Detecting time-evolving phenotypic topics via tensor factorization on electronic health records: Cardiovascular disease case study

Untangling the complexity of multimorbidity with machine learning

“…In another study, ( Zhao et al (2019) ) incorporated the time to the onset of cardiovascular disease as a dimension in the tensor. The phenotypes obtained from this approach, while temporally profiled, are specific to cardiovascular disease patients (see Fig.…”

“…Most factorisation methods used in the multimorbidity research are related to Non-negative Matrix and Tensor Factorisation approaches (i.e., NMF and NTF); both techniques take a non-negative input and factorise them into non-negative matrices, for which the multiplication should reconstruct the original input. The resulting patterns, due to non-negativity, are pushed to be sparse and, therefore, easier to interpret ( Hassaine et al, 2019 ; Ho et al, 2020 ; Wang et al, 2020b ; Afshar et al, 2019 ; Zhao et al, 2019 ). However, we know that some diseases are likely to have a suppression (negative) effect on each other; diseases A and B can be said to have suppressive effect on each other if the presence of disease A leads to lower risk (or prevention) of disease B, perhaps because of shared risk factors and clinical management (e.g.…”

AI and Big Data in Healthcare: Towards a More Comprehensive Research Framework for Multimorbidity

“…Zhao et al applied a modified non-negative tensor-factorization approach (a technique used for discovering latent object variables in image analysis) on eHRs data in order to identify phenotypic subtypes in patients at risk for cardiovascular disease. By combining ARM with the estimated risk of each subtype for the development of cardiovascular disease, these authors could identify some previously unknown phenotypes [ 100 ]. Nguyen et al developed a modified convolutional neural network (CNN) model for predicting the probability of hospital readmission, based on medical history information used as a sequence of concepts [ 101 ].…”

Deep phenotyping: Embracing complexity and temporality—Towards scalability, portability, and interoperability

“…Zhao et al applied a constrained non-negative tensor-factorization approach on electronic health records to detect temporal phenotypes of complex cardiovascular diseases (CVD) [29]. From a cohort of 12,380 CVD adults, they identified 14 subphenotypes.…”

“…First author NLP STND DL SRC SUB Disease Focus eMERGE OHDSI [14] Datta, S X [15] Liu, Q X X [16] Lyudovyk, O X X X Cancer X [17] Liu, C X X X [18] Hong, N X [19] Shang, N X X X [20] Hripcsak, G X X X [21] Ostropolets, A X X X [22] Reps, J X X X [23] Swerdel, J X X [24] Warner, J X Cancer X [25] Shen, F X X X X [26] Trace, JM X Parkinson's [27] Mate, S X [28] Meng, W Cancer [29] Zhao, J X X Cardiovascular X [30] Chen, X X X Rare disease [31] Xu, Z X X X Acute kidney injury [32] Zhang, L X X [33] Chen, P X TOTAL 6 12 6 6 3 5 6…”