in the electronic health record (EHR) is crucial, as they can influence the perception of subsequent clinicians reading the note. We propose a study that builds on previous work to examine documentation patterns in clinical notes from hospital birth admissions. The overall objectives of this study are to expand and refine our NLP system to accurately identify documentation patterns in EHR notes, investigate the associations between documentation patterns and morbidity outcomes for pregnant women and newborns, and inform discussions and interventions that lead to institutional change and improved patient outcomes. This secondary analysis will examine EHR notes for all ~35,000 pregnant women admitted to two NewYork-Presbyterian Hospitals from 2020 to 2024. We have assembled a multidisciplinary team with expertise in perinatal epidemiology, NLP and data science, obstetric medicine, linguistics, and community engagement to complete three aims. Aim 1: Expand and refine an existing NLP system to identify documentation patterns in obstetric clinical notes. Aim 2: Apply the NLP system to examine documentation patterns by patient demographic characteristics. Aim 3: Examine associations between documentation patterns and morbidity outcomes for pregnant women and newborns in our study sample, adjusting for sociodemographic, clinical, and care characteristics (e.g., pregnancy complications, length of stay). Outcomes include: administration of medications for preterm birth, low-risk cesarean birth, infection, hemorrhage, APGAR score, and intensive care admission. Study findings will inform future multilevel interventions, including: 1) clinical decision support to optimize documentation quality, 2) education and training of clinicians to improve documentation practices, and 3) realignment of institutional policies and processes to positively influence documentation quality and improve patient care quality.