COMPARING ACTIVE SURVEILLANCE AND WATCHFUL WAITING AND RADICAL TREATMENT USING MACHINE LEARNING MODELS AMONG PROSTATE CANCER PATIENTS.

In 2021, 59.6% of low-risk prostate cancer patients were under active surveillance as their first course of treatment. However, active surveillance and watchful waiting are difficult to define in population-based cohorts. The primary aim of our study is to develop and validate a population-level machine learning model for distinguishing active surveillance (AS) and watchful waiting (WW) in the conservative treatment group. A secondary aim is to investigate initial cancer management trends from 2004 to 2017 and the risk of chronic diseases among prostate cancer patients with different treatment modalities. A cohort of 18,134 cancer patients with prostate adenocarcinomas were diagnosed between 2004 and 2017 in Utah Cancer Registry. As a subset, 1,926 patients with available AS/WW information were diagnosed from 2010 to 2017 from the Surveillance, Epidemiology, and End Results Prostate with Watchful Waiting Database. Models were trained by four machine learning algorithms, and 10-fold cross-validation was performed. The area under the receiver operating curve, F-score, Brier score, and accuracy were used for model evaluation. Comorbidities diagnoses were identified from electronic medical records and statewide healthcare facilities data. Cox proportional hazard models were used to estimate hazard ratios (HRs) for the risk of chronic diseases. Logistic regression models performed better than other models in identifying AS from WW accurately. The developed model achieved a test area under the receiver operating curve of 0.73 (range 0.68-0.79), F-score of 0.79, accuracy of 0.71(range 0.66-0.76), and Brier score of 0.29 demonstrating good calibration, precision, and recall values. The top predictors were clinical factors, including Gleason Grade groups (p=0.0003), AJCC stage (p=0.017807), and T stage (p=0.0000109), and demographic characteristics, including race (p=0.007904) and birth year (p=0.046533). We noted a sharp increase in AS use between 2004 and 2016 among patients with low-risk prostate cancer and a moderate increase among intermediate-risk patients between 2008 and 2017. Compared to the AS group, radical treatment was associated with a lower risk of prostate cancer-specific mortality;but higher risks of Alzheimer's disease, anemia, glaucoma, hyperlipidemia, and hypertension. A machine learning approach accurately distinguished AS and WW groups in conservative treatment in this decision analytical model study. Our results provide insight into the necessity to separate AS and WW in population-based studies. [ABSTRACT FROM AUTHOR]