Predicting Wilson’s Disease Progression using Machine Learning with Real-World Electronic Health Records

Abstract

AbstractBACKGROUND & AIMSWilson’s disease (WD) is a rare genetic disorder causing excessive copper accumulation. Research on the natural history of WD is limited. Our objective was to identify predictors for WD progression to cirrhosis, liver failure, and death and to predict individual risk of progression to these endpoints at 1, 2, 3, and 5 years after WD diagnosis.METHODSA retrospective natural history cohort study of adult patients with first-recorded WD diagnosis was conducted using the US Optum EHR data between 1/1/2007 and 6/30/2020. LASSO Cox regression, Random Survival Forest (RSF), and XGBoost (XGB) models were used to identify important predictors for progression to cirrhosis, liver failure, and death. The strong predictors for each outcome identified through weighted average rankings across models and reviewed by clinical experts were used for patient-level prediction using RSF and XGB models. The resulting models were validated with an independent sample cohort. C-index and dynamic AUCs were used to evaluate model performance.RESULTSOver the study period, 310 out of 2,901 WD patients developed cirrhosis, 255 out of 3,251 developed liver failure, and 604 out of 3,559 died. Age at WD diagnosis, alcoholism, AST and bilirubin levels within 3 months of WD diagnosis, and neurologic and hepatic conditions were the most common predictors for progression to the study endpoints. XGB had a slight superior predictive performance compared with RSF and was then used to predict individual risks for progression to the study endpoints with the top ensemble predictors. The dynamic AUC was 0.78 at Year 1, 0.74 at Year 2, 0.72 at Year 3 and 0.72 at Year 5 for cirrhosis; 0.82 at Year 1, 0.78 at Year 2, and 0.77 at both Year 3 and Year 5 for liver failure; 0.81 at Year 1, 0.83 at Year 2, and 0.82 at both Year 3 and Year 5 for death.CONCLUSIONSThis study identified the most influential clinical predictors and assessed patient-level risk of WD progression using machine learning. Results from machine learning prognostic models will increase understanding of disease natural history and may help improve clinical trial design and guide individualized clinical care.