Machine Learning Models for Disease Activity Estimation

This technical report presents the development of machine learning models aimed at estimating validated measures of disease activity and symptom severity for four chronic conditions using real-world data (RWD). The objective is to assess the feasibility of creating these models based on clinical notes from RWD sources. The findings indicate that the application of these models significantly increases the number of patients available for analyses of disease progression and treatment response, with increases ranging from 1.5 to 19 times. The models demonstrated high correlation with collected disease severity scores, achieving area under the curve (AUC) values of 0.91 for both the Systemic Lupus Erythematosus Disease Activity Index and the Expanded Disability Status Scale, and 0.81 and 0.85 for the Patient Health Questionnaire-9 and New York Heart Association classification, respectively. The report emphasizes the importance of utilizing machine learning techniques to enhance the estimation of disease activity measures, suggesting that future research should further explore the strengths and limitations of these models in RWD studies.