OBJECTIVE: Although supervised machine learning is popular for information extraction from clinical notes, creating large annotated datasets requires extensive domain expertise and is time-consuming. Meanwhile, large language models (LLMs) have demonstrated promising transfer learning capability. In this study, we explored whether recent LLMs could reduce the need for large-scale data annotations.
METHODS: We curated a dataset of 769 breast cancer pathology reports, manually labeled with 12 categories, to compare zero-shot classification capability of the following LLMs: GPT-4, GPT-3.5, Starling, and ClinicalCamel, with task-specific supervised classification performance of 3 models: random forests, long short-term memory networks with attention (LSTM-Att), and the UCSF-BERT model.
RESULTS: Across all 12 tasks, the GPT-4 model performed either significantly better than or as well as the best supervised model, LSTM-Att (average macro F1-score of 0.86 vs 0.75), with advantage on tasks with high label imbalance. Other LLMs demonstrated poor performance. Frequent GPT-4 error categories included incorrect inferences from multiple samples and from history, and complex task design, and several LSTM-Att errors were related to poor generalization to the test set.
CONCLUSIONS: On tasks where large annotated datasets cannot be easily collected, LLMs can reduce the burden of data labeling. However, if the use of LLMs is prohibitive, the use of simpler models with large annotated datasets can provide comparable results.
CONCLUSIONS: GPT-4 demonstrated the potential to speed up the execution of clinical NLP studies by reducing the need for large annotated datasets. This may increase the utilization of NLP-based variables and outcomes in clinical studies.

BACKGROUND: Systematically screening published literature to determine the relevant publications to synthesize in a review is a time-consuming and difficult task. Large language models (LLMs) are an emerging technology with promising capabilities for the automation of language-related tasks that may be useful for such a purpose.
METHODS: LLMs were used as part of an automated system to evaluate the relevance of publications to a certain topic based on defined criteria and based on the title and abstract of each publication. A Python script was created to generate structured prompts consisting of text strings for instruction, title, abstract, and relevant criteria to be provided to an LLM. The relevance of a publication was evaluated by the LLM on a Likert scale (low relevance to high relevance). By specifying a threshold, different classifiers for inclusion/exclusion of publications could then be defined. The approach was used with four different openly available LLMs on ten published data sets of biomedical literature reviews and on a newly human-created data set for a hypothetical new systematic literature review.
RESULTS: The performance of the classifiers varied depending on the LLM being used and on the data set analyzed. Regarding sensitivity/specificity, the classifiers yielded 94.48%/31.78% for the FlanT5 model, 97.58%/19.12% for the OpenHermes-NeuralChat model, 81.93%/75.19% for the Mixtral model and 97.58%/38.34% for the Platypus 2 model on the ten published data sets. The same classifiers yielded 100% sensitivity at a specificity of 12.58%, 4.54%, 62.47%, and 24.74% on the newly created data set. Changing the standard settings of the approach (minor adaption of instruction prompt and/or changing the range of the Likert scale from 1-5 to 1-10) had a considerable impact on the performance.
CONCLUSIONS: LLMs can be used to evaluate the relevance of scientific publications to a certain review topic and classifiers based on such an approach show some promising results. To date, little is known about how well such systems would perform if used prospectively when conducting systematic literature reviews and what further implications this might have. However, it is likely that in the future researchers will increasingly use LLMs for evaluating and classifying scientific publications.

BACKGROUND: Large language models (LLMs) have demonstrated impressive performances in various medical domains, prompting an exploration of their potential utility within the high-demand setting of emergency department (ED) triage. This study evaluated the triage proficiency of different LLMs and ChatGPT, an LLM-based chatbot, compared to professionally trained ED staff and untrained personnel. We further explored whether LLM responses could guide untrained staff in effective triage.
OBJECTIVE: This study aimed to assess the efficacy of LLMs and the associated product ChatGPT in ED triage compared to personnel of varying training status and to investigate if the models\' responses can enhance the triage proficiency of untrained personnel.
METHODS: A total of 124 anonymized case vignettes were triaged by untrained doctors; different versions of currently available LLMs; ChatGPT; and professionally trained raters, who subsequently agreed on a consensus set according to the Manchester Triage System (MTS). The prototypical vignettes were adapted from cases at a tertiary ED in Germany. The main outcome was the level of agreement between raters\' MTS level assignments, measured via quadratic-weighted Cohen κ. The extent of over- and undertriage was also determined. Notably, instances of ChatGPT were prompted using zero-shot approaches without extensive background information on the MTS. The tested LLMs included raw GPT-4, Llama 3 70B, Gemini 1.5, and Mixtral 8x7b.
RESULTS: GPT-4-based ChatGPT and untrained doctors showed substantial agreement with the consensus triage of professional raters (κ=mean 0.67, SD 0.037 and κ=mean 0.68, SD 0.056, respectively), significantly exceeding the performance of GPT-3.5-based ChatGPT (κ=mean 0.54, SD 0.024; P<.001). When untrained doctors used this LLM for second-opinion triage, there was a slight but statistically insignificant performance increase (κ=mean 0.70, SD 0.047; P=.97). Other tested LLMs performed similar to or worse than GPT-4-based ChatGPT or showed odd triaging behavior with the used parameters. LLMs and ChatGPT models tended toward overtriage, whereas untrained doctors undertriaged.
CONCLUSIONS: While LLMs and the LLM-based product ChatGPT do not yet match professionally trained raters, their best models\' triage proficiency equals that of untrained ED doctors. In its current form, LLMs or ChatGPT thus did not demonstrate gold-standard performance in ED triage and, in the setting of this study, failed to significantly improve untrained doctors\' triage when used as decision support. Notable performance enhancements in newer LLM versions over older ones hint at future improvements with further technological development and specific training.

BACKGROUND: Large language models (LLMs) have the potential to support promising new applications in health informatics. However, practical data on sample size considerations for fine-tuning LLMs to perform specific tasks in biomedical and health policy contexts are lacking.
OBJECTIVE: This study aims to evaluate sample size and sample selection techniques for fine-tuning LLMs to support improved named entity recognition (NER) for a custom data set of conflicts of interest disclosure statements.
METHODS: A random sample of 200 disclosure statements was prepared for annotation. All \"PERSON\" and \"ORG\" entities were identified by each of the 2 raters, and once appropriate agreement was established, the annotators independently annotated an additional 290 disclosure statements. From the 490 annotated documents, 2500 stratified random samples in different size ranges were drawn. The 2500 training set subsamples were used to fine-tune a selection of language models across 2 model architectures (Bidirectional Encoder Representations from Transformers [BERT] and Generative Pre-trained Transformer [GPT]) for improved NER, and multiple regression was used to assess the relationship between sample size (sentences), entity density (entities per sentence [EPS]), and trained model performance (F1-score). Additionally, single-predictor threshold regression models were used to evaluate the possibility of diminishing marginal returns from increased sample size or entity density.
RESULTS: Fine-tuned models ranged in topline NER performance from F1-score=0.79 to F1-score=0.96 across architectures. Two-predictor multiple linear regression models were statistically significant with multiple R2 ranging from 0.6057 to 0.7896 (all P<.001). EPS and the number of sentences were significant predictors of F1-scores in all cases ( P<.001), except for the GPT-2_large model, where EPS was not a significant predictor (P=.184). Model thresholds indicate points of diminishing marginal return from increased training data set sample size measured by the number of sentences, with point estimates ranging from 439 sentences for RoBERTa_large to 527 sentences for GPT-2_large. Likewise, the threshold regression models indicate a diminishing marginal return for EPS with point estimates between 1.36 and 1.38.
CONCLUSIONS: Relatively modest sample sizes can be used to fine-tune LLMs for NER tasks applied to biomedical text, and training data entity density should representatively approximate entity density in production data. Training data quality and a model architecture\'s intended use (text generation vs text processing or classification) may be as, or more, important as training data volume and model parameter size.

UNASSIGNED: The generation of structured documents for clinical trials is a promising application of large language models (LLMs). We share opportunities, insights, and challenges from a competitive challenge that used LLMs for automating clinical trial documentation.
UNASSIGNED: As part of a challenge initiated by Pfizer (organizer), several teams (participant) created a pilot for generating summaries of safety tables for clinical study reports (CSRs). Our evaluation framework used automated metrics and expert reviews to assess the quality of AI-generated documents.
UNASSIGNED: The comparative analysis revealed differences in performance across solutions, particularly in factual accuracy and lean writing. Most participants employed prompt engineering with generative pre-trained transformer (GPT) models.
UNASSIGNED: We discuss areas for improvement, including better ingestion of tables, addition of context and fine-tuning.
UNASSIGNED: The challenge results demonstrate the potential of LLMs in automating table summarization in CSRs while also revealing the importance of human involvement and continued research to optimize this technology.

BACKGROUND: Dermatologic patient education materials (PEMs) are often written above the national average seventh- to eighth-grade reading level. ChatGPT-3.5, GPT-4, DermGPT, and DocsGPT are large language models (LLMs) that are responsive to user prompts. Our project assesses their use in generating dermatologic PEMs at specified reading levels.
OBJECTIVE: This study aims to assess the ability of select LLMs to generate PEMs for common and rare dermatologic conditions at unspecified and specified reading levels. Further, the study aims to assess the preservation of meaning across such LLM-generated PEMs, as assessed by dermatology resident trainees.
METHODS: The Flesch-Kincaid reading level (FKRL) of current American Academy of Dermatology PEMs was evaluated for 4 common (atopic dermatitis, acne vulgaris, psoriasis, and herpes zoster) and 4 rare (epidermolysis bullosa, bullous pemphigoid, lamellar ichthyosis, and lichen planus) dermatologic conditions. We prompted ChatGPT-3.5, GPT-4, DermGPT, and DocsGPT to \"Create a patient education handout about [condition] at a [FKRL]\" to iteratively generate 10 PEMs per condition at unspecified fifth- and seventh-grade FKRLs, evaluated with Microsoft Word readability statistics. The preservation of meaning across LLMs was assessed by 2 dermatology resident trainees.
RESULTS: The current American Academy of Dermatology PEMs had an average (SD) FKRL of 9.35 (1.26) and 9.50 (2.3) for common and rare diseases, respectively. For common diseases, the FKRLs of LLM-produced PEMs ranged between 9.8 and 11.21 (unspecified prompt), between 4.22 and 7.43 (fifth-grade prompt), and between 5.98 and 7.28 (seventh-grade prompt). For rare diseases, the FKRLs of LLM-produced PEMs ranged between 9.85 and 11.45 (unspecified prompt), between 4.22 and 7.43 (fifth-grade prompt), and between 5.98 and 7.28 (seventh-grade prompt). At the fifth-grade reading level, GPT-4 was better at producing PEMs for both common and rare conditions than ChatGPT-3.5 (P=.001 and P=.01, respectively), DermGPT (P<.001 and P=.03, respectively), and DocsGPT (P<.001 and P=.02, respectively). At the seventh-grade reading level, no significant difference was found between ChatGPT-3.5, GPT-4, DocsGPT, or DermGPT in producing PEMs for common conditions (all P>.05); however, for rare conditions, ChatGPT-3.5 and DocsGPT outperformed GPT-4 (P=.003 and P<.001, respectively). The preservation of meaning analysis revealed that for common conditions, DermGPT ranked the highest for overall ease of reading, patient understandability, and accuracy (14.75/15, 98%); for rare conditions, handouts generated by GPT-4 ranked the highest (14.5/15, 97%).
CONCLUSIONS: GPT-4 appeared to outperform ChatGPT-3.5, DocsGPT, and DermGPT at the fifth-grade FKRL for both common and rare conditions, although both ChatGPT-3.5 and DocsGPT performed better than GPT-4 at the seventh-grade FKRL for rare conditions. LLM-produced PEMs may reliably meet seventh-grade FKRLs for select common and rare dermatologic conditions and are easy to read, understandable for patients, and mostly accurate. LLMs may play a role in enhancing health literacy and disseminating accessible, understandable PEMs in dermatology.

A large language model (LLM) is a machine learning model inferred from text data that captures subtle patterns of language use in context. Modern LLMs are based on neural network architectures that incorporate transformer methods. They allow the model to relate words together through attention to multiple words in a text sequence. LLMs have been shown to be highly effective for a range of tasks in natural language processing (NLP), including classification and information extraction tasks and generative applications.
The aim of this adapted Delphi study was to collect researchers\' opinions on how LLMs might influence health care and on the strengths, weaknesses, opportunities, and threats of LLM use in health care.
We invited researchers in the fields of health informatics, nursing informatics, and medical NLP to share their opinions on LLM use in health care. We started the first round with open questions based on our strengths, weaknesses, opportunities, and threats framework. In the second and third round, the participants scored these items.
The first, second, and third rounds had 28, 23, and 21 participants, respectively. Almost all participants (26/28, 93% in round 1 and 20/21, 95% in round 3) were affiliated with academic institutions. Agreement was reached on 103 items related to use cases, benefits, risks, reliability, adoption aspects, and the future of LLMs in health care. Participants offered several use cases, including supporting clinical tasks, documentation tasks, and medical research and education, and agreed that LLM-based systems will act as health assistants for patient education. The agreed-upon benefits included increased efficiency in data handling and extraction, improved automation of processes, improved quality of health care services and overall health outcomes, provision of personalized care, accelerated diagnosis and treatment processes, and improved interaction between patients and health care professionals. In total, 5 risks to health care in general were identified: cybersecurity breaches, the potential for patient misinformation, ethical concerns, the likelihood of biased decision-making, and the risk associated with inaccurate communication. Overconfidence in LLM-based systems was recognized as a risk to the medical profession. The 6 agreed-upon privacy risks included the use of unregulated cloud services that compromise data security, exposure of sensitive patient data, breaches of confidentiality, fraudulent use of information, vulnerabilities in data storage and communication, and inappropriate access or use of patient data.
Future research related to LLMs should not only focus on testing their possibilities for NLP-related tasks but also consider the workflows the models could contribute to and the requirements regarding quality, integration, and regulations needed for successful implementation in practice.

Background Large language models (LLMs), such as ChatGPT-4, Gemini, and Microsoft Copilot, have been instrumental in various domains, including healthcare, where they enhance health literacy and aid in patient decision-making. Given the complexities involved in breast imaging procedures, accurate and comprehensible information is vital for patient engagement and compliance. This study aims to evaluate the readability and accuracy of the information provided by three prominent LLMs, ChatGPT-4, Gemini, and Microsoft Copilot, in response to frequently asked questions in breast imaging, assessing their potential to improve patient understanding and facilitate healthcare communication. Methodology We collected the most common questions on breast imaging from clinical practice and posed them to LLMs. We then evaluated the responses in terms of readability and accuracy. Responses from LLMs were analyzed for readability using the Flesch Reading Ease and Flesch-Kincaid Grade Level tests and for accuracy through a radiologist-developed Likert-type scale. Results The study found significant variations among LLMs. Gemini and Microsoft Copilot scored higher on readability scales (p < 0.001), indicating their responses were easier to understand. In contrast, ChatGPT-4 demonstrated greater accuracy in its responses (p < 0.001). Conclusions While LLMs such as ChatGPT-4 show promise in providing accurate responses, readability issues may limit their utility in patient education. Conversely, Gemini and Microsoft Copilot, despite being less accurate, are more accessible to a broader patient audience. Ongoing adjustments and evaluations of these models are essential to ensure they meet the diverse needs of patients, emphasizing the need for continuous improvement and oversight in the deployment of artificial intelligence technologies in healthcare.

BACKGROUND: Large language models (LLMs) are computational artificial intelligence systems with advanced natural language processing capabilities that have recently been popularized among health care students and educators due to their ability to provide real-time access to a vast amount of medical knowledge. The adoption of LLM technology into medical education and training has varied, and little empirical evidence exists to support its use in clinical teaching environments.
OBJECTIVE: The aim of the study is to identify and qualitatively evaluate potential use cases and limitations of LLM technology for real-time ward-based educational contexts.
METHODS: A brief, single-site exploratory evaluation of the publicly available ChatGPT-3.5 (OpenAI) was conducted by implementing the tool into the daily attending rounds of a general internal medicine inpatient service at a large urban academic medical center. ChatGPT was integrated into rounds via both structured and organic use, using the web-based \"chatbot\" style interface to interact with the LLM through conversational free-text and discrete queries. A qualitative approach using phenomenological inquiry was used to identify key insights related to the use of ChatGPT through analysis of ChatGPT conversation logs and associated shorthand notes from the clinical sessions.
RESULTS: Identified use cases for ChatGPT integration included addressing medical knowledge gaps through discrete medical knowledge inquiries, building differential diagnoses and engaging dual-process thinking, challenging medical axioms, using cognitive aids to support acute care decision-making, and improving complex care management by facilitating conversations with subspecialties. Potential additional uses included engaging in difficult conversations with patients, exploring ethical challenges and general medical ethics teaching, personal continuing medical education resources, developing ward-based teaching tools, supporting and automating clinical documentation, and supporting productivity and task management. LLM biases, misinformation, ethics, and health equity were identified as areas of concern and potential limitations to clinical and training use. A code of conduct on ethical and appropriate use was also developed to guide team usage on the wards.
CONCLUSIONS: Overall, ChatGPT offers a novel tool to enhance ward-based learning through rapid information querying, second-order content exploration, and engaged team discussion regarding generated responses. More research is needed to fully understand contexts for educational use, particularly regarding the risks and limitations of the tool in clinical settings and its impacts on trainee development.

OBJECTIVE: Automated identification of eligible patients is a bottleneck of clinical research. We propose Criteria2Query (C2Q) 3.0, a system that leverages GPT-4 for the semi-automatic transformation of clinical trial eligibility criteria text into executable clinical database queries.
METHODS: C2Q 3.0 integrated three GPT-4 prompts for concept extraction, SQL query generation, and reasoning. Each prompt was designed and evaluated separately. The concept extraction prompt was benchmarked against manual annotations from 20 clinical trials by two evaluators, who later also measured SQL generation accuracy and identified errors in GPT-generated SQL queries from 5 clinical trials. The reasoning prompt was assessed by three evaluators on four metrics: readability, correctness, coherence, and usefulness, using corrected SQL queries and an open-ended feedback questionnaire.
RESULTS: Out of 518 concepts from 20 clinical trials, GPT-4 achieved an F1-score of 0.891 in concept extraction. For SQL generation, 29 errors spanning seven categories were detected, with logic errors being the most common (n = 10; 34.48 %). Reasoning evaluations yielded a high coherence rating, with the mean score being 4.70 but relatively lower readability, with a mean of 3.95. Mean scores of correctness and usefulness were identified as 3.97 and 4.37, respectively.
CONCLUSIONS: GPT-4 significantly improves the accuracy of extracting clinical trial eligibility criteria concepts in C2Q 3.0. Continued research is warranted to ensure the reliability of large language models.