Research → Hypothesis-free signal detection and drug repurposing
Hypothesis-free signal detection and drug repurposing
Most pharmacoepidemiology begins with a question about a particular drug and a particular outcome. A complementary approach begins with the data: screening large numbers of drug–outcome pairs at once, without deciding in advance which ones are worth looking at. We apply data-driven methods — tree-based scan statistics, Bayesian approaches, disproportionality analysis, and machine learning — across spontaneous reporting systems, national claims databases, and global pharmacovigilance data to surface associations that a targeted study would never have been designed to find.
The same screening logic runs in two directions. Pointed at harm, it is post-marketing safety surveillance: detecting signals early enough to matter, and separating the ones worth pursuing from the noise that any large database produces. Pointed at benefit, it becomes drug repurposing — generating candidate hypotheses for therapeutic effects outside a drug's approved indication. In both directions, what comes out is not a conclusion but a question, which we then test with the comparative designs used elsewhere in the lab.
Selected work
- Bacillus Calmette-Guérin (BCG) vaccine safety surveillance in the Korea Adverse Event Reporting System using the tree-based scan statistic and conventional disproportionality-based algorithms. Vaccine, 2020.
- Detecting early safety signals of infliximab using machine learning algorithms in the Korea Adverse Event Reporting System. Scientific Reports, 2022.
- A real-world pharmacovigilance study on cardiovascular adverse events of tisagenlecleucel using a machine learning approach. Scientific Reports, 2024.
- Regional trends in the opioid epidemic using safety signals of drug abuse and dependence: a disproportionality analysis using VigiBase. Drugs — Real World Outcomes, 2026.