Prediction of the Outcome of Pakistani Heart Failure Patients by Various Supervised Machine Learning Methods

Abstract

Aim: To foresee the outcome of heart failure(HF) in Pakistani patients with potential predictors and through various machine learning (ML) methods.

Study design:The secondary data of Pakistani patients is taken from the UCI repository in which a cross-sectional, analytical study was planned.

Place and duration: This data was collected in April-December, 2015 at the Institute of Cardiology and Allied hospital Faisalabad-Pakistan.

Methodology: The data set consisted of299 patients distributed among male (194) and female patients (105). Ages, serum sodium (SS), serum creatinine (SC), gender, smoking, high blood pressure (HBP), ejection fraction (EF), anemia, platelets, Creatinine Phosphokinase (CPK), and diabetes were considered as the potential predictors for predicting the outcome of HF.The data set was analyzed with the help of various machine learning (ML) predictive models including Logistic regression (LR), K-nearest neighbor (KNN), and Decision trees (DT).

Results: The ages of the patients were within 60.833±11.894 years. Out of 299 patients, 129 were anemic, 105 had high blood pressure (HBP), and 96 had a smoking history. A statistical model was estimated by applying LR which assisted us in identifying the significant predictors. The sensitivity of the LRwas observed to be 92.1%, whereas 85.6% of the outcome of HF patients was correctly predicted by this model (LR) and DT achieved89.6% prediction accuracy.

Conclusion:Since HF is a substantial reason for deaths in Pakistan. Therefore, the identification of its potential risk factors and its accurate prediction by some modern tools are highly demanded. This study applied ML tools for the said task and concluded that among all the fitted ML models, DT predicted the correct outcome for HF patients proficiently.

Keywords: Heart failure, machine learning, logistic regression, k-nearest neighbor, decision trees