Early psoriatic arthritis (PsA) prediction and diagnosis remain challenging, and evidence supporting the utility of machine learning (ML) approaches is still limited. This study evaluates the effectiveness of ML in PsA prediction and diagnosis. PubMed, Cochrane Library, Embase, as well as Web of Science were searched until April 15, 2025. Tasks were categorized as either predictive or diagnostic, followed by subgroup analyses stratified according to the datasets. This systematic review included 29 studies, with 15 for predictive tasks and 14 for diagnostic tasks on PsA. The predictive model achieved c-index, sensitivity (SEN), and specificity (SPC) of 0.81 (95% CI 0.78–0.84), 0.77 (95% CI 0.65–0.85), and 0.75 (95% CI 0.55–0.89) in the training set. In the validation set, the c-index, SEN, and SPC were 0.80 (95% CI 0.74–0.85), 0.83 (95% CI 0.76–0.88), and 0.71 (95% CI 0.60–0.80). For the diagnostic tasks, the training set yielded a c-index of 0.85 (95% CI 0.80–0.89), with corresponding SEN and SPC of 0.89 (95% CI 0.77–0.96) and 0.89 (95% CI 0.72–0.96). In the validation set, the c-index was 0.78 (95% CI 0.73–0.83), SEN was 0.66 (95% CI 0.56–0.74), and SPC was 0.76 (95% CI 0.73–0.79). ML shows promising but moderate accuracy for predicting and diagnosing PsA, and that evidence is limited by reliance on internal validation and high risk of bias (RoB) in primary studies. Future research should prioritize external validation in multicenter datasets and assess clinical utility through decision curve analysis.
Sun et al. (2026) studied this question.