Rubik

• DOI: 10.1145/2783258.2783395
• Publication Date: 2015-08-07T15:38:27Z
• Authors: Yichen Wang; Robert Chen; Joydeep Ghosh; Joshua C. Denny; Abel Kho; You Chen; Bradley A. Malin; Jimeng Sun
• Keywords: 0202 electrical engineering, electronic engineering, information engineering; 02 engineering and technology; 3. Good health

Computational phenotyping is the process of converting heterogeneous electronic health records (EHRs) into meaningful clinical concepts. Unsupervised methods have potential to leverage a vast amount labeled EHR data for phenotype discovery. However, existing unsupervised do not incorporate current medical knowledge and cannot directly handle missing, or noisy data. We propose Rubik, constrained non-negative tensor factorization completion method phenotyping. Rubik incorporates 1) guidance constraints align with knowledge, 2) pairwise obtaining distinct, non-overlapping phenotypes. also has built-in that can significantly alleviate impact missing utilize Alternating Direction Method Multipliers (ADMM) framework completion, which be easily scaled through parallel computing. evaluate on two datasets, one contains 647,118 7,744 patients from an outpatient clinic, other public dataset containing 1,018,614 CMS claims 472,645 patients. Our results show discover more distinct phenotypes than baselines. In particular, by using constraints, sub-phenotypes several major diseases. runs around seven times faster state-of-the-art methods. Finally, scalable large datasets millions records.