Machine Learning-Based Prediction of Digoxin Toxicity in Heart Failure: A Multicenter Retrospective Study

Abstract

Digoxin toxicity (plasma digoxin concentration ≥0.9 ng/mL) is associated with worsening heart failure (HF). Decision tree (DT) analysis, a machine learning method, has a flowchart-like model where users can easily predict the risk of adverse drug reactions. The present study aimed to construct a flowchart using DT analysis that can be used by medical staff to predict digoxin toxicity. We conducted a multicenter retrospective study involving 333 adult patients with HF who received oral digoxin treatment. In this study, we employed a chi-squared automatic interaction detection algorithm to construct DT models. The dependent variable was set as the plasma digoxin concentration (≥ 0.9 ng/mL) in the trough during the steady state, and factors with p < 0.2 in the univariate analysis were set as the explanatory variables. Multivariate logistic regression analysis was conducted to validate the DT model. The accuracy and misclassification rates of the model were evaluated. In the DT analysis, patients with creatinine clearance <32 mL/min, daily digoxin dose ≥1.6 µg/kg, and left ventricular ejection fraction ≥50% showed a high incidence of digoxin toxicity (91.8%; 45/49). Multivariate logistic regression analysis revealed that creatinine clearance <32 mL/min and daily digoxin dose ≥1.6 µg/kg were independent risk factors. The accuracy and misclassification rates of the DT model were 88.2 and 46.2 2.7%, respectively. Although the flowchart created in this study needs further validation, it is straightforward and potentially useful for medical staff in determining the initial dose of digoxin in patients with HF.

Evaluation of a joint workshop on study design for hospital and community pharmacists: a retrospective cross-sectional survey

“…For multivariate logistic regression analysis, low involvement was used as the objective variable and factors that exhibited p < 0.05 in the univariate analysis. DT analysis was performed according to our previous study [ 6 ], based on the chi-squared automatic interaction detection algorithm. All statistical analyses were performed using SPSS Statistics version 27 (IBM Japan, Tokyo, Japan), and the significance level was set at p < 0.05.…”