Learning Multimorbidity Patterns from Electronic Health Records Using Non-negative Matrix Factorisation

Multimorbidity, or the presence of several medical conditions in the same\nindividual, has been increasing in the population, both in absolute and\nrelative terms. However, multimorbidity remains poorly understood, and the\nevidence from existing research to describe its burden, determinants and\nconsequences has been limited. Previous studies attempting to understand\nmultimorbidity patterns are often cross-sectional and do not explicitly account\nfor multimorbidity patterns' evolution over time; some of them are based on\nsmall datasets and/or use arbitrary and narrow age ranges; and those that\nemployed advanced models, usually lack appropriate benchmarking and\nvalidations. In this study, we (1) introduce a novel approach for using\nNon-negative Matrix Factorisation (NMF) for temporal phenotyping (i.e.,\nsimultaneously mining disease clusters and their trajectories); (2) provide\nquantitative metrics for the evaluation of disease clusters from such studies;\nand (3) demonstrate how the temporal characteristics of the disease clusters\nthat result from our model can help mine multimorbidity networks and generate\nnew hypotheses for the emergence of various multimorbidity patterns over time.\nWe trained and evaluated our models on one of the world's largest electronic\nhealth records (EHR), with 7 million patients, from which over 2 million where\nrelevant to this study.\n