Thailand

Instruction-tuned Large Language Models (LLMs) can perform a wide range of tasks given natural language instructions to do so, but they are sensitive to how such instructions are phrased. 
This issue is especially concerning in healthcare, as clinicians are unlikely to be experienced prompt engineers and the potential consequences of inaccurate outputs are heightened in this domain. 

This raises a practical question: How robust are instruction-tuned LLMs to natural variations in the instructions provided for clinical NLP tasks?
We collect prompts from medical doctors across a range of tasks and quantify the sensitivity of seven LLMs---some general, others specialized---to natural (i.e., non-adversarial) instruction phrasings. 
We find that performance varies substantially across all models, and that---perhaps surprisingly---domain-specific models explicitly trained on clinical data are especially brittle, compared to their general domain counterparts. 
Further, arbitrary phrasing differences can affect fairness, e.g., valid but distinct instructions for mortality prediction yield a range both in overall performance, and in terms of differences between demographic groups.

ACL 2024

Open (Clinical) LLMs are Sensitive to Instruction Phrasings

clinical instruction tuning

llms

fairness

robustness

workshop paper

### Welcome!
The 62nd Annual Meeting of the Association for Computational Linguistics (ACL 2024) will take place in Bangkok, Thailand from August 11th to 16th, 2024. Our Virtual Poster Sessions will take place online Thursday, August 22, 2024.

You are required to register for this event. **Please register [here](https://2024.aclweb.org/registration). **

If you have already registered, please check your inbox for an email from Underline granting you access to ACL 2024 content.

Please register!

The 62nd Annual Meeting of the Association for Computational Linguistics (ACL 2024) will take place in Bangkok, Thailand from August 11th to 16th, 2024. More information will be announced soon.

Large language models (LLMs) have shown remarkable performance on many tasks in different domains. However, their performance in contextual biomedical machine reading comprehension (MRC) has not been evaluated in depth. In this work, we evaluate GPT on four contextual biomedical MRC benchmarks. We experiment with different conventional prompting techniques as well as introduce our own novel prompting method. To solve some of the retrieval problems inherent to LLMs, we propose a prompting strategy named Implicit Retrieval Augmented Generation (RAG) that alleviates the need for using vector databases to retrieve important chunks in traditional RAG setups. Moreover, we report qualitative assessments on the natural language generation outputs from our approach. The results show that our new prompting technique is able to get the best performance in two out of four datasets and ranks second in rest of them. Experiments show that modern-day LLMs like GPT even in a zero-shot setting can outperform supervised models, leading to new state-of-the-art (SoTA) results on two of the benchmarks.

Can GPT Redefine Medical Understanding? Evaluating GPT on Biomedical Machine Reading Comprehension

The consequences of a healthcare data breach can be devastating for the patients, providers, and payers. The average financial impact of a data breach in recent months has been estimated to be close to USD 10 million. This is especially significant for healthcare organizations in India that are managing rapid digitization while still establishing data governance procedures that align with the letter and spirit of the law. Computer-based systems for de-identification of personal information are vulnerable to data drift, often rendering them ineffective in cross-institution settings. Therefore, a rigorous assessment of existing de-identification against local health datasets is imperative to support the safe adoption of digital health initiatives in India. Using a small set of de-identified patient discharge summaries provided by an Indian healthcare institution, in this paper, we report the nominal performance of de-identification algorithms (based on language models) trained on publicly available non-Indian datasets, pointing towards a lack of cross-institutional generalization. Similarly, experimentation with off-the-shelf de-identification systems reveals potential risks associated with the approach. To overcome data scarcity, we explore generating synthetic clinical reports (using publicly available and Indian summaries) by performing in-context learning over Large Language Models (LLMs). Our experiments demonstrate the use of generated reports as an effective strategy for creating high-performing de-identification systems with good generalization capabilities.

Generation and De-Identification of Indian Clinical Discharge Summaries using LLMs

Domain adaptation is a widely used method in natural language processing (NLP) to improve the performance of a language model within a specific domain. This method is particularly common in the biomedical domain, which sees regular publication of numerous scientific articles. PubMed, a significant corpus of text, is frequently used in the biomedical domain. The primary objective of this study is to explore whether refining a pre-training dataset using specific quality metrics for scientific papers can enhance the performance of the resulting model. To accomplish this, we employ two straightforward journal impact metrics and conduct experiments by continually pre-training BERT on various subsets of the complete PubMed training set, we then evaluate the resulting models on biomedical language understanding tasks from the BLURB benchmark. Our results show that pruning using journal impact metrics is not efficient. But we also show that pre-training using fewer abstracts (but with the same number of training steps) does not necessarily decrease the resulting model's performance.

Pre-training data selection for biomedical domain adaptation using journal impact metrics

This paper presents the first study for temporal relation extraction in a zero-shot setting focusing on biomedical text. We employ two types of prompts and five Large Language Models (LLMs; GPT-3.5, Mixtral, Llama 2, Gemma, and PMC-LLaMA) to obtain responses about the temporal relations between two events. Our experiments demonstrate that LLMs struggle in the zero-shot setting, performing worse than fine-tuned specialized models in terms of F1 score. This highlights the challenging nature of this task and underscores the need for further research to enhance the performance of LLMs in this context. We further contribute a novel comprehensive temporal analysis by calculating consistency scores for each LLM. Our findings reveal that LLMs face challenges in providing responses consistent with the temporal properties of uniqueness and transitivity. Moreover, we study the relation between the temporal consistency of an LLM and its accuracy, and whether the latter can be improved by solving temporal inconsistencies. 
Our analysis shows that even when temporal consistency is achieved, the predictions can remain inaccurate.

Analysing zero-shot temporal relation extraction on clinical notes using temporal consistency

While the popularity of large, versatile language models like ChatGPT continues to rise, the landscape shifts when considering open-source models tailored to specific domains. Moreover, many areas, such as clinical documents, suffer from a scarcity of training data, often amounting to only a few hundred instances. Additionally, in certain settings, such as hospitals, cloud-based solutions pose privacy concerns, necessitating the deployment of language models on traditional hardware, such as single GPUs or powerful CPUs. To address these complexities, we conduct extensive experiments on both clinical entity detection and relation extraction in clinical documents using 1B parameter models. Our study delves into traditional fine-tuning, continuous pre-training in the medical domain, and instruction-tuning methods, providing valuable insights into their effectiveness in a multilingual setting. Our results underscore the importance of domain-specific models and pre-training for clinical natural language processing tasks. Furthermore, data augmentation using cross-lingual information improves performance in most cases, highlighting the potential for multilingual enhancements.

Get the Best out of 1B LLMs: Insights from Information Extraction on Clinical Documents

This paper presents the setup and results of the second edition of the BioLaySumm shared task on the Lay Summarisation of Biomedical Research Articles, hosted at the BioNLP Workshop at ACL 2024. 
In this task edition, we aim to build on the first edition's success by further increasing research interest in this important task and encouraging participants to explore novel approaches that will help advance the state-of-the-art. 
Encouragingly, we found research interest in the task to be high, with this edition of the task attracting a total of 53 participating teams, a significant increase in engagement from the previous edition. Overall, our results show that a broad range of innovative approaches were adopted by task participants, with a predictable shift towards the use of Large Language Models (LLMs).

Overview of the BioLaySumm 2024 Shared Task on the Lay Summarization of Biomedical Research Articles

As the number of scientific publications is growing at a rapid pace, it is difficult for laypeople to keep track of and understand the latest scientific advances, especially in the biomedical domain. While the summarization of scientific publications has been widely studied, research on summarization targeting laypeople has remained scarce. In this study, considering the lengthy input of biomedical articles, we have developed a lay summarization system through an extract-then-summarize framework with large language models (LLMs) to summarize biomedical articles for laypeople. Using a fine-tuned GPT-3.5 model, our approach achieves the highest overall ranking and demonstrates the best relevance performance in the BioLaySumm 2024 shared task.

UIUC_BioNLP at BioLaySumm: An Extract-then-Summarize Approach Augmented with Wikipedia Knowledge for Biomedical Lay Summarization

This paper presents our approaches for the BioLaySumm 2024 Shared Task. We evaluate two methods for generating lay summaries based on biomedical articles: (1) fine-tuning the Longformer-Encoder-Decoder (LED) model, and (2) zero-shot and few-shot prompting on GPT-4. In the fine-tuning approach, we individually fine-tune the LED model using two datasets: PLOS and eLife. This process is conducted under two different settings: one utilizing 50% of the training dataset, and the other utilizing the entire 100% of the training dataset. We compare the results of both methods with GPT-4 in zero-shot and few-shot prompting. The experiment results demonstrate that fine-tuning with 100% of the training data achieves better performance than prompting with GPT-4. However, under data scarcity circumstances, prompting GPT-4 seems to be a better solution.

DeakinNLP at BioLaySumm: Evaluating Fine-tuning Longformer and GPT-4 Prompting for Biomedical Lay Summarization

This paper presents our contribution to the BioLaySumm 2024 shared task of the 23rd BioNLP Workshop. The task is to create a lay summary, given a biomedical research article and its technical summary. As the input to the system could be large, a Longformer Encoder-Decoder (LED) has been used. We continuously pre-trained a general domain LED model with biomedical data to adapt it to this specific domain. In the pre-training phase, several pre-training tasks were aggregated to inject linguistic knowledge and increase the abstractivity of the generated summaries. Since the distribution of samples between the two datasets, eLife and PLOS, is unbalanced, we fine-tuned two models: one for eLife and another for PLOS. To increase the quality of the lay summaries of the system, we developed a regression model that helps us rank the summaries generated by the summarization models. This regression model predicts the quality of the summary in three different aspects: Relevance, Readability, and Factuality. We present the results of our models and a study to measure the ranking capabilities of the regression model.

ELiRF-VRAIN at BioLaySumm: Boosting Lay Summarization Systems Performance with Ranking Models

The BioNLP ACL'24 Shared Task on Streamlining Discharge Documentation aims to reduce the administrative burden on clinicians by automating the creation of critical sections of patient discharge letters. This paper presents our approach using the Llama3 8B quantized model to generate the "Brief Hospital Course" and "Discharge Instructions" sections. We employ a zero-shot method combined with Retrieval-Augmented Generation (RAG) to produce concise, contextually accurate summaries. Our contributions include the development of a curated template-based approach to ensure reliability and consistency, as well as the integration of RAG for word count prediction. We also describe several unsuccessful experiments to provide insights into our pathway for the competition. Our results demonstrate the effectiveness and efficiency of our approach, achieving high scores across multiple evaluation metrics.

Premium content

Open (Clinical) LLMs are Sensitive to Instruction Phrasings

Downloads

Next from ACL 2024

Can GPT Redefine Medical Understanding? Evaluating GPT on Biomedical Machine Reading Comprehension

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES