Large-Scale mapping of psychotropic weight effects: LLM-Based event detection across 222,545 deduplicated real-world patient narratives

Psychotropic drug-induced weight change is clinically important but remains difficult to compare across large numbers of medications. We analysed 222,545 deduplicated user-generated reviews covering 176 psychotropic medications and acetaminophen as a negative control. A fine-tuned classifier identified reviews describing weight gain, weight loss, or no change, and a Qwen3-8B language model extracted quantified weight changes in kilograms. We derived a Net Weight Effect (NWE), defined as the proportion of reviews describing weight gain minus the proportion describing weight loss (range −1 to +1), and the mean weight change in kilograms, both with bootstrap 95% confidence intervals. Antipsychotics had the highest class-level weight liability (NWE + 0.154; mean +12.6 kg), whereas stimulants showed the most weight-reducing profile (NWE − 0.050; mean −7.1 kg). Olanzapine, mirtazapine, and risperidone ranked highest for weight gain, whereas bupropion/naltrexone and topiramate ranked lowest. Long-term use (>6 months) was associated with a larger mean gain than medium-term use (+8.8 kg vs +2.3 kg), and females reported weight gain more frequently than males. Per-drug rankings were highly concordant across the four source platforms (Kendall’s W = 0.83). In an exploratory receptor-affinity analysis, alpha-2 adrenergic affinity was the strongest correlate of NWE (rho = 0.594; permutation FDR p = 0.006), whereas H1 histamine affinity was the strongest correlate of mean weight change (rho = 0.638; permutation FDR p = 0.042). Large-scale AI-assisted analysis of patient narratives can therefore generate clinically useful real-world rankings of psychotropic weight liability while also highlighting mechanistically plausible, though correlational, pharmacological associations.

Authors

Institutions

Publication Details

Journal
Translational Psychiatry
Published
2026-11-16
DOI
https://doi.org/10.1038/s41398-026-04551-z
Primary Topic
Pharmacology and Obesity Treatment
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Large-Scale mapping of psychotropic weight effects: LLM-Based event detection across 222,545 deduplicated real-world patient narratives

Julien Godet, Hervé Javelot, Taïoh Yokoyama, Johan H. G. Natter et al.
Translational Psychiatry
Pharmacology and Obesity Treatment
article

Large-Scale mapping of psychotropic weight effects: LLM-Based event detection across 222,545 deduplicated real-world patient narratives

Julien Godet, Hervé Javelot, Taïoh Yokoyama, Johan H. G. Natter, Michel Bruno
article en

Abstract

Psychotropic drug-induced weight change is clinically important but remains difficult to compare across large numbers of medications. We analysed 222,545 deduplicated user-generated reviews covering 176 psychotropic medications and acetaminophen as a negative control. A fine-tuned classifier identified reviews describing weight gain, weight loss, or no change, and a Qwen3-8B language model extracted quantified weight changes in kilograms. We derived a Net Weight Effect (NWE), defined as the proportion of reviews describing weight gain minus the proportion describing weight loss (range −1 to +1), and the mean weight change in kilograms, both with bootstrap 95% confidence intervals. Antipsychotics had the highest class-level weight liability (NWE + 0.154; mean +12.6 kg), whereas stimulants showed the most weight-reducing profile (NWE − 0.050; mean −7.1 kg). Olanzapine, mirtazapine, and risperidone ranked highest for weight gain, whereas bupropion/naltrexone and topiramate ranked lowest. Long-term use (>6 months) was associated with a larger mean gain than medium-term use (+8.8 kg vs +2.3 kg), and females reported weight gain more frequently than males. Per-drug rankings were highly concordant across the four source platforms (Kendall’s W = 0.83). In an exploratory receptor-affinity analysis, alpha-2 adrenergic affinity was the strongest correlate of NWE (rho = 0.594; permutation FDR p = 0.006), whereas H1 histamine affinity was the strongest correlate of mean weight change (rho = 0.638; permutation FDR p = 0.042). Large-scale AI-assisted analysis of patient narratives can therefore generate clinically useful real-world rankings of psychotropic weight liability while also highlighting mechanistically plausible, though correlational, pharmacological associations.

Translational Psychiatry
Centre National de la Recherche Scientifique (FR), Laboratoire des Sciences de l'Ingénieur, de l'Informatique et de l'Imagerie (FR), Observatoire Régional de la Santé d’Alsace (FR), Hôpitaux Universitaires de Strasbourg (FR), Laboratoire de pharmacologie et de toxicologie neurocardiovasculaire (FR), Université de Strasbourg (FR)
Good health and well-being
Pharmacology and Obesity Treatment
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon