Evaluating Large Language Models on Medical Evidence Summarization

Abstract

Recent advances in large language models (LLMs) have demonstrated remarkable successes in zero- and few-shot performance on various downstream tasks, paving the way for applications in high-stakes domains. In this study, we systematically examine the capabilities and limitations of LLMs, specifically GPT-3.5 and ChatGPT, in performing zero-shot medical evidence summarization across six clinical domains. We conduct both automatic and human evaluations, covering several dimensions of summary quality. Our study has demonstrated that automatic metrics often do not strongly correlate with the quality of summaries. Furthermore, informed by our human evaluations, we define a terminology of error types for medical evidence summarization. Our findings reveal that LLMs could be susceptible to generating factually inconsistent summaries and making overly convincing or uncertain statements, leading to potential harm due to misinformation. Moreover, we find that models struggle to identify the salient information and are more error-prone when summarizing over longer textual contexts.

No Time for ChitChat? Artificial intelligence Chatbots as a tool to identify research priorities in hip and knee arthroplasty.

“…[34] Further studies have shed light on the reliability of content generated by Chatbots when summarising medical literature reviews, highlighting the concerning possibility of Chatbots producing unfounded statements or even fabricated information. [35] Similar instances have been observed outside of healthcare, as exempli ed by a recent court case where a legal representative resorted to the use of fabricated citations from ChatGPT3 in support of their argument.…”

Can Large Language Models Provide Emergency Medical Help Where There Is No Ambulance? A Comparative Study on Large Language Model Understanding of Emergency Medical Scenarios in Resource-Constrained Settings

“…Another study on LLM outputs for medical evidence summarization tasks also employed both automatic and human evaluation. They defined summary quality based on coherence, factual consistency, comprehensiveness, and harmfulness Tang et al (2023). The researchers concluded that automatic metrics often do not strongly correlate with the quality of summaries.…”

Comparative Study on Automated Reference Summary Generation using BERT Models and ROUGE Score Assessment

“…Automatic text summarization is a sub-area of text mining in which a system determines the most informative information in the original text to produce a summary for certain tasks and users. To generate a summary, researchers have developed summarization systems for different purposes (i.e., single document summarization (Liu et al, 2019b), multi-document summarization (Chen et al, 2023), aspect-based opinion summarization (Wu et al, 2016), query-focused summarization (Baumel et al, 2016), update summarization (Delort, & Alfonseca, 2012), and cross-language document summarization (Wan, 2011) to summarize different text genres such as product reviews (Yu et al, 2016), news articles (Huang et al, 2011), political text (Sharevski et al, 2021), meeting text (Oya et al, 2014), scientific articles (Altmami, & Menai, 2022), online debates (Sanchan et al, 2017;Sanchan, Bontcheva, & Aker, 2020), and medical data (Abacha et al, 2021;Tang et al, 2023) with the assistant of Artificial Intelligence i.e., ChatGPT in their summarization task. Later, the generated summaries will be assessed against various criteria such as informativeness, text coherence, readability, and understandability.…”