Missing Value Imputation Methods for Electronic Health Records

Electronic health records (EHR) are patient-level information, e.g., laboratory tests and questionnaires, stored in electronic format. Compared to physical records, the EHR alternative allows patients access their data easily helps staff with management procedural tasks such as information sharing across different organizations. Moreover, this type of is commonly used by researchers for predictive classification purposes, employing statistical machine learning methods. However, missingness a phenomenon that observed very frequently measurements. Even though often significant, it usually treated poorly either case deletion or simple methods, resulting suboptimal and/or inaccurate results. This happens because k-nearest neighbors (kNN) mean/mode imputation, fail most cases incorporate complex relationships define these medical datasets. To address limitations, paper we test improve state-of-the-art missing imputation models practices. We propose new value method based on denoising autoencoders (DAE) kNN pre-imputation task. optimize training methodology re-applying every N epochs using variable xmlns:xlink="http://www.w3.org/1999/xlink">k each time yield more accurate also revise approach generative adversarial network (GAN). Using baseline, introduce improvements regarding both architecture procedure. These compared ones employed within clinical research studies task post-imputation prediction. Results show our proposed deep approaches outperform standard baselines, yielding better