Predicting Moderate-to-Severe Plaque Psoriasis From Prescription and Procedure Electronic Health Record Data
Author(s)
Rasouliyan L, Althoff A, Kumar V, Black D
OMNY Health, Atlanta, GA, USA
Presentation Documents
OBJECTIVES: The objective of this research was to utilize only prescription and procedure data from electronic health records (EHRs) to distinguish moderate and severe plaque psoriasis (PsO) from mild PsO in patients from specialty dermatology networks in the United States. METHODS: Patients from 6 specialty dermatology networks within the OMNY Health real-world data platform from 2017 to 2024 with an assessment of body surface area (BSA) directly associated with PsO were included. Encounters were categorized into mild (BSA < 3%), moderate (BSA 3-10%), and severe (BSA > 10%) PsO; moderate and severe categories were combined. Logistic regression with 5-fold cross validation was employed where moderate/severe status was modeled as a function of prescriptions (topical corticosteroids, other topical agents, oral corticosteroids, cyclosporine, methotrexate, tapinarof, apremilast, deucravacitinib, biologics) and procedure codes (phototherapy, moderate or complex disease management) that occurred at the same encounter. Final variables were selected based on their strengths of association with the outcome. Model performance was assessed by the cross-validated area under the receiver-operating characteristic (AUROC) curve. RESULTS: Of 298,326 PsO patients, 126,494 had at least 1 BSA assessment comprising 328,107 unique encounters. Final selected variables positively associated with moderate/severe PsO status were prescriptions for topical corticosteroids, oral corticosteroids, other topical agents, and apremilast and procedure codes for phototherapy and moderate or complex disease management. The cross-validated AUROC achieved was 0.61, indicating poor to fair discrimination between moderate/severe and mild disease severity. CONCLUSIONS: The ability to identify moderate and severe PsO patients in EHR data is important to study advanced therapies in the real-world setting. Same-day prescription and procedure data alone may not be sufficient to distinguish moderate/severe from mild PsO. Further research to develop a more discriminative algorithm may include integrating patient history, demographics, and natural language processing of clinical notes.
Conference/Value in Health Info
2024-05, ISPOR 2024, Atlanta, GA, USA
Value in Health, Volume 27, Issue 6, S1 (June 2024)
Code
CO23
Topic
Clinical Outcomes, Epidemiology & Public Health, Methodological & Statistical Research, Study Approaches
Topic Subcategory
Artificial Intelligence, Machine Learning, Predictive Analytics, Clinical Outcomes Assessment, Disease Classification & Coding, Electronic Medical & Health Records
Disease
Systemic Disorders/Conditions (Anesthesia, Auto-Immune Disorders (n.e.c.), Hematological Disorders (non-oncologic), Pain)