A Claims-Based Algorithm for Identifying Hidradenitis Suppurativa Severity.

Purpose: Information on severity of hidradenitis suppurativa (HS) is not available in administrative claims databases. Accurately identifying HS severity in claims data is important for identifying treatment effect modification by severity. We sought to develop and validate a claims-based algorithm to identify patients with mild, moderate, or severe HS. Methods: Mass General Brigham (MGB) electronic health records (EHR) were linked to Medicaid claims data in the US from October 2016 through December 2019 to identify 350 patients aged 10 years and older with an ICD-10 diagnosis code for HS (L73.1). Chart review determined HS severity. A multinomial LASSO regression within a 70% training sample determined the most influential claims-based variables out of 30 candidates associated with mild, moderate, or severe HS. This model was used to calculate the positive predictive values (PPVs) for each level of HS within the hold-out testing sample. Results: The study cohort was predominantly female (81%) aged 18-45 years (74%) with 26% White and 21% Black patients. We identified 72 patients with mild/uncertain HS, 173 with moderate HS, and 105 with severe HS. One ICD-10 diagnosis of HS had a PPV of 89%, which was further improved to 100% when also requiring the concurrent use of a systemic medication for HS. The PPV was 20% for mild/uncertain, 54% for moderate and 67% for severe HS. When combining severity into mild/moderate versus severe the PPV was 71%, indicating that among those classified as severe, 71% were truly severe. Conclusion: The claims-based algorithm has a reasonable performance in identifying severe HS but had limitations distinguishing moderate and mild HS. The algorithm performed best at distinguishing severity when combining mild and moderate versus severe HS.