Objectives: Malnutrition remains the leading cause of child mortality in Tanzania, with over 34% of children under 5 years of age affected by stunting and approximately 5% experiencing acute malnutrition. This study aimed to develop a machine learning model to predict fetal nutritional status using maternal and clinical data, thereby enabling early risk identification for health workers and parents and facilitating timely intervention. To enhance practical applicability, the model was deployed within a mobile application to provide accessible, real-time predictions that support prompt clinical and behavioral responses.Methods: Using a dataset collected in Tanzania, the performance of multiple binary classification algorithms—logistic regression, multi-layer perceptron, random forest, extreme gradient boosting, and light gradient boosting machine (LightGBM)—was compared using the geometric mean and F-measure. These models were trained on clinical data from 11,703 pregnant women to predict fetal nutritional status based on maternal and clinical variables.Results: The results indicated that the LightGBM algorithm achieved the best overall performance in predicting fetal nutritional status. The most influential predictors included maternal age, weight, fetal age, hemoglobin level, number of meals per day, medical history, and education level. Additionally, 93% of respondents reported satisfaction with the application’s predictive functionality, supporting its potential utility for early intervention in low-resource settings.Conclusions: These findings highlight the potential of data-driven approaches to address public health challenges in maternal and child health. The proposed model may enable healthcare providers to make timely, informed decisions that improve maternal and fetal outcomes, ultimately contributing to the reduction of child malnutrition in Tanzania.
Objectives: This study aimed to apply machine learning (ML) techniques to predict postoperative length of stay (LOS) in patients with hip fracture. Because LOS varies widely across individuals owing to complex clinical factors, accurate prediction remains challenging. To address this challenge, an enhanced two-stage approach was developed and compared with a conventional one-stage approach.Methods: Data from 3,118 surgically treated patients with hip fracture were extracted from a hospital information system. Demographic and perioperative variables were analyzed, and the dataset was divided into training and test sets at a 70:30 ratio. ML algorithms were applied using a two-stage modeling approach. In the first stage, a classification model categorized LOS as short stay or long stay. In the second stage, regression models predicted the number of hospital days within each group. Performance was evaluated using accuracy, precision, recall, and F1-score for classification and mean absolute error (MAE), root mean square error (RMSE), and mean relative error (MRE) for regression.Results: In the one-stage approach, the support vector machine model showed the lowest prediction error, with an MAE of 2.20, RMSE of 3.18, and MRE of 0.42. The two-stage approach, which integrated classification and regression, outperformed the onestage approach, achieving an MAE of 1.46, RMSE of 1.86, and MRE of 0.35.Conclusions: The two-stage approach outperformed the one-stage approach, suggesting that LOS stratification improves prediction accuracy. This improvement may help hospitals anticipate resource needs, plan postoperative care, and manage bed allocation more effectively.
Objectives: Parkinson’s disease (PD) is a chronic neurodegenerative disorder characterized by central nervous system dysfunction. Early identification may enable prompt treatment and help slow the progression of disabling symptoms. Previous studies have reported that clinical assessments based on visual interpretation may be insufficiently accurate and often miss early PD. Therefore, this study aimed to develop an automated single-photon emission computed tomography-based model for binary classification of healthy controls and patients with early PD.Methods: We analyzed 514 DaTSCAN images from the Parkinson’s Progression Markers Initiative database, using one unique scan per individual. The workflow comprised three main stages: image processing; computation of 23 features, including radial, threshold, boundary, and striatal binding ratio features; and image classification using machine learning algorithms.Results: The medium Gaussian support vector machine achieved an accuracy of 97.09% ± 1.53% (95% confidence interval, 95.11%–99.07%), sensitivity of 98.18% ± 1.48%, specificity of 93.85% ± 3.44%, and area under the receiver operating characteristic curve of 98.49% ± 1.31%. This performance was significantly higher than that of the three-dimensional convolutional neural network baseline model (accuracy, 94.55% ± 2.98%; p = 0.045), while requiring substantially less training time.Conclusions: In this dataset, carefully designed feature engineering combined with classical machine learning outperformed the deep-learning baseline when the training data were limited and the selected features aligned with clinical diagnostic criteria. This approach achieved high accuracy for early PD detection and may provide computational efficiency and interpretability suitable for clinical implementation.
Objectives: Inherited retinal dystrophies are characterized by high genetic and phenotypic heterogeneity, and their clinical progression may alternate between latent and active phases. Identifying the onset of the active phase may support earlier intervention for inflammatory retinal degeneration. Plasma proteomics has shown potential for characterizing predictive biomarkers in retinal diseases, but its application remains experimental. This study aimed to develop a methodological simulation exercise to evaluate the performance of machine learning (ML) models in distinguishing active and latent phases using an artificially generated dataset.Methods: An artificial dataset of 500 samples was created, and plasma proteomic profiles were generated for each sample using arbitrary values. Sample classification was based on a pathway activation score. Four ML models were tested: support vector machine, random forest, logistic regression, and extreme gradient boosting. Each model was trained across a range of hyperparameters.Results: Logistic regression achieved the best performance, with an accuracy of 0.73, precision of 0.73, and F1-score of 0.73.Conclusions: This simulation study showed that synthetic proteomic datasets can be used to evaluate ML approaches for distinguishing active and latent phases of retinal dystrophies when real data are scarce. Synthetic data can support the creation of targeted datasets for proteins associated with retinal dystrophies, helping to address the limited availability of suitable open-source data.
Objectives: Machine learning (ML) techniques have shown strong potential for predicting infant mortality (IM), but their application in the Indian context remains limited. This study aimed to use ML algorithms to predict IM in India using a large national survey database.Methods: Data were analyzed from the National Family Health Survey-5, 2019–2021, a large cross-sectional survey. Random forest, decision tree, adaptive boosting, logistic regression, and naïve Bayes models were implemented using Weka version 3.8.3. Model performance was evaluated using accuracy, precision, F1-score, Matthews correlation coefficient, and area under the curve (AUC).Results: Compared with logistic regression, which achieved 65.2% accuracy, the random forest and decision tree models showed higher predictive accuracy, at 74.1% and 73.2%, respectively. Their AUCs were 0.80 and 0.79, respectively, compared with 0.69 for logistic regression. The models identified birth order, maternal education, twin birth, wealth index, cooking fuel use, and age at first birth as the six strongest predictors of IM.Conclusions: In this analysis, random forest and decision tree models outperformed logistic regression in predicting IM. These findings underscore the relevance of sociodemographic and economic disparities and support the use of ML algorithms for risk prediction and targeted interventions aimed at reducing IM.
Objectives: The hospital standardised mortality ratio (HSMR) is the ratio of the observed number of hospital deaths to the expected number of deaths, with the latter estimated using statistical models that adjust for available case-mix factors. This study aimed to develop and validate in-hospital or 30-day post-discharge mortality prediction models for 40 diagnosis groups within the HSMR framework, using penalized logistic regression (pLR) and random forest (RF), and to compare the performance of these two approaches.Methods: We analysed 1,144,890 hospital admissions from 14 Malaysian state hospitals between 2012 and 2016. Separate models were developed for each diagnosis group using nine administrative features, including age, comorbidities, and admission category. Model performance was evaluated using mean Brier scores and Cstatistics across multiple bootstrapped datasets to obtain less biased performance estimates. Aggregate expected mortality counts were also compared with observed counts.Results: The overall observed mortality rate was 10.2%. The pLR models consistently showed better discrimination and calibration than the RF models, with lower Brier scores and higher C-statistics across the 40 diagnosis groups. On average, the C-statistic for pLR exceeded that for RF by 0.062. Although the RF model produced aggregate mortality predictions that were numerically closer to the observed counts, it showed high variance and poorer probabilistic calibration than pLR.Conclusions: The pLR model tended to underestimate mortality more than RF but still demonstrated better calibration and discrimination, making it the preferable model for HSMR analysis in this dataset.
Objectives: Brain magnetic resonance imaging (MRI) contains identifiable facial and cranial features, creating privacy risks that can limit secondary research use. This review examines current MRI de-identification technologies, quantitative validation methods, and governance frameworks to identify practical strategies for preserving data utility while protecting patient privacy. Methods: A descriptive narrative review was conducted across technical and policy domains. Studies of facial deidentification were analyzed according to the tools used, validation procedures, and downstream analytic performance. The reviewed approaches included traditional defacing, refacing, and deep-learning-based anonymization. Evaluation frameworks used the structural similarity index measure (SSIM), Dice similarity coefficient (DSC), intraclass correlation coefficient (ICC), and the paired t-test to quantify both privacy preservation and analytic fidelity. A parallel policy analysis compared the Health Insurance Portability and Accountability Act (HIPAA), the General Data Protection Regulation (GDPR), Japan’s Act on Anonymized Medical Information, Taiwan’s Personal Data Protection Act, and South Korea’s Personal Information Protection Act and 2024 Health Data Use Guidelines to assess policy convergence and institutional consistency. Results: Visual inspection studies reported that FreeSurfer preserved cortical anatomy but incompletely removed facial features, whereas FSL_deface overmasked some nonfacial regions. Artificial intelligence (AI)-based recognition tests achieved 28%–38% accuracy on defaced data, confirming measurable residual re-identification risk. Quantitative assessments identified segmentation degradation, including a DSC decrease from 0.970 to 0.918, and regional volumetric variability, including a hippocampal ICC of 0.742, with p < 0.05. Generative adversarial network-based refacing improved perceptual similarity, with SSIM values >0.7, but retained subtle facial geometry. The governance analysis indicated that HIPAA and GDPR provide established standards, whereas South Korea’s Data Review Board oversight remains discretionary and nonuniform, limiting reproducibility across institutions. Conclusions: MRI de-identification requires integrated pipelines that combine AI-based facial masking and metadata cleansing with standardized evaluation metrics and enforceable review protocols.
Objectives: In prehospital and in-hospital emergency care, artificial intelligence (AI)-enabled emergency medical services (EMS) systems are reshaping clinical workflows through real-time triage, automated hospital assignment, telemedicine, and biometric monitoring. However, their implementation in Korea has revealed substantial legal and regulatory barriers. This study analyzed the structural incompatibilities between AI-enabled EMS and Korea’s current legal framework, with particular attention to the Medical Service Act, Emergency Medical Service Act, and Personal Information Protection Act.Methods: We conducted a statutory review and case analysis of pilot projects, including the Chungbuk Smart EMS Project, the AI HOTline in Ilsan, and Yonsei’s AI-Ambulance model. We also performed a comparative assessment of foreign regulatory frameworks, including the US Health Insurance Portability and Accountability Act and the EU General Data Protection Regulation.Results: Key barriers included the restriction of remote care to interphysician communication under Article 34 of the Medical Service Act, the lack of legal mandates for real-time hospital capacity sharing under the Emergency Medical Service Act, and limits on real-time sensitive-data processing caused by vague emergency clauses in the Personal Information Protection Act. These barriers constrain the scalability of AI-enabled EMS in Korea.Conclusions: To support lawful and secure integration of AI-driven emergency care in Korea, we propose legislative revisions that establish emergency exceptions for remote AI-supported guidance, require hospital resource disclosure, and provide safe-harbor protections for healthcare professionals who rely on certified AI tools. A harmonized legal infrastructure is essential for realizing the potential of digital transformation in emergency care.
Objectives: This study aimed to develop a custom GPT-based, knowledge-guided documentation support system that drafts emergency department (ED) interdepartmental consultation requests from minimal free-text input and evaluate its feasibility, usability, and workflow integration.Methods: The system generates a patient summary, recommended consulting department, structured consultation draft, and clinical reference notes from brief natural language input, including age, chief concern, and key findings. Two public AI Hub datasets, the Essential Medical Knowledge Dataset (Dataset No. 71875) and the Specialized Medical Knowledge Dataset (Dataset No. 71874), were used to guide prompt construction and constrain output structure rather than for model training. Department recommendation concordance was assessed using 22 de-identified ED consultation scenarios purposively selected across multiple specialties. The documented department was removed from the input and compared with the system’s recommendation. A post-use survey of 10 emergency physicians and residents assessed workflow utility, clinical appropriateness, usability, reflection of relevant information, perceived patient safety, adoption intention, and editing requirements using 5-point Likert scales.Results: In the 22-case pilot set, the system’s recommended department aligned with the historically documented department in all cases under controlled conditions. Given the small sample size and study design, this finding represents preliminary evidence of workflow alignment rather than definitive validation of clinical performance.Conclusions: The proposed system demonstrates the feasibility of large language modelenabled documentation support for generating standardized ED consultation requests, potentially reducing documentation burden. The system is intended to support, not replace, clinical decision-making, and clinician review remains essential. With further validation, this approach may be adaptable to other clinical referral contexts.
Objectives: Type 2 diabetes mellitus (T2DM) is a prevalent chronic metabolic disorder associated with serious complications, including nephropathy, cardiovascular disease, retinopathy, and neuropathy. Given its increasing incidence and the complexity of associated factors—such as obesity, metabolic syndrome, and sedentary lifestyle—accurate identification is essential. This study aimed to evaluate and compare the performance of several machine learning algorithms to identify key associated factors and detect individuals with T2DM within this dataset.Methods: A publicly available dataset from Kaggle, comprising health records of 99,982 individuals, was used. Five supervised machine learning models were evaluated: Bayesian ridge regression, logistic regression, extreme gradient boosting (XGBoost), artificial neural networks, and random forest. Each model was trained and evaluated to assess classification performance. Performance was measured using the area under the receiver operating characteristic curve (AUC–ROC) and accuracy. SHapley Additive Explanations (SHAP) values were used to interpret model outputs and identify the most influential features.Results: Among the five models, XGBoost demonstrated the highest performance, achieving an accuracy of 96% and an AUC–ROC of 0.98. SHAP analysis identified hemoglobin A1c, blood glucose, age, body mass index, and sex as the most influential predictors of T2DM. Conclusion: XGBoost was the most effective algorithm for identifying individuals with T2DM in this dataset. It also provided insights into the relative importance of clinical features, supporting more precise classification. However, results should be interpreted with caution until validated in independent cohorts.
Objectives This study evaluated the feasibility of voice recognition-based electronic medical record (EMR) documentation for periodontal probing in dentistry, particularly emphasizing Korean-English bilingual speech patterns and real-world clinical conditions. Methods Experiments were conducted in a dental chair setting during routine clinical hours. Environmental noise levels were measured, and two microphone types (stationary and pin-type) were evaluated. Periodontal probing phrases composed of three-digit numbers and positional terms were used for speech recognition. Consistent with common clinical practice in Korea, numerical values were spoken in Korean, whereas positional terms were spoken in English. Two speech-to-text application programming interfaces, Google Cloud Speech-to-Text and Naver Clova Speech Recognition, were assessed. Recognition accuracy was evaluated for both numerical components and complete bilingual phrases. Results The mean environmental noise level was 60.65 dB and was minimally influenced by activity at adjacent dental chairs. The stationary microphone failed to capture speech effectively, whereas the pin-type microphone demonstrated stable recognition performance. For three-digit number recognition, accuracy was 88.3% with Google and 96.8% with Naver. For full-phrase recognition, complete matching was achieved in 36.7% of cases for Google and 52.5% for Naver. Partial recognition occurred more frequently for numerical components than for English positional terms. Conclusions Voice recognition-based EMR documentation for periodontal probing demonstrated preliminary feasibility in a dental clinical environment; however, performance was influenced by Korean-English bilingual speech patterns. These findings suggest that bilingual speech characteristics should be considered when implementing voice recognition systems in dental EMR workflows. Further optimization is required before routine clinical application.
OBJECTIVES:This study aimed to evaluate and compare the accuracy of multiple artificial intelligence (AI) models (ChatGPT 5.2 Pro, Gemini 3 Fast, Claude 4.5 Sonnet, and Microsoft Copilot) in detecting orthodontic malocclusion features in standardized multiview intraoral photographs. The reference standard was assessment by an orthodontist. METHODS:A cross-sectional observational study was conducted using five standardized intraoral photographs (frontal, right lateral, left lateral, maxillary occlusal, and mandibular occlusal) obtained from 50 children aged 9-12 years. The following eight malocclusion parameters were assessed: anterior crowding, diastema, overjet, overbite, molar relationship, canine relationship, crossbite, and dental arch symmetry. Diagnostic accuracy and agreement between each AI model and the orthodontist were evaluated using Cohen's kappa (κ) and the area under the receiver operating characteristic curve (AUC). RESULTS:Agreement between the AI models and the orthodontist ranged from poor to moderate across all orthodontic domains, with Cohen's κ values ranging from -0.15 to 0.63. Visually prominent alignment features, including anterior crowding and diastema, demonstrated comparatively higher agreement (κ, 0.00-0.63) and discriminatory performance, with AUC values ranging from 0.56 to 0.85. In contrast, parameters requiring precise spatial interpretation, such as sagittal relationships, overbite, crossbite, and arch morphology, showed consistently low agreement (κ, -0.15 to 0.38) and poor to near-random classification performance, with AUC values predominantly ranging from 0.41 to 0.70 and, in some cases, approaching 0.50. CONCLUSIONS:Current multimodal AI models demonstrate limited, parameter-dependent accuracy in detecting orthodontic malocclusions from intraoral photographs. These findings emphasize the limitations of general-purpose AI systems for orthodontic decision support and highlight the need for task-specific models trained on clinically annotated datasets.
Objectives Large language models (LLMs) are increasingly being used for medical question-answering (QA) tasks. However, most models are trained primarily on English-language data, which limits their effectiveness in non-English clinical contexts. In healthcare settings such as Korea, LLMs adapted to both the local language and the medical domain are needed. This study aimed to evaluate the performance of Korean-native and multilingual LLMs fine-tuned on Korean arthritis-related medical QA data and to examine the impact of language and domain adaptation. Methods A dataset of 5,451 Korean QA pairs related to arthritis was constructed from a public medical corpus (AI Hub). Five LLMs (Mi:dm, EXAONE, Kanana, HyperCLOVAX, and LLaMA) were fine-tuned under identical conditions using Low-Rank Adaptation with 4-bit quantization. Model performance was evaluated on 597 validation samples using BERTScore-F1 and SBERT similarity, along with a qualitative evaluation of clinical correctness, safety, and response completeness. Results EXAONE and HyperCLOVAX showed comparable quantitative performance in semantic accuracy and contextual consistency. Mi:dm achieved lower similarity-based scores than EXAONE and HyperCLOVAX but showed the highest performance in the qualitative evaluation, particularly for clinical correctness. Kanana exhibited moderate performance with limited domain adaptability. LLaMA showed the lowest performance in Korean medical QA, although it achieved the largest relative improvement, indicating challenges in adaptation to Korean clinical contexts. Conclusions Domain-specific fine-tuning and Korean-oriented model design improved performance in Korean arthritis medical QA. EXAONE and HyperCLOVAX achieved the highest semantic similarity, whereas Mi:dm demonstrated superior clinical correctness and safety in the qualitative evaluation. General multilingual LLaMA remained limited despite substantial gains, supporting the development of disease-specific Korean medical LLMs.
OBJECTIVES:Standardizing medication concepts across heterogeneous vocabularies is essential for interoperable analytics and observational research. In the Observational Medical Outcomes Partnership (OMOP) Common Data Model, local drug codes must be mapped to standardized RxNorm concepts, but automated mapping is challenging because drug strings encode clinically critical attributes, including strength, dosage form/route, release characteristics, and brand. METHODS:We propose THIRAWAT (Terminology Harmonization using Late-Interaction Reranker With Alignment-tuned Transformers), a fine-tuned ColBERTv1 late-interaction reranker, and embed it within THIRAWAT Mapper, a retrieval-reranking pipeline with deterministic tie-breaking and stable ordering. Candidate generation used approximate nearest-neighbor retrieval with a bi-encoder (SapBERT-XLMR or BioLORD-2023). Candidates were reranked by THIRAWAT models that were fine-tuned using one-sided MaxSim and scored at inference using our adapted Bidirectional MaxSim (BiMaxSim) pooling. Finally, a deterministic tie-breaker extracted clinically salient cues, including strength, dosage form/route, release characteristics, and bracketed brand annotations, to resolve near-ties reproducibly. RESULTS:We evaluated three mapping settings: Branded Drugs, Clinical Drugs, and Thai Medicines Terminology (TMT). Using SapBERT-XLMR retrieval with THIRAWAT-Sap-BERT reranking and deterministic tie-breaking, THIRAWAT Mapper achieved MRR@100 values of 0.954 (95% confidence interval [CI], 0.921-0.983), 0.898 (95% CI, 0.866-0.925), and 0.912 (95% CI, 0.891-0.931), outperforming a lexical term frequency-inverse document frequency baseline (0.491, 0.216, and 0.143, respectively). Hits@1 improved to 0.942 (95% CI, 0.899-0.978), 0.859 (95% CI, 0.817-0.898), and 0.868 (95% CI, 0.838-0.896), respectively. CONCLUSIONS:BiMaxSim and deterministic tie-breaking improved drug mapping to RxNorm while preserving an efficient runtime profile and stable ordering. Overall, THIRAWAT Mapper offers a pragmatic combination of learned semantic matching and deterministic lexical constraints. Models and code are available on Hugging Face (https://huggingface.co/collections/sidataplus/thirawat) and GitHub (https://github.com/sidataplus/THIRAWAT-mapper).
OBJECTIVES:This study evaluated the feasibility and performance of a multi-agent (MA) system designed to support early sepsis management in intensive care units. The system integrates three specialized agents-sepsis management, antibiotic recommendation, and guideline compliance-to provide evidence-based recommendations at T = 0 hours (before culture results), extending prior single-case findings across 10 diverse cases. METHODS:The MA system was powered by Palmyra- Med 70B (selected for superior MedQA performance [average score, 85.9]) and compared with GPT-3.5 Turbo and GPT-4o mini (all at a temperature of 0.25). It used retrieval-augmented generation (RAG) with ChromaDB (2021 Surviving Sepsis Campaign, over 20 high-impact manuscripts [reviews published 2018-2025] on sepsis etiologies, and other relevant sources). Eight cases from the MIMIC-IV demo and two cases from the literature were formatted as vignettes. RAG used the BAAI/bge-base-en-v1.5 embedding model with cosine similarity (threshold, 0.75) and top-5 chunks. Performance was assessed via TruLens (groundedness, approximately 0.62) and by two intensivists using a standardized questionnaire. RESULTS:The system generated guideline-compliant recommendations (e.g., prompt surgical debridement plus meropenem and vancomycin for necrotizing fasciitis). Hallucinations occurred in three of 10 cases (e.g., "altered mental status"). Expert agreement was quantified by a Cohen kappa of 0.26. Programmatic and expert assessments showed negligible correlation. CONCLUSIONS:In this exploratory study, the MA system shows preliminary promise for early sepsis support but requires human oversight to mitigate hallucinations. Code is available in GitHub; further validation is needed.
Objectives Migrating legacy on-premise electronic health record (EHR) systems in tertiary hospitals to modern cloud-native platforms presents technical and strategic challenges. We aimed to establish an optimized roadmap for transitioning legacy monolithic systems to a microservice architecture-based cloud-native EHR (MCEHR). Methods We conducted a 3-month strategic assessment and case study based on the modernization requirements of a global healthcare provider. The methodology incorporated semi-structured interviews with key stakeholders, including clinical informatics officers and system architects, to identify critical pain points such as .NET 4.0 end-of-support risks and performance bottlenecks during peak clinical hours. Phased hybrid migration was adopted, analyzing over 20 TB of Oracle-based legacy data and evaluating the technical feasibility of transitioning to .NET 9 and RESTful APIs. To ensure clinical safety, a proof-of-concept (PoC) environment was developed to simulate high-concurrency clinical workloads, emphasizing system resilience and transaction integrity during intensive order-entry periods. Results The transition to .NET 9 and MCEHR demonstrated 100% transaction integrity across 1,012 complex clinical test cases. Frontend and backend modernization showed high feasibility; however, migration of business logic embedded within legacy Oracle views represented a primary technical bottleneck, necessitating targeted decoupling. The PoC confirmed that RESTful API-based services maintained stable throughput under heavy concurrent loads, significantly reducing the risk of system-induced delays in clinical workflows. Conclusions Transitioning to an MCEHR architecture is complex but strategically essential. The proposed task force team roadmap outlines staged upgrades incorporating core technology modernization (.NET 9 and RESTful APIs), selective business component migration, and parallel DevOps adoption.
Objectives Large language models (LLMs) demonstrate remarkable potential in healthcare communication. However, whether they can process complex, high-volume medical information, such as stroke-related content, remains insufficiently validated. This study aimed to evaluate the natural language processing capabilities of LLMs in handling such content and to develop an evaluation instrument. Methods A survey compared educational materials generated by two LLMs (ChatGPT 4.0 and Claude 3) with neurologist-authored content on stroke. The materials were based on two clinical scenarios representing distinct stroke etiologies: cardioembolism and large-artery atherosclerosis. They were evaluated in terms of accuracy, legality, ethics, comprehensiveness, and information delivery. Scores for comprehensiveness and information delivery were compared according to participants’ agreement with the use of LLMs in healthcare. Results ChatGPT received the highest scores across all domains, except for legality in Scenario 2. In Scenario 1, the ranking for accuracy and summarization of clinical information was, from highest to lowest, ChatGPT, Claude, and the neurologist (η2 = 0.140, p < 0.001; η2 = 0.175, p < 0.001). The same hierarchy was observed in Scenario 2 for accuracy (η2 = 0.077, p < 0.001) and summarization (η2 = 0.194, p < 0.001). Participants who agreed with the use of LLMs in healthcare assigned higher scores for the comprehensiveness (Scenario 1, p = 0.005; Scenario 2, p = 0.007) and information delivery (Scenario 1, p = 0.003; Scenario 2, p = 0.026) of ChatGPT-generated materials than participants who did not agree. Conclusions LLMs demonstrated adequate capability to convey complex content, such as stroke-related information, in an accessible and understandable manner for non-experts.
