Incorporating Ensembled Stacked Approach into Automated Machine Learning in an Attempt to Predict Acute Ischemic Heart Disease in Patients with Atypical Chest Pain: Secondary Analysis of a Single-Center Retrospective Cohort Study

Abstract

Background, Rationale & ObjectiveConsiderable research has been carried out to explore the adoption of novel methodologies in a bid to optimize the management protocols for acute ischemic heart disease (AIHD) that is prevalent worldwide and has significant morbidity and mortality associated with it. With Artificial intelligence in general and machine learning in particular exhibiting superior predictive and prognositicative capabilities, when compared with conventional statistical techniques, we explored automated Machine Learning to predict AIHD in such patients presenting with atypical chest pain. MethodologyThe study population comprised 3,833 patients presenting with chest pain who visited the ED of a tertiary-care hospital from January 2014 to December 2018. Such patients who exhibited typical chest pain were excluded. (Mean age = 60.3; Female subjects = 45.6%; Diagnosed with AIHD = 11.8%). Variables included a plethora of variables obtained on a routine ED evaluation including vital signs and laboratory data. The current state of the art for automated Machine Learning was adopted to develop classification models using algorithms including Extreme Gradiant Boosting, Random Forest and Neural Network. The models were stacked after implementation of hyperparameter tuning. Ensemble approach, which is the amalgamation of two or more than two algorithmic models to develop such a model which is better than either of its computive components, was superimposed on stacked models. The macro-weighted average Area under the Response-operating Curve (mwA-AUROC) gauged the discriminating ability of models with the highest value of 1 indicating perfect discriminative classification ability. ResultsAn ensemble of stacked Extra Trees, Neural Network, CatBoost, Light Gradient Boosted Machine, Random Forest algorithmic models achieved an mwA-AUROC of 0.78 and an accuracy of 87.9%. (Figure 1) Our model outperformed those developed by Kim KH et al. The ensemble model exhibited the least log loss among all of the developed algorithmic models. ConclusionSuch predictive models, when deployed on cloud, have the capability of significantly influencing the prognosticating protocols for AIHD by providing instantaneous predictions. Such an optimization of the risk stratification protocols achieved via incorporation of aML would serve to improve the management of not only the respective patient but also that of the resource allocation. This would ultimately translate into significant reduction of morbidity and mortality associated with AIHD.These authors contributed equally to data curation, development of methodology, statistical analyses and writing with subsequent reviewing and editing of the abstract draft.These authors, indicated alphabetically, contributed equally to statistical analyses and validation.