An AdaBoost-Inspired Ensemble Method for ADR Signal Detection

• Published in: 2020 IEEE International Conference on Healthcare Informatics (ICHI) pp. 1 - 2
• Main Authors: Lin, Wen-Yang, Dai, I
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.11.2020
• Subjects: AdaBoost; ADR Signal Detection; Adverse Drug Reaction; Drugs; Ensemble Learning; Medical services; Monitoring; Organizations; Random forests; Signal detection; SRS; Stacking
• DOI: 10.1109/ICHI48887.2020.9374305

Spontaneous Reporting System (SRS) is the major mechanism employed for monitoring Adverse Drug Reaction (ADR), also the main repository for detecting suspect ADR signals. Most research organizations rely on a single ADR detection method to make decision. Although some organizations such as the British MHRA and EU have used the aggregative indicator method to combine several different rules based on the rule of thumb, or like some China scholars suggested using a simple voting ensemble method, there is no research comprehensively and in-depth investigating how to synergize different methods and showing any experimental results. In this paper, we propose an ensemble of ADR signal detectors that adopts the operating principle of AdaBoost in ensemble learning. The proposed method integrates the advantages from different ADR detection methods and automatically adjust the weight of each ADR detection method to improve the overall detection performance. Experiments conducted using FAERS datasets showed that our method significantly outperforms simple voting as well as random forest ensemble.