Personalized prediction of psoriasis relapse post-biologic discontinuation: a machine learning-driven population cohort study

Abstract

Background: Identifying the risk of psoriasis relapse after discontinuing biologics can help optimize treatment strategies, potentially reducing relapse rates and alleviating the burden of disease management.

Objective: To develop and validate a personalized prediction model for psoriasis relapse following the discontinuation of biologics.

Methods: This study enrolled patients who achieved remission following biologic therapy. Relapse predictors were identified using the Boruta algorithm combined with multivariate Cox regression. A nomogram and an online calculator were created to aid in the visualization and computation of outcomes. The model's performance was thoroughly assessed using Receiver Operating Characteristic (ROC) curves, Area Under the Curve (AUC), C-statistics, calibration plots, and Decision Curve Analysis (DCA).

Results: The study included 597 patients, with 534 in the derivation cohort and 63 in the validation cohort. Anxiety, disease duration, prior biologic treatments, treatment duration, time to achieve PASI 75, and maximum PASI response were identified as influential factors for relapse and were incorporated into the model. Both internal and external evaluations indicate that the model exhibits good predictive accuracy.

Conclusion: A multivariate model leveraging standard clinical data can relatively accurately predict the risk of psoriasis relapse post-biologic discontinuation, guiding personalized treatment strategies.

Keywords: Psoriasis; biologics; machine learning; prediction model; relapse.