Impact of the COVID-19 pandemic on outcomes in malignant bowel obstruction: A National Inpatient analysis using explainable machine learning (2020–2023).

Abstract

e13572 Background: Malignant bowel obstruction (MBO) is a high-mortality complication of advanced abdominal and pelvic malignancies, most commonly ovarian, colorectal, pancreatic, and gastric cancers. Obstruction typically arises from peritoneal carcinomatosis, direct tumor invasion, or extrinsic compression, and frequently reflects advanced-stage disease. The COVID-19 pandemic disrupted oncology screening and care delivery. We utilized explainable artificial intelligence (XAI) to develop a mortality prediction model and to characterize temporal changes in mortality risk factors from the peak-pandemic period (2020–2021) to the post-pandemic recovery period (2022–2023). Methods: We performed a retrospective analysis of the National Inpatient Sample (NIS) from 2020 to 2023. MBO was defined by ICD-10-CM intestinal obstruction codes in hospitalizations with gastrointestinal, genitourinary, or pelvic malignancies, including colorectal, pancreatic, gastric, ovarian, uterine, and cervical cancers. An XGBoost model was trained (70:30 split) to predict in-hospital mortality using 45 clinical, demographic, hospital, and socioeconomic features. Model interpretability was assessed using SHapley Additive exPlanations (SHAP) to rank feature importance. Temporal comparisons were performed between 2020–2021 and 2022–2023. Results: Among 84,472 MBO hospitalizations, the XGBoost model achieved an AUC of 0.85 (95% CI, 0.83–0.87), outperforming the Charlson Comorbidity Index (AUC 0.68). In 2020–2021, active COVID-19 infection was a leading contributor to mortality risk. In 2022–2023, the relative contribution of COVID-19 decreased, while features consistent with advanced disease and physiologic compromise (e.g., peritoneal carcinomatosis, cachexia, sepsis) increased in importance, suggesting a shift toward higher acuity at presentation. Socioeconomic disparities persisted; the lowest household income quartile remained a top-10 predictor of mortality across all years (p < 0.05). Conclusions: This XAI framework provides a robust tool for mortality risk stratification in MBO across diverse abdominal and pelvic tumor types. Shifts in feature importance from 2020–2021 to 2022–2023 suggest evolving drivers of mortality in emergency oncology admissions following pandemic-related disruptions in care.