JMIR Publications

ABSTRACT

Background:

With the prevalence of cardiovascular disease (CVD) increasing, early prediction and accurate assessment of heart failure (HF) risk is crucial to meet the clinical demand.

Objective:

We sought to develop machine-learning models based on real-world electronic health records (EHRs) to predict one-year in-hospital mortality, the use of positive inotropic agents and one-year all-cause readmission rate.

Methods:

For this single-center study, we recruited patients with newly diagnosed HF hospitalized between December 2010 and August 2018 at the First Affiliated Hospital of Dalian Medical University. The models were constructed for a population set (90:10) by using 79 variables during the first hospitalization. Logistic regression (LR), a support vector machine (SVM), an artificial neural network (ANN), random forest (RF), and eXtreme Boosting (XGBoost) were investigated for outcome predictions.

Results:

Of the 13,602 patients with HF enrolled, 537 (3.95%) were died within one year, and 2779 patients (20.43%) had history of positive inotropic agents. The performance of predictive models for one-year in-hospital mortality (AUCs 0.92-0.99), positive inotropic medication (AUCs 0.86-0.94) and one-year readmission rates (AUCs 0.62-0.89) were improved by ML algorithms. A decision tree of mortality risk was created and stratified by single variables at levels of high-sensitive cardiac troponin (hs-cTnl) (<0.068μg/L), followed by lymphocyte percentage (< 14.688%) and neutrophil count (4.870 x 109/L).

Conclusions:

ML techniques based on a large scale of clinical variables can improve outcome predictions for patients with HF. The mortality decision tree may contribute to guide better clinical risk assessment and clinical decision making.