Comparative Analysis of Machine Learning Models for Risk Prediction of Prostate Cancer in African Ancestry: A Systematic Review

Abstract

Men of African ancestry experience disproportionately higher rates of prostate cancer (PCa) incidence, aggressiveness, and mortality compared to other populations. Although machine learning (ML) has advanced cancer risk prediction, most models have been developed using non-African datasets, limiting their clinical relevance and equity for African populations. This study systematically reviewed and compared ML models for PCa risk prediction among individuals of African descent, assessing model performance, data sources, and population-specific adaptations. A comprehensive literature search was conducted following PRISMA guidelines across PubMed, Scopus, and Web of Science for studies published between 2020 and 2025. Of 248 identified records, 12 studies met the inclusion criteria. Data were extracted on ML techniques, model performance metrics, utilised biomarkers, and population representation. ML models incorporating African-specific biomarkers (e.g., PVT1, PCA3) and genomic variants (e.g., 8q24, HOXB13) demonstrated enhanced predictive performance, achieving area under the curve (AUC) values greater than 0.80 and reducing false negative rates by 18-22%. Multi-modal models integrating genomic, histopathological, and clinical data outperformed single-source approaches. However, the underrepresentation of African cohorts and limited external validation remain major limitations. Tailoring ML models to African populations significantly improves PCa risk prediction. Addressing data diversity, enhancing external validation, and mitigating algorithmic bias are crucial steps toward equitable and effective clinical implementation.

Ähnliche Arbeiten