Skip to main content

Interpretable and Explainable Machine Learning Models for Patient Risk Prediction

A converging theme of Luo Lab's vision spanning different application domains (i.e., medical natural language processing, time series analysis, computational phenotyping and integrative omics) is to build AI/machine learning models that are not only accurate but also interpretable, by exploring techniques such as graph mining, factorization models, and interpretable deep learning techniques. Some of the focus areas are as follows.

Mortality Risk: ICU mortality risk prediction may help clinicians take effective interventions to improve patient outcome. Existing machine learning approaches often face challenges in integrating a comprehensive panel of physiologic variables and presenting to clinicians interpretable models. We aim to improve both accuracy and interpretability of prediction models by introducing Subgraph Augmented Non-negative Matrix Factorization (SANMF) on ICU physiologic time series. SANMF converts time series into a graph representation and applies frequent subgraph mining to automatically extract temporal trends. We then apply non-negative matrix factorization to group trends in a way that approximates patient pathophysiologic states. Trend groups are then used as features in training a logistic regression model for mortality risk prediction, and are also ranked according to their contribution to mortality risk. We evaluated SANMF against four empirical models on the task of predicting mortality or survival 30 days after discharge from ICU using the observed physiologic measurements between 12 and 24 hours after admission. SANMF outperforms all comparison models, and in particular, demonstrates an improvement in AUC (0.848 vs. 0.827, p<0.002) compared to a state-of-the-art machine learning method that uses manual feature engineering. Feature analysis was performed to illuminate insights and benefits of subgraph groups in mortality risk prediction.

Readmission Risk: ICU readmission risk prediction may help physicians to re-evaluate the patient’s physical conditions before patients are discharged and avoid preventable readmissions. ICU readmission prediction models are often built based on physiological variables. Intuitively, snapshot measurements, especially the last measurements, are effective predictors that are widely used by researchers. However, methods that only use snapshot measurements neglect predictive information contained in the trends of physiological and medication variables. We find strong predictors with ability of capturing detailed temporal trends of variables for 30-day readmission risk and build prediction models with high accuracy. Using the SANMF framework, we train a logistic regression model to predict 30-day ICU readmission risk based on snapshot measurements, grouped physiological trends and medication trends. Our dataset consists of 1170 patients who are alive 30 days after discharge from ICU and have at least 12 h of data. In the dataset, 860 patients were not readmitted and 310 were readmitted, within 30 days after discharge. Our model outperforms all comparison models, and shows an improvement in the area under the receiver operating characteristic curve (AUC) of almost 4% from the best comparison model. Grouped physiological and medication trends carry predictive information for ICU readmission risk. In order to build predictive models with higher accuracy, we should add grouped physiological and medication trends as complementary features to snapshot measurements.

Acute Kidney Injury (AKI) Risk: For structured health data, we have also performed extensive case studies on predictive modeling of AKI onset for ICU patients. In particular, we used structured data including physiologic variables, vitals, medications, ICD codes etc. for AKI early prediction and obtained AUCs up to 0.796. We experimented with unstructured clinical notes alone for AKI early prediction and achieved an AUC up to 0.779. We also combined CNN and multi-layer perceptron (MLP) models to integrate both structured EHR and unstructured clinical notes to improve AKI early prediction and obtained an AUC of 0.835. Furthermore, we used a memory network-based deep learning approach to discover predictive AKI sub-phenotypes using structured and unstructured health data of patients before AKI diagnosis. Our approach identified three distinct sub-phenotypes: sub-phenotype I is with an average age of 63.03±17.25 years, and is characterized by mild loss of kidney excretory function (Serum Creatinne (SCr) 1.55 ± 0.34 mg/dL, estimated Glomerular Filtration Rate Test (eGFR) 107.65±54.98 mL/min/1.73m2). These patients are more likely to develop stage I AKI. Sub-phenotype II is with average age 66.81±10.43 years, and was characterized by severe loss of kidney excretory function (SCr 1.96 ± 0.49 mg/dL, eGFR 82.19 ± 55.92 mL/min/1.73m2). These patients are more likely to develop stage III AKI. Sub-phenotype III is with average age 65.07±11.32 years, and was characterized moderate loss of kidney excretory function and thus more likely to develop stage II AKI (SCr 1.69 ± 0.32 mg/dL, eGFR 93.97 ± 56.53 mL/min/1.73m2).