Abstract: BACKGROUND: Hemophilia is an inherited X-linked bleeding disorder, characterized by the deficiency of coagulation factor VIII (hemophilia A) or factor IX (hemophilia B), leading to recurrent bleeding, chronic joint damage, and functional impairment. Beyond clinical manifestations, hemophilia substantially affects health-related quality of life (HRQoL). While HRQoL assessment is increasingly emphasized, predictive modeling approaches integrating clinical and patient-reported outcomes in hemophilia remain limited. OBJECTIVES: To explore the feasibility of data-driven predictive models for estimating HRQoL outcomes in adults with hemophilia using EuroQol five-dimension five-level (EQ-5D-5L)-derived measures. METHODOLOGY: This pilot, cross-sectional, observational study included adults (≥18 years) with mild or moderate hemophilia A or B attending a tertiary care hemophilia treatment center. HRQoL was assessed using the EQ-5D-5L questionnaire. For exploratory modeling, HRQoL was dichotomized into lower and higher categories based on the median utility score of the study sample. Demographic, clinical, and treatment-related variables were entered as predictors, with EQ-5D domain scores used exclusively for outcome derivation to avoid circularity. Supervised machine-learning models (logistic regression and random forest) were developed and evaluated using internal cross-validation. Model performance was assessed using accuracy, sensitivity, precision, and area under the receiver operating characteristic curve. RESULTS: Fifty participants were included (mean age 38.6 years), with 52% having mild and 48% moderate hemophilia. Poorer HRQoL was more frequent among individuals with moderate disease, target-joint pain, and mobility limitation. Logistic regression demonstrated high sensitivity for identifying the individuals with lower HRQoL, while Random Forest achieved higher overall accuracy and precision. Feature-importance analysis highlighted pain burden, mobility limitation, and disease severity as key contributors to HRQoL classification, with psychological well-being also showing relevance. CONCLUSION: This pilot study demonstrates the feasibility of applying supervised machine-learning models for HRQoL risk stratification in hemophilia using routinely collected clinical data. Disease severity, pain burden, mobility limitation, and psychosocial factors emerged as important determinants of HRQoL. These findings support the potential role of data-driven approaches in complementing traditional clinical assessment and guiding patient-centered care, warranting validation in larger, multicenter cohorts.
Krishnasamy et al. (2026) studied this question.