JMIR Publications

ABSTRACT

Background:

Machine learning offers new solutions to predicting life-threatening, unpredictable amiodarone-induced thyroid dysfunction. Traditional regression approaches for amiodarone-induced thyroid adverse effects without time-series consideration of features have yielded suboptimal predictions.

Objective:

To develop and validate the machine learning models for forecasting individualized amiodarone-induced thyroid dysfunction risk and to optimize a machine learning-based risk stratification scheme by a resampling method and readjusting the decision thresholds.

Methods:

The clinical research database included patients receiving amiodarone from January 2013 to December 2017. The training set was composed of data from Taipei Medical University Hospital (TMUH) and Wan Fang Hospital (WFH), while data from Taipei Medical University Shuang Ho Hospital (SHH) were used as the external test set. This study constructed 16 machine learning models, using eXtreme Gradient Boosting, AdaBoost, K-Nearest Neighbor, and logistic regression models along with the original and three resampling methods, oversampling with Borderline-Synthesized Minority Oversampling Technique, undersampling edited nearest neighbor, and over- and undersampling hybrid methods. The model performance was compared using the accuracy, precision, recall, F1 score, geometric mean, area under the curve of the receiver operating characteristic curve (AUROC), and the area under the precision-recall curve (AUPRC). After determining the best model, the decision threshold was readjusted to decide the best cutoff value.

Results:

The training set contained 4075 patients from TMUH and WFH, while the external test set was 2422 patients from SHH, with 583 (14.3%) developing amiodarone-induced thyroid dysfunction. Within the external test set, 275 (11.4%) were included in the amiodarone-induced thyroid dysfunction group. The eXtreme Gradient Boosting oversampling machine learning model demonstrated the best predictive outcomes among all 16 models. The accuracy, precision, recall, F1 score, G-mean, AUPRC, and AUROC were 0.923, 0.632, 0.756, 0.688, 0.845, 0.751, and 0.934, respectively. After readjusting the cutoff to the best cutoff value, the F1 score reached 0.699.

Conclusions:

Machine learning models combined with resampling methods can predict amiodarone-induced thyroid dysfunction and serve as a supportive tool for individualized risk prediction and clinical decision support tools.