Comprehensive Summary
This article evaluates the validity of EPICS, an AI-driven scoring system designed to more accurately measure moderate to severe symptoms of Erythrodermic psoriasis for better matched solutions of care for each patient. This retrospective cohort study included patients at Peking Union Medical College Hospital between 2005 and 2022, who were at least 12 years old, with a definitive diagnosis of psoriasis, and generalized erythematous plaques covering at least 90% body surface area. This new EPICS scoring system combined both acute and systemic systems and lab tests to then test machine learning models to delineate which factors best predicted severity. The severity of edema was graded by 4 levels and measurements from key symptoms and laboratory markers were standardized and verified with medical records. K-means clustering categorized the disease severity into mild vs moderate-to-severe. The SHapley Additive exPlanations method was used to calculate values for all features and helped identify the top 10 accurate indicators to assess disease severity. The findings revealed that the EPICS was more reliable in comparison to the 2018-EPSA in predicting whether patients would need systemic therapy and length of hospitalization stay. Overall, EPICS provides an efficient, data driven tool to help doctors in quickly assessing disease severity and making beneficial treatment decisions.
Outcomes and Implications
With the recent improvements in AI to make diagnostic decisions, the question of its reliability and practicality is critical. In the past, there are models such as PASI and BSA that take into account surface area, however, both lack the consideration of systemic implications of EP and are largely time consuming, subjective tools. The study concluded that EPICS has the potential to serve as an extremely helpful tool in dermatology, to predict severity in erythrodermic psoriasis for improved treatment decisions and personalized care. This study’s findings may not be generalized beyond this single-center, retrospective cohort so external validation is needed and long-term follow up would allow evaluation of survival and quality of life.