A machine-learning analysis of 278 people in topiramate trials found that a genetic score predicting time until alcohol relapse, combined with baseline drinking metrics, identified individuals likely to respond well to the medication. Pre-treatment drinking severity remained the strongest predictor, though genetic data added meaningful information to the model .
Topiramate is one of the few medications with evidence supporting its use in alcohol use disorder (AUD), but not everyone responds equally well to treatment. This new analysis used machine learning to decode which features, including genetic data, could forecast who would benefit most from the drug.
The researchers analyzed data from 278 participants in topiramate randomized controlled trials. They built predictive models using 23 variables: four polygenic risk scores (PRS) related to alcohol use, nine baseline drinking measures (like drinks per day and heavy drinking frequency), and 10 clinical and sociodemographic factors. Polygenic risk scores are calculated from hundreds of thousands of genetic variants associated with a trait; in this case, researchers used four different PRS, including one specifically predicting time until relapse after treatment initiation.
After applying machine-learning methods to identify the most informative predictors, the final model narrowed down to just three variables: average drinks per day at baseline, percentage of heavy drinking days, and the "Time Until Relapse" polygenic risk score. This streamlined model explained about 30% of the variance in topiramate response (bias-corrected R2 = 0.30). The two drinking measures consistently ranked as the strongest predictors across all model iterations, but the genetic score for time until relapse added meaningful predictive power beyond drinking metrics alone.
The analysis identified "Likely Responders" (LRs) as individuals in the top four quintiles of predicted response where topiramate outperformed placebo in the trial data. These comprised roughly 80% of the sample. Compared with "Unlikely Responders," the LR group had lower baseline drinking severity and higher values on the "Time Until Relapse" PRS. The genetic marker's contribution was substantial enough to influence both who was classified as a likely responder and the magnitude of observed treatment effects. The findings suggest that incorporating genetic data into clinical decision-making algorithms could help clinicians identify candidates most apt to benefit from topiramate before treatment begins.
If you or someone you know is considering topiramate for alcohol use disorder, these results don't yet translate into a clinical test you can order. The study is a proof-of-concept showing that genetic information can improve prediction models beyond standard clinical assessment. Here's what matters in practice:
Current clinical approach remains primary. Your doctor will still evaluate baseline drinking severity, medical history, and other factors as the main criteria for prescribing decisions. Polygenic risk scores are not yet integrated into routine care for AUD treatment selection.
Genetic predisposition is one thread in a complex picture. The model explained 30% of treatment response variation, meaning 70% is determined by other factors: medication adherence, psychosocial support, environment, concurrent treatments, and individual biology not captured by these genetic variants. A good genetic score doesn't guarantee response; a less favorable score doesn't mean the medication won't work.
If topiramate is prescribed, engagement matters most. The strongest predictors were pre-treatment drinking metrics, not genetics. This underscores that baseline severity and drinking patterns drive much of the response signal. Behavioral interventions, alcohol-reduction strategies, and support networks remain essential alongside any medication.
Future precision prescribing may become available. As genetic databases grow and machine-learning models are prospectively validated (tested on new patients), clinicians may eventually have access to tools combining genetic and clinical data to refine treatment selection. This study provides an early map toward that goal but is not yet clinical practice.
| Characteristic | Detail |
|---|---|
| Study Type | Randomized controlled trial data analyzed with machine learning |
| Sample Size | 278 participants |
| Primary Outcome | Topiramate treatment response predicted by polygenic risk scores, drinking measures, and clinical variables |
| Key Predictors Retained | Average drinks per day, percent heavy drinking days, Time Until Relapse PRS |
| Model Performance | Bias-corrected R2 = 0.30 |
| Likely Responders | Top four quintiles of predicted response (80% of sample) |
| Journal | Alcohol and Alcoholism (Oxford, Oxfordshire) |
| PubMed ID | 42822858 |
Schacht, J. P., et al. "Polygenic risk scores as predictors of topiramate's effect in treating alcohol use disorder." *Alcohol and Alcoholism (Oxford, Oxfordshire)*, 2025. PubMed: 42822858
ProtocolEngine provides general health information based on published research. This is not medical advice. Consult a healthcare professional before starting any supplement or health protocol.