Early diagnosis of persons with von Willebrand disease using a machine learning algorithm and real-world data

Abstract

ABSTRACT Background Von Willebrand disease (VWD) is underdiagnosed, often delaying treatment. VWD claims coding is limited, including no severity qualifiers; improved identification methods for VWD are needed. This study’s aim: identify and characterize undiagnosed symptomatic persons with VWD in the US using medical insurance claims to develop predictive machine learning (ML) models. Research design andmethods Diagnosed andpotentially undiagnosed VWD cohorts were defined using Komodo longitudinal USclaims data (January 2015-March 2020). ML models were built using keycharacteristics predictive of VWD diagnosis from the diagnosed cohort. Two MLmodels predicted VWD diagnosis with the highest accuracy in females (randomforest; 84%) and males (gradient boosting machine; 85%). Undiagnosed personssuspected to have VWD were identified using an 80% cutoff probability; profilesof key characteristics were constructed. Results The trained MLmodels were applied to the undiagnosed cohort (28,463 females; 20439 males)with suspected VWD. 52% of undiagnosed females had heavy menstrual bleeding, akey pre-diagnosis symptom. Undiagnosed males tended to have more frequentmedical procedures, hospitalizations, and emergency room visits compared withundiagnosed females. Conclusions ML algorithmssuccessfully identified potentially undiagnosed symptomatic people with VWD,although many may remain undiagnosed and undertreated. External validation ofthe algorithms is recommended.