Machine Learning Models and Risk Scoring System for Survival in Bladder Cancer Patients With Type 2 Diabetes: A 20‐Year Territory‐Wide Cohort Study

Purpose To identify the prognostic risk factors specific to diabetic bladder cancer patients is essential for improving individualized treatment strategies and survival outcomes. This study aimed to identify these risk factors and develop a predictive risk score system for survival. Methods Data from 2120 bladder cancer patients with T2DM were retrieved from the Hong Kong Hospital Authority Data Collaboration Laboratory (2000–2020). Univariable and multivariable Cox regressions were used to identify survival‐associated prognostic factors. Five machine learning algorithms—Cox proportional hazards, CoxPH with lasso penalty, CoxBoost, survival tree (ST), and random survival forest (RSF)—were applied to develop the predictive model. A risk score system was created using the AutoScore‐Survival package for risk stratification. Results Multivariable Cox regression identified significant predictors of mortality: age at cancer diagnosis, diabetes duration, triglycerides, and HbA1c levels. The use of antilipid drugs was associated with reduced mortality risk. The RSF model demonstrated the best predictive performance. The predefined risk score system classified 27.2% of patients as high‐risk, with a 5‐year survival probability of 25.4%, compared to 62.4% for the low‐risk group. The AUCs for 1, 3, and 5 years were 0.713, 0.749, and 0.782, respectively, with a C‐index of 0.716 in the training set. Conclusion This study identified key risk factors affecting the prognosis of diabetic bladder cancer patients and developed a validated risk score system for patient stratification. Further research is needed to validate this system and to explore the underlying biological mechanisms, particularly the potential role of antilipid drugs in improving patient outcomes.