Similarity Computing on Electronic Health Records
Authors:
Abstract
Similarity computing on real world applications like Electronic Health Records (EHRs) can reveal numerous interesting knowledge. Similarity measures the closeness between comparable things such as patients. Like similarity computing amongst Intensive Care Unit (ICU) patients can create various benefits, such as case based patient retrieval, unearthing of similar patient groups. However, many classical methods such as euclidean distance, cosine similarity can’t be directly applicable as similarity computing in EHRs is subjective and in many cases conditional. Also, many intrinsic relationships between the data are lost due to poor data representation and conversion. To address these challenges, firstly, we propose structural network representation for EHRs to preserve inherent relationship. And, to make them more comparable, we do data enrichment e.g. adding abstract information. Then, we extract different similarity feature sets to generate different similarity metrics and retrieve top similar patients. Finally, we perform experiment which shows promising results over classical methods.