Evolutionary algorithms create novel Comorbidities Index for prostate cancer mortality prediction

A team of researchers led by Davide Farinati has applied Population-Based Bio-Inspired Algorithms (PBBIAs)—including Genetic Algorithms, FST-PSO, and symbolic regression with GPLearn—to derive a novel, data-driven Comorbidities Index tailored for prostate cancer (PCa) patients being considered for radical prostatectomy (RP). The traditional Charlson Comorbidities Index (CCI), widely used to estimate 10-year mortality, relies on weights that may no longer reflect modern prognoses. This limitation is critical in PCa, where radical treatment should only be offered to patients with at least a decade of life expectancy. The new framework recalibrates comorbidity weights and evolves interpretable symbolic formulations optimized for survival discrimination.

Results show that the best-performing strategies—GA, FST-PSO, and SLIM—outperform both the original CCI and the PCCI (a PCa-specific variant) by up to 0.1 in Concordance Index, particularly when PCa-specific variables are included. GPLearn produces compact, interpretable models with competitive accuracy. This computational approach provides clinicians with an updated, personalized tool to more accurately estimate competing mortality risks, potentially reducing overtreatment in low-risk PCa patients who would not benefit from surgery. The study is published on arXiv as arXiv:2605.31213.