Journal of Medical Internet Research

Correction: Association Between Wearable Device Adoption and Health-Related Lifestyle Behaviors: Retrospective Cohort Study

2026-10-28T14:00:29-04:00

Wearable devices are increasingly adopted for personal health monitoring, but evidence on their long-term associations with health-related lifestyle behaviors in real-world population settings remains limited. This study examined the longitudinal association between wearable device adoption and engagement in healthrelated lifestyle behaviors using a nationally representative panel dataset from South Korea. We analyzed data from the 2016 and 2022 waves of the Korea Media Panel survey. Health-related lifestyle behaviors in the physical, social, and cultural domains were operationalized as estimated annual activity counts based on self-reported frequency measures. We used a difference-in-differences framework with generalized estimating equations to compare changes in these behaviors between new wearable adopters and nonadopters adjusting for demographic and socioeconomic characteristics. Relative changes were estimated using Poisson models with a log link, and subgroup analyses were conducted to explore variation across sociodemographic groups. As a sensitivity analysis, inverse probability of treatment weighting was additionally applied to assess the robustness of the findings to observed baseline imbalance. Wearable device adoption was associated with greater increases in total, physical, and cultural health-related lifestyle activities over time. In the difference-in-differences model, adopters showed greater relative increases in total activity (rate ratio [RR] 1.24, 95% CI 1.08-1.35), physical activity (RR 1.36, 95% CI 1.12-1.64), and cultural activity (RR 1.78, 95% CI 1.31-2.42) than nonadopters. Subgroup analyses showed limited evidence of consistent heterogeneity and should be interpreted cautiously. Sensitivity analyses using inverse probability of treatment weighting showed overall patterns broadly similar to those of the primary analyses. In this nationally representative panel study, wearable device adoption was associated with greater increases in total, physical, and cultural health-related lifestyle activities over time, whereas no clear association was observed for social activity. These findings should be interpreted as associative rather than causal given the observational design and the inability to directly assess parallel trends.

Neck-Specific Exercises With Internet Support Versus Standard Neck-Specific Exercise for the Treatment of Chronic Whiplash-Associated Disorders: Cost-Effectiveness Analysis

2026-10-28T13:30:03-04:00

Background: There is currently scarce information about the cost-effectiveness of interventions for individuals with chronic whiplash-associated disorders (WAD) or of internet-based exercise programs for neck pain, including WAD. Objective: This study aimed to investigate the cost-effectiveness of a 3-month neck-specific exercise program with internet support in combination with 4 physiotherapy visits (neck-specific exercise with internet support [NSEIT]) compared with the same exercises performed twice a week (24 times) at a physiotherapy clinic (neck-specific exercise [NSE]) for individuals with chronic WAD. Methods: This was a prospective, multicenter, randomized controlled trial of the cost-effectiveness of the interventions among 140 individuals with chronic moderate to severe WAD with 3- and 15-month follow-up. The main outcome was the incremental cost-effectiveness ratio (ICER). Results: At the 15-month follow-up, the total costs calculated for NSEIT and NSE were SEK 12,882 (95% CI 17,406.30-8358.42) and SEK 40,087 (95% CI 51,764.32-28,410.96), respectively (SEK 1=US $0.10 as of October 6, 2026). The health-related quality of life (HRQoL) score at 15 months was 0.662 (95% CI 0.73-0.59) and 0.711 (95% CI 0.77-0.65) for NSEIT and NSE, respectively. The ICER for HRQoL was SEK 555,209.80 per point when the NSEIT group was compared with NSE. The internet-based intervention was less costly and less effective than NSE at the 15-month follow-up. The cost-effectiveness acceptability curve showed that at a willingness to pay threshold of SEK 500,000 per point, there was a moderate probability (0.57) of the intervention being cost-effective with a loss of HRQoL. One-way sensitivity analysis showed that the health care costs of individuals receiving NSE and NSEIT affected the ICER. Conclusions: NSEIT intervention—an eHealth intervention for chronic WAD—was less costly and less effective compared with NSE. Health care costs associated with both interventions affected the ICER; therefore, patients’ overall well-being should also be considered when treating chronic WAD. Even though the cost effectiveness acceptability curve showed that the cost savings from NSEIT outweighed the lost health outcomes, the reliability of the findings was limited by the exclusion of presenteeism costs, patients’ self-reported health care use, and the small difference in health outcomes. Trial Registration: ClinicalTrials.gov NCT03022812; https://clinicaltrials.gov/study/NCT03022812

AI in Long-Term Care: A Closed-Loop Sociotechnical Reliability Framework

2026-10-28T07:00:03-04:00

AI is increasingly used in long-term care (LTC), yet reliability is often judged at the level of individual models or devices. This is insufficient in LTC, where residents vary in capacity, autonomy, and preferences. Their ability to question AI outputs may fluctuate, and care delivery depends on distributed roles and finite operational capacity. We define sociotechnical reliability as the capacity of an AI-enabled care system to consistently translate resident-relevant information into appropriate and timely care actions under real-world LTC conditions, with failures that are detectable and correctable across the care pathway. In this Viewpoint, we draw on professional experience overseeing AI initiatives across LTC and hospital settings, examined in relation to a purposively mapped interdisciplinary literature, to inform the development of a reliability framework. Our aim is to reframe AI reliability in LTC as an end-to-end care system property and to provide a framework for identifying where reliability is gained, lost, and evaluated. We propose 5 reliability functions: sensing, inference, verification and contextualization, response and escalation, and learning and adaptation, with governance and accountability operating across all 5 functions. The framework highlights recurrent failures, including incomplete sensing, nonportable inference, overreliance on or dismissal of AI outputs during verification, unclear ownership or delayed response, and repeated failures without correction. The framework also makes reliability empirically examinable through questions about where the pathway fails, who has the authority to verify and act, whether response demands are sustainable within available capacity, and whether failures become visible to governance and inform learning and adaptation. Our central viewpoint is that AI in LTC should be judged not by model performance alone but by whether the surrounding care system can reliably translate AI-generated information into care.

Real-Time Influenza Surveillance Using DeepSeek-Augmented BERT Classification on Chinese Social Media: Infodemiology Study

2026-10-27T15:30:22-04:00

Background: Traditional influenza surveillance systems rely on sentinel hospitals and often experience a 1‐ to 2-week reporting lag, limiting timely public health response. Social media platforms provide real-time data streams, but extracting reliable epidemiological signals remains challenging due to substantial linguistic noise. Although large language models (LLMs) improve semantic extraction, their computational cost limits routine deployment in resource-constrained public health settings. Objective: This study proposes and evaluates a hybrid framework that distills knowledge from an LLM (DeepSeek) into a lightweight bidirectional encoder representations from transformers (BERT)–based classifier for real-time influenza surveillance using Weibo data from Guangdong Province, China, during the 2025 influenza season. The study further assesses the temporal alignment between the derived signal and official influenza-like illness (ILI) surveillance data. Methods: A total of 389,959 Weibo posts from epidemiological weeks 7 to 50 (February 10 to December 14, 2025) were collected using a domain-specific keyword strategy. After multistage filtering, 99,158 posts were retained for analysis. A DeepSeek-Chat model was used to generate pseudo-labels for 2500 posts via zero-shot prompting, which were evaluated against expert annotations (Cohen κ=0.846) and used to fine-tune a Chinese-RoBERTa-WWM-ext classifier (a robustly optimized BERT pretraining approach). A chronological train-validation-test split was applied to reduce temporal leakage. The trained model was used to construct a weekly Weibo Flu Index. This index was compared with keyword-based and Term Frequency–Inverse Document Frequency (TF-IDF) + logistic regression baselines using time-lag Pearson correlation (0‐5 wk) against official ILI% data. Savitzky-Golay smoothing and second-derivative analysis were applied to identify epidemic inflection points and peak timing. Results: The classifier achieved an accuracy of 85.1% and an -score of 0.800 on a chronologically held-out test set, identifying 29,779 posts (30.0%) as ILI self-reports among 99,158 posts. The BERT-derived index showed the strongest correlation with official ILI% at lag 0 (=0.523, 95% CI 0.268‐0.710; ²=0.274) and lag 1 (=0.446, 95% CI 0.169‐0.658), compared with the keyword baseline (lag 0: =0.481, ²=0.231) and TF-IDF baseline (lag 0: =0.185, =0.034). Inflection analysis indicated that the Weibo-derived signal preceded both the epidemic rise (week 44 vs week 45) and peak (week 48 vs week 49) in official ILI data, corresponding to an approximate 1-week lead at both transition points. Conclusions: The results suggest that a DeepSeek-augmented lightweight BERT classifier can extract meaningful influenza-related signals from noisy Chinese social media data and produce a surveillance index that shows an approximate 1-week lead relative to official ILI reporting, with improved performance over keyword-based and TF-IDF approaches. This framework provides a proof-of-concept for cost-effective real-time influenza nowcasting using social media data.

Evaluating the Performance of Five Leading Large Language Models in Generating Health Education Materials for Patients With Infertility: Comparative Analysis

2026-10-27T12:00:30-04:00

Background: Patients with infertility often have inadequate reproductive health literacy (RHL), while existing health education materials may be difficult to read, overly complex, and insufficiently tailored to individual needs. In recent years, large language models (LLMs) have created new opportunities to improve health education through their ability to generate and interpret medical information. However, their applicability and content quality in infertility education remain insufficiently evaluated. Objective: This study systematically evaluated the readability and content quality of LLM-generated health education materials for patients with infertility, identified model strengths and limitations, and provided evidence to support their standardized use in reproductive health. Methods: This study selected DeepSeek-R1, ChatGPT (GPT-5.2 Plus), Kimi 2.5, Qwen 3, and Gemini 3 as the study subjects. Based on the theoretical framework of RHL, this study developed a standardized question list comprising 6 dimensions and 93 items, and used unified prompts to generate health education texts in both Chinese and English. In this study, the Datayze Readability Analyzer was used to assess readability via the Flesch Reading Ease, Gunning Fog Index, and Flesch-Kincaid Grade Level (FKGL). Meanwhile, the content quality was evaluated in a double-blinded manner by reproductive medicine experts across 3 dimensions: accuracy, comprehensiveness, and humanistic care. Repeated-measures analysis of variance was used to compare performance across the LLMs, and Pearson correlation analysis was conducted to examine the associations between readability and content quality. A radar chart was used to evaluate the overall performance of each model. Results: The 5 LLMs differed significantly in both text readability and content quality. In terms of readability, Gemini 3 and Qwen 3 performed best, with significantly higher FRE scores than the other 3 models. DeepSeek-R1 and Kimi 2.5 performed least favorably, with FKGL scores corresponding to a ninth-grade reading level. ChatGPT (GPT-5.2 Plus) showed intermediate performance. In terms of content quality, Kimi 2.5 performed best across accuracy, comprehensiveness, and humanistic care, followed by DeepSeek-R1. ChatGPT (GPT-5.2 Plus) performed least favorably. Analysis revealed significant negative correlations between readability and content comprehensiveness for texts generated by Kimi 2.5, indicating a trade-off between these 2 attributes. The composite radar chart analysis showed that Kimi 2.5 achieved the highest score (11,594.33), whereas ChatGPT (GPT-5.2 Plus) had the lowest score (5555.83). Conclusions: Different LLMs showed significant differences in performance in generating health education texts for patients with infertility. In some models, readability was negatively correlated with content comprehensiveness. Under the conditions of this study, Kimi 2.5 demonstrated strong overall performance. However, no model achieved the best performance across all evaluation dimensions. Future applications of LLMs in reproductive health education will require domain-specific optimization, context-specific adaptation, and systematic quality control to support more accurate, comprehensive, and patient-centered communication.

Development and Pilot Evaluation of a Standardized and Accessible Rubric (OMNI) for Assessing Generative AI in Health Care Contexts: Four-Stage Mixed Methods Study

2026-10-27T10:30:24-04:00

Background: The rapid adoption of generative AI in the community and health care necessitates reliable, human-led evaluation. Objective: This study aimed to develop a standardized rubric to support health care professionals, administrators, and researchers in evaluating generative AI outputs in health contexts. Methods: A rubric was developed through a 4-stage mixed methods study: literature synthesis, expert refinement, pilot application, and post-pilot feedback. The rubric was informed by 69 publications and refined through iterative feedback from a multidisciplinary expert panel (n=23). The pilot rubric was applied to 4 health care scenarios: responses to health-related Reddit questions, generation of health-related images and videos, responses to health professional continuing education questions, and automated physician-to-physician clinical documentation generation. Via these scenarios, 15 clinicians (medical practitioners, pharmacists, and medical oncologists) evaluated 250 input-output pairs using domain-level ratings and qualitative comments. The evaluated outputs included 200 AI-generated responses across the 4 health care scenarios and 50 physician-tagged replies in the health-related Reddit questions scenario. Results: The final rubric, OMNI, comprised 10 evaluation domains (Accuracy, Safety, Recognizes Uncertainty, Completeness, Communicates Clearly, Empathetic, Context Aware, Referencing, Follows Instructions, and Avoids Coerciveness) assessed on 4-point Likert scales, supported by detailed instructional and example materials. Via the 4 pilot scenarios, assessors generated 9600 domain-level ratings and 4010 qualitative comments. The mean pairwise agreement for the 4-point Likert groups ranged from 30.8% to 53.4%, increasing to 59.3% to 79.6% when collapsed into broad agreement categories; Gwet’s AC2 ranged from 0.49 to 0.80. OMNI differentiated strengths and weaknesses between models within scenarios, while recurrent limitations in Referencing, Recognizes Uncertainty, Completeness, Empathetic, and Accuracy were identified across scenarios. Multimodal image and video outputs had the highest occurrence of negative rating domains. Assessors reported OMNI was usable and indicated they would reuse it. Conclusions: OMNI is a standardized, accessible rubric for expert-led evaluation of generative AI outputs in health contexts. Pilot application demonstrated that OMNI could differentiate AI performance across models and domains while generating structured feedback. Further work is needed to assess its applicability to purpose-built health AI tools, agentic systems, and organizational evaluation settings.

Implementing Digital Care Pathways and Changes in Contact Volume, Contact Distribution, and Outpatient Costs in Specialized Health Care: Retrospective Longitudinal Observational Study

2026-10-27T08:00:02-04:00

Background: Health care systems increasingly promote patient engagement and self-management through digital health services. Digital care pathways, which support coordinated care delivery across the patient journey, have emerged as a promising approach, although evidence regarding their impact on resource use and costs remains limited. Objective: The aim of this study was to examine the implementation and impact of digital care pathways specifically designed to reduce or replace physical contacts or phone consultations in specialized health care. Methods: We analyzed the 10 most widely used digital care pathways by user volume. The selection criteria were (1) more than 1500 users per digital care pathway and (2) the aim of replacing or reducing in-person contacts. Descriptive analyses quantified digital contact volumes, average contacts and costs per patient, and inflation-adjusted expenditures using the consumer price index, both overall and by patient group. Ecologic group-level correlations were examined to assess whether the 2024 proportion of digital contacts was associated with changes in physical contacts and total costs from 2018 to 2024. Results: Digital care pathways were progressively adopted between 2018 and 2024, increasing from no use to accounting for 19.1% (34,837/182,188) of all outpatient contacts among the studied patient groups. Physical visits declined from 74.8% (99,201/132,622) to 54.1% (98,642/182,188), while contacts per patient rose by 7.6% on average. Conditions with the highest uptake of digital pathways showed the strongest reductions in the share of physical visits per patient, with declines of up to 39.7% (0.73/0.44). Average outpatient costs per patient decreased by 18.2% (€421/€515; €1=US $1.12 as of Oct 16, 2026) in real terms, with a 39.3% (€233/€384) reduction in the costs of physical contacts and a 15.4% (€135/€117) increase in the costs of phone visits. Digital pathway contacts accounted for 12.6% (€53/€421 per patient) of total outpatient costs in 2024; however, only 8.8% of the variance in total cost levels was explained by digital pathway adoption, suggesting that the shift toward digital care pathways did not substantially drive overall cost changes. Conclusions: Digital care pathways expanded rapidly, reaching nearly one-fifth of all outpatient contacts by 2024 in the included patient groups and contributing to a notable reduction in physical visits. However, their overall cost impact remains uncertain, as there have also been other factors impacting the development of outpatient costs. Trial Registration:

At Your Fingertips and on Your Mind? Digital Health Engagement and Psychological Distress

2026-10-26T15:30:25-04:00

Effects of Temporal Information Structure on Causal Explanations Generated by Large Language Models: Experimental Study Using Lung Cancer Incidence Data

2026-10-26T14:00:20-04:00

Background: Large language models (LLMs) are increasingly applied to causal explanation in public health and epidemiology. However, recent studies suggest that LLMs exhibit instability in processing temporal information, which raises concerns about their suitability for generating coherent causal explanations. Despite this, little is known about how distinct components of temporal information influence LLM-generated causal explanations in epidemiological contexts. Objective: This study aimed to examine how event ordering and the validity and presence of explicit temporal information affect LLM-generated causal explanations of long-term lung cancer incidence trends. Specifically, we evaluated their effects on evidence grounding, causal alignment, and disease specificity. Methods: We used a 2×3 factorial design manipulating event ordering (chronological vs randomized) and explicit temporal information (accurate, inaccurate, or absent). Five LLMs (GPT-5, Gemini 3.5 Flash, Claude Opus 4.8, Qwen3-32B, and Llama 3.3 70B Instruct) generated 600 causal explanations of long-term lung cancer incidence trends using Surveillance, Epidemiology, and End Results (SEER) data and PubMed-derived event statements. The effects of ordering and information were assessed using regression and factorial ANOVA. Results: Randomized ordering showed higher evidence grounding than chronological ordering (adjusted odds ratio [aOR] 2.53, 95% CI 1.89-3.37; P<.001). Causal alignment was higher under chronological than randomized ordering across all 3 information conditions, with larger differences in the decreasing phase. Disease specificity was higher under randomized ordering only under accurate information. Conclusions: More accurate or structured temporal information did not necessarily improve LLM-generated causal explanations. Randomized ordering was associated with greater evidence utilization, whereas chronological ordering supported causal alignment across all information conditions, particularly for decreasing trajectories. These findings suggest that temporal context should be structured not only to provide temporal information but also to support the intended characteristics of the resulting explanation.

Generative Large Language Models for Psychiatric Concept Detection in Spanish-Language Electronic Health Records: Development and Evaluation Study

2026-10-26T13:15:02-04:00

Background: The accurate detection of clinical symptoms in electronic health records (EHRs) is pivotal for advancing large-scale genetic and longitudinal studies in psychiatry. Free-text clinical notes are an essential source of symptom-level information, particularly in psychiatry; however, the automated extraction of symptoms from clinical text remains challenging. Objective: We aimed to evaluate the ability of open-source generative large language models (LLMs) to detect psychiatric symptom-level concepts from clinical text in Spanish and to develop a new fine-tuned, open-source LLM that can be shared without risk of exposing patient information. Methods: We tested 11 open-source generative LLMs for their ability to detect 109 psychiatric concepts from clinical text, using annotated EHR notes from a psychiatric hospital in Colombia. The LLMs were evaluated both “out-of-the-box” and after fine-tuning and compared against a traditional natural language processing (tNLP) method developed from the same data. To develop a fine-tuned LLM that can be shared without risk of exposing patient information, we created a novel synthetic dataset, leveraging real concept annotations while generating fully synthetic clinical text, and fine-tuned a top-performing LLM on these data. Fine-tuned model generalizability was assessed through cross-site evaluation using data from a second independent hospital in Colombia. Results: We show that while base LLM performance was poor to moderate (zero-shot: 0.2-0.6 macro-F₁; few-shot: 0.2-0.74 macro-F₁), it improved significantly after fine-tuning (0.75-0.86 macro-F₁), with several fine-tuned LLMs outperforming the tNLP method. In total, 100 psychiatric concepts could be reliably detected (F₁>0.8) using either a fine-tuned LLM or tNLP. Using the synthetic dataset, we developed Mistral-small-psych, a new LLM capable of detecting psychiatric concepts from Spanish clinical text with performance comparable to models trained on real EHR data (macro-F₁=0.79) while remaining fully distributable. In cross-site evaluations, most LLMs generalized well, with small out-of-domain performance losses (macro-F₁ decrease of 0.02-0.16). Conclusions: This work provides the first large-scale, cross-institutional evaluation of state-of-the-art LLMs for psychiatric concept detection in Spanish EHRs, introduces a privacy-preserving pathway for building shareable clinical natural language processing models, and presents a new open model for extracting symptom-level psychiatric information from Spanish clinical text.