Machine Learning Approach to Predict In-Hospital Mortality in Patients Admitted for Peripheral Artery Disease in the...

Area of Science:

• Cardiovascular Medicine
• Medical Informatics
• Health Services Research

Background:

• Peripheral artery disease (PAD) impacts over 10 million individuals in the US, carrying significant risks of poor outcomes and premature mortality.
• The increasing use of machine learning (ML) on large datasets offers potential for predicting clinical outcomes in complex diseases like PAD.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting in-hospital mortality in patients hospitalized with a primary diagnosis of peripheral artery disease.
• To identify critical predictors of in-hospital mortality within this patient population using a national inpatient database.

Main Methods:

• Utilized inpatient data from the 2016-2019 National Inpatient Sample, identifying 150,921 patients with PAD diagnoses and procedures.
• Trained four machine learning models (logistic regression, random forest, light gradient boosting, extreme gradient boosting) to predict in-hospital death.
• Included patient characteristics, comorbidities, procedures, and hospital factors as predictive variables.

Main Results:

• In-hospital mortality was observed in 1.8% of the study cohort.
• All four ML models demonstrated comparable performance, with Area Under the Curve (AUC) ranging from 0.83 to 0.85.
• Key predictors of mortality included the total number of diagnoses and procedures, advanced age, endovascular revascularization, congestive heart failure, diabetes, and diabetes with complications.

Conclusions:

• Machine learning models are feasible and effective for predicting in-hospital mortality in patients with peripheral artery disease.
• These models can identify high-risk patients, enabling targeted interventions and personalized treatment strategies to improve outcomes.