Imputation of Clinical Covariates in Time Series

• DOI: 10.48550/arxiv.1812.00418
• Publication Date: 2018-01-01
• Authors: Dimitris Bertsimas; Agni Orfanoudaki; Colin Pawlowski
• Keywords: Imputation (statistics); Categorical variable

Missing data is a common problem in real-world settings and particularly relevant healthcare applications where researchers use Electronic Health Records (EHR) results of observational studies to apply analytics methods. This issue becomes even more prominent for longitudinal sets, multiple instances the same individual correspond different observations time. Standard imputation methods do not take into account patient specific information incorporated multivariate panel data. We introduce novel algorithm MedImpute that addresses this problem, extending flexible framework OptImpute suggested by Bertsimas et al. (2018). Our provides imputations sets with missing continuous categorical features, we present formulation implement scalable first-order $K$-NN model. test performance our on from Framingham Heart Study when are completely at random (MCAR). demonstrate leads significant improvements both accuracy downstream model AUC compared state-of-the-art