OBJECTIVES: To review prognostic models for predicting the onset of psoriatic arthritis in psoriasis patients using machine learning techniques.

METHODS: The review included articles published from January 1982 to October 2023, sourced from Pubmed, Embase, Web of Science, and Epistemonikos databases. The systematic review was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. Studies published in the English language, reporting outcomes related to predicting psoriatic arthritis in psoriasis patients using machine learning, were included. The extraction was conducted using the CHARMS checklist and risk of bias was assessed by PROBAST checklist. The discriminative ability of models was compared using Area Under the Receiver Operating Characteristic Curve (AUROCC).

RESULTS: A total of 13 studies were included in the review from an initial pool of 98 studies identified through abstract and full-text screening, as well as bibliographic search. These comprised of 6 studies focusing on clinical features, 4 studies on molecular profiles, 2 studies combining clinical features and molecular profiles, and 1 study assessing musculoskeletal performances. Out of the 13 studies, only 4 (30.77%) conducted external validation of their proposed models. Ten studies demonstrated good discriminative ability (AUROCC ≥ 0.7) in predicting psoriatic arthritis. Models based on molecular profiles of differentially expressed genes (DEGs) and a combination of DEGs and differentially expressed proteins (DEPs) showed excellent discrimination (AUROCC=1). One study reported their model’s accuracy as high as 99.82%, while the remaining two studies had models with AUROCC < 0.7, indicating poor discrimination.

CONCLUSIONS: Machine learning has proven to be an effective tool in predicting the onset of psoriatic arthritis in patients with psoriasis. While a majority of the models displayed good discriminative ability, the importance of external validation is emphasized to enhance the clinical utility of these models.