Prognostic prediction models for psoriatic arthritis in psoriasis: A systematic review.
Fatimah Khoda, Yasmin Khan, William W Y Haw et al.
✦ AI-curated · Sources linked
In 30 seconds
This systematic review evaluated the predictive performance, risk of bias and reporting quality of models estimating psoriatic arthritis risk in people with psoriasis. Fifteen studies contributed 490 models, using demographic, genetic, clinical, patient-reported and lifestyle predictors; nine studies used machine-learning methods. Discrimination varied widely, with C-statistics ranging from 0.53 to 0.99. Calibration was modest where reported, and only three studies externally validated their models; discrimination declined in those validations, raising concerns about generalisability.
Key findings
- Across 15 studies, 490 models were assessed, with C-statistics ranging from 0.53 to 0.99.
- Nine studies used machine-learning methods; the others used traditional statistical approaches, including regression and risk scores.
- Only three studies performed external validation, where discriminative performance declined.
- Calibration was modest where reported; sensitivity and specificity appeared in seven studies, often with inconsistent thresholds and reporting.
Why it matters
Risk prediction could help identify people with psoriasis who may develop psoriatic arthritis, but the wide range in model discrimination and limited external validation make performance uncertain across settings. The review highlights the need for transparent reporting and validation in independent cohorts before models can reliably inform risk assessment.
What to keep in mind
Only three studies externally validated their models, and discrimination declined in those validations. Calibration was modest where reported, while sensitivity and specificity reporting was inconsistent.