Clustering Longitudinal Clinical Marker Trajectories from Electronic Health Data: Applications to Phenotyping and Endotype Discovery

By Peter F. Schulam, Fredrick Wigley, Suchi Saria. 2015

116 citations recorded by the source corpus.

Diseases such as autism, cardiovascular disease, and the autoimmune disorders are difficult to treat because of the remarkable degree of variation among affected individuals. Subtyping research seeks to refine the definition of such complex, multi-organ diseases by identifying homogeneous patient subgroups. In this paper, we propose the Probabilistic Subtyping Model (PSM) to identify subgroups based on clustering individual clinical severity markers. This task is challenging due to the presence of nuisance variability — variations in measurements that are not due to disease subtype — which, if not accounted for, generate biased estimates for the group-level trajectories. Measurement sparsity and irregular sampling patterns pose additional challenges in clustering such data. PSM uses a hierarchical model to account for these different sources of variability. Our experiments demonstrate that by accounting for nuisance variability, PSM is able to more accurately model the marker data. We also discuss novel subtypes discovered using PSM and the resulting clinical hypotheses that are now the subject of follow up clinical experiments.

Subfields: Artificial Intelligence, Health Information Management, Molecular Biology

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Health Information Management

In Molecular Biology

Citations in the corpus

Sample-Efficient Reinforcement Learning via Counterfactual-Based Data Augmentation (2020, background): “(1)More sophisticated clustering methods may lead to improved results [31].”
Learning to Diagnose with LSTM Recurrent Neural Networks (2015, background): “…of research applies machine learning to temporal clinical data for tasks including artifact removal (Aleks et al., 2009; Quinn et al., 2009), early detection and prediction (Stanculescu et al., 2014a; Henry et al., 2015), and clustering and subtyping (Marlin et al., 2012; Schulam et al., 2015).”
A Framework for Individualizing Predictions of Disease Trajectories by Exploiting Multi-Resolution Structure (2015, background): “Others, more recently, have developed models for disease trajectories (see [15, 16] and references within) but these focus on retrospective analysis to discover disease etiology rather than dynamical prediction.”
Machine Learning and Decision Support in Critical Care (2016, unknown): “[167] took a different approach to a time-series clustering model, in which they defined a set of generative linear prototype functions to describe the behavior of individual clinical features over time for patients with scleroderma (a connective tissue disease).”
Interpretable Deep Models for ICU Outcome Prediction (2016, background): “This has resulted in development of novel and sophisticated machine learning solutions 3,4,5,6,7,8 .”
Unsupervised Representation Learning for Time Series with Temporal Neighborhood Coding (2021, background): “…discovering subgroups of patients with similar underlying disease mechanisms from temporal clinical data (Lasko et al., 2013; Suresh et al., 2018; Schulam et al., 2015), and disease progression modeling, for learning the hidden vector of comorbidities representing a disease over time (Wang et…”
Predicting Complications in Critical Care Using Heterogeneous Clinical Data (2016, background): “[28] studied phenotypes for scleroderma (an autoimmune disease) with the aim of discovering disease subtypes using physiological data.”
Deep Computational Phenotyping (2015, background): “Most work on phenotyping from clinical time series has focused on variations of Gaussian processes and related models [27, 22] and focus on discovering cluster structure (e.”
Sparse Multi-Output Gaussian Processes for Medical Time Series Prediction (2017, methodology): “We showed that our method, MedGP, improves performance for online prediction of 24 clinical covariates as compared with independent univariate GPs, a naive method of propagating the previous observation, and an earlier state-of-the-art approach, PSM (Schulam et al., 2015).”
Mining Electronic Health Records: A Survey (2017, background): “…and Hripcsak 2013] Cohort Identiﬁcation Descritive Atemporal [Roque et al. 2011; Bauer-Mehren et al. 2013] Descritive Temporal [Gotz et al. 2014; Schulam et al. 2015; Lasko et al. 2013; Wang et al. 2015a] Cross-Sectional [Wang et al. 2013b; Peissig et al. 2012; Pathak et al. 2012; Nadkarni et…”
Large-Scale Discovery of Disease-Disease and Disease-Gene Associations (2016, methodology): “Deep learning has also been applied to the task of phenotyping 30 , as well as graph mining 31 and clustering 34 , used to identify patient subgroups based on individual clinical markers.”