Background:Large language models (LLMs) are rapidly emerging in health care, offering opportunities in decision support, education, and research, but raising critical concerns about safety, reliability, and ethics. Although several guidelines for trustworthy AI exist in business and technology, few systematic reviews have applied them to medical contexts. Objective:This study aimed to conduct a systematic review of LLM research in health care, applying the AI Guidelines for Business as a framework across 11 domains, including safety, reliability, ethics, transparency, fairness, inclusiveness, privacy, security, robustness, data quality, and verifiability. Methods:Following PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) 2020 guidelines (retrospectively registered on the Open Science Framework; DOI 10.17605/OSF.IO/P4KSB), the PubMed, Scopus, Web of Science, arXiv, and IEEE Xplore databases were searched on January 15, 2025. Records were screened in 2 stages by 3 reviewers (with records retained only upon unanimous agreement). A total of 247 studies were included, of which 211 (85.4%) contributed quantitative values. Eligible studies were classified across 11 trustworthy AI domains. Heterogeneous metrics were summarized within metric families; when multiple models were evaluated, the mean across models was used as the primary estimate, with best, median, and primary-model sensitivity analyses. The LLM-assisted categorization (GPT-5 mini) was validated by using an automated internal consistency check, and 95% CIs were estimated by using cluster bootstrap on study-level values. Results:Of the 25,156 records, 247 (1.0%) studies were included, and of these, 211 (85.4%) contributed quantitative values. Evaluation concentrated on accuracy (143/247, 57.9%) and fairness and inclusiveness (93/247, 37.7%), followed by data quality (47/247, 19.0%) and prevention of misinformation (44/247, 17.8%). Normalized performance was moderate to high (accuracy mean 0.73, 95% CI 0.7-0.76; data quality: 0.64; prevention of misinformation: 0.82). Selecting the best-performing model inflated domain means by up to 0.05. Privacy protection (2/247, 0.8%) and security assurance (0/247, 0.0%) were almost entirely absent. Domain assignments were recoverable from objective metric types in 98.9% of values (Cohen κ=0.985). Conclusions:Current evaluations emphasize accuracy while underreporting privacy, security, robustness, explainability, and verifiability. This finding reflects gaps in reporting rather than demonstrated poor performance, underscoring the need for comprehensive, guideline-based, multidomain evaluation before deployment in high-stakes clinical settings.
Background:Financial toxicity can contribute to adverse health and care-access outcomes among US veterans, yet scalable methods to identify individuals at elevated risk remain limited. Public health informatics frameworks may enable the translation of patient-reported financial risk signals into streamlined screening, risk stratification, and care-navigation workflows. Objective:This study aimed to examine concept-level indicators of financial literacy and financial toxicity among US veterans and explore how these findings could inform future informatics-enabled screening strategies for identifying subgroups at increased risk of health-related financial strain. Methods:We conducted an exploratory cross-sectional survey of 88 US veterans from 2024 to 2025. Financial literacy was assessed using 3 benchmark items from the National Financial Capability Study. Financial toxicity was assessed using items aligned with domains reflected in the Comprehensive Score for Financial Toxicity framework, including difficulty affording care, reduced or quit work, borrowing money or using savings for care, and treatment-adherence impact. Analyses included descriptive statistics, Fisher exact tests, unadjusted logistic regression, and a minimally adjusted sensitivity model for work disruption, controlling for age and education. Results:Female veterans had lower rates of high financial literacy than male veterans (15/29, 52% vs 48/59, 81%; P=.006) and lower correct-response rates on compound interest (10/29, 35% vs 36/59, 61%; P=.02) and inflation (14/29, 48% vs 43/59, 73%; P=.03). Black veterans had lower correct-response rates than non-Black veterans on inflation (11/24, 46% vs 46/64, 72%; P=.03) and retirement strategy (15/24, 63% vs 56/64, 88%; P=.014), although composite high-literacy rates did not differ significantly by race. In unadjusted models among participants with complete outcome data (n=75), lower financial literacy was directionally associated with higher odds of all 4 financial toxicity outcomes, with the clearest association observed for work disruption (odds ratio 0.56 per 1-point increase in financial literacy score, 95% CI 0.33-0.95; P=.03). Black female veterans reported elevated financial toxicity across multiple domains. Financial support program use was low overall (29%). Conclusions:These findings suggest that financial literacy may be a marker of vulnerability to financial toxicity among veterans, but observed associations should be regarded as preliminary and hypothesis-generating. The results identify concept-level financial literacy domains and work disruption as candidate signals for future screening evaluation. Future research should evaluate whether brief screening, financial literacy assessment, and benefit-navigation strategies improve identification, referral, adherence, and downstream financial and health-related outcomes in larger and more representative veteran populations.
Background:The transition of social health care tools from paper-based to web-based formats is increasingly common in social and health care settings. While digital delivery may improve accessibility, flexibility, and scalability, such transitions require careful adaptation of content, design, and workflows. However, evidence on how these transitions are carried out in practice, and which challenges and strategies are involved, remains fragmented. Objective:This scoping review aimed to map how social health care tools have transitioned from paper-based to web-based formats and to synthesize evidence on tool characteristics and target groups, adaptations made during the transition, strategies and methodologies used, and reported challenges and solutions. Methods:A scoping review was conducted following the Arksey and O'Malley framework. Systematic searches were performed in PubMed, PsycINFO, Scopus, and Web of Science to identify English-language empirical studies describing the transition of social health care tools from paper-based to web-based formats. Social health care tools were defined as technologies supporting communication, reflection, or informed decision-making among patients, family caregivers, and/or health care professionals. Two reviewers independently screened studies and extracted data using a standardized form. Targeted gray literature searches were conducted to supplement missing information about the included tools. Findings were synthesized narratively and presented in tables. Results:Eleven tools were included. Most were delivered as web-based platforms (n=8), with additional tools delivered as mobile apps (n=2) or in a game-based format (n=1). The tools targeted diverse populations, including people with chronic or life-limiting conditions, family caregivers, adolescents, parents, pregnant women, and health care professionals. Reported reasons for transitioning to digital formats included improving accessibility and reach (n=7), enhancing user engagement through interactive features (n=5), and increasing efficiency or scalability (n=3). Across tools, core content was largely retained, while adaptations mainly focused on format and delivery, such as multimedia elements, interactive modules, tailored feedback, and usability improvements. User-centered and participatory design approaches were commonly reported, sometimes combined with iterative or agile development methods. Common challenges included usability issues, low engagement, technical integration difficulties, administrative delays, and resource constraints, with solutions focusing on iterative redesign, clearer instructions, increased interactivity, and hybrid paper-digital approaches. Conclusions:This scoping review examines the process of transitioning social health care tools from paper-based to web-based formats, rather than focusing solely on effectiveness outcomes. It synthesizes how adaptations are made, which strategies are used, and which challenges arise during digitalization. The findings identify common adaptation patterns and gaps in reporting, particularly regarding target group-specific design and equity considerations. These insights may support developers, researchers, and policymakers in planning more usable, inclusive, and context-sensitive digital transitions in social and health care settings.
BackgroundSmartphone-based imaging has shown promise in teledentistry; yet, patient comfort during its use compared with intraoral cameras remains underexplored. ObjectiveThis study aimed to compare self-perceived comfort during oral evaluations using smartphones and intraoral cameras in a teledentistry setting. MethodsA convenience sample of 192 participants aged 6 years and older completed a survey and underwent 2 procedures: trained researchers captured 3 photographs of the anterior teeth with a Samsung smartphone, and the same teeth were examined with a MouthWatch intraoral camera. Comfort was rated by participants before and after each procedure using a pictorial 7-point Likert scale, and data were analyzed with nonparametric tests. ResultsComfort decreased in 15% (29/192) of the participants following intraoral camera examination (P=.001), while no significant change was observed after smartphone imaging (P=.52). The difference in comfort change between smartphone imaging and intraoral camera examination was marginal (P=.05). Among the small subgroup of youths (aged 6-19 years; n=13, 6.8%), 38% (5/13) reported a reduction in comfort after smartphone imaging, compared with 10% (18/179) among adults (P=.02). In addition, significantly more participants without tooth wear reported a reduction in comfort after smartphone imaging (5/12, 42%), compared with 10% (18/180) among those with tooth wear (P=.02). ConclusionsOverall, the findings indicate generally comparable comfort levels between the 2 approaches, with a marginal difference favoring smartphone imaging. While the potential for broader use of smartphone-based imaging in a teledentistry setting still requires further research, targeted education and aligned institutional and insurance policies may enable its scalable, sustainable integration that enhances patient experience.
Background:Phenylketonuria (PKU) is the most common inborn error of amino acid metabolism, and if untreated, leads to severe neurocognitive impairment. Over the past 2 decades, treatment strategies have evolved from strict dietary phenylalanine restriction to include pharmacological therapies such as tetrahydrobiopterin and, more recently, enzyme substitution with pegvaliase. Despite these advances, significant heterogeneity exists in global research priorities, collaboration patterns, and the translation of emerging therapies into clinical practice. A systematic overview of the field's development, thematic shifts, and remaining knowledge gaps is currently lacking. Objective:This study aimed to provide a systematic bibliometric analysis of global treatment research on PKU from 2000 to 2025. The aim was to quantify publication trends, collaboration patterns, thematic evolution, and research gaps, thereby informing future scientific and clinical directions. Methods:A search of the Web of Science Core Collection was performed on September 13, 2025. The search initially identified 1877 records. After screening, 1462 English-language articles and reviews were included. Publication trends were analyzed using Microsoft Excel (version 16.101), while VOSviewer 1.6.20 and CiteSpace 6.4R1 were used to visualize country- and institutional-level collaborations, journal networks, keyword co-occurrence, citation bursts, and thematic clusters. Statistical charts were generated with GraphPad Prism 10.2.1. Results:Annual publication output demonstrated an overall upward trend, peaking in 2022 with 111 publications. The United States led in both publication volume (375 studies) and total citations (11,142 citations), maintaining strong collaborative ties with several European countries, particularly the Netherlands and the United Kingdom. China ranked seventh globally in publication volume, although its citation impact remains comparatively limited. Key institutions, including the University of Groningen and Birmingham Children's Hospital, as well as prominent scholars such as Francjan J van Spronsen and Anita MacDonald, have occupied central positions in the global PKU treatment research collaboration network over the study period. High-frequency and high-centrality keywords, such as "phenylalanine," "dietary treatment," and "tetrahydrobiopterin," highlighted continued emphasis on metabolic control and targeted therapies. Keyword burst analysis revealed a gradual shift from conventional dietary management toward enzyme replacement therapies, investigations of neurocognitive outcomes, and precision medicine-oriented approaches. Conclusions:Over the past 25 years, PKU treatment research has progressed from foundational dietary interventions to molecular mechanistic studies and individualized therapeutic strategies. Future research should prioritize longitudinal multiomics investigations, targeted metabolic correction technologies, gene-based therapeutic approaches, and enhanced international collaboration, particularly to strengthen diagnosis and management capacities in low- and middle-income regions. Such efforts will be critical to advancing global standards of PKU care.
Abstract Background AI has become an essential component of modern health care delivery in Epic (Epic Systems Corporation) electronic medical record (EMR) systems, supporting predictive analytics, diagnostic decision-making, and population health management. Despite these advancements, evidence reveals that AI algorithms can perpetuate or even amplify existing health inequities through biased training data and flawed model design. Such algorithmic bias poses ethical challenges for health care leadership, regulatory compliance, and executive communication, especially in ensuring patient equity, transparency, and public accountability. Objective This conceptual paper examines how algorithmic bias in Epic’s AI modules influences executive decision-making, organizational communication, and trust within health care systems. It integrates organizational communication theory and public health informatics research to propose a framework for ethical, transparent, and equitable communication in AI-integrated health care settings. Methods Drawing upon the ethical communication and algorithmic trust framework (ECATF), this paper synthesizes interdisciplinary literature on AI bias, data governance, and leadership communication. The framework explains how transparent executive communication creates stakeholder trust in the context of bias identification and regulatory oversight. Results (Conceptual Findings) This conceptual analysis suggests that algorithmic bias influences leadership communication, equity framing, and governance strategies in AI-integrated health care systems. Incorporating AI bias auditing alongside regulatory monitoring and public education initiatives may support fairness, accountability, and health literacy across communities. Conclusions As AI continues to shape health care leadership and policy, ongoing evaluation, ethical foresight, and regulatory vigilance are essential. Transparency, collaborative governance, and adaptive education will be necessary to ensure that AI supports equitable innovation rather than reinforcing unintended harm.
Background:Robust and reliable health information systems (HISs) are foundational to equitable health care delivery in resource-constrained settings. Yet, HISs often exhibit significant fragmentation and complexity, which stem from many factors, including inadequate infrastructure, limited and unevenly allocated financial resources, expertise gaps, and a lack of integrated systems. At the same time, advances in modern HISs and digital technologies, such as electronic medical records (EMRs), present opportunities for addressing these limitations and supporting evidence-based health systems if well implemented and sustained. However, limited attention has been paid to how modern and resilient HISs can be effectively sustained in fragile, resource-constrained settings. Objective:This empirical study seeks to identify pathways for strengthening the resilience, adaptability, and contextual-fit of EMRs in fragile, resource-constrained settings such as Haiti. Methods:Using a qualitative research methodology, the study used semistructured interviews with purposive sampling of key informants, including frontline doctors, nurses, and IT or data specialists. Interview participants were selected for their expertise and capacity to offer insightful and varied viewpoints on the topic. Transcripts were analyzed using an inductive approach to identify key emerging themes. Results:The findings of this investigation reveal that implementing resilient EMR in limited-resource settings, such as Haiti, requires a comprehensive strategy that accounts for ecosystemic challenges from interdependent systems (electricity, sociopolitical instability, internet connectivity), as well as the intrinsic complexities of legacy systems. Conclusions:Drawing on empirical evidence, the study identifies a set of protective strategies that, if effectively implemented, may enhance the resilience and adaptability of EMRs. These strategies include prioritizing integration and interoperability between systems, building redundancy to mitigate cascading failures derived from other interdependent systems, strengthening staffing to support system use, right-sizing the paper footprint (eg, improving handwriting-to-text scanning and digitization solutions), and embracing technological innovations.
Unlabelled:Rapid AI integration has introduced novel psychosocial stressors. Little is known about AI-specific clinical impacts in resource-limited settings. This study aimed to assess AI-associated distress prevalence and nature in a Nigerian primary care and psychiatry clinic. Retrospective audit of 28 consecutive patients with anxiety, depression, or stress (April-August 2025) at J-Shalom Hospital, a primary care and psychiatry clinic in Ibadan. Technology-related stressor questions adapted from the AIAS (Artificial Intelligence Anxiety Scale) were incorporated into routine clinical interviews; C-SSRS (Columbia-Suicide Severity Rating Scale) evaluated suicidality as part of standard clinical practice. A total of 67.9% (19/28; 95% CI 49.3%-82.1%) reported technology stress; 42.9% (12/28; 95% CI 26.5%-60.9%) identified AI-specific stressors. Chatbot distress was most common (n=7/12, 58.3%). Three patients (25.0%; 95% CI 8.9%-53.2%) reported worsening suicidal ideation following distressing chatbot interactions characterized by perceived rejection or invalidation. AI stressors are emerging clinical presentations. The chatbot-suicidality link demands urgent regulatory attention.
Unlabelled:Completion of the HEDIS (Healthcare Effectiveness Data and Information Set) Childhood Immunization Status Combination 10 (Combo 10) measure among US children aged 24-35 months declined from 53.7% in 2021 to 44.6% in 2023, with a statistically significant survey-weighted annual trend. An explainable machine learning approach identified influenza vaccination and rotavirus series completion as the strongest component-level drivers of Combo 10 completion, supporting targeted public health quality improvement.
Background:Digital health care technologies, including mobile applications and telemedicine platforms, have transformed how medical professionals communicate and deliver care. Remote consultation by doctors plays a vital role in ensuring access to appropriate expertise, particularly in medically underserved or geographically remote areas. However, the diversity in technological modalities, devices, and patterns of use across specialties and regions has not been systematically mapped. Objective:This study aimed to explore the current status and characteristics of teleconsultations among medical providers and specialists, focusing on device use, consultation modalities, clinical specialties, and regional differences. Through this approach, we aimed to provide a comprehensive overview of technological and practical trends in mobile health (mHealth) and telemedicine. Methods:A systematic scoping review was conducted in accordance with PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews) guidelines using the MEDLINE and Embase databases. The search covered studies published up to February 2026, with no restrictions on the publication year. Studies meeting the predefined inclusion criteria were also included. Results:A total of 255 citations were screened, and 99 articles were included. Studies were analyzed according to consultation method, target, and regional characteristics. Of these, 81 out of 99 (81.8%) articles were in the doctor-to-doctor category. Email, web, or app-based platforms were the most common. In addition, 86 out of 99 studies used medical images, most frequently photographs. Orthopedics and dermatology were the most frequently involved specialties, followed by internal medicine. Regarding the region, 60 of 99 studies were domestic, and the other 39 studies were international. Rural-to-urban domestic consultations comprised 27 out of 99 (27.3%) studies, whereas consultations from low- and middle-income countries and high-income countries accounted for 11 of 99 (11%) studies. Conclusions:This review examined doctor-to-doctor and doctor-to-patient consultations with doctor involvement. Specialties in which medical images are central, such as orthopedics and dermatology, were more frequently represented than in other fields. This highlights disparities in the use of teleconsultation across clinical disciplines and suggests that addressing these imbalances is essential for broader adoption. Furthermore, the findings indicated a progressive shift from videoconference-based interactions to mobile and app-based platforms, reflecting ongoing technological advancements. Optimizing the integration of these digital tools and promoting equitable access are critical for enhancing the quality and reach of teleconsultation practices in future digital health systems.
BackgroundLyme disease (LD) is the most common vector-borne disease in the United States. It is difficult to diagnose because it can mimic numerous other conditions, and testing protocols may not be sufficient. Although the Centers for Disease Control and Prevention (CDC) recommend a 2-tiered serologic testing approach for LD diagnosis, many patients are diagnosed clinically based on criteria such as the erythema migrans rash and, particularly when the rash is not present, various symptom patterns, exposure history, other information, and clinician observations. Against this backdrop, online symptom checkers, using artificial intelligence (AI) processing techniques, are increasingly used to obtain diagnostic information and resources. With LD as a use case, this research applied a modified capabilities approach to explore the relative effectiveness and utility of AI-based tools in application and comparison to serologically (CDC+) and clinically based diagnoses. ObjectiveThe overarching goal of this research was to provide a baseline exploration of AI-assisted diagnostic tools relative to more traditional medical assessment approaches in detecting complex infectious diseases. To assess the potential diagnostic utility (DU) of online symptom checker platforms with LD as a use case, this study aimed to (1) evaluate platform performance in identifying LD across different diagnostic cohorts; (2) compare symptom patterns and severity distributions among online LD diagnoses; and (3) identify the most frequently co-occurring conditions potentially misclassified as LD or vice versa. MethodsData were drawn from a limited structured survey of patients with confirmed or probable LD, including diagnostic pathways (CDC+ and/or clinical), symptom profiles and severity, treatment history, and time to diagnosis. These patient cases were then entered into 3 leading AI-based symptom checker platforms—MediFind, Isabel, and WebMD—to examine diagnostic performance. Descriptive analytics, logistic regressions, and postestimation analyses were used to identify patterns of DU and interaction effects among cohort type, symptom severity, and AI-based platforms. This survey-based research was not intended to serve as an experiment or clinical trial. ResultsDU varied significantly across platforms and symptom severity thresholds. DU improved at higher symptom severity thresholds (≥3; P<.001) and was slightly higher among clinically diagnosed cohorts compared to CDC+ cohorts (P=.21). Marginal analyses revealed that clinically diagnosed respondents were more sensitive to changes in severity, but with different levels of platform consistency across conditions. Note that DU was analyzed principally for exploratory purposes. ConclusionsThis research serves as a preliminary and directive step for expanded data collection and a larger, more comprehensive study. The findings suggest that AI-based symptom checkers may supplement early diagnostic reasoning in complex conditions such as LD, particularly when symptom severity is high. Inconsistency across platforms and diagnostic categories highlights the need for algorithmic refinement and standardized validation frameworks to enhance diagnostic reliability in AI-based tools.
BackgroundChina has initiated a national stroke program since 2011. Understanding the magnitude of stroke burden in China is crucial for modulating related policies. ObjectiveThis study aimed to analyze stroke status in China from the 2023 Global Burden of Disease study. MethodsThe 2023 Global Burden of Disease study was the source of the estimation of stroke burden. The data of deaths, incidence, prevalence, disability-adjusted life years (DALYs), years lived with disability (YLDs), and years of life lost attributable to stroke and its subtypes from 1990 to 2023 were analyzed. The mortality-to-incidence ratio was calculated to quantify the proportion of incident stroke cases that result in deaths. DALY decomposition analysis was applied to quantify how stroke burden has shifted between mortality and long-term disability over time. Sex-specific characteristics were also compared. ResultsFrom 1990 to 2023, absolute stroke incidence (1.87-4.23 million) and prevalence (11.56-26.68 million) increased substantially, with ischemic stroke as the primary driver. Age-standardized incidence declined from 246.15 to 192.19 per 100,000 populations, while mortality significantly reduced from 302.72 to 92.83 per 100,000 populations. DALYs and years of life lost decreased 9.3% and 11%, respectively, while YLDs increased more than double. Mortality-to-incidence ratio fell for all subtypes except subarachnoid hemorrhage, indicating improved survival. Male individuals bore a heavier burden in total stroke and ischemic stroke, while subarachnoid hemorrhage showed female predominance in prevalence and YLDs. DALY decomposition revealed that YLDs contribution rose from 5.8% to 14.5%. ConclusionsFrom 1990 to 2023, despite declining age-standardized rates, stroke burden in China grew due to ischemic stroke and population changes. The shifting burden toward disability and persistent sex disparities highlights the need for tailored prevention, rehabilitation, and gender-specific interventions.
Background:The evaluation of health technologies is fundamental to the sustainability of the Brazilian National Health System (SUS), especially in highly complex and costly procedures such as computed tomography (CT). Objective:This study analyzed the financial efficiency of CT scanner use in the SUS in the state of Rio Grande do Norte (Brazil), considering the distribution of supply, the profile of providers, and the idle capacity of the public network. Methods:A total of 49,061 CT scan requests registered in the state system "RegulaRN Ambulatorial" between October 2023 and January 2025 were examined. Subsequently, data cleaning and grouping were performed. Results:The results indicated that 66.8% (24,769/37,089) of requests were from patients with cancer, 61.2% (22,703/37,089) were women, predominantly from municipalities with higher population densities, such as Natal and Mossoró, with an average waiting time of 44.86 (SD 88.43) days. More than 36,000 (99%) requests for the examinations were performed by private and philanthropic hospitals contracted by SUS, while public units with CT scanners accounted for less than 1% of production, evidencing significant underutilization. Conclusions:The findings suggest that strategically shifting demand to the public health system can increase efficiency, reduce costs and waiting times, and promote greater equity in access, contributing empirical evidence for more rational public policies on the use of diagnostic technologies.
Background:Mobile health (mHealth) represents a modality of teledentistry that has the potential to improve access to dental care. Given that patient reactions to dental procedures can influence both clinician experience and care delivery, assessing patient discomfort when smartphones are used to capture dental images for teledentistry examinations is crucial. Objective:This study aimed to explore patient discomfort from the perspective of dental professionals using smartphone-based photography in teledentistry. Methods:A qualitative study was conducted through group interviews with a sample (N=10) of dental professionals, all of whom had experience capturing dental photos using smartphones equipped with an mHealth app at dental clinics and research facilities in Thailand and the United States. Audio-recorded interviews were transcribed, coded through consensus, and analyzed thematically. Results:The dental professionals, including dental specialists, general dentists, dental therapists, and dental students, reported minimal to no patient discomfort during smartphone-based dental photography. Key factors contributing to patient comfort during teledentistry encounters included clear communication, informed consent, and reassurances regarding privacy and data security. Conclusions:The findings suggest that providing patients with clear information and managing expectations can help reduce discomfort in teledentistry encounters. Improving communication strategies may enhance patient comfort, support the adoption of mHealth practices, and optimize interactions between patients and health care providers. Future research directions are indicated, such as directly assessing patient discomfort and identifying strategies to further minimize discomfort in teledentistry. Additionally, expanding teledentistry training in dental education and professional development will better equip dental professionals to effectively use this technology, ultimately improving accessibility and patient-centered care in dentistry.
Abstract BackgroundPublic opinion, which may be influenced by personal experiences, news, and social media, can impact compliance with public health measures (PHMs) during health emergencies. Artificial intelligence (AI) tools offer opportunities to analyze public opinion in real time during health emergencies. However, their performance in accurately identifying sentiment and themes in health-related online content remains unclear. ObjectiveThis study aimed to evaluate the performance of natural language processing–based and large language model (LLM)–based AI tools when compared to human coding for sentiment analysis, topic modeling, and thematic analysis of public health datasets. Tools were selected to reflect those available to public health analysts and decision-makers. MethodsData were collected via Google Alerts (GA) and social media posts from X (formerly known as Twitter) relevant to COVID-19 mitigation PHMs from December 2022 to February 2023. Following relevance screening, the sentiment of the complete datasets was analyzed by a human rater, with descriptive statistics used to summarize the overall sentiment profile. Subsets of 400 GA articles and 400 tweets were manually coded for sentiment by 2 human raters. Results were compared with outputs from 5 AI tools, including VADER (Valence Aware Dictionary and Sentiment Reasoner), SentimentGI, SentimentQDAP, Microsoft Azure, and OpenAI’s ChatGPT-4. Topic modeling of the GA and X datasets was conducted using latent Dirichlet allocation in R and zero-shot prompting in ChatGPT-4 and compared with manual topic summaries. Thematic analysis of positive and negative sentiment datasets was conducted by a human rater and ChatGPT-4, with outputs evaluated for proficiency and reasonableness. The sentiment of the entire datasets was analyzed by a human rater, and descriptive statistics were calculated. ResultsOf 2227 GA results and 3484 tweets, 58% (n=1238) and 71% (n=2473), respectively, were relevant to PHMs. Human-coded sentiment analysis showed mostly neutral reporting in the news media, while social media expressed more polarized views. Across both datasets, AI tools demonstrated poor concordance with human-coded sentiment (Cohen κ <0.5 for all tools and sentiment categories). Topic modeling with ChatGPT-4 aligned more closely with human-rated topics than latent Dirichlet allocation, and of the 20 LLM-generated thematic outputs, 13 were rated proficient, and 7 were rated partially proficient. LLM outputs provided coherent, high-level summaries but lacked contextual insight. Human and LLM thematic analyses both identified themes of vaccine effectiveness, debate regarding PHMs, and public trust. ConclusionsAccessible AI tools demonstrate limited reliability for sentiment classification of health-related online text but show promise for rapid thematic exploration when combined with human oversight. These tools could complement traditional qualitative research in the context of health emergencies; however, they require human review to enhance the accuracy of interpretation. Further research is needed for non-English datasets.
BackgroundPersonas, fictional profiles representing user segments, play an important role in human-centered design, ensuring tools are tailored to the needs of users. Although public health organizations often develop information systems to promote population health, human-centered design methods and personas are generally underused in public health informatics projects. ObjectiveThis study aims to present a novel, mixed methods approach to developing data-driven personas for use in public health information system design, leveraging 2 statewide surveys conducted in Washington State. The aim is to produce realistic, representative, and actionable personas that reflect the diversity of a state population and support user-centered design in public health initiatives. MethodsQuantitative (cluster analysis) and qualitative (thematic review and quote extraction) methods were applied to 2 statewide survey datasets: (1) a statewide knowledge, attitudes, and practices survey (N=1103) which used random, address-based sampling, and (2) a subset of the knowledge, attitudes, and practices respondents (N=143), which included more targeted questions on opinions and preferences related to public health information systems. Characteristics examined included demographics, technological readiness, opinions about public health policies, and experience using online health tools. ResultsK-prototype clustering resulted in 5 clusters. These 5 clusters were studied using both quantitative and qualitative analysis of key factors of the Washington State population to build 13 personas. Each persona represents a different population demographic, varying levels of technological readiness and attitudes toward public health policies, and differing experiences with online health tools. Persona descriptions are further elucidated with a short profile and 2-3 quotes. ConclusionsThis study offers a scalable and adaptable framework for persona development in public health, demonstrating how existing datasets can be transformed into effective design tools. Through a mixed methods approach, personas that reflect the diverse needs, preferences, and behaviors of Washington State residents were created. These personas can enhance the design, development, and evaluation of public health information systems by centering on user experience. Persona development and the methods described here can be used in future public health informatics projects to assist in formative research, guide design and development, inform usability testing, and shape communication strategies. By bridging the gap between large-scale data and user-centered design, this approach provides a practical model for making public health technologies more aligned with community needs.
BackgroundOral nicotine pouches (ONPs), such as Zyn, have gained popularity among young people; however, their portrayal on social media remains under-studied. Instagram memes, a widely shared form of digital communication, may shape young people’s perceptions about ONPs and contribute to the widespread acceptance of ONP use. ObjectiveThis study examines the thematic content of Instagram memes related to ONPs to understand how these products are represented online. MethodsThe content of Instagram memes tagged with ONP-related hashtags—#oralnicotinepouch, #zyn, #on, #velo, and #nicotinepouch—was systematically analyzed. After screening, a total of 244 photo- and text-based memes were included in the final dataset. Using a structured coding framework, 3 researchers categorized the memes into key themes using NVivo software. ResultsThree dominant themes emerged: (1) the Zyn community (35.6%)—memes fostered a sense of belonging among users; (2) marketing and branding (27.8%)—humorous critiques of product advertising and accessibility; and (3) perceived consequences of use (13.9%)—memes highlighted perceived positive or negative consequences of ONP use. Engagement metrics revealed high levels of interaction, with the Zyn community theme garnering the most user engagement. ConclusionsONP-related Instagram memes are primarily focused on community identity, humor, and marketing, with community-centered content receiving the highest engagement. These findings indicate that social belonging and humor are central to the online representation of ONPs.
Background:Older adults often access traditional media, such as newspapers, magazines, television, and radio, for health information. However, compared with older adults without frailty, older adults with frailty experience greater declines in physical functions and mental health (including depressive symptoms), as well as social functioning, due to reduced interaction with others, which limits their access to these sources of information. Objective:This study aimed to identify the health information sources that are less accessible to participants with frailty than to those without frailty. Methods:A cross-sectional web-based survey was conducted among independent Japanese adults aged ≥75 years. We assessed frailty using the Questionnaire of Medical Checkup for Old-Old, with a score of ≥4 indicating frailty. Participants were asked whether they had accessed any health information source in the past year, including medical institutions, family members, friends or acquaintances, neighbors, government agencies, long-term care or welfare services, television, radio, the internet, magazines, newspapers, or books. The primary explanatory variable was frailty status. Covariates included age, sex, income, education, living arrangements, and health literacy, measured using the eHealth Literacy Scale. Results:In total, 1032 participants (n=518, 50.2% male; median age: 77 y) were analyzed. Multivariable logistic regression analysis revealed that participants with frailty had significantly less access to the following sources of information compared to individuals without frailty: family (odds ratio [OR] 0.69, 95% CI 0.50-0.95), friends/acquaintances (OR 0.70, 95% CI 0.51-0.98), radio (OR 0.50, 95% CI 0.31-0.79), and newspapers (OR 0.66, 95% CI 0.50-0.88). Sex-based subgroup analyses revealed no significant interaction effects, indicating no heterogeneity in the findings. Conclusions:Older adults with frailty were less likely to obtain health information from interpersonal and traditional media sources than did individuals without frailty. Health information providers need to devise strategies for delivering accurate information and improving usability to enable older adults with frailty to proactively access diverse health information.
