ABSTRACT
We present Language-Interfaced Fine-Tuning (LIFT) in application to COVID-19 patient survival classification. LIFT describes translating tabular Electronic Health Records (EHRs) into text inputs for transformer neural networks. We study LIFT with a dataset of 5,371 COVID-19 patients. We focus on the predictive task of survival classification utilizing demographic and medical history features. We begin by presenting information about our dataset. We preface our investigation in text-based transformers by reporting the performances of conventional machine learning models such as Logistic Regression and Random Forest classifiers. We also present the results of a few configurations of tabular input-based Deep Multilayer Perceptron (MLP) networks. 86% of the patients in our database survived in the measured time window. Thus, predictive models are heavily biased to predict that a patient will survive. We emphasize that this problem of Class Imbalance was a major challenge in developing these models. Our balanced sampling strategy from examples in the majority and minority classes is crucial to achieving even reasonable predictive performance. For this reason, we also report performance based on Precision, Recall, and F-score metrics, in addition to Accuracy. Having established baselines with tabular inputs, we then shift our focus to the prompts for translating from tabular to text inputs. We report the performance of 5 prompts. The LIFT model achieves an F-score on the held-out test set of 0.21, slightly behind the Deep MLP with Tabular Features score of 0.23. Both models outperform the Random Forest with Tabular Features at 0.15. We believe that LIFT is a very exciting direction for machine learning in healthcare applications because text-based inputs enables us to take advantage of recent advances in Transfer Learning and Retrieval-Augmented Learning. This study illustrates the effectiveness of converting tabular EHRs to text inputs and utilizing transformer neural networks for prediction. © 2022 IEEE.
ABSTRACT
Patient care in emergency rooms can utilize urgency labeling to facilitate resource allocation. With COVID-19 care, one of the most important indicators of care urgency is the severity of respiratory illness. We present an early analysis of 5,584 patient records, of whom 5,371 (96.2%) have returned a positive COVID-19 test, to understand how well we can predict the severity of a respiratory illness given other features describing a patient using Deep Learning methods. The goal of our work is to illustrate the connection of our COVID-19 patient dataset with Deep Learning techniques, setting the stage for future work. The features in our dataset include when COVID-19 symptoms began, age, height, weight, demographics, and pre-existing conditions, to give a quick preview. We report train-test performance of a Deep Multi-Layer Perceptron (MLP) to predict the severity of respiratory analysis on a one-hot encoded scale of 5 labels. This 5-level scale is a truncation of our available labels, which we plan to extend and include in future work. We utilize a high-level of Dropout in order to avoid overfitting with our Deep Learning model. Further, we particularly study the impact of class imbalance on this dataset (Johnson and Khoshgoftaar 2019). We find that Random Oversampling (ROS) is an effective solution for decreasing minority class false negatives, as well as increasing overall accuracy. Readers will understand the performance of Deep Learning, with Dropout and ROS, to predict the severity of a COVID-19 pa-tient’s respiratory illness in which patients are described with Tabular Electronic Health Records (EHR). © 2022 by the authors. All rights reserved.