Computational Phenotype Discovery Using Unsupervised Feature Learning over Noisy, Sparse, and Irregular Clinical Data

By Thomas A. Lasko, Joshua Charles Denny, Mia A. Levy. 2013

317 citations recorded by the source corpus.

Inferring precise phenotypic patterns from population-scale clinical data is a core computational task in the development of precision, personalized medicine. The traditional approach uses supervised learning, in which an expert designates which patterns to look for (by specifying the learning task and the class labels), and where to look for them (by specifying the input variables). While appropriate for individual tasks, this approach scales poorly and misses the patterns that we don’t think to look for. Unsupervised feature learning overcomes these limitations by identifying patterns (or features) that collectively form a compact and expressive representation of the source data, with no need for expert input or labeled examples. Its rising popularity is driven by new deep learning methods, which have produced high-profile successes on difficult standardized problems of object recognition in images. Here we introduce its use for phenotype discovery in clinical data. This use is challenging because the largest source of clinical data – Electronic Medical Records – typically contains noisy, sparse, and irregularly timed observations, rendering them poor substrates for deep learning methods. Our approach couples dirty clinical data to deep learning architecture via longitudinal probability densities inferred using Gaussian process regression. From episodic, longitudinal sequences of serum uric acid measurements in 4368 individuals we produced continuous phenotypic features that suggest multiple population subtypes, and that accurately distinguished (0.97 AUC) the uric-acid signatures of gout vs. acute leukemia despite not being optimized for the task. The unsupervised features were as accurate as gold-standard features engineered by an expert with complete knowledge of the domain, the classification task, and the class labels. Our findings demonstrate the potential for achieving computational phenotype discovery at population scale. We expect such data-driven phenotypes to expose unknown disease variants and subtypes and to provide rich targets for genetic association studies.

Subfields: Artificial Intelligence, Complementary and alternative medicine, Molecular Biology, Representation & Self-Supervised Learning, Medical & Health AI

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 Complementary and alternative medicine

In Molecular Biology

In Representation & Self-Supervised Learning

In Medical & Health AI

Citations in the corpus

Distilling Knowledge from Deep Networks with Applications to Healthcare Domain (2015, background): “Recent works [9, 28] have used deep learning models to achieve state-of-the-art performance on computational phenotype prediction problems.”
Natural Language Processing for EHR-Based Computational Phenotyping (2018, background): “Unsupervised learning, on the other hand, can automatically classify phenotypes without extra annotations by experts [105, 140, 141].”
Learning to Diagnose with LSTM Recurrent Neural Networks (2015, background): “Recently, feedforward networks have been applied to medical time series in sliding window fashion to classify cases of gout, leukemia (Lasko et al., 2013), and critical illness (Che et al.”
Using recurrent neural network models for early detection of heart failure onset (2016, background): “Both works, 28,29 however, focused on learning patterns from clinical records rather than predicting a clinical event.”
Scalable Joint Models for Reliable Uncertainty-Aware Event Prediction (2017, methodology): “[59] use GPs to model univariate longitudinal data and train autoencoders on the GP predictions to”
Mining Electronic Health Records: A Survey (2017, methodology): “Lasko et al. [Lasko et al. 2013] used deep learning for phenotype discovery in clinical data.”
Machine Learning and Decision Support in Critical Care (2016, background): “[190] applied a deep learningbased approach to unsupervised learning of phenotypical features in longitudinal sequences of serum uric acid measurements.”
Interpretable Predictions of Clinical Outcomes with An Attention-based Recurrent Neural Network (2017, background): “were the first to apply deep learning in health care, and they demonstrated the capability of deep learning to generalize patterns from serum uric acid measurements with autoencoders [21].”
Interpretable Deep Models for ICU Outcome Prediction (2016, methodology): “Among these methods, deep learning models (e.g., multilayer neural networks) have achieved the state-of-the-art performance on several tasks, such as computational phenotype discovery 9,10 and predictive modeling 8,11 .”
Sparse Multi-Output Gaussian Processes for Medical Time Series Prediction (2017, methodology): “Previous work used a single-output GP regression model to smooth and impute each covariate independently (Stegle et al., 2008; Lasko et al., 2013).”
$\mathtt {Deepr}$: A Convolutional Net for Medical Records (2016, methodology): “In [37], time-series are modeled using autoencoders (an unsupervised FFN) to discover meaningful phenotypes.”