ABSTRACT
INTRODUCTION: Medical decision-making is crucial for effective treatment, especially in psychiatry where diagnosis often relies on subjective patient reports and a lack of high-specificity symptoms. Artificial intelligence (AI), particularly Large Language Models (LLMs) like GPT, has emerged as a promising tool to enhance diagnostic accuracy in psychiatry. This comparative study explores the diagnostic capabilities of several AI models, including Aya, GPT-3.5, GPT-4, GPT-3.5 clinical assistant (CA), Nemotron, and Nemotron CA, using clinical cases from the DSM-5. METHODS: We curated 20 clinical cases from the DSM-5 Clinical Cases book, covering a wide range of psychiatric diagnoses. Four advanced AI models (GPT-3.5 Turbo, GPT-4, Aya, Nemotron) were tested using prompts to elicit detailed diagnoses and reasoning. The models' performances were evaluated based on accuracy and quality of reasoning, with additional analysis using the Retrieval Augmented Generation (RAG) methodology for models accessing the DSM-5 text. RESULTS: The AI models showed varied diagnostic accuracy, with GPT-3.5 and GPT-4 performing notably better than Aya and Nemotron in terms of both accuracy and reasoning quality. While models struggled with specific disorders such as cyclothymic and disruptive mood dysregulation disorders, others excelled, particularly in diagnosing psychotic and bipolar disorders. Statistical analysis highlighted significant differences in accuracy and reasoning, emphasizing the superiority of the GPT models. DISCUSSION: The application of AI in psychiatry offers potential improvements in diagnostic accuracy. The superior performance of the GPT models can be attributed to their advanced natural language processing capabilities and extensive training on diverse text data, enabling more effective interpretation of psychiatric language. However, models like Aya and Nemotron showed limitations in reasoning, indicating a need for further refinement in their training and application. CONCLUSION: AI holds significant promise for enhancing psychiatric diagnostics, with certain models demonstrating high potential in interpreting complex clinical descriptions accurately. Future research should focus on expanding the dataset and integrating multimodal data to further enhance the diagnostic capabilities of AI in psychiatry.
ABSTRACT
Effective communication hinges on a mutual understanding of word meaning in different contexts. We recorded brain activity using electrocorticography during spontaneous, face-to-face conversations in five pairs of epilepsy patients. We developed a model-based coupling framework that aligns brain activity in both speaker and listener to a shared embedding space from a large language model (LLM). The context-sensitive LLM embeddings allow us to track the exchange of linguistic information, word by word, from one brain to another in natural conversations. Linguistic content emerges in the speaker's brain before word articulation and rapidly re-emerges in the listener's brain after word articulation. The contextual embeddings better capture word-by-word neural alignment between speaker and listener than syntactic and articulatory models. Our findings indicate that the contextual embeddings learned by LLMs can serve as an explicit numerical model of the shared, context-rich meaning space humans use to communicate their thoughts to one another.
ABSTRACT
Nitrogen is an essential macronutrient critical for plant growth and productivity. Plants have the capacity to uptake inorganic nitrate and ammonium, with nitrate playing a crucial role as a signaling molecule in various cellular processes. The availability of nitrate and the signaling pathways involved finely tune the processes of nitrate uptake and assimilation. NIN-like proteins (NLPs), a group of transcription factors belonging to the RWP-RK gene family, act as major nitrate sensors and are implicated in the primary nitrate response (PNR) within the nucleus of both non-leguminous and leguminous plants through their RWP-RK domains. In leguminous plants, NLPs are indispensable for the initiation and development of nitrogen-fixing nodules in symbiosis with rhizobia. Moreover, NLPs play pivotal roles in plant responses to abiotic stresses, including drought and cold. Recent studies have identified NLP homologs in oomycete pathogens, suggesting their potential involvement in pathogenesis and virulence. This review article delves into the conservation of RWP-RK genes, examining their significance and implications across different plant species. The focus lies on the role of NLPs as nitrate sensors, investigating their involvement in various processes, including rhizobial symbiosis in both leguminous and non-leguminous plants. Additionally, the multifaceted functions of NLPs in abiotic stress responses, developmental processes, and interactions with plant pathogens are explored. By comprehensively analyzing the role of NLPs in nitrate signaling and their broader implications for plant growth and development, this review sheds light on the intricate mechanisms underlying nitrogen sensing and signaling in various plant lineages.
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
BACKGROUND: Artificial intelligence (AI) chatbots, such as ChatGPT, have made significant progress. These chatbots, particularly popular among health care professionals and patients, are transforming patient education and disease experience with personalized information. Accurate, timely patient education is crucial for informed decision-making, especially regarding prostate-specific antigen screening and treatment options. However, the accuracy and reliability of AI chatbots' medical information must be rigorously evaluated. Studies testing ChatGPT's knowledge of prostate cancer are emerging, but there is a need for ongoing evaluation to ensure the quality and safety of information provided to patients. OBJECTIVE: This study aims to evaluate the quality, accuracy, and readability of ChatGPT-4's responses to common prostate cancer questions posed by patients. METHODS: Overall, 8 questions were formulated with an inductive approach based on information topics in peer-reviewed literature and Google Trends data. Adapted versions of the Patient Education Materials Assessment Tool for AI (PEMAT-AI), Global Quality Score, and DISCERN-AI tools were used by 4 independent reviewers to assess the quality of the AI responses. The 8 AI outputs were judged by 7 expert urologists, using an assessment framework developed to assess accuracy, safety, appropriateness, actionability, and effectiveness. The AI responses' readability was assessed using established algorithms (Flesch Reading Ease score, Gunning Fog Index, Flesch-Kincaid Grade Level, The Coleman-Liau Index, and Simple Measure of Gobbledygook [SMOG] Index). A brief tool (Reference Assessment AI [REF-AI]) was developed to analyze the references provided by AI outputs, assessing for reference hallucination, relevance, and quality of references. RESULTS: The PEMAT-AI understandability score was very good (mean 79.44%, SD 10.44%), the DISCERN-AI rating was scored as "good" quality (mean 13.88, SD 0.93), and the Global Quality Score was high (mean 4.46/5, SD 0.50). Natural Language Assessment Tool for AI had pooled mean accuracy of 3.96 (SD 0.91), safety of 4.32 (SD 0.86), appropriateness of 4.45 (SD 0.81), actionability of 4.05 (SD 1.15), and effectiveness of 4.09 (SD 0.98). The readability algorithm consensus was "difficult to read" (Flesch Reading Ease score mean 45.97, SD 8.69; Gunning Fog Index mean 14.55, SD 4.79), averaging an 11th-grade reading level, equivalent to 15- to 17-year-olds (Flesch-Kincaid Grade Level mean 12.12, SD 4.34; The Coleman-Liau Index mean 12.75, SD 1.98; SMOG Index mean 11.06, SD 3.20). REF-AI identified 2 reference hallucinations, while the majority (28/30, 93%) of references appropriately supplemented the text. Most references (26/30, 86%) were from reputable government organizations, while a handful were direct citations from scientific literature. CONCLUSIONS: Our analysis found that ChatGPT-4 provides generally good responses to common prostate cancer queries, making it a potentially valuable tool for patient education in prostate cancer care. Objective quality assessment tools indicated that the natural language processing outputs were generally reliable and appropriate, but there is room for improvement.
Subject(s)
Patient Education as Topic , Prostatic Neoplasms , Humans , Male , Patient Education as Topic/methods , Artificial IntelligenceABSTRACT
RATIONALE AND OBJECTIVES: The process of generating radiology reports is often time-consuming and labor-intensive, prone to incompleteness, heterogeneity, and errors. By employing natural language processing (NLP)-based techniques, this study explores the potential for enhancing the efficiency of radiology report generation through the remarkable capabilities of ChatGPT (Generative Pre-training Transformer), a prominent large language model (LLM). MATERIALS AND METHODS: Using a sample of 1000 records from the Medical Information Mart for Intensive Care (MIMIC) Chest X-ray Database, this investigation employed Claude.ai to extract initial radiological report keywords. ChatGPT then generated radiology reports using a consistent 3-step prompt template outline. Various lexical and sentence similarity techniques were employed to evaluate the correspondence between the AI assistant-generated reports and reference reports authored by medical professionals. RESULTS: Results showed varying performance among NLP models, with Bart (Bidirectional and Auto-Regressive Transformers) and XLM (Cross-lingual Language Model) displaying high proficiency (mean similarity scores up to 99.3%), closely mirroring physician reports. Conversely, DeBERTa (Decoding-enhanced BERT with disentangled attention) and sequence-matching models scored lower, indicating less alignment with medical language. In the Impression section, the Word-Embedding model excelled with a mean similarity of 84.4%, while others like the Jaccard index showed lower performance. CONCLUSION: Overall, the study highlights significant variations across NLP models in their ability to generate radiology reports consistent with medical professionals' language. Pairwise comparisons and Kruskal-Wallis tests confirmed these differences, emphasizing the need for careful selection and evaluation of NLP models in radiology report generation. This research underscores the potential of ChatGPT to streamline and improve the radiology reporting process, with implications for enhancing efficiency and accuracy in clinical practice.
ABSTRACT
BACKGROUND: Although uncertainties exist regarding implementation, artificial intelligence-driven generative language models (GLMs) have enormous potential in medicine. Deployment of GLMs could improve patient comprehension of clinical texts and improve low health literacy. OBJECTIVE: The goal of this study is to evaluate the potential of ChatGPT-3.5 and GPT-4 to tailor the complexity of medical information to patient-specific input education level, which is crucial if it is to serve as a tool in addressing low health literacy. METHODS: Input templates related to 2 prevalent chronic diseases-type II diabetes and hypertension-were designed. Each clinical vignette was adjusted for hypothetical patient education levels to evaluate output personalization. To assess the success of a GLM (GPT-3.5 and GPT-4) in tailoring output writing, the readability of pre- and posttransformation outputs were quantified using the Flesch reading ease score (FKRE) and the Flesch-Kincaid grade level (FKGL). RESULTS: Responses (n=80) were generated using GPT-3.5 and GPT-4 across 2 clinical vignettes. For GPT-3.5, FKRE means were 57.75 (SD 4.75), 51.28 (SD 5.14), 32.28 (SD 4.52), and 28.31 (SD 5.22) for 6th grade, 8th grade, high school, and bachelor's, respectively; FKGL mean scores were 9.08 (SD 0.90), 10.27 (SD 1.06), 13.4 (SD 0.80), and 13.74 (SD 1.18). GPT-3.5 only aligned with the prespecified education levels at the bachelor's degree. Conversely, GPT-4's FKRE mean scores were 74.54 (SD 2.6), 71.25 (SD 4.96), 47.61 (SD 6.13), and 13.71 (SD 5.77), with FKGL mean scores of 6.3 (SD 0.73), 6.7 (SD 1.11), 11.09 (SD 1.26), and 17.03 (SD 1.11) for the same respective education levels. GPT-4 met the target readability for all groups except the 6th-grade FKRE average. Both GLMs produced outputs with statistically significant differences (P<.001; 8th grade P<.001; high school P<.001; bachelors P=.003; FKGL: 6th grade P=.001; 8th grade P<.001; high school P<.001; bachelors P<.001) between mean FKRE and FKGL across input education levels. CONCLUSIONS: GLMs can change the structure and readability of medical text outputs according to input-specified education. However, GLMs categorize input education designation into 3 broad tiers of output readability: easy (6th and 8th grade), medium (high school), and difficult (bachelor's degree). This is the first result to suggest that there are broader boundaries in the success of GLMs in output text simplification. Future research must establish how GLMs can reliably personalize medical texts to prespecified education levels to enable a broader impact on health care literacy.
ABSTRACT
With the widespread adoption of next-generation sequencing technologies, the speed and convenience of genome sequencing have significantly improved, and many biological genomes have been sequenced. However, during the assembly of small genomes, we still face a series of challenges, including repetitive fragments, inverted repeats, low sequencing coverage, and the limitations of sequencing technologies. These challenges lead to unknown gaps in small genomes, hindering complete genome assembly. Although there are many existing assembly software options, they do not fully utilize the potential of artificial intelligence technologies, resulting in limited improvement in gap filling. Here, we propose a novel method, DLGapCloser, based on deep learning, aimed at assisting traditional tools in further filling gaps in small genomes. Firstly, we created four datasets based on the original genomes of Saccharomyces cerevisiae, Schizosaccharomyces pombe, Neurospora crassa, and Micromonas pusilla. To further extract effective information from the gene sequences, we also added homologous genomes to enrich the datasets. Secondly, we proposed the DGCNet model, which effectively extracts features and learns context from sequences flanking gaps. Addressing issues with early pruning and high memory usage in the Beam Search algorithm, we developed a new prediction algorithm, Wave-Beam Search. This algorithm alternates between expansion and contraction phases, enhancing efficiency and accuracy. Experimental results showed that the Wave-Beam Search algorithm improved the gap-filling performance of assembly tools by 7.35%, 28.57%, 42.85%, and 8.33% on the original results. Finally, we established new gap-filling standards and created and implemented a novel evaluation method. Validation on the genomes of Saccharomyces cerevisiae, Schizosaccharomyces pombe, Neurospora crassa, and Micromonas pusilla showed that DLGapCloser increased the number of filled gaps by 8.05%, 15.3%, 1.4%, and 7% compared to traditional assembly tools.
Subject(s)
Neural Networks, Computer , Algorithms , Deep Learning , Genome, Fungal , Saccharomyces cerevisiae/genetics , Schizosaccharomyces/genetics , High-Throughput Nucleotide Sequencing/methods , Neurospora crassa/genetics , Software , Genomics/methods , Sequence Analysis, DNA/methodsABSTRACT
The Languages of the Indian subcontinent are less represented in current NLP literature. To mitigate this gap, we present the IndicDialogue dataset, which contains subtitles and dialogues in 10 major Indic languages: Hindi, Bengali, Marathi, Telugu, Tamil, Urdu, Odia, Sindhi, Nepali, and Assamese. This dataset is sourced from OpenSubtitles.org, with subtitles pre-processed to remove irrelevant tags, timestamps, square brackets, and links, ensuring the retention of relevant dialogues in JSONL files. The IndicDialogue dataset comprises 7750 raw subtitle files (SRT), 11 JSONL files, 6,853,518 dialogues, and 42,188,569 words. It is designed to serve as a foundation for language model pre-training for low-resource languages, enabling a wide range of downstream tasks including word embeddings, topic modeling, conversation synthesis, neural machine translation, and text summarization.
ABSTRACT
BACKGROUND: Over the past 2 years, researchers have used various medical licensing examinations to test whether ChatGPT (OpenAI) possesses accurate medical knowledge. The performance of each version of ChatGPT on the medical licensing examination in multiple environments showed remarkable differences. At this stage, there is still a lack of a comprehensive understanding of the variability in ChatGPT's performance on different medical licensing examinations. OBJECTIVE: In this study, we reviewed all studies on ChatGPT performance in medical licensing examinations up to March 2024. This review aims to contribute to the evolving discourse on artificial intelligence (AI) in medical education by providing a comprehensive analysis of the performance of ChatGPT in various environments. The insights gained from this systematic review will guide educators, policymakers, and technical experts to effectively and judiciously use AI in medical education. METHODS: We searched the literature published between January 1, 2022, and March 29, 2024, by searching query strings in Web of Science, PubMed, and Scopus. Two authors screened the literature according to the inclusion and exclusion criteria, extracted data, and independently assessed the quality of the literature concerning Quality Assessment of Diagnostic Accuracy Studies-2. We conducted both qualitative and quantitative analyses. RESULTS: A total of 45 studies on the performance of different versions of ChatGPT in medical licensing examinations were included in this study. GPT-4 achieved an overall accuracy rate of 81% (95% CI 78-84; P<.01), significantly surpassing the 58% (95% CI 53-63; P<.01) accuracy rate of GPT-3.5. GPT-4 passed the medical examinations in 26 of 29 cases, outperforming the average scores of medical students in 13 of 17 cases. Translating the examination questions into English improved GPT-3.5's performance but did not affect GPT-4. GPT-3.5 showed no difference in performance between examinations from English-speaking and non-English-speaking countries (P=.72), but GPT-4 performed better on examinations from English-speaking countries significantly (P=.02). Any type of prompt could significantly improve GPT-3.5's (P=.03) and GPT-4's (P<.01) performance. GPT-3.5 performed better on short-text questions than on long-text questions. The difficulty of the questions affected the performance of GPT-3.5 and GPT-4. In image-based multiple-choice questions (MCQs), ChatGPT's accuracy rate ranges from 13.1% to 100%. ChatGPT performed significantly worse on open-ended questions than on MCQs. CONCLUSIONS: GPT-4 demonstrates considerable potential for future use in medical education. However, due to its insufficient accuracy, inconsistent performance, and the challenges posed by differing medical policies and knowledge across countries, GPT-4 is not yet suitable for use in medical education. TRIAL REGISTRATION: PROSPERO CRD42024506687; https://www.crd.york.ac.uk/prospero/display_record.php?RecordID=506687.
Subject(s)
Educational Measurement , Licensure, Medical , Humans , Licensure, Medical/standards , Licensure, Medical/statistics & numerical data , Educational Measurement/methods , Educational Measurement/standards , Educational Measurement/statistics & numerical data , Clinical Competence/statistics & numerical data , Clinical Competence/standards , Artificial Intelligence , Education, Medical/standardsABSTRACT
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: 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
BACKGROUND: OpenAI's ChatGPT is a pioneering artificial intelligence (AI) in the field of natural language processing, and it holds significant potential in medicine for providing treatment advice. Additionally, recent studies have demonstrated promising results using ChatGPT for emergency medicine triage. However, its diagnostic accuracy in the emergency department (ED) has not yet been evaluated. OBJECTIVE: This study compares the diagnostic accuracy of ChatGPT with GPT-3.5 and GPT-4 and primary treating resident physicians in an ED setting. METHODS: Among 100 adults admitted to our ED in January 2023 with internal medicine issues, the diagnostic accuracy was assessed by comparing the diagnoses made by ED resident physicians and those made by ChatGPT with GPT-3.5 or GPT-4 against the final hospital discharge diagnosis, using a point system for grading accuracy. RESULTS: The study enrolled 100 patients with a median age of 72 (IQR 58.5-82.0) years who were admitted to our internal medicine ED primarily for cardiovascular, endocrine, gastrointestinal, or infectious diseases. GPT-4 outperformed both GPT-3.5 (P<.001) and ED resident physicians (P=.01) in diagnostic accuracy for internal medicine emergencies. Furthermore, across various disease subgroups, GPT-4 consistently outperformed GPT-3.5 and resident physicians. It demonstrated significant superiority in cardiovascular (GPT-4 vs ED physicians: P=.03) and endocrine or gastrointestinal diseases (GPT-4 vs GPT-3.5: P=.01). However, in other categories, the differences were not statistically significant. CONCLUSIONS: In this study, which compared the diagnostic accuracy of GPT-3.5, GPT-4, and ED resident physicians against a discharge diagnosis gold standard, GPT-4 outperformed both the resident physicians and its predecessor, GPT-3.5. Despite the retrospective design of the study and its limited sample size, the results underscore the potential of AI as a supportive diagnostic tool in ED settings.
Subject(s)
Emergency Service, Hospital , Humans , Emergency Service, Hospital/statistics & numerical data , Retrospective Studies , Aged , Female , Middle Aged , Male , Aged, 80 and over , Artificial Intelligence , Physicians/statistics & numerical data , Natural Language Processing , Triage/methodsABSTRACT
BACKGROUND: Commonly offered as supportive care, therapist-led online support groups (OSGs) are a cost-effective way to provide support to individuals affected by cancer. One important indicator of a successful OSG session is group cohesion; however, monitoring group cohesion can be challenging due to the lack of nonverbal cues and in-person interactions in text-based OSGs. The Artificial Intelligence-based Co-Facilitator (AICF) was designed to contextually identify therapeutic outcomes from conversations and produce real-time analytics. OBJECTIVE: The aim of this study was to develop a method to train and evaluate AICF's capacity to monitor group cohesion. METHODS: AICF used a text classification approach to extract the mentions of group cohesion within conversations. A sample of data was annotated by human scorers, which was used as the training data to build the classification model. The annotations were further supported by finding contextually similar group cohesion expressions using word embedding models as well. AICF performance was also compared against the natural language processing software Linguistic Inquiry Word Count (LIWC). RESULTS: AICF was trained on 80,000 messages obtained from Cancer Chat Canada. We tested AICF on 34,048 messages. Human experts scored 6797 (20%) of the messages to evaluate the ability of AICF to classify group cohesion. Results showed that machine learning algorithms combined with human input could detect group cohesion, a clinically meaningful indicator of effective OSGs. After retraining with human input, AICF reached an F1-score of 0.82. AICF performed slightly better at identifying group cohesion compared to LIWC. CONCLUSIONS: AICF has the potential to assist therapists by detecting discord in the group amenable to real-time intervention. Overall, AICF presents a unique opportunity to strengthen patient-centered care in web-based settings by attending to individual needs. INTERNATIONAL REGISTERED REPORT IDENTIFIER (IRRID): RR2-10.2196/21453.
ABSTRACT
BACKGROUND: Machine learning has advanced medical event prediction, mostly using private data. The public MIMIC-3 (Medical Information Mart for Intensive Care III) data set, which contains detailed data on over 40,000 intensive care unit patients, stands out as it can help develop better models including structured and textual data. OBJECTIVE: This study aimed to build and test a machine learning model using the MIMIC-3 data set to determine the effectiveness of information extracted from electronic medical record text using a named entity recognition, specifically QuickUMLS, for predicting important medical events. Using the prediction of extended-spectrum ß-lactamase (ESBL)-producing bacterial infections as an example, this study shows how open data sources and simple technology can be useful for making clinically meaningful predictions. METHODS: The MIMIC-3 data set, including demographics, vital signs, laboratory results, and textual data, such as discharge summaries, was used. This study specifically targeted patients diagnosed with Klebsiella pneumoniae or Escherichia coli infection. Predictions were based on ESBL-producing bacterial standards and the minimum inhibitory concentration criteria. Both the structured data and extracted patient histories were used as predictors. In total, 2 models, an L1-regularized logistic regression model and a LightGBM model, were evaluated using the receiver operating characteristic area under the curve (ROC-AUC) and the precision-recall curve area under the curve (PR-AUC). RESULTS: Of 46,520 MIMIC-3 patients, 4046 were identified with bacterial cultures, indicating the presence of K pneumoniae or E coli. After excluding patients who lacked discharge summary text, 3614 patients remained. The L1-penalized model, with variables from only the structured data, displayed a ROC-AUC of 0.646 and a PR-AUC of 0.307. The LightGBM model, combining structured and textual data, achieved a ROC-AUC of 0.707 and a PR-AUC of 0.369. Key contributors to the LightGBM model included patient age, duration since hospital admission, and specific medical history such as diabetes. The structured data-based model showed improved performance compared to the reference models. Performance was further improved when textual medical history was included. Compared to other models predicting drug-resistant bacteria, the results of this study ranked in the middle. Some misidentifications, potentially due to the limitations of QuickUMLS, may have affected the accuracy of the model. CONCLUSIONS: This study successfully developed a predictive model for ESBL-producing bacterial infections using the MIMIC-3 data set, yielding results consistent with existing literature. This model stands out for its transparency and reliance on open data and open-named entity recognition technology. The performance of the model was enhanced using textual information. With advancements in natural language processing tools such as BERT and GPT, the extraction of medical data from text holds substantial potential for future model optimization.
ABSTRACT
BACKGROUND: In the United States, 1 in 5 adults currently serves as a family caregiver for an individual with a serious illness or disability. Unlike professional caregivers, family caregivers often assume this role without formal preparation or training. Thus, there is an urgent need to enhance the capacity of family caregivers to provide quality care. Leveraging technology as an educational tool or an adjunct to care is a promising approach that has the potential to enhance the learning and caregiving capabilities of family caregivers. Large language models (LLMs) can potentially be used as a foundation technology for supporting caregivers. An LLM can be categorized as a foundation model (FM), which is a large-scale model trained on a broad data set that can be adapted to a range of different domain tasks. Despite their potential, FMs have the critical weakness of "hallucination," where the models generate information that can be misleading or inaccurate. Information reliability is essential when language models are deployed as front-line help tools for caregivers. OBJECTIVE: This study aimed to (1) develop a reliable caregiving language model (CaLM) by using FMs and a caregiving knowledge base, (2) develop an accessible CaLM using a small FM that requires fewer computing resources, and (3) evaluate the model's performance compared with a large FM. METHODS: We developed a CaLM using the retrieval augmented generation (RAG) framework combined with FM fine-tuning for improving the quality of FM answers by grounding the model on a caregiving knowledge base. The key components of the CaLM are the caregiving knowledge base, a fine-tuned FM, and a retriever module. We used 2 small FMs as candidates for the foundation of the CaLM (LLaMA [large language model Meta AI] 2 and Falcon with 7 billion parameters) and adopted a large FM (GPT-3.5 with an estimated 175 billion parameters) as a benchmark. We developed the caregiving knowledge base by gathering various types of documents from the internet. We focused on caregivers of individuals with Alzheimer disease and related dementias. We evaluated the models' performances using the benchmark metrics commonly used in evaluating language models and their reliability for providing accurate references with their answers. RESULTS: The RAG framework improved the performance of all FMs used in this study across all measures. As expected, the large FM performed better than the small FMs across all metrics. Interestingly, the small fine-tuned FMs with RAG performed significantly better than GPT 3.5 across all metrics. The fine-tuned LLaMA 2 with a small FM performed better than GPT 3.5 (even with RAG) in returning references with the answers. CONCLUSIONS: The study shows that a reliable and accessible CaLM can be developed using small FMs with a knowledge base specific to the caregiving domain.
ABSTRACT
Evidence-based medicine (EBM) emerged from McMaster University in the 1980-1990s, which emphasizes the integration of the best research evidence with clinical expertise and patient values. The Health Information Research Unit (HiRU) was created at McMaster University in 1985 to support EBM. Early on, digital health informatics took the form of teaching clinicians how to search MEDLINE with modems and phone lines. Searching and retrieval of published articles were transformed as electronic platforms provided greater access to clinically relevant studies, systematic reviews, and clinical practice guidelines, with PubMed playing a pivotal role. In the early 2000s, the HiRU introduced Clinical Queries-validated search filters derived from the curated, gold-standard, human-appraised Hedges dataset-to enhance the precision of searches, allowing clinicians to hone their queries based on study design, population, and outcomes. Currently, almost 1 million articles are added to PubMed annually. To filter through this volume of heterogenous publications for clinically important articles, the HiRU team and other researchers have been applying classical machine learning, deep learning, and, increasingly, large language models (LLMs). These approaches are built upon the foundation of gold-standard annotated datasets and humans in the loop for active machine learning. In this viewpoint, we explore the evolution of health informatics in supporting evidence search and retrieval processes over the past 25+ years within the HiRU, including the evolving roles of LLMs and responsible artificial intelligence, as we continue to facilitate the dissemination of knowledge, enabling clinicians to integrate the best available evidence into their clinical practice.
Subject(s)
Evidence-Based Medicine , Medical Informatics , Medical Informatics/methods , Medical Informatics/trends , Humans , History, 20th Century , History, 21st Century , Machine LearningABSTRACT
The advent of large language models (LLMs) such as ChatGPT has potential implications for psychological therapies such as cognitive behavioral therapy (CBT). We systematically investigated whether LLMs could recognize an unhelpful thought, examine its validity, and reframe it to a more helpful one. LLMs currently have the potential to offer reasonable suggestions for the identification and reframing of unhelpful thoughts but should not be relied on to lead CBT delivery.
ABSTRACT
Background: The introduction of ChatGPT by OpenAI has garnered significant attention. Among its capabilities, paraphrasing stands out. Objective: This study aims to investigate the satisfactory levels of plagiarism in the paraphrased text produced by this chatbot. Methods: Three texts of varying lengths were presented to ChatGPT. ChatGPT was then instructed to paraphrase the provided texts using five different prompts. In the subsequent stage of the study, the texts were divided into separate paragraphs, and ChatGPT was requested to paraphrase each paragraph individually. Lastly, in the third stage, ChatGPT was asked to paraphrase the texts it had previously generated. Results: The average plagiarism rate in the texts generated by ChatGPT was 45% (SD 10%). ChatGPT exhibited a substantial reduction in plagiarism for the provided texts (mean difference -0.51, 95% CI -0.54 to -0.48; P<.001). Furthermore, when comparing the second attempt with the initial attempt, a significant decrease in the plagiarism rate was observed (mean difference -0.06, 95% CI -0.08 to -0.03; P<.001). The number of paragraphs in the texts demonstrated a noteworthy association with the percentage of plagiarism, with texts consisting of a single paragraph exhibiting the lowest plagiarism rate (P<.001). Conclusions: Although ChatGPT demonstrates a notable reduction of plagiarism within texts, the existing levels of plagiarism remain relatively high. This underscores a crucial caution for researchers when incorporating this chatbot into their work.