Machine Learning for Detection of Safety Signals From Spontaneous Reporting System Data: Example of Nivolumab and
Ji-Hwan Bae1, Yeon-Hee Baek1, Jeong-Eun Lee1
1School of Pharmacy, Sungkyunkwan University, Suwon-si, South Korea.
Frontiers in Pharmacology
|February 25, 2021
Summary
Machine learning, specifically gradient boosting machine (GBM), effectively detects adverse drug reaction (ADR) signals for anticancer agents nivolumab and docetaxel. GBM outperformed traditional methods, identifying more novel safety signals.
Area of Science:
- Pharmacovigilance
- Machine Learning in Drug Safety
- Oncology Drug Surveillance
Background:
- Traditional methods for detecting adverse drug reaction (ADR) signals have limitations.
- The application of machine learning (ML) for ADR signal detection requires further evaluation.
- Nivolumab and docetaxel represent newer and established anticancer agents, respectively.
Purpose of the Study:
- To assess the feasibility of ML algorithms in identifying ADR signals for nivolumab and docetaxel.
- To compare the performance of ML algorithms against traditional disproportionality analysis methods.
- To evaluate the capability of ML in discovering novel ADR signals.
Main Methods:
- A safety surveillance study utilized the Korea national spontaneous reporting database (2009-2018).
- Machine learning algorithms (gradient boosting machine [GBM], random forest [RF]) and traditional methods (reporting odds ratio [ROR], information component [IC]) were trained and evaluated.
- Performance was assessed using the area under the curve (AUC), comparing predictions against known ADRs and then applying methods to unknown ADR datasets.
Main Results:
- GBM demonstrated superior predictive performance with the highest AUC values (0.97 for nivolumab, 0.93 for docetaxel).
- GBM identified significantly more novel safety signals for both nivolumab (24 vs. ROR, 9 vs. IC) and docetaxel (82 vs. ROR, 76 vs. IC) compared to traditional methods.
- Random forest also showed strong performance, while ROR and IC had considerably lower AUC values.
Conclusions:
- Machine learning, particularly GBM, is a highly effective tool for ADR signal detection in pharmacovigilance.
- GBM significantly enhances the ability to identify new and previously unknown safety signals for anticancer drugs.
- ML algorithms offer a more powerful approach to drug safety surveillance compared to conventional disproportionality analyses.
