Imputation of Missing Diagnosis of Diabetes in an Administrative EMR System

Abstract

Administrative electronic medical records (EMRs) contain rich patient data and are an important data source for health informatics studies. Prevalent in such EMRs, poor/missing diagnosis coding is intractable while can be mitigated by imputation techniques. In this work, based on an administrative EMR database in Singapore, we adopted popular machine learning methods to model the relations between diseases and healthcare utilization features, and used the model to impute missing diagnosis of diabetes. Further, this was partially validated with supplementary clinical data. The structured method in this work can be easily extended to other diseases and would benefit other works in health services and research.