JMIR Publications

ABSTRACT

Background:

Despite advances in understanding and treating non-ST-elevation acute coronary syndrome (NSTE-ACS), patients still experience high rates of adverse outcomes, especially in non-ST-segment elevation myocardial infarction (NSTEMI), a leading cause of cardiovascular mortality. Existing risk models are outdated due to changing patient profiles. Developing new, tailored risk calculators using machine learning (ML) is crucial to accurately predict in-hospital mortality (IHM) at various diagnostic stages and ultimately improve patient outcomes.

Objective:

To develop predictive models for IHM in patients with NSTE-ACS using ML methods and predictor sets obtained during the diagnostic process.

Methods:

Data were collected at the Research Institute of Cardiology, a branch of the Federal State Budgetary Scientific Institution "Tomsk National Research Medical Center of the Russian Academy of Sciences". The dataset comprised medical records of 1,144 patients with NSTE-ACS. For model development, we employed multivariable logistic regression, Random Forest, eXtreme Gradient Boosting (XGBoost), and CatBoost.

Results:

A key feature of the developed models is their practical applicability at different stages of the diagnostic process, utilizing the predictors available at each specific stage. Predictive accuracy increased with the expansion of the variable set, reaching an area under the curve (ROC-AUC) of 0.94 at the final stage. The SHAP - method was used to identify the most significant predictors for mortality risk (Killip class of acute heart failure, patient age, Charlson comorbidity index, and creatinine and hematocrit levels).

Conclusions:

The use of ML methods allowed the development of prognostic models for IHM in NSTE-ACS patients at various stages of the diagnostic process. Clinical Trial: This research is a retrospective observational study, which does not require mandatory registration as defined by the ICMJE.