Accepted for/Published in: JMIR Medical Informatics
Date Submitted: May 22, 2025
Date Accepted: May 5, 2026
Warning: This is an author submission that is not peer-reviewed or edited. Preprints - unless they show as "accepted" - should not be relied on to guide clinical practice or health-related behavior and should not be reported in news media as established information.
Classification of Randomised Controlled Trial Abstracts by Intervention Type using Zero-Shot and Llama 3 large language models
ABSTRACT
Background:
Artificial intelligence has gained relevance due to its potential to reduce the work burden in evidence synthesis or bibliometric projects. While main focus has been lately on the use of instruct Large Language Models, Zero-shot classification models haven’t been tested for such task. These models are large language models trained on large datasets of labelled data able to categorize a text among proposed labels, irrespective of the text domain or the topic. They are relatively small, able to run on consumer grade computers, and almost parameter free.
Objective:
In our study, we use abstracts of randomized clinical trials in rheumatology as a case example to evaluate their performance in classifying types of interventions and compare it to classification obtained with Llama 3 8B and a human gold standard.
Methods:
We classified all rheumatology RCT abstracts published between 2009 and 2022 (n=1,055) as “drug” or “non-drug” using two zero-shot text classification models (DeBERTa and BART) and few-shot prompting using LLama3 8B. Different labelling of categories provided to the zero shot classification models and different prompts provided to Llama3 were tested. Performance was evaluated using accuracy and predictive value of both categories against a human gold standard.
Results:
Most RCTs (43.9%) assessed pharmacological interventions. The DeBERTa model achieved the highest accuracy (88.1% [95%CI: 86-90%]) when using the “drug” and “non-drug” labels. Llama 3 and few-shot prompting had slightly higher accuracy and predictive values. Both zero-shot and Llama 3 models had performance on par with a human without experience in evidence synthesis (86.9% [95%CI: 83.6-87.8%] accuracy). Misclassifications happened for trials where the intervention was harder to classify, such as procedures (e.g., intra-articular injections), food compounds, vitamins, supplements, or biological treatments.
Conclusions:
This study shows the potential of Zero-shot classification models for simple classification tasks, which had an accuracy comparable to an untrained human. These models are a potential tool to streamline systematic review tasks for bibliometric studies in classifying abstracts by replacing one reviewer. Clinical Trial: NOT APPLICABLE
Citation
Request queued. Please wait while the file is being generated. It may take some time.
Copyright
© The authors. All rights reserved. This is a privileged document currently under peer-review/community review (or an accepted/rejected manuscript). Authors have provided JMIR Publications with an exclusive license to publish this preprint on it's website for review and ahead-of-print citation purposes only. While the final peer-reviewed paper may be licensed under a cc-by license on publication, at this stage authors and publisher expressively prohibit redistribution of this draft paper other than for review purposes.