301 Psoriatic arthritis risk in psoriasis patients in the Corrona® Psoriasis Registry

Abstract

Background: Understanding characteristics that predict the onset of psoriatic arthritis (PsA) in psoriasis (PsO) patients would help effectively direct treatment. We aimed to develop a model to predict the 2-year PsA risk for real-world PsO patients. Methods: Patients in the prospective, multicenter, non-interventional Corrona Psoriasis Registry without PsA at enrollment and with a 24-month follow-up visit (FU) were included. Demographic and clinical variables collected at enrollment were used to construct logistic regression models to predict PsA diagnosis at FU. Data were randomly partitioned into training (70%) and testing (30%) sets. Models were developed using stepwise forward selection, backward elimination, and elastic net to select predictors. Performance was compared using area under the receiver-operating-characteristic curve (AUC), sensitivity (SE), and specificity (SP). Results: 1489 patients were analyzed (n=1042, training; n=447, test). In the training set, mean age was 49 years, 43% were female, and 119 (11%) developed PsA; those who developed PsA had higher mean Psoriasis Epidemiology Screening Tool (PEST) scores (2.8 vs 1.8), and patient-reported fatigue (33 vs 25, VAS-100) and skin pain (23 vs 19, VAS-100) at enrollment vs those who did not. Nine unique models were constructed; PEST and BMI were common in all. In the testing set, the most predictive model included PEST, BMI, modified Rheumatic Disease Comorbidity Index, work status, alcohol use, and fatigue (AUC= 68.9%, SE=82.9%, SP=48.8%). A more parsimonious model (PEST and BMI only) performed similarly (AUC=68.8%; SE=92.7%, SP=36.5%). Conclusions: While predictive ability was limited, and further refinement and external validation are needed, our findings provide insight for developing a tool to evaluate PsA risk in PsO patients.