From Nigerian Cancer Registries to Global Impact: Multi-site Validation of an AI Platform for Early Prostate Cancer Prediction in African Men

PURPOSE Prostate cancer is the leading cancer diagnosed and cause of cancer death among Nigerian men; inequities are intensified by late presentation, fragmented records, PSA/MRI/pathology heterogeneity, and limited externally validated AI in African populations. The AICaP study aims to validate a locally grounded, AI-driven prostate cancer risk model in Nigeria and generate evidence for trustworthy, fair, and scalable early detection in LMIC settings. METHODS AICaP is a multi-site retrospective validation study across cancer registries and affiliated clinical systems in Ekiti, Abuja/FCT, Kano, Calabar, and Enugu. Eligible participants are men aged 40–75 with PSA testing, MRI/ultrasound where available, biopsy/pathology records, and longitudinal clinical data from 2010–2024; complementary controls include negative-biopsy and age/site-matched population controls. The Durotimi AI platform applies temporal machine-learning, including a GNN-LSTM framework/architecture, to structured clinical, laboratory, imaging-report, pathology, comorbidity, and encounter data to generate a 0–1 risk probability and clinically interpretable low/medium/high risk strata. Ground truth is biopsy-confirmed prostate adenocarcinoma, with aggressive disease defined by Gleason ≥8, Grade Group 4–5, or early metastatic presentation. The sample size is 5,000–10,000 retrospective records plus 5,000–15,000 controls. Evaluation will include AUC-ROC/AUC-PR, sensitivity, specificity, PPV/NPV, calibration, Brier score, decision-curve analysis, site-level external holdout validation, temporal robustness, and subgroup performance by age, region, ethnicity, urban/rural status, and available socioeconomic proxies. RESULTS The study has established a harmonized Nigerian data abstraction, de-identification, linkage, governance, and quality-control framework for multi-modal AI validation across geographically diverse registry settings. The framework directly addresses local discovery challenges: PSA assay variation, sparse imaging availability, verification bias, missing-not-at-random screening histories, and site-level data drift. Expected outputs include validated discrimination and calibration estimates, subgroup fairness analysis, clinically relevant risk thresholds, and a scalable implementation pathway for registry-to-clinic AI evaluation. CONCLUSION AICaP translates Nigerian registry and clinical data into globally relevant evidence for equitable AI in prostate cancer early detection. By centering African populations, local governance, and practical workflow constraints, the study offers a replicable model for transforming locally generated cancer intelligence into generalizable global cancer-control impact.