Meta-analysis of artificial intelligence models for predicting prostate cancer outcomes

BackgroundProstate cancer is the sixth most common cause of cancer-related death in males globally and is often identified as a secondary malignancy. Early detection and appropriate management increase the patient’s life expectancy. In predicting prostate cancer outcomes, artificial intelligence models are currently employed increasingly, although they are not widely applicable due to an unclear understanding of their performances. This systematic review and meta-analysis examined the pooled predictive performances of these models using area under the curve (AUC) metrics across different prediction endpoints: overall survival (OS), progression-free survival (PFS), toxicity or quality of life (TQL), recurrence distance metastasis (RDM), and treatment response (TR).MethodsPubMed and Scopus were systematically searched for studies that reported prediction models for prostate cancer outcomes. In accordance with PRISMA criteria, we carried out a systematic review of prediction models for prostate cancer, with a registered protocol in PROSPERO (CRD42025611480). Using the Prediction Model Quality Score (PMQS), qualified studies were assessed for methodological quality. We conducted a random-effects meta-analysis, pooling AUCs for each prediction endpoint; heterogeneity was assessed using I², τ², and Q statistics. Funnel plots, Egger’s regression test, and the trim-and-fill method were used to assess publication bias.ResultsA total of 144 studies were selected based on the eligibility criteria for systematic review, but only 85 were included in the meta-analysis. The overall pooled AUC in all prediction endpoints were 0.808, 0.792, 0.845, 0.835, and 0.805 for OS, PFS, RDM, TR, and TQL, respectively, indicating good predictive performance. Funnel plots showed no significant asymmetry.ConclusionThe models evaluated in this review demonstrated over 75% accuracy in predicting prostate cancer with varied performance across prediction endpoints. Standardized model reporting and robust external validation are needed to improve reproducibility and clinical translation.Systematic Review Registrationhttps://www.crd.york.ac.uk/PROSPERO/view/CRD42025611480, identifier CRD42025611480.