Limestone: High-throughput candidate phenotype generation via tensor factorization

By Joyce Ho, Joydeep Ghosh, Steven R. Steinhubl, Walter F. Stewart, Joshua Charles Denny, Bradley A. Malin, Jimeng Sun. 2014

164 citations recorded by the source corpus.

The rapidly increasing availability of electronic health records (EHRs) from multiple heterogeneous sources has spearheaded the adoption of data-driven approaches for improved clinical research, decision making, prognosis, and patient management. Unfortunately, EHR data do not always directly and reliably map to medical concepts that clinical researchers need or use. Some recent studies have focused on EHR-derived phenotyping, which aims at mapping the EHR data to specific medical concepts; however, most of these approaches require labor intensive supervision from experienced clinical professionals. Furthermore, existing approaches are often disease-centric and specialized to the idiosyncrasies of the information technology and/or business practices of a single healthcare organization. In this paper, we propose Limestone, a nonnegative tensor factorization method to derive phenotype candidates with virtually no human supervision. Limestone represents the data source interactions naturally using tensors (a generalization of matrices). In particular, we investigate the interaction of diagnoses and medications among patients. The resulting tensor factors are reported as phenotype candidates that automatically reveal patient clusters on specific diagnoses and medications. Using the proposed method, multiple phenotypes can be identified simultaneously from data. We demonstrate the capability of Limestone on a cohort of 31,815 patient records from the Geisinger Health System. The dataset spans 7 years of longitudinal patient records and was initially constructed for a heart failure onset prediction study. Our experiments demonstrate the robustness, stability, and the conciseness of Limestone-derived phenotypes. Our results show that using only 40 phenotypes, we can outperform the original 640 features (169 diagnosis categories and 471 medication types) to achieve an area under the receiver operator characteristic curve (AUC) of 0.720 (95% CI 0.715 to 0.725). Moreover, in consultation with a medical expert, we confirmed 82% of the top 50 candidates automatically extracted by Limestone are clinically meaningful.

Subfields: Artificial Intelligence, Computational Mathematics

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

1. Fei Wang
2. Jimeng Sun
3. Yuan Luo
4. Harini Suresh
5. John V. Guttag
6. Xiaoqian Jiang
7. Zexian Zeng
8. Hwanjo Yu

In Computational Mathematics

1. Xiaoqian Jiang
2. Evangelos E. Papalexakis
3. Jimeng Sun
4. Hwanjo Yu
5. Yejin Kim
6. Fei Wang
7. Yuan Luo
8. R. Vuduc

Citations in the corpus

• Federated Tensor Factorization for Computational Phenotyping (2017, methodology): “Overcoming several disadvantages of the previous methods, tensor factorization methods have shown great potential in discovering meaningful phenotypes from complicated and heterogeneous health records [13, 14, 28].”
• Natural Language Processing for EHR-Based Computational Phenotyping (2018, methodology): “[46], they applied tensor factorization [47], [48] on medication orders to generate phenotypes without supervision.”
• Risk Prediction with Electronic Health Records: A Deep Learning Approach (2016, methodology): “[10, 16] proposed a nonnegative tensor factorization based approach for phenotype extraction from those EHR tensors.”
• Discriminative and Distinct Phenotyping by Constrained Tensor Factorization (2017, methodology): “We only focused on diagnosis and prescription data as previous phenotyping definition 29–31 , but we can extend the tensor to a high order ( > 3) to utilize additional data such as laboratory results and procedures.”
• L G ] 1 1 A pr 2 01 7 Federated Tensor Factorization for Computational Phenotyping (2018, methodology): “Overcoming several disadvantages of the previous methods, tensor factorization methods have shown great potential in discovering meaningful phenotypes from complicated and heterogeneous health records [13, 14, 28].”
• SUSTain: Scalable Unsupervised Scoring for Tensors and its Application to Phenotyping (2018, methodology): “Factorization methods have been successfully used for EHR-based unsupervised phenotyping [21, 22, 24, 41, 42, 50].”
• Tensor Factorization for Precision Medicine in Heart Failure with Preserved Ejection Fraction (2017, unknown): “[42] studied the problem of heart failure onset prediction with clinically meaningful sub-tensors.”
• Learning Tasks for Multitask Learning: Heterogenous Patient Populations in the ICU (2018, background): “With the growing availability of large, high-dimensional clinical data, many works have proposed approaches to learning patient phenotypes [5, 20, 21, 33].”
• Large-Scale Discovery of Disease-Disease and Disease-Gene Associations (2016, background): “Further, Ho et al. 33 formulated patient EHRs as tensors, where each dimension is represented by a different medical event, and the use of non-negative tensor factorization in the identification of phenotypes.”
• COPA: Constrained PARAFAC2 for Sparse & Large Datasets (2018, unknown): “1 INTRODUCTION Tensor factorization encompasses a set of powerful analytic methods that been successfully applied in many application domains: social network analysis, urban planning, text mining, and health analytics [2, 13, 14, 17, 18, 25].”