2. UroToday Home
  3. Recent Abstracts
  4. Urologic Oncology
  5. Bladder Cancer

Machine Learning Decision Support Model for Radical Cystectomy Discharge Planning - Beyond the Abstract

Radical cystectomy (RC) remains one of the most morbid and costly oncologic surgeries with high readmission and complication rates.1  Studies have shown that appropriate and timely discharge placement is key to reducing this burden.2  However, determining optimal discharge location in the perioperative period remains challenging even with established risk stratification methods such as the 11-point modified Frailty Index (FI), the Charlson Comorbidity Index (CCI), and American Society of Anesthesiologists (ASA) class. These indices, while easy to use and validated for their intended purposes, have been shown to have poor prognostic ability in the nuanced process of RC discharge planning, which requires the incorporation of a much wider range of patient data. Historically, such large multivariate clinical models were impractically tedious to use and difficult to build. However, the proliferation of the electronic medical record (EMR) and machine learning techniques are rapidly changing that paradigm. Clinical models incorporating dozens of patient variables can now seamlessly integrate into the EMR, providing real-time insights for clinicians.3


We believe this approach can add significant value to RC discharge planning, which can be both a clinically and logistically challenging process that benefits from additional lead-time. Our study aimed to use contemporary machine learning techniques (gradient boosted trees) to build a predictive model for non-home discharge among patients recovering from radical cystectomy. Our model provides clinicians with a prediction probability based on 38 pre-operative variables and updates predictions based on 18 post-operative “in-house” complications. It was trained on almost 12,000 patients from the nationally validated ACS-NSQIP dataset and demonstrates an area under the receiver operating characteristic curve (AUC) of 0.80, a recall of 76%, and a precision of 21% in unseen test data. These results show strong predictive power in identifying patients that require elevated care at discharge and perform far better than traditional indices (FI, CCI, ASA) tested on the same data.4

While our model requires further validation and calibration as well as a strategy for implementation, we believe this proof-of-concept work demonstrates the utility of machine learning models to synthesize a wide range of data to provide novel insight into post-operative RC care. Discharge planning, especially for highly morbid procedures, is often more of an art than a science. Evidence-based tools are needed to provide timely guidance to ensure patients receive the care they need. As we strive to provide more evidence-based care to our patients, we would be remiss not to leverage EMR data in this way. We hope this work provides the basis for further radical cystectomy clinical decision support models and encourages future work in this space.

Written by: Calvin C. Zhao, MD1 & Richard S. Matulewicz, MD2

  1. New York University Grossman School of Medicine, Department of Urology, New York, NY
  2. Memorial Sloan Kettering Cancer Center, Department of Surgery, Urology Service, New York, NY

References:

  1. Shabsigh A, Korets R, Vora KC, et al. Defining early morbidity of radical cystectomy for patients with bladder cancer using a standardized reporting methodology. Eur Urol. Jan 2009;55(1):164-74. doi:10.1016/j.eururo.2008.07.031
  2. Hu M, Jacobs BL, Montgomery JS, et al. Sharpening the focus on causes and timing of readmission after radical cystectomy for bladder cancer. Cancer. May 1 2014;120(9):1409-16. doi:10.1002/cncr.28586
  3. Razavian N, Major VJ, Sudarshan M, et al. A validated, real-time prediction model for favorable outcomes in hospitalized COVID-19 patients. NPJ Digit Med. 2020;3:130. doi:10.1038/s41746-020-00343-x
  4. Meng X, Press B, Renson A, et al. Discriminative Ability of Commonly Used Indexes to Predict Adverse Outcomes After Radical Cystectomy: Comparison of Demographic Data, American Society of Anesthesiologists, Modified Charlson Comorbidity Index, and Modified Frailty Index. Clinical Genitourinary Cancer. 2018/08/01/ 2018;16(4):e843-e850. doi:https://doi.org/10.1016/j.clgc.2018.02.009

Read the Abstract