Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

A novel language model for predicting serious adverse event outcomes in clinical trials from their prospective registrations

Created by
  • Haebom

Author

Qixuan Hu, Xumou Zhang, Jinman Kim, Florence Bourgeois, Adam G. Dunn

Outline

This paper evaluated a method for predicting the incidence of serious adverse events (SAEs) using only clinical trial registration data from ClinicalTrials.gov. Analyzing data from 22,107 two-arm, parallel-group comparative trials, we developed two models: a classification model to predict whether the experimental group would have a higher rate of SAEs than the control group, and a regression model to predict the rate of SAEs in the control group. Pretrained language models, such as ClinicalT5 and BioBERT, were used for feature extraction, and a sliding window technique was used to process long trial descriptions. The optimal model (ClinicalT5+Transformer+MLP) achieved an area under the curve (AUC) of 77.6% for predicting the group with a higher rate of SAEs and an 18.6% root mean square error (RMSE) for predicting the rate of SAEs in the control group. The sliding window technique outperformed the direct comparison method.

Takeaways, Limitations

•
Takeaways:
◦
We demonstrate that summary outcome data from ClinicalTrials.gov can be used to predict the incidence of SAEs in clinical trials.
◦
Suggests the possibility of identifying discrepancies between expected and reported safety outcomes using predictive models.
◦
We present a method for effectively processing semantic representations of clinical trial narratives using pre-trained language models and sliding window techniques.
◦
Has the potential to contribute to improving clinical trial design and monitoring.
•
Limitations:
◦
The model's prediction accuracy is not perfect, and the AUC and RMSE values still have room for improvement.
◦
The data used in the analysis were limited to data registered in ClinicalTrials.gov, so further research is needed to determine generalizability.
◦
Generalization performance validation is needed for various types of clinical trials and SAEs.
◦
Further research is needed on the interpretability of predictive models.
👍
Made with Slashpage