Predicting Toxicities and Survival Outcomes in De Novo Metastatic Hormone-Sensitive Prostate Cancer Using Clinical...

Area of Science:

• Oncology
• Medical Informatics

Background:

• Conventional prognostic factors for metastatic hormone-sensitive prostate cancer (mHSPC) are assessed at diagnosis.
• Variations in vital signs and laboratory parameters during treatment may predict prognosis and toxicity.

Purpose of the Study:

• To investigate the prognostic value of dynamic changes in vital signs and laboratory parameters during systemic treatment for mHSPC.
• To compare the performance of machine learning models using static versus dynamic variables for predicting outcomes and toxicities.

Main Methods:

• Retrospective study of 363 de novo mHSPC patients (2014-2023).
• Systematic collection of clinical and laboratory data, grading variations using CTCAE V5.0.
• Cox regression and machine learning models (SVC, Random Forest, LGBM) were used to analyze dynamic variables' impact on progression-free survival (PFS), overall survival (OS), and organ-specific toxicities.

Main Results:

• Dynamic models did not improve prediction of single organ-specific toxicities.
• Hematological, liver, kidney toxicity, and electrolyte disturbances were associated with shorter PFS and/or OS.
• Increasing alkaline phosphatase, decreasing albumin, and hyponatremia predicted shorter OS.
• Integrating static and dynamic variables significantly improved ML model ability to identify poor responders (PFS prediction AUC 0.91-0.94).

Conclusions:

• Integrating dynamic prognostic factors with conventional ones can enhance patient stratification and survival prediction in mHSPC.
• Further multicenter validation studies are necessary to confirm these findings.