Identifying racial disparities in adverse dispositions following major surgery for prostate cancer using machine learning.
Atulya Aman Khosla,Nitya Batra,Mohammad Arfat Ganiyani,Karan Jatwani,Rohit Singh,Lakshmi Prajwala Chedella Venkata,Mukesh Roy,Venkataraghavan Ramamoorthy,Muni B. Rubens,Anshul Saxena,Rohan Garje,Ishmael A. Jaiyesimi
DOI: https://doi.org/10.1200/jco.2024.42.4_suppl.269
IF: 45.3
2024-02-01
Journal of Clinical Oncology
Abstract:269 Background: Machine learning has shown potential in accurately predicting outcomes after urologic surgery, allowing for improved patient selection, risk stratification, and predicting discharge dispositions. Healthcare disparities pose a major public health challenge, leading to unequal health outcomes for different patient groups. Successful measures to mitigate these disparities rely on precise recognition of the demographic groups impacted. In this study, we utilized AI and ML models to predict such adverse outcomes so that early intervention may lead to timely and optimal prostate cancer care. Methods: Using the 2016–2021 ACS National Surgical Quality Improvement Program (NSQIP) data, we identified 84,064 patients who underwent prostatectomy or surgeries related to excision of the prostate. We used Gradient Boosted Trees, Random Forests, and Deep Learning to examine the association between various variables from demographic, clinical, preoperative labs, past medical history domains, and non-home discharge disposition after cancer surgery. Accuracy rates were utilized to compare various models. Results: The median [Q1, Q3] age of the sample was 65.0 [59.0, 69.0] years. The total median [Q1, Q3] hospital length of stay and operation time of the sample was 1.0 [1.0, 2.0] days and 200.0 [153.0, 257.0] minutes, respectively. White non-Hispanic (WNH) patients constituted 62% (n = 52,444) of the sample, and Black non-Hispanic (BNH) were 11% (n = 9547). Smoking prevalence was higher among BNH cases (18.6%) as compared to WNH cases (10.9%). The proportion of cases requiring hypertension medication was 66.6% among BNH compared to 52.6% among WNH. The 30-day readmission rates for WNH, BNH, and White Hispanic (WH) cases were 6.4%, 6%, and 5.6%, respectively. The mortality rates for WNH, BNH, and WH cases were 0.3%, 0.3%, and 0.2%, respectively. The non-home discharge rates for WNH, BNH, and WH cases were 19.8%, 20.6%, and 22.4%, respectively. The machine learning model results showed that the Deep Learning model performed best in predicting discharge disposition after prostatectomy for prostate cancer, with an accuracy of 80.48%. Race was present in the list of important factors, indicating that race had a significant association with the outcome. Conclusions: Our study demonstrated that using a national registry, ML models can accurately identify patient populations affected by healthcare disparities. We found that racial disparities in adverse dispositions persisted after adjusting for covariates. Interventions aimed at reducing these disparities are warranted to improve the quality of life of this subgroup of patients. [Table: see text]
oncology