ABSTRACT
OBJECTIVES: Head and neck squamous cell carcinoma (HNSCC) is a complex malignancy that requires a multidisciplinary tumor board approach for individual treatment planning. In recent years, artificial intelligence tools have emerged to assist healthcare professionals in making informed treatment decisions. This study investigates the application of the newly published LLM Claude 3 Opus compared to the currently most advanced LLM ChatGPT 4.0 for the diagnosis and therapy planning of primary HNSCC. The results were compared to that of a conventional multidisciplinary tumor board; (2) Materials and Methods: We conducted a study in March 2024 on 50 consecutive primary head and neck cancer cases. The diagnostics and MDT recommendations were compared to the Claude 3 Opus and ChatGPT 4.0 recommendations for each patient and rated by two independent reviewers for the following parameters: clinical recommendation, explanation, and summarization in addition to the Artificial Intelligence Performance Instrument (AIPI); (3) Results: In this study, Claude 3 achieved better scores for the diagnostic workup of patients than ChatGPT 4.0 and provided treatment recommendations involving surgery, chemotherapy, and radiation therapy. In terms of clinical recommendations, explanation and summarization Claude 3 scored similar to ChatGPT 4.0, listing treatment recommendations which were congruent with the MDT, but failed to cite the source of the information; (4) Conclusion: This study is the first analysis of Claude 3 for primary head and neck cancer cases and demonstrates a superior performance in the diagnosis of HNSCC than ChatGPT 4.0 and similar results for therapy recommendations. This marks the advent of a newly launched advanced AI model that may be superior to ChatGPT 4.0 for the assessment of primary head and neck cancer cases and may assist in the clinical diagnostic and MDT setting.
ABSTRACT
Healthcare 4.0 describes the future transformation of the healthcare sector driven by the combination of digital technologies, such as artificial intelligence (AI), big data and the Internet of Medical Things, enabling the advancement of precision medicine. This overview article addresses various areas such as large language models (LLM), diagnostics and robotics, shedding light on the positive aspects of Healthcare 4.0 and showcasing exciting methods and application examples in cardiology. It delves into the broad knowledge base and enormous potential of LLMs, highlighting their immediate benefits as digital assistants or for administrative tasks. In diagnostics, the increasing usefulness of wearables is emphasized and an AI for predicting heart filling pressures based on cardiac magnetic resonance imaging (MRI) is introduced. Additionally, it discusses the revolutionary methodology of a digital simulation of the physical heart (digital twin). Finally, it addresses both regulatory frameworks and a brief vision of data-driven healthcare delivery, explaining the need for investments in technical personnel and infrastructure to achieve a more effective medicine.
ABSTRACT
BACKGROUND: Although history taking is fundamental for diagnosing medical conditions, teaching and providing feedback on the skill can be challenging due to resource constraints. Virtual simulated patients and web-based chatbots have thus emerged as educational tools, with recent advancements in artificial intelligence (AI) such as large language models (LLMs) enhancing their realism and potential to provide feedback. OBJECTIVE: In our study, we aimed to evaluate the effectiveness of a Generative Pretrained Transformer (GPT) 4 model to provide structured feedback on medical students' performance in history taking with a simulated patient. METHODS: We conducted a prospective study involving medical students performing history taking with a GPT-powered chatbot. To that end, we designed a chatbot to simulate patients' responses and provide immediate feedback on the comprehensiveness of the students' history taking. Students' interactions with the chatbot were analyzed, and feedback from the chatbot was compared with feedback from a human rater. We measured interrater reliability and performed a descriptive analysis to assess the quality of feedback. RESULTS: Most of the study's participants were in their third year of medical school. A total of 1894 question-answer pairs from 106 conversations were included in our analysis. GPT-4's role-play and responses were medically plausible in more than 99% of cases. Interrater reliability between GPT-4 and the human rater showed "almost perfect" agreement (Cohen κ=0.832). Less agreement (κ<0.6) detected for 8 out of 45 feedback categories highlighted topics about which the model's assessments were overly specific or diverged from human judgement. CONCLUSIONS: The GPT model was effective in providing structured feedback on history-taking dialogs provided by medical students. Although we unraveled some limitations regarding the specificity of feedback for certain feedback categories, the overall high agreement with human raters suggests that LLMs can be a valuable tool for medical education. Our findings, thus, advocate the careful integration of AI-driven feedback mechanisms in medical training and highlight important aspects when LLMs are used in that context.
Subject(s)
Medical History Taking , Patient Simulation , Students, Medical , Humans , Prospective Studies , Medical History Taking/methods , Medical History Taking/standards , Students, Medical/psychology , Female , Male , Clinical Competence/standards , Artificial Intelligence , Feedback , Reproducibility of Results , Education, Medical, Undergraduate/methodsABSTRACT
The availability of data from profiling of cancer patients with multiomics is rapidly increasing. However, integrative analysis of such data for personalized target identification is not trivial. Multiomics2Targets is a platform that enables users to upload transcriptomics, proteomics, and phosphoproteomics data matrices collected from the same cohort of cancer patients. After uploading the data, Multiomics2Targets produces a report that resembles a research publication. The uploaded matrices are processed, analyzed, and visualized using the tools Enrichr, KEA3, ChEA3, Expression2Kinases, and TargetRanger to identify and prioritize proteins, genes, and transcripts as potential targets. Figures and tables, as well as descriptions of the methods and results, are automatically generated. Reports include an abstract, introduction, methods, results, discussion, conclusions, and references and are exportable as citable PDFs and Jupyter Notebooks. Multiomics2Targets is applied to analyze version 3 of the Clinical Proteomic Tumor Analysis Consortium (CPTAC3) pan-cancer cohort, identifying potential targets for each CPTAC3 cancer subtype. Multiomics2Targets is available from https://multiomics2targets.maayanlab.cloud/.
ABSTRACT
Background: With the increasing application of large language models like ChatGPT in various industries, its potential in the medical domain, especially in standardized examinations, has become a focal point of research. Objective: The aim of this study is to assess the clinical performance of ChatGPT, focusing on its accuracy and reliability in the Chinese National Medical Licensing Examination (CNMLE). Methods: The CNMLE 2022 question set, consisting of 500 single-answer multiple choices questions, were reclassified into 15 medical subspecialties. Each question was tested 8 to 12 times in Chinese on the OpenAI platform from April 24 to May 15, 2023. Three key factors were considered: the version of GPT-3.5 and 4.0, the prompt's designation of system roles tailored to medical subspecialties, and repetition for coherence. A passing accuracy threshold was established as 60%. The χ2 tests and κ values were employed to evaluate the model's accuracy and consistency. Results: GPT-4.0 achieved a passing accuracy of 72.7%, which was significantly higher than that of GPT-3.5 (54%; P<.001). The variability rate of repeated responses from GPT-4.0 was lower than that of GPT-3.5 (9% vs 19.5%; P<.001). However, both models showed relatively good response coherence, with κ values of 0.778 and 0.610, respectively. System roles numerically increased accuracy for both GPT-4.0 (0.3%-3.7%) and GPT-3.5 (1.3%-4.5%), and reduced variability by 1.7% and 1.8%, respectively (P>.05). In subgroup analysis, ChatGPT achieved comparable accuracy among different question types (P>.05). GPT-4.0 surpassed the accuracy threshold in 14 of 15 subspecialties, while GPT-3.5 did so in 7 of 15 on the first response. Conclusions: GPT-4.0 passed the CNMLE and outperformed GPT-3.5 in key areas such as accuracy, consistency, and medical subspecialty expertise. Adding a system role insignificantly enhanced the model's reliability and answer coherence. GPT-4.0 showed promising potential in medical education and clinical practice, meriting further study.
Subject(s)
Educational Measurement , Licensure, Medical , Humans , China , Educational Measurement/methods , Educational Measurement/standards , Reproducibility of Results , Clinical Competence/standardsABSTRACT
Background: Teaching medical students the skills required to acquire, interpret, apply, and communicate clinical information is an integral part of medical education. A crucial aspect of this process involves providing students with feedback regarding the quality of their free-text clinical notes. Objective: The goal of this study was to assess the ability of ChatGPT 3.5, a large language model, to score medical students' free-text history and physical notes. Methods: This is a single-institution, retrospective study. Standardized patients learned a prespecified clinical case and, acting as the patient, interacted with medical students. Each student wrote a free-text history and physical note of their interaction. The students' notes were scored independently by the standardized patients and ChatGPT using a prespecified scoring rubric that consisted of 85 case elements. The measure of accuracy was percent correct. Results: The study population consisted of 168 first-year medical students. There was a total of 14,280 scores. The ChatGPT incorrect scoring rate was 1.0%, and the standardized patient incorrect scoring rate was 7.2%. The ChatGPT error rate was 86%, lower than the standardized patient error rate. The ChatGPT mean incorrect scoring rate of 12 (SD 11) was significantly lower than the standardized patient mean incorrect scoring rate of 85 (SD 74; P=.002). Conclusions: ChatGPT demonstrated a significantly lower error rate compared to standardized patients. This is the first study to assess the ability of a generative pretrained transformer (GPT) program to score medical students' standardized patient-based free-text clinical notes. It is expected that, in the near future, large language models will provide real-time feedback to practicing physicians regarding their free-text notes. GPT artificial intelligence programs represent an important advance in medical education and medical practice.
Subject(s)
Students, Medical , Humans , Retrospective Studies , Education, Medical, Undergraduate/methods , Educational Measurement/methods , Language , Medical History Taking/methods , Medical History Taking/standards , Clinical Competence/standards , MaleABSTRACT
Cervical spondylosis is the most common degenerative spinal disorder in modern societies. Patients require a great deal of medical knowledge, and large language models (LLMs) offer patients a novel and convenient tool for accessing medical advice. In this study, we collected the most frequently asked questions by patients with cervical spondylosis in clinical work and internet consultations. The accuracy of the answers provided by LLMs was evaluated and graded by 3 experienced spinal surgeons. Comparative analysis of responses showed that all LLMs could provide satisfactory results, and that among them, GPT-4 had the highest accuracy rate. Variation across each section in all LLMs revealed their ability boundaries and the development direction of artificial intelligence.
ABSTRACT
BACKGROUND: Recent advancements in artificial intelligence (AI) and large language models (LLMs) have shown potential in medical fields, including dermatology. With the introduction of image analysis capabilities in LLMs, their application in dermatological diagnostics has garnered significant interest. These capabilities are enabled by the integration of computer vision techniques into the underlying architecture of LLMs. OBJECTIVE: This study aimed to compare the diagnostic performance of Claude 3 Opus and ChatGPT with GPT-4 in analyzing dermoscopic images for melanoma detection, providing insights into their strengths and limitations. METHODS: We randomly selected 100 histopathology-confirmed dermoscopic images (50 malignant, 50 benign) from the International Skin Imaging Collaboration (ISIC) archive using a computer-generated randomization process. The ISIC archive was chosen due to its comprehensive and well-annotated collection of dermoscopic images, ensuring a diverse and representative sample. Images were included if they were dermoscopic images of melanocytic lesions with histopathologically confirmed diagnoses. Each model was given the same prompt, instructing it to provide the top 3 differential diagnoses for each image, ranked by likelihood. Primary diagnosis accuracy, accuracy of the top 3 differential diagnoses, and malignancy discrimination ability were assessed. The McNemar test was chosen to compare the diagnostic performance of the 2 models, as it is suitable for analyzing paired nominal data. RESULTS: In the primary diagnosis, Claude 3 Opus achieved 54.9% sensitivity (95% CI 44.08%-65.37%), 57.14% specificity (95% CI 46.31%-67.46%), and 56% accuracy (95% CI 46.22%-65.42%), while ChatGPT demonstrated 56.86% sensitivity (95% CI 45.99%-67.21%), 38.78% specificity (95% CI 28.77%-49.59%), and 48% accuracy (95% CI 38.37%-57.75%). The McNemar test showed no significant difference between the 2 models (P=.17). For the top 3 differential diagnoses, Claude 3 Opus and ChatGPT included the correct diagnosis in 76% (95% CI 66.33%-83.77%) and 78% (95% CI 68.46%-85.45%) of cases, respectively. The McNemar test showed no significant difference (P=.56). In malignancy discrimination, Claude 3 Opus outperformed ChatGPT with 47.06% sensitivity, 81.63% specificity, and 64% accuracy, compared to 45.1%, 42.86%, and 44%, respectively. The McNemar test showed a significant difference (P<.001). Claude 3 Opus had an odds ratio of 3.951 (95% CI 1.685-9.263) in discriminating malignancy, while ChatGPT-4 had an odds ratio of 0.616 (95% CI 0.297-1.278). CONCLUSIONS: Our study highlights the potential of LLMs in assisting dermatologists but also reveals their limitations. Both models made errors in diagnosing melanoma and benign lesions. These findings underscore the need for developing robust, transparent, and clinically validated AI models through collaborative efforts between AI researchers, dermatologists, and other health care professionals. While AI can provide valuable insights, it cannot yet replace the expertise of trained clinicians.
ABSTRACT
A good theory of mathematical beauty is more practical than any current observation, as new predictions about physical reality can be self-consistently verified. This belief applies to the current status of understanding deep neural networks including large language models and even the biological intelligence. Toy models provide a metaphor of physical reality, allowing mathematically formulating the reality (i.e., the so-called theory), which can be updated as more conjectures are justified or refuted. One does not need to present all details in a model, but rather, more abstract models are constructed, as complex systems such as the brains or deep networks have many sloppy dimensions but much less stiff dimensions that strongly impact macroscopic observables. This type of bottom-up mechanistic modeling is still promising in the modern era of understanding the natural or artificial intelligence. Here, we shed light on eight challenges in developing theory of intelligence following this theoretical paradigm. Theses challenges are representation learning, generalization, adversarial robustness, continual learning, causal learning, internal model of the brain, next-token prediction, and the mechanics of subjective experience.
ABSTRACT
BACKGROUND: The rapid evolution of ChatGPT has generated substantial interest and led to extensive discussions in both public and academic domains, particularly in the context of medical education. OBJECTIVE: This study aimed to evaluate ChatGPT's performance in a pulmonology examination through a comparative analysis with that of third-year medical students. METHODS: In this cross-sectional study, we conducted a comparative analysis with 2 distinct groups. The first group comprised 244 third-year medical students who had previously taken our institution's 2020 pulmonology examination, which was conducted in French. The second group involved ChatGPT-3.5 in 2 separate sets of conversations: without contextualization (V1) and with contextualization (V2). In both V1 and V2, ChatGPT received the same set of questions administered to the students. RESULTS: V1 demonstrated exceptional proficiency in radiology, microbiology, and thoracic surgery, surpassing the majority of medical students in these domains. However, it faced challenges in pathology, pharmacology, and clinical pneumology. In contrast, V2 consistently delivered more accurate responses across various question categories, regardless of the specialization. ChatGPT exhibited suboptimal performance in multiple choice questions compared to medical students. V2 excelled in responding to structured open-ended questions. Both ChatGPT conversations, particularly V2, outperformed students in addressing questions of low and intermediate difficulty. Interestingly, students showcased enhanced proficiency when confronted with highly challenging questions. V1 fell short of passing the examination. Conversely, V2 successfully achieved examination success, outperforming 139 (62.1%) medical students. CONCLUSIONS: While ChatGPT has access to a comprehensive web-based data set, its performance closely mirrors that of an average medical student. Outcomes are influenced by question format, item complexity, and contextual nuances. The model faces challenges in medical contexts requiring information synthesis, advanced analytical aptitude, and clinical judgment, as well as in non-English language assessments and when confronted with data outside mainstream internet sources.
Subject(s)
Educational Measurement , Pulmonary Medicine , Students, Medical , Humans , Cross-Sectional Studies , Pulmonary Medicine/education , Students, Medical/statistics & numerical data , Educational Measurement/methods , Education, Medical, Undergraduate/methods , Male , Aptitude , Female , Clinical CompetenceABSTRACT
BACKGROUND: Discharge letters are a critical component in the continuity of care between specialists and primary care providers. However, these letters are time-consuming to write, underprioritized in comparison to direct clinical care, and are often tasked to junior doctors. Prior studies assessing the quality of discharge summaries written for inpatient hospital admissions show inadequacies in many domains. Large language models such as GPT have the ability to summarize large volumes of unstructured free text such as electronic medical records and have the potential to automate such tasks, providing time savings and consistency in quality. OBJECTIVE: The aim of this study was to assess the performance of GPT-4 in generating discharge letters written from urology specialist outpatient clinics to primary care providers and to compare their quality against letters written by junior clinicians. METHODS: Fictional electronic records were written by physicians simulating 5 common urology outpatient cases with long-term follow-up. Records comprised simulated consultation notes, referral letters and replies, and relevant discharge summaries from inpatient admissions. GPT-4 was tasked to write discharge letters for these cases with a specified target audience of primary care providers who would be continuing the patient's care. Prompts were written for safety, content, and style. Concurrently, junior clinicians were provided with the same case records and instructional prompts. GPT-4 output was assessed for instances of hallucination. A blinded panel of primary care physicians then evaluated the letters using a standardized questionnaire tool. RESULTS: GPT-4 outperformed human counterparts in information provision (mean 4.32, SD 0.95 vs 3.70, SD 1.27; P=.03) and had no instances of hallucination. There were no statistically significant differences in the mean clarity (4.16, SD 0.95 vs 3.68, SD 1.24; P=.12), collegiality (4.36, SD 1.00 vs 3.84, SD 1.22; P=.05), conciseness (3.60, SD 1.12 vs 3.64, SD 1.27; P=.71), follow-up recommendations (4.16, SD 1.03 vs 3.72, SD 1.13; P=.08), and overall satisfaction (3.96, SD 1.14 vs 3.62, SD 1.34; P=.36) between the letters generated by GPT-4 and humans, respectively. CONCLUSIONS: Discharge letters written by GPT-4 had equivalent quality to those written by junior clinicians, without any hallucinations. This study provides a proof of concept that large language models can be useful and safe tools in clinical documentation.
Subject(s)
Patient Discharge , Humans , Patient Discharge/standards , Electronic Health Records/standards , Single-Blind Method , LanguageABSTRACT
BACKGROUND: Limited data indicated the performance of large language model (LLM) taking on the role of doctors. We aimed to investigate the potential for ChatGPT-3.5 and New Bing Chat acting as doctors using thyroid nodules as an example. METHODS: A total of 145 patients with thyroid nodules were included for generating questions. Each question was entered into chatbot of ChatGPT-3.5 and New Bing Chat five times and five responses were acquired respectively. These responses were compared with answers given by five junior doctors. Responses from five senior doctors were regarded as gold standard. Accuracy and reproducibility of responses from ChatGPT-3.5 and New Bing Chat were evaluated. RESULTS: The accuracy of ChatGPT-3.5 and New Bing Chat in answering Q2, Q3, Q5 were lower than that of junior doctors (all P < 0.05), while both LLMs were comparable to junior doctors when answering Q4 and Q6. In terms of "high reproducibility and accuracy", ChatGPT-3.5 outperformed New Bing Chat in Q1 and Q5 (P < 0.001 and P = 0.008, respectively), but showed no significant difference in Q2, Q3, Q4, and Q6 (P > 0.05 for all). New Bing Chat generated higher accuracy than ChatGPT-3.5 (72.41% vs 58.62%) (P = 0.003) in decision making of thyroid nodules, and both were less accurate than junior doctors (89.66%, P < 0.001 for both). CONCLUSIONS: The exploration of ChatGPT-3.5 and New Bing Chat in the diagnosis and management of thyroid nodules illustrates that LLMs currently demonstrate the potential for medical applications, but do not yet reach the clinical decision-making capacity of doctors.
ABSTRACT
BACKGROUND: Despite the significance and prevalence of acute respiratory distress syndrome (ARDS), its detection remains highly variable and inconsistent. In this work, we aim to develop an algorithm (ARDSFlag) to automate the diagnosis of ARDS based on the Berlin definition. We also aim to develop a visualization tool that helps clinicians efficiently assess ARDS criteria. METHODS: ARDSFlag applies machine learning (ML) and natural language processing (NLP) techniques to evaluate Berlin criteria by incorporating structured and unstructured data in an electronic health record (EHR) system. The study cohort includes 19,534 ICU admissions in the Medical Information Mart for Intensive Care III (MIMIC-III) database. The output is the ARDS diagnosis, onset time, and severity. RESULTS: ARDSFlag includes separate text classifiers trained using large training sets to find evidence of bilateral infiltrates in radiology reports (accuracy of 91.9%±0.5%) and heart failure/fluid overload in radiology reports (accuracy 86.1%±0.5%) and echocardiogram notes (accuracy 98.4%±0.3%). A test set of 300 cases, which was blindly and independently labeled for ARDS by two groups of clinicians, shows that ARDSFlag generates an overall accuracy of 89.0% (specificity = 91.7%, recall = 80.3%, and precision = 75.0%) in detecting ARDS cases. CONCLUSION: To our best knowledge, this is the first study to focus on developing a method to automate the detection of ARDS. Some studies have developed and used other methods to answer other research questions. Expectedly, ARDSFlag generates a significantly higher performance in all accuracy measures compared to those methods.
Subject(s)
Algorithms , Electronic Health Records , Machine Learning , Natural Language Processing , Respiratory Distress Syndrome , Humans , Respiratory Distress Syndrome/diagnosis , Intensive Care Units , Middle Aged , Male , FemaleABSTRACT
BACKGROUND: Suicide is a leading cause of death worldwide. Journalistic reporting guidelines were created to curb the impact of unsafe reporting; however, how suicide is framed in news reports may differ by important characteristics such as the circumstances and the decedent's gender. OBJECTIVE: This study aimed to examine the degree to which news media reports of suicides are framed using stigmatized or glorified language and differences in such framing by gender and circumstance of suicide. METHODS: We analyzed 200 news articles regarding suicides and applied the validated Stigma of Suicide Scale to identify stigmatized and glorified language. We assessed linguistic similarity with 2 widely used metrics, cosine similarity and mutual information scores, using a machine learning-based large language model. RESULTS: News reports of male suicides were framed more similarly to stigmatizing (P<.001) and glorifying (P=.005) language than reports of female suicides. Considering the circumstances of suicide, mutual information scores indicated that differences in the use of stigmatizing or glorifying language by gender were most pronounced for articles attributing legal (0.155), relationship (0.268), or mental health problems (0.251) as the cause. CONCLUSIONS: Linguistic differences, by gender, in stigmatizing or glorifying language when reporting suicide may exacerbate suicide disparities.
Subject(s)
Mass Media , Social Stigma , Suicide , Humans , Female , Male , Suicide/psychology , Suicide/statistics & numerical data , Mass Media/statistics & numerical data , Sex Factors , AdultABSTRACT
Metastatic breast cancer (MBC) continues to be a leading cause of cancer-related deaths among women. This work introduces an innovative non-invasive breast cancer classification model designed to improve the identification of cancer metastases. While this study marks the initial exploration into predicting MBC, additional investigations are essential to validate the occurrence of MBC. Our approach combines the strengths of large language models (LLMs), specifically the bidirectional encoder representations from transformers (BERT) model, with the powerful capabilities of graph neural networks (GNNs) to predict MBC patients based on their histopathology reports. This paper introduces a BERT-GNN approach for metastatic breast cancer prediction (BG-MBC) that integrates graph information derived from the BERT model. In this model, nodes are constructed from patient medical records, while BERT embeddings are employed to vectorise representations of the words in histopathology reports, thereby capturing semantic information crucial for classification by employing three distinct approaches (namely univariate selection, extra trees classifier for feature importance, and Shapley values to identify the features that have the most significant impact). Identifying the most crucial 30 features out of 676 generated as embeddings during model training, our model further enhances its predictive capabilities. The BG-MBC model achieves outstanding accuracy, with a detection rate of 0.98 and an area under curve (AUC) of 0.98, in identifying MBC patients. This remarkable performance is credited to the model's utilisation of attention scores generated by the LLM from histopathology reports, effectively capturing pertinent features for classification.
ABSTRACT
PURPOSE: This study aimed to develop a retrained large language model (LLM) tailored to the needs of HN cancer patients treated with radiotherapy, with emphasis on symptom management and survivorship care. METHODS: A comprehensive external database was curated for training ChatGPT-4, integrating expert-identified consensus guidelines on supportive care for HN patients and correspondences from physicians and nurses within our institution's electronic medical records for 90 HN patients. The performance of our model was evaluated using 20 patient post-treatment inquiries that were then assessed by three Board certified radiation oncologists (RadOncs). The rating of the model was assessed on a scale of 1 (strongly disagree) to 5 (strongly agree) based on accuracy, clarity of response, completeness s, and relevance. RESULTS: The average scores for the 20 tested questions were 4.25 for accuracy, 4.35 for clarity, 4.22 for completeness, and 4.32 for relevance, on a 5-point scale. Overall, 91.67% (220 out of 240) of assessments received scores of 3 or higher, and 83.33% (200 out of 240) received scores of 4 or higher. CONCLUSION: The custom-trained model demonstrates high accuracy in providing support to HN patients offering evidence-based information and guidance on their symptom management and survivorship care.
ABSTRACT
Background: Large language models (LLMs) have achieved great progress in natural language processing tasks and demonstrated the potential for use in clinical applications. Despite their capabilities, LLMs in the medical domain are prone to generating hallucinations (not fully reliable responses). Hallucinations in LLMs' responses create substantial risks, potentially threatening patients' physical safety. Thus, to perceive and prevent this safety risk, it is essential to evaluate LLMs in the medical domain and build a systematic evaluation. Objective: We developed a comprehensive evaluation system, MedGPTEval, composed of criteria, medical data sets in Chinese, and publicly available benchmarks. Methods: First, a set of evaluation criteria was designed based on a comprehensive literature review. Second, existing candidate criteria were optimized by using a Delphi method with 5 experts in medicine and engineering. Third, 3 clinical experts designed medical data sets to interact with LLMs. Finally, benchmarking experiments were conducted on the data sets. The responses generated by chatbots based on LLMs were recorded for blind evaluations by 5 licensed medical experts. The evaluation criteria that were obtained covered medical professional capabilities, social comprehensive capabilities, contextual capabilities, and computational robustness, with 16 detailed indicators. The medical data sets include 27 medical dialogues and 7 case reports in Chinese. Three chatbots were evaluated: ChatGPT by OpenAI; ERNIE Bot by Baidu, Inc; and Doctor PuJiang (Dr PJ) by Shanghai Artificial Intelligence Laboratory. Results: Dr PJ outperformed ChatGPT and ERNIE Bot in the multiple-turn medical dialogues and case report scenarios. Dr PJ also outperformed ChatGPT in the semantic consistency rate and complete error rate category, indicating better robustness. However, Dr PJ had slightly lower scores in medical professional capabilities compared with ChatGPT in the multiple-turn dialogue scenario. Conclusions: MedGPTEval provides comprehensive criteria to evaluate chatbots by LLMs in the medical domain, open-source data sets, and benchmarks assessing 3 LLMs. Experimental results demonstrate that Dr PJ outperforms ChatGPT and ERNIE Bot in social and professional contexts. Therefore, such an assessment system can be easily adopted by researchers in this community to augment an open-source data set.
ABSTRACT
PURPOSE: Large language models (LLMs) are a form of artificial intelligence (AI) that uses deep learning techniques to understand, summarize and generate content. The potential benefits of LLMs in healthcare is predicted to be immense. The objective of this study was to examine the quality of patient information leaflets (PILs) produced by 3 LLMs on urological topics. METHODS: Prompts were created to generate PILs from 3 LLMs: ChatGPT-4, PaLM 2 (Google Bard) and Llama 2 (Meta) across four urology topics (circumcision, nephrectomy, overactive bladder syndrome, and transurethral resection of the prostate). PILs were evaluated using a quality assessment checklist. PIL readability was assessed by the Average Reading Level Consensus Calculator. RESULTS: PILs generated by PaLM 2 had the highest overall average quality score (3.58), followed by Llama 2 (3.34) and ChatGPT-4 (3.08). PaLM 2 generated PILs were of the highest quality in all topics except TURP and was the only LLM to include images. Medical inaccuracies were present in all generated content including instances of significant error. Readability analysis identified PaLM 2 generated PILs as the simplest (age 14-15 average reading level). Llama 2 PILs were the most difficult (age 16-17 average). CONCLUSION: While LLMs can generate PILs that may help reduce healthcare professional workload, generated content requires clinician input for accuracy and inclusion of health literacy aids, such as images. LLM-generated PILs were above the average reading level for adults, necessitating improvement in LLM algorithms and/or prompt design. How satisfied patients are to LLM-generated PILs remains to be evaluated.
Subject(s)
Artificial Intelligence , Urology , Humans , Patient Education as Topic/methods , Language , Urologic Diseases/surgeryABSTRACT
BACKGROUND AND OBJECTIVES: Integrating large language models (LLMs) such as GPT-4 Turbo into diagnostic imaging faces a significant challenge, with current misdiagnosis rates ranging from 30-50%. This study evaluates how prompt engineering and confidence thresholds can improve diagnostic accuracy in neuroradiology. METHODS: We analyze 751 neuroradiology cases from the American Journal of Neuroradiology using GPT-4 Turbo with customized prompts to improve diagnostic precision. RESULTS: Initially, GPT-4 Turbo achieved a baseline diagnostic accuracy of 55.1%. By reformatting responses to list five diagnostic candidates and applying a 90% confidence threshold, the highest precision of the diagnosis increased to 72.9%, with the candidate list providing the correct diagnosis at 85.9%, reducing the misdiagnosis rate to 14.1%. However, this threshold reduced the number of cases that responded. CONCLUSIONS: Strategic prompt engineering and high confidence thresholds significantly reduce misdiagnoses and improve the precision of the LLM diagnostic in neuroradiology. More research is needed to optimize these approaches for broader clinical implementation, balancing accuracy and utility.