A Predictive Model Development Framework Based on Machine-Learning and Stacking Ensemble Technique for Patient Prognosis

This research designed a machine-learning framework to predict patient death in hospitals, utilizing real-world clinical data from patients diagnosed with pneumonia. Predicting in-hospital mortality among high-risk patients is a significant challenge in healthcare domain. Specifically, estimating the risk of complications in many pneumonia patients hinges on the analysis of pathologic and radiologic data, a process that is both expensive and labor-intensive. In this work, we introduce a machine learning approach designed to predict mortality among high-risk pneumonia patients using real clinical historical data, including diagnosis codes, prescription codes, and lab test codes, from individuals hospitalized with pneumonia. Specifically, the model was developed using the data converted into a 'Common data model' format, standardizing hospital records into a unified structure. We demonstrate two types of predictive models: 1) traditional machine learning algorithm-based models and 2) hybrid stacking-based ensemble models. The mortality prediction model for approximately 2,540 patients diagnosed with pneumonia was evaluated in terms of precision, recall, F1-measure, and area under the curve (AUC). The experimental results show that our methods outperform standard classification models and, particularly, identify which factors based on the features input into the model influence the risk of death in patients. Since this model is based on the Common Data Model, it has the advantage of being extendable to other hospitals that hold this type of data in the future.