Disparities in inpatient treatment delivery for gynecologic cancers: A national analysis using machine learning.

Abstract

e13694 Background: Patients with gynecologic malignancies are frequently hospitalized for cancer-related and unrelated indications. Treatment varies by cancer type: endometrial and ovarian cancers are managed with surgery and adjuvant chemotherapy; early cervical cancer requires surgery while locally advanced disease requires chemoradiation; vulvar and vaginal cancers require surgical resection with or without radiation. We used machine learning to identify factors associated with treatment receipt among hospitalized patients. Methods: We queried the National Inpatient Sample (2016-2021) for hospitalizations with cervical, ovarian, endometrial, vulvar, or vaginal cancer. Treatment was defined as gynecologic surgery, staging/debulking, lymphadenectomy, chemotherapy, or radiotherapy during admission. A gradient boosting classifier predicted treatment receipt with SHAP analysis for feature importance. Results: Among 188,709 admissions (mean age 62; 66% White, 15% Black, 11% Hispanic), 34% included treatment. Among treated patients, 67% received gynecologic surgery, 47% oophorectomy/salpingectomy, 17% lymphadenectomy, 9% omentectomy, 4% chemotherapy, and 1% radiotherapy. Treatment rates by cancer: endometrial 38%, ovarian 33%, cervical 26%, vulvar 17%. Elective admissions were more likely to include treatment than non-elective (77% vs 13%). Treatment was more frequent at urban teaching vs rural hospitals (37% vs 16%) and among privately insured vs Medicare patients (44% vs 28%). Model AUC: 0.906. Top predictors: admission type, teaching status, age, cancer type, payer. Conclusions: Treatment delivery varies substantially by admission context and hospital setting. These system-level factors may represent targets for improving care coordination and access.