Predicting Onset of Psoriatic Arthritis in Patients With Psoriasis

The Psoriasis Epidemiology Screening Tool (PEST) and body mass index (BMI) predicted psoriatic arthritis (PsA) onset in patients with psoriasis during 2 years, according to data from a prospective cohort study published in the Journal of the American Academy of Dermatology.

Investigators obtained data from the CorEvitas Psoriasis Registry, a prospective, multicenter, noninterventional registry for adult patients with psoriasis under the care of a dermatologist. They analyzed data from patients without dermatologist-identified PsA at enrollment, with a subsequent diagnosis of dermatologist-identified PsA, and with a 24-month follow-up visit. Investigators had access to patient characteristics, demographics, and clinical data. The main outcome was a new diagnosis of PsA during the 24-month follow-up period.

To quantify effect sizes for differences between groups, D was calculated. Investigators constructed 14 models to analyze enrollment visit variables to predict dermatologist-identified PsA at 24 months: 8 unregularized logistic regression models (LR1-8) and 6 regularized elastic net (EN1-6) models. To identify which models were best at prediction, performance was compared using area under the receiver-operating-characteristic curve (AUC), sensitivity, specificity, Brier score, Youden’s J, and Mathews’ correlation coefficient (MCC). The optimal discrimination threshold was that which maximized Youden’s J (sensitivity + specificity -1).

There were 1489 patients included in the final analysis: 1042 in the training data set for model development and 447 in the testing data set for internal validation. Overall, the mean age at enrollment was 49.3 years and 42.7% of patients were women. Patients with PsA had a higher mean BMI than those who did not (D =.305).

Patients with PsA had higher modified Rheumatic Disease Comorbidity Index (mRDCI) scores (D =.273), higher mean PEST scores (D =.679), and higher rates of fatigue and skin pain. BMI and PEST were included in the variables selected for all models. The most predictive model, EN5, included PEST, BMI, fatigue, mRDCI, work status (part time), and alcohol use. Its AUC was 68.9%, sensitivity was 82.9% and specificity was 48.8%, respectively. Another model (LR1) with only PEST and BMI performed similarly, researchers noted. Both models had low specificity and AUC values, indicating a high false positive rate and weak discrimination.

The study was limited by possible misclassification bias since PsA diagnoses were not confirmed by rheumatologists, and that treatments were not included in the models.

Study authors wrote that their research is the first step in developing “a model that could potentially be used to predict future PsA and be easily adapted for routine use in the dermatology clinic,” hopefully leading to early disease identification and improved patient outcomes.

Disclosure: Several study authors declared affiliations with biotech, pharmaceutical, and/or device companies. This research was supported CorEvitas, LLC. Please see the original reference for a full list of disclosures.