JMIR Publications

ABSTRACT

Background:

Acute myocardial infarction (AMI) is one of the most severe cardiovascular diseases and is associated with high risk of in-hospital mortality. However, the current deep learning models for in-hospital mortality prediction are lack of interpretability.

Objective:

This study aimed to establish an explainable deep learning model to provide individualized in-hospital mortality prediction and risk factor assessment in patients with AMI.

Methods:

In this retrospective multicentre study, we used data for consecutive patients hospitalized with AMI from Chongqing University Central Hospital between July 2016 and December 2022 and eICU Collaborative Research Database (eICU-CRD). These patients were randomly divided into training (70%) and internal testing (30%) datasets. In addition, patients with AMI from Medical Information Mart for Intensive Care (MIMIC-IV) database were used for external validation. Deep learning models were used to predict in-hospital mortality in patients with AMI, and compared with linear and tree-based models. The SHapley Additive exPlanations (SHAP) method was utilized to explain the best-performing model to quantify and visualize the features that drive predictions.

Results:

11,055 patients with AMI who were admitted to Chongqing University Central Hospital or included in the eICU-CRD were randomly divided into a training dataset of 7,668 (70%) patients and a testing dataset of 3,287 (30%) patients. 9,355 patients from MIMIC-IV database were used for independent external validation. In-hospital mortality occurred in 670 (8.7%), 287 (8.7%), and 853 (9.1%) patients in the training, test, and external validation cohorts, respectively. The Self-Attention and Intersample Attention Transformer (SAINT) model performed best in both the test set and the external validation set among the 9 prediction models, with the highest AUC of 0.86 (95% CI 0.84, 0.88) and 0.85 (95% CI 0.84, 0.87), respectively. Older age, high heart rate, and low body temperature were the three most important predictors of increased mortality, according to the explanations of the SAINT model.

Conclusions:

The developed explainable deep learning model could provide estimates of mortality and visual contribution of the features to the prediction for a patient with AMI. The explanations suggested that more attention should be given to these modifiable risk factors related to vital signs, biochemistry, and metabolism in the treatment of AMI.