Original Research
Sunday, November 12, 2023
12:45 PM–02:15 PM
Abstract
Introduction:
Machine learning with time series data, unlike one-snap data collection, predicts sunitinib- and sorafenib-induced thyroid dysfunction on a real-time basis to aid the cancer therapy race against time.
Research Question or Hypothesis:
The study purpose was to develop machine learning models for sunitinib- and sorafenib-induced thyroid dysfunction using time-series data collection with threshold adjustment and web-based application.
Study Design:
This retrospective study included patients newly prescribed sunitinib or sorafenib from the de-identified clinical research database of Taipei Medical University.
Methods:
Time-series data of derivation and temporal validation cohorts were collected at baseline, and in the 1 st , 2 nd , 3 rd , 4 th , 5 th , 6 th , 9 th , 12 th , 18 th , 24 th , 30 th , and 36 th months after the index date. Logistic regression, random forest, Adaptive Boosting, Light Gradient Boosting Machine, and Gradient Boosting Decision Tree were employed. Performance was compared by accuracy, precision, recall, f1 score, the area under the receiver operating characteristic curve, and the area under the precision-recall curve. The optimal threshold was selected based on the maximum f1 score, and the model was further integrated into a web-based application.
Results:
The training cohort contained 609 patients, with 8.54% of cases, whereas 8.08% of cases occurred in the temporal validation cohort of 198 patients. The Gradient Boosting Decision Tree without resampling outperformed other models, with an AUPRC, AUROC, and f1 score of 0.60, 0.88, and 0.583, respectively. Higher cholesterol levels, longer days of medication use, and clear cell adenocarcinoma increased the risk of thyroid dysfunction. A web-based application was developed with predictive probability generated by the model.
Conclusion:
The best-performing Gradient Boosting Decision Tree without resampling model with time-series data can serve as an online progressive drug surveillance system for predicting sunitinib- and sorafenib-induced thyroid dysfunction.
Presenting Author
Hsiang-Yin Chen Pharm.D, M.S.Wan-Fang Hospital, Taipei Medical University
Authors
Yi-En Ku B.S.
Taipei Medical University
Wen-Nung Lie PhD.
National Chung Cheng University
Fan-Ying Chan M.S.
Taipei Medical University