Development and validation of a machine learning model for online predicting the risk of in heart failure: based on the routine blood test and their derived parameters.

Abstract

Background: Heart failure (HF), a core component of cardiovascular diseases, is characterized by high morbidity and mortality worldwide. By collecting and analyzing routine blood data, machine learning models were built to identify the patterns of changes in blood indicators related to HF.

Methods: We conducted a statistical analysis of routine blood data from 226 patients who visited Zhejiang Provincial Hospital of Traditional Chinese Medicine (Hubin) between May 1, 2024, and June 30, 2024. The patients were divided into an experimental group (HF patients) and a normal control group. Additionally, 211 patients from the Qiantang and Xixi centers formed an independent external validation cohort. This study used both univariate and multivariate analyses to identify the risk factors associated with HF. Variables associated with HF were selected using LASSO regression analysis. In addition, eight different machine learning algorithms were applied for prediction, and the prediction performances of these algorithms were comprehensively evaluated using the receiver operating characteristic curve, area under the curve (AUC), calibration curve analysis, and decision curve analysis and confusion matrix.

Conclusions: Using LASSO regression analysis, leukocyte, neutrophil, red blood cell, hemoglobin, platelet, and monocyte-to-lymphocyte ratios were identified as risk factors for HF. Among the evaluated models, the random forest model exhibited the best performance. In the validation cohort, the area under the curve (AUC) of the model was 0.948, while that of the test cohort was 1.000. The calibration curve revealed good agreement between the actual and predicted probabilities, whereas the decision curve showed the significant clinical application of the model. Additionally, the AUC of the model in the external independent test cohort was 0.945.

Discussion: We used an online predictive tool to develop a predictive machine-learning model. The main purpose of this model was to predict the probability of developing HF in the future. This prediction can provide strong support and references for clinicians when making decisions. This online forecasting tool not only processes a large amount of data but also continuously optimizes and adjusts the accuracy of the model according to the latest medical research and clinical data. We hope to identify high-risk patients for early intervention to reduce the incidence of HF and improve their quality of life.