Prognostic prediction models for psoriatic arthritis in psoriasis: A systematic review

BACKGROUND: Psoriatic arthritis is a chronic inflammatory arthropathy that affects up to 22% of individuals with psoriasis. Identifying patients at higher risk of developing this disease presents a window of opportunity for dermatologists to consider treatments that can limit its progression. OBJECTIVES: This study aims to systematically identify and evaluate the predictive performance, risk of bias and reporting quality of prognostic prediction models developed to estimate the risk of psoriatic arthritis in patients with psoriasis. METHOD: A systematic search was conducted across Medline, Embase, Cochrane, Web of Science, pre-print servers and grey literature, with the search completed on 15 October 2025. Data extraction was guided by the Checklist for Critical Appraisal and Data Extraction for Systematic Reviews for Prediction Modelling Studies (CHARMS). The Prediction model Risk of Bias Assessment Tool (PROBAST) and the Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD-AI) were used to evaluate risk of bias and reporting, respectively. RESULTS: Fifteen studies were included after full-text screening, comprising of 490 models incorporating a variety of predictors. The most frequently used predictors were demographic variables (age, gender, BMI), genetic markers (HLA-B27, HLA-B08, HLA-C*06), clinician and patient-reported outcomes (PASI, DLQI), clinical features and lifestyle factors. Nine studies used machine learning methods, such as Bayesian probabilistic methods and random forest, whilst the remainder used traditional statistical approaches such as regression and risk scores. The C-statistic ranged from 0.53 to 0.99. Calibration, where reported, was modest and sensitivity and specificity were reported in seven studies, often with inconsistent thresholds and reporting. Only three studies performed external validation, which reduced discriminative performance, raising concerns about generalizability. CONCLUSION: Future studies should incorporate a variety of predictors to develop comprehensive models that reflect the aetiology of Psoriatic arthritis whilst ensuring adequate reporting and validation in external cohorts to allow assessment of model performance.