JMIR AI最新文献

Background: Large language model (LLM)-based conversational agents have been increasingly used in digital health interventions. However, their specific application to physical activity (PA) and cognitive training-two critical well-being domains-has not been systematically mapped. In fact, these domains share an important need for personalized, adaptive support and conversational engagement, making them relevant targets for examining how LLM-based agents are currently conceptualized and deployed.

Objective: This scoping review aimed to map the extent, characteristics, and design practices of LLM-based conversational agents supporting PA or cognitive training, specifically analyzing their application contexts, social roles, and technological features.

Methods: Following PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews) guidelines, we searched Web of Science, Scopus, PubMed, ACM Digital Library, and IEEE Xplore for studies published between January 2018 and December 2024. We included eligible studies that described LLM-based conversational agents designed for PA or cognitive training. Two reviewers independently screened records and extracted data. Descriptive synthesis and framework analysis were used to characterize intervention domains, agent roles, prompting strategies, model types, and reported outcomes.

Results: Of 357 records screened, 10 studies met eligibility criteria (7 on PA and 3 on cognitive training). Applications predominantly involved coaching roles for PA and companion or scaffolding roles in cognitive domains. The agent landscape was dominated by proprietary LLMs (GPT-3.5, GPT-4, and Bard), with limited use of open-weight models. Prompt engineering emerged as a central yet inconsistently documented design mechanism. Reported outcomes mainly focused on perceived usefulness, engagement, or content quality, with few quantitative behavioral outcomes.

Conclusions: LLM-based conversational agents have demonstrated early promise for supporting PA and emerging approaches to cognitive training, yet the current evidence remains exploratory and methodologically limited. Key challenges persist, including inconsistent reporting of prompts, reliance on proprietary models with limited reproducibility, and a lack of standardized outcome measures. More rigorous and transparently documented evaluations of these tools are required to strengthen the evidence base and guide future development.

Background: Low back pain (LBP) is a leading cause of disability worldwide, affecting people of all ages while showing increasing prevalence among younger demographics. Patients may present with different symptoms and treatment responses despite identical magnetic resonance imaging results, making it difficult to determine whether surgical and medical interventions are appropriate.

Objective: This study aimed to develop SPOILS (Software to Predict Outcome in Lumbar Spondylosis), an artificial intelligence-based decision support tool that merges clinical intelligence and radiomics to generate customized therapy plans for patients with LBP.

Methods: The SPOILS system used deep learning models to perform automated segmentation, enabling the extraction of geometrical parameters, including disk height, disk width, vertebrae height, vertebrae width, canal diameter, disk height index, signal intensity, and disk volume. A labeled dataset was created using expert-verified Pfirrmann and spondylosis severity gradings to address the clinical issues stemming from manual grading variability and subjectivity. Machine learning algorithms were used with this combined dataset to predict outcomes and recommend personalized treatment plans.

Results: The DeepLabV3+ segmentation model with a ResNet50 encoder achieved 95.5% accuracy, which increased to 98.7% after 8-fold cross-validation and simultaneously improved precision (96.95%), recall (97.1%), Dice coefficient (96.9%), and intersection over union (IoU; 94.8%). The convolutional neural network with MobileNetV2 achieved 97.84% accuracy and 96.76% IoU for spondylosis severity prediction after cross-validation. The Gradient Boost classifier demonstrated the best results with geometrical data by achieving 91.65% accuracy and 84.59% IoU.

Conclusions: SPOILS introduced an innovative method to customize LBP treatment through the combination of artificial intelligence technology with radiological data and clinical expertise.

Background: Type 2 diabetes mellitus (T2D) is a rapidly growing global health concern requiring innovative treatment methods. Ozempic (semaglutide), a glucagon-like peptide-1 receptor agonist, has proven consistent effectiveness in lowering blood glucose levels, supporting weight loss, and minimizing cardiovascular complications. In parallel, artificial intelligence (AI) elevates diabetes care yet complements these efforts by converting raw data from wearable devices, electronic health records, and medical imaging into practical insights for efficient, tailored, and customized treatment plans.

Objective: The objective of this systematic review is to examine current evidence of AI-driven methods to optimize Ozempic-based T2D therapy.

Methods: A total of 18 peer-reviewed articles were identified, revealing four dominant thematic clusters: (1) patient stratification and risk prediction, (2) AI-enhanced imaging for body composition changes, (3) cardiovascular and metabolic risk assessment, and (4) personalized AI-driven dosage.

Results: Across multiple metrics, such as glycated hemoglobin reduction, weight loss, cardiovascular benefits, and adverse event mitigation, AI-based approaches outperformed standard fixed-dose regimens. A theoretical framework is proposed for AI-Ozempic integration, with continuous data collection, AI processing, clinical decision support, real-time support, and real-time feedback and modeling iteration refinement cycles.

Conclusions: Significant gaps remain a persistent challenge, including the need for large-scale randomized controlled trials, longer follow-up periods, explainable AI models, regulatory validation, and practical strategies for routine clinical implementation. The findings emphasize the AI's potential to transform semaglutide therapy while delineating important paths for future research.

Background: Artificial intelligence (AI) models are increasingly being used in medical education. Although models like ChatGPT have previously demonstrated strong performance on United States Medical Licensing Examination (USMLE)-style questions, newer AI tools with enhanced capabilities are now available, necessitating comparative evaluations of their accuracy and reliability across different medical domains and question formats.

Objective: This study aimed to evaluate and compare the performance of 5 publicly available AI models: Grok, ChatGPT-4, Copilot, Gemini, and DeepSeek, on the USMLE Step 1 free 120-question set, assessing their accuracy and consistency across question types and medical subjects.

Methods: This cross-sectional observational study was conducted between February 10 and March 5, 2025. Each of the 119 USMLE-style questions (excluding 1 audio-based item) was presented to each AI model by using a standardized prompt cycle. Models answered each question 3 times to assess confidence and consistency. Questions were categorized as text-based or image-based and as case-based or information-based. Statistical analysis was performed using chi-square and Fisher exact tests, with Bonferroni adjustment for pairwise comparisons.

Results: Grok achieved the highest score (109/119, 91.6%), followed by Copilot (101/119, 84.9%), Gemini (100/119, 84%), ChatGPT-4 (95/119, 79.8%), and DeepSeek (86/119, 72.3%). DeepSeek's lower score was due to an inability to process visual media, resulting in 0% accuracy on image-based items. When limited to text-only questions (n=96), DeepSeek's accuracy increased to 89.6% (86/96), matching Copilot. Grok showed the highest accuracy on image-based (21/23, 91.3%) and case-based questions (70/78, 89.7%), with statistically significant differences observed between Grok and DeepSeek on case-based items (P=.01). The models performed best in biostatistics and epidemiology (5.8/6, 96.7%) and worst in musculoskeletal, skin, and connective tissue (4.4/7, 62.9%). Grok maintained 100% consistency in responses, while Copilot demonstrated the most self-correction (112/119, 94.1% consistency), improving its accuracy to 89.9% (107/119) on the third attempt.

Conclusions: AI models showed varying strengths across domains, with Grok demonstrating the highest accuracy and consistency in this dataset, particularly for image-based and reasoning-heavy questions. Although ChatGPT-4 remains widely used, newer models like Grok and Copilot also performed competitively. Continuous evaluation is essential as AI tools rapidly evolve.

Background: High-quality nursing services are essential for improving patient satisfaction and health outcomes. Today, artificial intelligence (AI) applications such as ChatGPT offer potential solutions to enhance patient education and assist nurses in providing more accurate and personalized information. Despite its promising potential in nursing education, concerns regarding information accuracy, privacy, and ethical considerations must be addressed.

Objective: This scoping review aimed to map the current evidence on the use of ChatGPT (OpenAI) as an educational resource in nursing practice, focusing on its educational functions, reported outcomes, and implementation challenges.

Methods: The literature search was conducted using 3 databases (PubMed, Scopus, and ProQuest). Following the Population, Concept, and Context framework. Inclusion criteria encompassed studies published between 2019 and 2025, in English, and available in full text. The PRISMA-ScR (Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews) guideline was used to guide the screening and selection process.

Results: We included 20 articles and synthesized four main findings: (1) AI in patient education and simplification of medical information, (2) AI in clinical decision-making and patient monitoring, (3) AI in nursing education, and (4) challenges and prospects of AI in nursing. Across studies, commonly reported limitations involved response accuracy inconsistencies, ethical concerns, and the absence of standardized implementation guidelines.

Conclusions: ChatGPT shows promise as an adjunct educational resource in nursing practice, particularly for information accessibility and learner engagement. Nevertheless, its use requires professional oversight, ethical safeguards, and further implementation-focused research.

Background: Artificial intelligence (AI) tools are being developed in a rapidly evolving technology. The convergence of ethical, technical, and research methods' considerations is crucial for multidisciplinary teams aiming to produce effective AI tools. The success of these tools postdeployment hinges on the intricate interplay between the AI system's development on its output through rigorous decision-making processes and stakeholders' capacity to act on the AI's recommendations.

Objective: This paper synthesizes ethical, technical, and epidemiological considerations for all involved in artificial intelligence tool production (ETEPAI), based on established guidelines, checklists, and frameworks.

Methods: Relevant guidelines, checklists, frameworks, and expert recommendations were systematically identified and synthesized into ETEPAI, an ethical, technical, and epidemiological framework for AI tool development in health care.

Results: From 30 reviewed frameworks, ETEPAI integrates critical considerations across 4 stages (design, development, deployment, and postdeployment) and 3 domains (ethics, technical, and epidemiological), providing a compact yet comprehensive guide. It includes probing questions, key indicators, and common pitfalls to support high-quality, ethically sound, and clinically relevant AI tools. ETEPAI aligns with European Union trustworthiness standards and is supported by a research proposal template and supplementary references to aid implementation and adoption. We present probing questions and critical pointers across 4 stages from the design, development, deployment, and postdeployment, highlighting their relevance in health care settings. The designing stage aligns with epidemiologic research methodologies, while the development stage emphasizes transparent project execution. Deployment and postdeployment stages focus on real-world implementation. Additionally included are common pitfalls and challenges to emphasize the importance of due attention to the importance of ETEPAI considerations to avoid serious consequences.

Conclusions: Applying ETEPAI ensures comprehensive, complete, compact, and crisp consideration from conception to execution, promoting high-quality, ethically sound, and clinically relevant AI tools. The brevity and conciseness of ETEPAI might be adequate for trained personnel and serve as clear signposts to unprepared stakeholders.

Background: Despite the significant post-COVID-19 pandemic surge in research using symptom data and machine learning (ML) for patient screening, data on patient trajectories and epidemiological conditions, although crucial, have remained underused.

Objective: This study aimed to enhance the performance of ML models for COVID-19 screening by incorporating mobility and epidemic information in addition to patient symptom data.

Methods: Data, including daily self-reported symptoms, location information, and test results, were collected from 48,798 individuals using a smartphone app. These data were then combined with Our World in Data and national government epidemic information to train 5 ML-based screening models to classify patient infection status. The models were logistic regression, extreme gradient boosting, light gradient boosting machine, tabular data network, and Google AutoML.

Results: The addition of mobility and epidemic data significantly improved the performance of all 5 models. The highest area under the receiver operating characteristic curve score increased from 0.8712 without mobility and epidemic data to 0.9104 with mobility and epidemic data. This highlights the considerable impact of external information on enhancing the performance of ML models.

Conclusions: This study demonstrated the potential of using mobility and epidemic data, such as location information and epidemic data, in combination with patient symptom data to improve the accuracy of ML models for diagnosing COVID-19. Considering additional contextual information can enhance the ability to screen for COVID-19.