Feature extraction from unequal length heterogeneous EHR time series via dynamic time warping and tensor decomposition

Electronic Health Records (EHR) data is routinely generated patient data that can provide useful information for analytical tasks such as disease detection and clinical event prediction. However, temporal EHR data such as physiological vital signs and lab test results are particularly challenging. Temporal EHR features typically have different sampling frequencies; such examples include heart rate (measured almost continuously) and blood test results (a few times during a patient’s entire stay). Different patients also have different length of stays. Existing approaches for temporal EHR sequence extraction either ignore the temporal pattern within features, or use a predefined window to select a section of the sequences without taking into account all the information. We propose a novel approach to tackle the issue of irregularly sampled, unequal length EHR time series using dynamic time warping and tensor decomposition. We use DTW to learn the pairwise distances for each temporal feature among the patient cohort and stack the distance matrices into a tensor. We then decompose the tensor to learn the latent structure, which is consequently used for patient representation. Finally, we use the patient representation for in-hospital mortality prediction. We illustrate our method on two cohorts from the MIMIC-III database: the sepsis and the acute kidney failure cohorts. We show that our method produces outstanding classification performance in terms of AUROC, AUPRC and accuracy compared with the baseline methods: LSTM and DTW-KNN. In the end we provide a detailed analysis on the feature importance for the interpretability of our method.

[1]  Eamonn J. Keogh,et al.  Making Time-Series Classification More Accurate Using Learned Constraints , 2004, SDM.

[2]  Jun Wang,et al.  Generalizing DTW to the multi-dimensional case requires an adaptive approach , 2016, Data Mining and Knowledge Discovery.

[3]  S. Cha,et al.  A model predicting short-term mortality in patients with advanced liver cirrhosis and concomitant infection , 2018, Medicine.

[4]  J. Parrillo,et al.  Serum lactate as a predictor of mortality in patients with infection , 2007, Intensive Care Medicine.

[5]  Jun S. Kim,et al.  An attention based deep learning model of clinical events in the intensive care unit , 2019, PloS one.

[6]  Karsten Borgwardt,et al.  Early Recognition of Sepsis with Gaussian Process Temporal Convolutional Networks and Dynamic Time Warping , 2019, MLHC.

[7]  Hongying Ni,et al.  Urine output on ICU entry is associated with hospital mortality in unselected critically ill patients , 2014, Journal of Nephrology.

[8]  Tamara G. Kolda,et al.  Tensor Decompositions and Applications , 2009, SIAM Rev..

[9]  Toni Giorgino,et al.  Computing and Visualizing Dynamic Time Warping Alignments in R: The dtw Package , 2009 .

[10]  Chonghui Guo,et al.  An evaluation of time series summary statistics as features for clinical prediction tasks , 2020, BMC Medical Informatics and Decision Making.

[11]  R. Campbell,et al.  Analysis and prediction of unplanned intensive care unit readmission using recurrent neural networks with long short-term memory , 2018, bioRxiv.

[12]  Vit Niennattrakul,et al.  Learning DTW Global Constraint for Time Series Classification , 2009, ArXiv.

[13]  Ricard Gavaldà,et al.  Clustering Patients with Tensor Decomposition , 2017, MLHC.

[14]  Mihaela van der Schaar,et al.  Forecasting Individualized Disease Trajectories using Interpretable Deep Learning , 2018, ArXiv.

[15]  C. Kruse,et al.  Challenges and Opportunities of Big Data in Health Care: A Systematic Review , 2016, JMIR medical informatics.

[16]  Weihui Dai,et al.  Dynamic Time Warping: Itakura vs Sakoe-Chiba , 2019, 2019 IEEE International Symposium on INnovations in Intelligent SysTems and Applications (INISTA).

[17]  K. Souliotis,et al.  Real world, big data cost of pharmaceutical treatment for rheumatoid arthritis in Greece , 2019, PloS one.

[18]  Li Zhou,et al.  Machine learning approaches to predict peak demand days of cardiovascular admissions considering environmental exposure , 2020, BMC Medical Informatics and Decision Making.

[19]  Peter Szolovits,et al.  MIMIC-III, a freely accessible critical care database , 2016, Scientific Data.

[20]  Peter Szolovits,et al.  Predicting ICU Mortality Risk by Grouping Temporal Trends from a Multivariate Panel of Physiologic Measurements , 2016, AAAI.

[21]  T. McKeever,et al.  Predicting 30-day mortality in patients with sepsis: An exploratory analysis of process of care and patient characteristics , 2018, Journal of the Intensive Care Society.

[22]  Geoffrey I. Webb,et al.  FastEE: Fast Ensembles of Elastic Distances for time series classification , 2019, Data Mining and Knowledge Discovery.

[23]  Milos Hauskrecht,et al.  Monitoring ICU Mortality Risk with A Long Short-Term Memory Recurrent Neural Network , 2019, PSB.

[24]  Jimeng Sun,et al.  COPA: Constrained PARAFAC2 for Sparse & Large Datasets , 2018, CIKM.

[25]  Le Zhang,et al.  An Effective Patient Representation Learning for Time-series Prediction Tasks Based on EHRs , 2018, 2018 IEEE International Conference on Bioinformatics and Biomedicine (BIBM).

[26]  Yan Liu,et al.  Benchmarking deep learning models on large healthcare datasets , 2018, J. Biomed. Informatics.

[27]  Eamonn J. Keogh,et al.  Scaling up Dynamic Time Warping to Massive Dataset , 1999, PKDD.

[28]  Jimeng Sun,et al.  Opportunities and challenges in developing deep learning models using electronic health records data: a systematic review , 2018, J. Am. Medical Informatics Assoc..

[29]  Stan Salvador,et al.  FastDTW: Toward Accurate Dynamic Time Warping in Linear Time and Space , 2004 .

[30]  J. Moon,et al.  Predicting Mortality of Critically Ill Patients by Blood Glucose Levels , 2013, Diabetes & metabolism journal.

[31]  Stephan Günnemann,et al.  Introduction to Tensor Decompositions and their Applications in Machine Learning , 2017, ArXiv.

[32]  Jimeng Sun,et al.  Granite: Diversified, Sparse Tensor Factorization for Electronic Health Record-Based Phenotyping , 2017, 2017 IEEE International Conference on Healthcare Informatics (ICHI).

[33]  Germain Forestier,et al.  Judicious setting of Dynamic Time Warping's window width allows more accurate classification of time series , 2017, 2017 IEEE International Conference on Big Data (Big Data).

[34]  Rajvir Singh,et al.  Factors that predict 1-month mortality in patients with pregnancy-specific liver disease. , 2014, Clinical Gastroenterology and Hepatology.

[35]  L. Chiche,et al.  Fio2 and acute respiratory distress syndrome definition during lung protective ventilation* , 2009, Critical care medicine.

[36]  Vince D. Calhoun,et al.  Tensor-based fusion of EEG and FMRI to understand neurological changes in schizophrenia , 2016, 2017 IEEE International Symposium on Circuits and Systems (ISCAS).

[37]  Iryna Gurevych,et al.  Optimal Hyperparameters for Deep LSTM-Networks for Sequence Labeling Tasks , 2017, ArXiv.

[38]  Fei Wang,et al.  An RNN Architecture with Dynamic Temporal Matching for Personalized Predictions of Parkinson's Disease , 2017, SDM.

[39]  Chien-Lung Chan,et al.  Good Mortality Prediction by Glasgow Coma Scale for Neurosurgical Patients , 2010, Journal of the Chinese Medical Association : JCMA.

[40]  Harini Suresh,et al.  Learning Tasks for Multitask Learning: Heterogenous Patient Populations in the ICU , 2018, KDD.

[41]  Alfredo Vellido,et al.  Sepsis mortality prediction with the Quotient Basis Kernel , 2014, Artif. Intell. Medicine.

[42]  H. Kiers Towards a standardized notation and terminology in multiway analysis , 2000 .

[43]  Jimeng Sun,et al.  Limestone: High-throughput candidate phenotype generation via tensor factorization , 2014, J. Biomed. Informatics.

[44]  Aram Galstyan,et al.  Multitask learning and benchmarking with clinical time series data , 2017, Scientific Data.

[45]  S. Lemeshow,et al.  A new Simplified Acute Physiology Score (SAPS II) based on a European/North American multicenter study , 1993 .

[46]  Yaohang Li,et al.  A Disease Inference Method Based on Symptom Extraction and Bidirectional Long Short Term Memory networks. , 2020, Methods.

[47]  Nikos D. Sidiropoulos,et al.  Tensor Decomposition for Signal Processing and Machine Learning , 2016, IEEE Transactions on Signal Processing.

[48]  Jimeng Sun,et al.  Marble: high-throughput phenotyping from electronic health records via sparse nonnegative tensor factorization , 2014, KDD.

[49]  Fei Wang,et al.  SPARTan: Scalable PARAFAC2 for Large & Sparse Data , 2017, KDD.

[50]  S. Chiba,et al.  Dynamic programming algorithm optimization for spoken word recognition , 1978 .

[51]  M. Legrand,et al.  Understanding urine output in critically ill patients , 2011, Annals of intensive care.

[52]  James Large,et al.  The Great Time Series Classification Bake Off: An Experimental Evaluation of Recently Proposed Algorithms. Extended Version , 2016, ArXiv.

[53]  M. Assuncao,et al.  Blood Lactate Levels Cutoff and Mortality Prediction in Sepsis—Time for a Reappraisal? a Retrospective Cohort Study , 2016, Shock.

[54]  Philip Chan,et al.  Toward accurate dynamic time warping in linear time and space , 2007, Intell. Data Anal..

[55]  Hadi Fanaee-T,et al.  Event and Anomaly Detection Using Tucker3 Decomposition , 2014, ArXiv.

[56]  F. Itakura,et al.  Minimum prediction residual principle applied to speech recognition , 1975 .

[57]  Hagen Malberg,et al.  Predicting sepsis with a recurrent neural network using the MIMIC III database , 2019, Comput. Biol. Medicine.

[58]  Joydeep Ghosh,et al.  PIVETed-Granite: Computational Phenotypes through Constrained Tensor Factorization , 2018, ArXiv.

[59]  S. Lemeshow,et al.  A new Simplified Acute Physiology Score (SAPS II) based on a European/North American multicenter study. , 1993, JAMA.

[60]  Zachary Chase Lipton The mythos of model interpretability , 2016, ACM Queue.