Machine learning models to predict the warfarin discharge dosage using clinical information of inpatients from South Korea

Abstract

As warfarin has a narrow therapeutic window and obvious response variability among individuals, it is difficult to rapidly determine personalized warfarin dosage. Adverse drug events(ADE) resulting from warfarin overdose can be critical, so that typically physicians adjust the warfarin dosage through the INR monitoring twice a week when starting warfarin. Our study aimed to develop machine learning (ML) models that predicts the discharge dosage of warfarin as the initial warfarin dosage using clinical data derived from electronic medical records within 2 days of hospitalization. During this retrospective study, adult patients who were prescribed warfarin at Asan Medical Center (AMC) between January 1, 2018, and October 31, 2020, were recruited as a model development cohort (n = 3168). Additionally, we created an external validation dataset (n = 891) from a Medical Information Mart for Intensive Care III (MIMIC-III). Variables for a model prediction were selected based on the clinical rationale that turned out to be associated with warfarin dosage, such as bleeding. The discharge dosage of warfarin was used the study outcome, because we assumed that patients achieved target INR at discharge. In this study, four ML models that predicted the warfarin discharge dosage were developed. We evaluated the model performance using the mean absolute error (MAE) and prediction accuracy. Finally, we compared the accuracy of the predictions of our models and the predictions of physicians for 40 data point to verify a clinical relevance of the models. The MAEs obtained using the internal validation set were as follows: XGBoost, 0.9; artificial neural network, 0.9; random forest, 1.0; linear regression, 1.0; and physicians, 1.3. As a result, our models had better prediction accuracy than the physicians, who have difficulty determining the warfarin discharge dosage using clinical information obtained within 2 days of hospitalization. We not only conducted the internal validation but also external validation. In conclusion, our ML model could help physicians predict the warfarin discharge dosage as the initial warfarin dosage from Korean population. However, conducting a successfully external validation in a further work is required for the application of the models.

Figure 1

Overview of the workflow. Our single-center study complied with the workflow in the following order: (1) based on clinical rationales, we extracted 17 features associated with warfarin from the electronic medical record (EMR) database up to the second day of hospitalization; (2) we conducted data pre-processing, such as missing value imputation, outlier filtering, and normalization, and created the final dataset; (3) the models that predict the warfarin discharge dosage using the dataset were developed. (4) The discharge dose was used as ground-truth to calculate the error with physicians initial dose and model predictions. The yellow circles represent the one day during hospitalization. The green and orange circle indicate the first warfarin dosage after hospitalization and the warfarin dosage at discharge, respectively.

Figure 2

Cohort diagram. AMC Asan medical center, EHR electronic health record, MIMIC-III medical information mart for intensive care III.

Figure 3

Bar plot representing the MAE. Performance abilities of the models based on the MAE were visualized using the internal validation (n = 634) and external validation (n = 891) sets. The unit of error is milligram (mg).

Figure 4

SHAP beeswarm plot of the features affecting the predictions of the XGBoost model. Features are ranked in descending order based on the absolute value of their influence on the XGBoost model. The x-axis indicates SHAP values. Each dot denotes a data point. Colors represent high values (red) or low values (blue) of specific data points. SHAP Shapley additive explanations.

Figure 5

SHAP waterfall plot. The x-axis represents the individual warfarin dosage prediction of the models. The y-axis represents the input features of the models. E[f(x)] (2.714) represents the baseline value, which is the model output of the entire dataset, and f(x) represents the individual model output for each patient. Each arrow indicates whether a specific feature increased (red) or decreased (blue) the warfarin dosage. SHAP Shapley additive explanations.

Figure 6

Distribution of warfarin dosage predicted by the physicians and models. We identified the predictions of five physicians and that of our model using 40 data points. The actual warfarin dosage at discharge (yellow) was distributed evenly within 1–5 mg. The average prediction dosage of our model (green) was 2 $.$ 6 mg, and the maximum prediction dosage of our model was 3 mg. The prediction accuracy of model and the physicians are 50% and 23%, respectively. WFR warfarin, DC discharge.