Comorbidity Scores and Machine Learning Methods Can Improve Risk Assessment in Radical Cystectomy for Bladder Cancer.

Area of Science:

• Urology
• Oncology
• Data Science in Medicine

Background:

• Pre-operative risk assessment for radical cystectomy (RC) presents a significant challenge, particularly for elderly patients.
• Accurate prediction of post-operative outcomes is crucial for surgical planning and patient management.

Purpose of the Study:

• To evaluate the predictive capabilities of established comorbidity indices for mortality and morbidity following RC.
• To assess the enhanced predictive performance of machine learning models incorporating comorbidity indices and clinical parameters.

Main Methods:

• Retrospective analysis of 392 patients undergoing open RC, recording complication and mortality rates.
• Evaluation of age-adjusted Charlson Comorbidity Index (aCCI), Elixhauser Index (EI), American Society of Anesthesiologists (ASA) physical status, and Gagne's Comorbidity Index (GCI) using regression.
• Investigation of various machine learning models, including Gaussian Naïve Bayes, logistic regression, neural networks, decision trees, and random forests.

Main Results:

• The aCCI, ASA, and GCI demonstrated significant predictive value for complications and mortality (p<0.01).
• The EI did not show significant predictive ability.
• Area Under the Receiver Operating Characteristic Curves (AUROCs) indicated good prediction of mortality for aCCI (0.81) and ASA (0.78), but poor prediction for complications.
• Machine learning models combining ASA, age, BMI, and sex showed improved prediction, with Gaussian Naïve Bayes (0.79) and logistic regression (0.76) performing best.

Conclusions:

• The ASA and aCCI effectively predict mortality after RC but are less accurate for predicting complications.
• Integrating comorbidity indices with clinical parameters within machine learning frameworks offers a promising approach for improving risk stratification in RC patients.