Development and validation of an explainable artificial intelligence model to predict in-hospital adverse events for patients with acute myocardial infarction
Nov 2026· PLOS Digital Health· Vol 5· 0 citations· 29 references
Medicine
Abstract
In-hospital adverse events after acute myocardial infarction (AMI) can occur abruptly and adversely affect short-term prognosis. We developed and externally validated explainable machine learning models to predict a composite of in-hospital atrial fibrillation (AF), ventricular fibrillation (VF), cardiac arrest (CA), or death using routinely available admission data. We retrospectively analyzed 7,496 patients from the MIMIC-IV database as the development cohort and 2,776 patients from Xuanwu Hospital as the external validation cohort. Predictors included demographics, medical history, and admission laboratory tests. Logistic regression, random forest, extreme gradient boosting (XGBoost), multilayer perceptron, and TabNet models were evaluated using five-fold cross-validation. Performance was assessed using discrimination, classification, calibration, and decision curve analysis. Latent class analysis identified three outcome clusters with distinct event-probability profiles. XGBoost achieved the best performance for the composite outcome in the development cohort (AUROC, 0.798). Predictive performance was moderate to good for AF and in-hospital death, with AUROCs of up to 0.834 and 0.870 in the development cohort and 0.803 and 0.849 in the external validation cohort, respectively. Discrimination was lower for VF and CA. SHapley Additive exPlanations highlighted neutrophil and leukocyte counts, N-terminal pro-B-type natriuretic peptide, renal function, glucose, glycated hemoglobin A1c, and prior AF as important predictors. These explainable models may support early risk stratification and help guide monitoring intensity and timely intervention in patients with AMI.
It is found that Llama 3 delivers comparable quality to leading language models such as GPT-4 on a plethora of tasks, and performs competitively with the state-of-the-art on image, video, and speech recognition tasks.
This paper investigates developing real-world perception for active agents, proposes Gibson Environment for this purpose, and showcases a set of perceptual tasks learned therein.
F. Xia, Amir Zamir, Zhi-Yang He et al.· 2018 IEEE/CVF Conference on...· 1,069 citations· ⚡112
GAOKAO-Bench is introduced, an intuitive benchmark that employs questions from the Chinese GAOKAO examination as test samples, including both subjective and objective questions that contribute a robust evaluation benchmark for future large language models and offers valuable insights into the advantages and limitations...
Xiaotian Zhang, Chun-yan Li, Yi Zong et al.· arXiv.org· 216 citations· ⚡17
An algorithm is proposed which leverages disentangled semantic factors to generate adversarial perturbation by altering controlled semantic attributes to fool the learner towards various "adversarial" targets.
Haonan Qiu, Chaowei Xiao, Lei Yang et al.· arXiv.org· 208 citations· ⚡17
This work proposes a HALO that directly maximizes the utility of generations instead of maximizing the log-likelihood of preferences, and suggests that the best loss depends on the inductive biases most appropriate for a given setting, an oft-overlooked consideration.
This work investigates the possibilities of using LLMs in a resume screening setting via a document retrieval framework that simulates job candidate selection and finds that the MTEs are biased, significantly favoring White-associated names in 85% of cases and female-associated names in only 11.1% of cases.