Background: The future of artificial intelligence in medicine includes the use of machine learning and large language models to improve diagnostic accuracy, as a point-of-care tool, at the time of admission to an acute care hospital. The large language model, ChatGPT-4, has been shown to diagnose complex medical conditions with accuracies comparable to experienced clinicians, however, most published studies involved curated cases or examination-like questions and are not point-of-care. To test the hypothesis that ChatGPT-4 can make an accurate medical diagnosis using real-world medical cases and a convenient cut and paste strategy, we performed a retrospective case study involving critically ill patients admitted to a community hospital medical intensive care unit. Methods: A redacted H&P was essentially cut and pasted into ChatGPT-4 with uniform instructions to make a leading diagnosis and a list of 5 possibilities as a differential diagnosis. All features that could be used to identify patients were removed to ensure privacy and HIPAA compliance. The ChatGPT-4 diagnoses were compared with critical care physician diagnoses using a blinded longitudinal chart review as the ground truth diagnosis. Results: A total of 120 randomly selected cases were included in the study. The diagnostic accuracy was 88.3% for physicians and 85.0% for ChatGPT-4, with no significant difference by McNemar testing (p-value of 0.249). The agreement between physician diagnosis and ChatGPT-4 diagnosis was moderate, 0.57 (95% CI: 0.35–0.79), based on Cohen's kappa statistic. Conclusion: These results suggest that ChatGTP-4 achieved diagnostic accuracy comparable to board certified physicians in the context of critically ill patients admitted to a community medical intensive care unit. Furthermore, the agreement was only moderate, suggesting that there may be complementary ways of combining the diagnostic acumen of physicians and ChatGPT-4 to improve overall accuracy. A prospective study would be necessary to determine if ChatGPT-4 could improve patient outcomes as a point-of-care tool at the time of admission.

Clostridium difficile infection (CDI) is one of the rising public health threats in the United States. It has imposed significant morbidity and mortality in the elderly population. However, the burden of the disease in the young population is unclear. This study aimed to identify hospitalization trends and outcomes of CDI in the young population. We obtained data from the National (Nationwide) Inpatient Sample (NIS) for hospitalizations with CDI between 2007 and 2017. We used the International Classification of Diseases Ninth Edition-Clinical Modification (ICD-9-CM) and ICD-10-CM to identify CDI and other diagnoses of interest. The primary outcome of our study was to identify the temporal trends and demographic characteristics of patients aged less than 50 years old hospitalized with CDI. The secondary outcomes were in-hospital mortality, length of hospital stay (LOS), and discharge dispositions. We utilized the Cochran Armitage trend test and multivariable survey logistic regression models to analyze the trends and outcomes. From 2007 to 2017, CDI was present among 1,158,047 hospitalized patients. The majority (84.04%) of the patients were ≥50 years old versus 15.95% of patients <50 years old. From 2007 to 2017, there was a significant increase in CDI among <50-year-old hospitalized patients (12.6% from 2007 to 18.1% in 2017; p < 0.001). In trend analysis by ethnicities, among patients <50 years old, there was an increasing trend in Caucasians (63.9% versus 67.9%; p < 0.001) and Asian females (58.4% versus 62.6%; p < 0.001). We observed an increased trend of discharge to home (91.3% vs 95.8%; p < 0.001) in association with a decrease in discharge to facility (8.3% vs 4%; p < 0.001). The average LOS from 2007 to 2017 was 5 ± 0.03 days, which remained stable during the study period. The proportion of young (<50 years old) hospitalized patients with CDI has been steadily increasing over the past decade. Our findings might represent new epidemiological trends related to non-traditional risk factors. Future CDI surveillance should extend to the young population to confirm our findings, and the study of emerging risk factors is required to better understand the increasing CDI hospitalization in the young population.

Machine learning (ML) has immense potential in healthcare, but its implementation is often restricted to those with coding expertise. This study explores the use of automatic ML (AutoML) platforms, specifically Google Cloud’s Vertex AI, to enable clinicians and medical trainees to develop and deploy ML models without programming knowledge. We describe a step-by-step approach to building a polyp detection model using publicly available colonoscopy images from the Kvasir dataset. By leveraging AutoML’s intuitive interface, we trained and deployed a model capable of detecting polyps with an average precision of 100%, a precision of 64.5%, and a recall of 54.5%. The entire process required minimal technical expertise, making it accessible for residents and non-programmers interested in ML applications in clinical practice. While AutoML offers a simplified approach to ML model development, it has inherent limitations, including reduced flexibility, lack of hyperparameter tuning, and the necessity for high-quality labeled datasets. Despite these constraints, our results demonstrate that AutoML platforms can serve as a valuable tool for clinicians to prototype ML solutions for healthcare applications. By providing a no-code approach to artificial intelligence (AI) model development, this study highlights how medical professionals can integrate ML into practice without requiring programming expertise, fostering innovation and accessibility in clinical AI research.

Background: High-value care emphasizes services offering significant health benefits, aligning with patient preferences while minimizing costs and interventions of little benefit. This approach is increasingly vital in a healthcare environment constrained by finite resources and rising costs. Large Language Models (LLMs) like Generative Pretrained transformer-4 (GPT-4), with their vast data processing capabilities, offer a promising avenue for supporting healthcare providers in making evidence-based, high-value care decisions. Aims: This study aims to evaluate the performance of OpenAI’s GPT-4 in providing responses to high-value care clinical scenarios within internal medicine, assessing its accuracy, relevance, and reasoning against established medical guidelines and literature. Materials and Methods: An observational study was conducted using MKSAP-19’s high-value care questions, comparing GPT-4’s responses to the correct answers based on established studies, trials, and guidelines. The study did not involve real patient data, circumventing the need for Institutional Review Board (IRB) approval. Performance metrics focused on the accuracy, relevance, and consistency of GPT-4’s answers. Results: GPT-4 demonstrated a 74.4% accuracy rate (32 out of 43 questions) across a range of high-value care clinical scenarios, including image-based questions. Errors made by GPT-4 were similar to those by medical residents using MKSAP-19, suggesting areas for model improvement and potential educational applications. The study detailed GPT-4’s decision-making pattern, emphasizing its clinical reasoning capabilities. Conclusion: The findings suggest that GPT-4 can significantly support high-value care in internal medicine by providing accurate, evidence-based responses to complex clinical scenarios. Despite its limitations, including a 25.6% error rate and the scope of its training data, GPT-4’s performance indicates its potential as both a clinical and educational tool in healthcare.

Aim: This study aims to evaluate the performance of large language models (LLMs), specifically OpenAI’s ChatGPT and Google Bard, in answering gastroenterology clinical questions from the Medical Knowledge Self-Assessment Program-19 (MKSAP-19), thereby assessing their potential utility in clinical decision-making within the field of gastroenterology. Materials and Methods: A comparative analysis was conducted using a dataset of 50 gastroenterology questions from MKSAP-19, assessing the ability of ChatGPT and Bard to provide correct answers without prior training or access to MKSAP-19 materials. The performance of each LLM was evaluated based on the percentage of correct answers, with a passing score set at 50%. Results: ChatGPT outperformed Bard, achieving a 68% success rate in answering the questions correctly, compared to Bard’s 44%. ChatGPT attempted all questions, while Bard abstained from answering two. The analysis also identified specific areas where both LLMs struggled, indicating gaps in their clinical reasoning capabilities. Conclusions: ChatGPT demonstrated a higher efficacy in clinical decision-making for gastroenterology questions than Bard, suggesting the potential of LLMs as supplementary tools in clinical settings. However, the limitations of LLMs, including their inability to interpret images and consider real-life factors such as social determinants of health, highlight the need for further development before they can independently guide medical decisions.