Exploring Mortality and Prognostic Factors of Heart Failure with In-Hospital and Emergency Patients by Electronic...

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning

Background:

• Advanced heart failure (HF) significantly impacts patient quality of life, leading to distressing symptoms and increased healthcare utilization.
• Understanding prognostic factors is crucial for managing advanced HF and improving patient outcomes.

Purpose of the Study:

• To investigate the relationship between mortality and potential prognostic factors in hospitalized and emergency patients with HF.
• To apply machine learning models for predicting HF mortality risk.

Main Methods:

• A case series study involving 3871 HF patients from 2014 to 2021.
• Data pre-processing of electronic medical records, including ICD codes and laboratory data.
• Application of various machine learning models (logistic regression, Cox regression, Random Forest, SVM, Adaboost, Naïve Bayes) and Shapley Additive Explanations for risk factor identification.

Main Results:

• Intensive care unit (ICU) history within one week (OR: 9.765) and prothrombin time (OR: 1.193) were significantly associated with HF mortality.
• Machine learning models demonstrated high performance (AUROC >0.87), with Naïve Bayes excelling in specificity and precision.
• Ensemble learning identified age, ICU history within one week, and respiratory rate as top mortality risk factors.

Conclusions:

• Machine learning models effectively predict HF mortality by analyzing clinical risk factors.
• Identifying key predictors like ICU history and prothrombin time aids in clinical decision-making for HF patient monitoring.
• This approach can enhance the quality of care for HF patients in hospital settings.