Analysis of In-Hospital Morality Prediction by Machine Learning

Abstract

　Previous reports have indicated that machine learning/AI can predict the prognosis of hospitalized patients with high accuracy. However, it has been unclear for which types of hospitalizations these methods were effective. Therefore, this study aimed to analyze the features of the mortality prediction models by machine learning/AI. Compared the characteristics of hospital admissions that result in high mortality prediction probabilities with those that result in low probabilities, we aim to identify considerations for clinical use of prognostic AI and insights for future model improvements.