%0 Journal Article %@ 2564-1891 %I JMIR Publications %V 5 %N %P e57455 %T Infodemic Versus Viral Information Spread: Key Differences and Open Challenges %A Cinelli,Matteo %A Gesualdo,Francesco %K infodemic %K information spreading %K infodemiology %K misinformation %K artificial intelligence %K information virality %K public health %K multidisciplinary %K data science %K AI %K difference %K challenge %D 2025 %7 7.5.2025 %9 %J JMIR Infodemiology %G English %X As we move beyond the COVID-19 pandemic, the risk of future infodemics remains significant, driven by emerging health crises and the increasing influence of artificial intelligence in the information ecosystem. During periods of apparent stability, proactive efforts to advance infodemiology are essential for enhancing preparedness and improving public health outcomes. This requires a thorough examination of the foundations of this evolving discipline, particularly in understanding how to accurately identify an infodemic at the appropriate time and scale, and how to distinguish it from other processes of viral information spread, both within and outside the realm of public health. In this paper, we integrate expertise from data science and public health to examine the key differences between information production during an infodemic and viral information spread. We explore both clear and subtle distinctions, including context and contingency (ie, the association of an infodemic and viral information spread with a health crisis); information dynamics in terms of volume, spread, and predictability; the role of misinformation and information voids; societal impact; and mitigation strategies. By analyzing these differences, we highlight challenges and open questions. These include whether an infodemic is solely associated with pandemics or whether it could arise from other health emergencies; if infodemics are limited to health-related issues or if they could emerge from crises initially unrelated to health (like climate events); and whether infodemics are exclusively global phenomena or if they can occur on national or local scales. Finally, we propose directions for future quantitative research to help the scientific community more robustly differentiate between these phenomena and develop tailored management strategies. %R 10.2196/57455 %U https://infodemiology.jmir.org/2025/1/e57455 %U https://doi.org/10.2196/57455 %0 Journal Article %@ 2564-1891 %I JMIR Publications %V 5 %N %P e62703 %T Identifying Misinformation About Unproven Cancer Treatments on Social Media Using User-Friendly Linguistic Characteristics: Content Analysis %A Fridman,Ilona %A Boyles,Dahlia %A Chheda,Ria %A Baldwin-SoRelle,Carrie %A Smith,Angela B %A Elston Lafata,Jennifer %+ Lineberger Comprehensive Cancer Center, University of North Carolina, 450 West Dr, Chapel Hill, NC, 27599, United States, 1 6469028137, ilona_fridman@med.unc.edu %K linguistic characteristics %K linguistic features %K cancer %K Linguistic Inquiry and Word Count %K misinformation %K X %K Twitter %K cancer %K alternative therapy %K oncology %K social media %K natural language processing %K machine learning %K synthesis %K review methodology %K search %K literature review %D 2025 %7 12.2.2025 %9 Original Paper %J JMIR Infodemiology %G English %X Background: Health misinformation, prevalent in social media, poses a significant threat to individuals, particularly those dealing with serious illnesses such as cancer. The current recommendations for users on how to avoid cancer misinformation are challenging because they require users to have research skills. Objective: This study addresses this problem by identifying user-friendly characteristics of misinformation that could be easily observed by users to help them flag misinformation on social media. Methods: Using a structured review of the literature on algorithmic misinformation detection across political, social, and computer science, we assembled linguistic characteristics associated with misinformation. We then collected datasets by mining X (previously known as Twitter) posts using keywords related to unproven cancer therapies and cancer center usernames. This search, coupled with manual labeling, allowed us to create a dataset with misinformation and 2 control datasets. We used natural language processing to model linguistic characteristics within these datasets. Two experiments with 2 control datasets used predictive modeling and Lasso regression to evaluate the effectiveness of linguistic characteristics in identifying misinformation. Results: User-friendly linguistic characteristics were extracted from 88 papers. The short-listed characteristics did not yield optimal results in the first experiment but predicted misinformation with an accuracy of 73% in the second experiment, in which posts with misinformation were compared with posts from health care systems. The linguistic characteristics that consistently negatively predicted misinformation included tentative language, location, URLs, and hashtags, while numbers, absolute language, and certainty expressions consistently predicted misinformation positively. Conclusions: This analysis resulted in user-friendly recommendations, such as exercising caution when encountering social media posts featuring unwavering assurances or specific numbers lacking references. Future studies should test the efficacy of the recommendations among information users. %M 39938078 %R 10.2196/62703 %U https://infodemiology.jmir.org/2025/1/e62703 %U https://doi.org/10.2196/62703 %U http://www.ncbi.nlm.nih.gov/pubmed/39938078 %0 Journal Article %@ 2561-326X %I JMIR Publications %V 8 %N %P e60024 %T Impact of Artificial Intelligence–Generated Content Labels On Perceived Accuracy, Message Credibility, and Sharing Intentions for Misinformation: Web-Based, Randomized, Controlled Experiment %A Li,Fan %A Yang,Ya %+ School of Journalism and Communication, Beijing Normal University, NO.19, Xinjiekouwai Street, Haidian District, Beijing, 100875, China, 86 18810305219, yangya@bnu.edu.cn %K generative AI %K artificial intelligence %K ChatGPT %K AIGC label %K misinformation %K perceived accuracy %K message credibility %K sharing intention %K social media %K health information %D 2024 %7 24.12.2024 %9 Original Paper %J JMIR Form Res %G English %X Background: The proliferation of generative artificial intelligence (AI), such as ChatGPT, has added complexity and richness to the virtual environment by increasing the presence of AI-generated content (AIGC). Although social media platforms such as TikTok have begun labeling AIGC to facilitate the ability for users to distinguish it from human-generated content, little research has been performed to examine the effect of these AIGC labels. Objective: This study investigated the impact of AIGC labels on perceived accuracy, message credibility, and sharing intention for misinformation through a web-based experimental design, aiming to refine the strategic application of AIGC labels. Methods: The study conducted a 2×2×2 mixed experimental design, using the AIGC labels (presence vs absence) as the between-subjects factor and information type (accurate vs inaccurate) and content category (for-profit vs not-for-profit) as within-subjects factors. Participants, recruited via the Credamo platform, were randomly assigned to either an experimental group (with labels) or a control group (without labels). Each participant evaluated 4 sets of content, providing feedback on perceived accuracy, message credibility, and sharing intention for misinformation. Statistical analyses were performed using SPSS version 29 and included repeated-measures ANOVA and simple effects analysis, with significance set at P<.05. Results: As of April 2024, this study recruited a total of 957 participants, and after screening, 400 participants each were allocated to the experimental and control groups. The main effects of AIGC labels were not significant for perceived accuracy, message credibility, or sharing intention. However, the main effects of information type were significant for all 3 dependent variables (P<.001), as were the effects of content category (P<.001). There were significant differences in interaction effects among the 3 variables. For perceived accuracy, the interaction between information type and content category was significant (P=.005). For message credibility, the interaction between information type and content category was significant (P<.001). Regarding sharing intention, both the interaction between information type and content category (P<.001) and the interaction between information type and AIGC labels (P=.008) were significant. Conclusions: This study found that AIGC labels minimally affect perceived accuracy, message credibility, or sharing intention but help distinguish AIGC from human-generated content. The labels do not negatively impact users’ perceptions of platform content, indicating their potential for fact-checking and governance. However, AIGC labeling applications should vary by information type; they can slightly enhance sharing intention and perceived accuracy for misinformation. This highlights the need for more nuanced strategies for AIGC labels, necessitating further research. %R 10.2196/60024 %U https://formative.jmir.org/2024/1/e60024 %U https://doi.org/10.2196/60024 %0 Journal Article %@ 2564-1891 %I JMIR Publications %V 4 %N %P e57748 %T The Complex Interaction Between Sleep-Related Information, Misinformation, and Sleep Health: Call for Comprehensive Research on Sleep Infodemiology and Infoveillance %A Bragazzi,Nicola Luigi %A Garbarino,Sergio %+ Human Nutrition Unit, Department of Food and Drugs, University of Parma, Via Volturno 39, Parma, 43125, Italy, 39 0521 903121, robertobragazzi@gmail.com %K sleep health %K sleep-related clinical public health %K sleep information %K health information %K infodemiology %K infoveillance %K social media %K myth %K misconception %K circadian %K chronobiology %K insomnia %K eHealth %K digital health %K public health informatics %K sleep data %K health data %K well-being %K patient information %K lifestyle %D 2024 %7 13.12.2024 %9 Viewpoint %J JMIR Infodemiology %G English %X The complex interplay between sleep-related information—both accurate and misleading—and its impact on clinical public health is an emerging area of concern. Lack of awareness of the importance of sleep, and inadequate information related to sleep, combined with misinformation about sleep, disseminated through social media, nonexpert advice, commercial interests, and other sources, can distort individuals’ understanding of healthy sleep practices. Such misinformation can lead to the adoption of unhealthy sleep behaviors, reducing sleep quality and exacerbating sleep disorders. Simultaneously, poor sleep itself impairs critical cognitive functions, such as memory consolidation, emotional regulation, and decision-making. These impairments can heighten individuals’ vulnerability to misinformation, creating a vicious cycle that further entrenches poor sleep habits and unhealthy behaviors. Sleep deprivation is known to reduce the ability to critically evaluate information, increase suggestibility, and enhance emotional reactivity, making individuals more prone to accepting persuasive but inaccurate information. This cycle of misinformation and poor sleep creates a clinical public health issue that goes beyond individual well-being, influencing occupational performance, societal productivity, and even broader clinical public health decision-making. The effects are felt across various sectors, from health care systems burdened by sleep-related issues to workplaces impacted by decreased productivity due to sleep deficiencies. The need for comprehensive clinical public health initiatives to combat this cycle is critical. These efforts must promote sleep literacy, increase awareness of sleep’s role in cognitive resilience, and correct widespread sleep myths. Digital tools and technologies, such as sleep-tracking devices and artificial intelligence–powered apps, can play a role in educating the public and enhancing the accessibility of accurate, evidence-based sleep information. However, these tools must be carefully designed to avoid the spread of misinformation through algorithmic biases. Furthermore, research into the cognitive impacts of sleep deprivation should be leveraged to develop strategies that enhance societal resilience against misinformation. Sleep infodemiology and infoveillance, which involve tracking and analyzing the distribution of sleep-related information across digital platforms, offer valuable methodologies for identifying and addressing the spread of misinformation in real time. Addressing this issue requires a multidisciplinary approach, involving collaboration between sleep scientists, health care providers, educators, policy makers, and digital platform regulators. By promoting healthy sleep practices and debunking myths, it is possible to disrupt the feedback loop between poor sleep and misinformation, leading to improved individual health, better decision-making, and stronger societal outcomes. %M 39475424 %R 10.2196/57748 %U https://infodemiology.jmir.org/2024/1/e57748 %U https://doi.org/10.2196/57748 %U http://www.ncbi.nlm.nih.gov/pubmed/39475424 %0 Journal Article %@ 2564-1891 %I JMIR Publications %V 4 %N %P e60678 %T Evaluating the Influence of Role-Playing Prompts on ChatGPT’s Misinformation Detection Accuracy: Quantitative Study %A Haupt,Michael Robert %A Yang,Luning %A Purnat,Tina %A Mackey,Tim %+ Global Health Program, Department of Anthropology, University of California, San Diego, 9500 Gilman Drive, La Jolla, CA, 92093, United States, 1 858 534 4145, tkmackey@ucsd.edu %K large language models %K ChatGPT %K artificial intelligence %K AI %K experiment %K prompt engineering %K role-playing %K social identity %K misinformation detection %K COVID-19 %D 2024 %7 26.9.2024 %9 Original Paper %J JMIR Infodemiology %G English %X Background: During the COVID-19 pandemic, the rapid spread of misinformation on social media created significant public health challenges. Large language models (LLMs), pretrained on extensive textual data, have shown potential in detecting misinformation, but their performance can be influenced by factors such as prompt engineering (ie, modifying LLM requests to assess changes in output). One form of prompt engineering is role-playing, where, upon request, OpenAI’s ChatGPT imitates specific social roles or identities. This research examines how ChatGPT’s accuracy in detecting COVID-19–related misinformation is affected when it is assigned social identities in the request prompt. Understanding how LLMs respond to different identity cues can inform messaging campaigns, ensuring effective use in public health communications. Objective: This study investigates the impact of role-playing prompts on ChatGPT’s accuracy in detecting misinformation. This study also assesses differences in performance when misinformation is explicitly stated versus implied, based on contextual knowledge, and examines the reasoning given by ChatGPT for classification decisions. Methods: Overall, 36 real-world tweets about COVID-19 collected in September 2021 were categorized into misinformation, sentiment (opinions aligned vs unaligned with public health guidelines), corrections, and neutral reporting. ChatGPT was tested with prompts incorporating different combinations of multiple social identities (ie, political beliefs, education levels, locality, religiosity, and personality traits), resulting in 51,840 runs. Two control conditions were used to compare results: prompts with no identities and those including only political identity. Results: The findings reveal that including social identities in prompts reduces average detection accuracy, with a notable drop from 68.1% (SD 41.2%; no identities) to 29.3% (SD 31.6%; all identities included). Prompts with only political identity resulted in the lowest accuracy (19.2%, SD 29.2%). ChatGPT was also able to distinguish between sentiments expressing opinions not aligned with public health guidelines from misinformation making declarative statements. There were no consistent differences in performance between explicit and implicit misinformation requiring contextual knowledge. While the findings show that the inclusion of identities decreased detection accuracy, it remains uncertain whether ChatGPT adopts views aligned with social identities: when assigned a conservative identity, ChatGPT identified misinformation with nearly the same accuracy as it did when assigned a liberal identity. While political identity was mentioned most frequently in ChatGPT’s explanations for its classification decisions, the rationales for classifications were inconsistent across study conditions, and contradictory explanations were provided in some instances. Conclusions: These results indicate that ChatGPT’s ability to classify misinformation is negatively impacted when role-playing social identities, highlighting the complexity of integrating human biases and perspectives in LLMs. This points to the need for human oversight in the use of LLMs for misinformation detection. Further research is needed to understand how LLMs weigh social identities in prompt-based tasks and explore their application in different cultural contexts. %M 39326035 %R 10.2196/60678 %U https://infodemiology.jmir.org/2024/1/e60678 %U https://doi.org/10.2196/60678 %U http://www.ncbi.nlm.nih.gov/pubmed/39326035 %0 Journal Article %@ 2564-1891 %I JMIR Publications %V 4 %N %P e59641 %T Large Language Models Can Enable Inductive Thematic Analysis of a Social Media Corpus in a Single Prompt: Human Validation Study %A Deiner,Michael S %A Honcharov,Vlad %A Li,Jiawei %A Mackey,Tim K %A Porco,Travis C %A Sarkar,Urmimala %+ Departments of Ophthalmology, Epidemiology and Biostatistics, Global Health Sciences, and Francis I Proctor Foundation, University of California San Francisco, 490 Illinois St, 2nd Floor, San Francisco, CA, 94158, United States, 1 415 476 4101, travis.porco@ucsf.edu %K generative large language model %K generative pretrained transformer %K GPT %K Claude %K Twitter %K X formerly known as Twitter %K social media %K inductive content analysis %K COVID-19 %K vaccine hesitancy %K infodemiology %D 2024 %7 29.8.2024 %9 Original Paper %J JMIR Infodemiology %G English %X Background: Manually analyzing public health–related content from social media provides valuable insights into the beliefs, attitudes, and behaviors of individuals, shedding light on trends and patterns that can inform public understanding, policy decisions, targeted interventions, and communication strategies. Unfortunately, the time and effort needed from well-trained human subject matter experts makes extensive manual social media listening unfeasible. Generative large language models (LLMs) can potentially summarize and interpret large amounts of text, but it is unclear to what extent LLMs can glean subtle health-related meanings in large sets of social media posts and reasonably report health-related themes. Objective: We aimed to assess the feasibility of using LLMs for topic model selection or inductive thematic analysis of large contents of social media posts by attempting to answer the following question: Can LLMs conduct topic model selection and inductive thematic analysis as effectively as humans did in a prior manual study, or at least reasonably, as judged by subject matter experts? Methods: We asked the same research question and used the same set of social media content for both the LLM selection of relevant topics and the LLM analysis of themes as was conducted manually in a published study about vaccine rhetoric. We used the results from that study as background for this LLM experiment by comparing the results from the prior manual human analyses with the analyses from 3 LLMs: GPT4-32K, Claude-instant-100K, and Claude-2-100K. We also assessed if multiple LLMs had equivalent ability and assessed the consistency of repeated analysis from each LLM. Results: The LLMs generally gave high rankings to the topics chosen previously by humans as most relevant. We reject a null hypothesis (P<.001, overall comparison) and conclude that these LLMs are more likely to include the human-rated top 5 content areas in their top rankings than would occur by chance. Regarding theme identification, LLMs identified several themes similar to those identified by humans, with very low hallucination rates. Variability occurred between LLMs and between test runs of an individual LLM. Despite not consistently matching the human-generated themes, subject matter experts found themes generated by the LLMs were still reasonable and relevant. Conclusions: LLMs can effectively and efficiently process large social media–based health-related data sets. LLMs can extract themes from such data that human subject matter experts deem reasonable. However, we were unable to show that the LLMs we tested can replicate the depth of analysis from human subject matter experts by consistently extracting the same themes from the same data. There is vast potential, once better validated, for automated LLM-based real-time social listening for common and rare health conditions, informing public health understanding of the public’s interests and concerns and determining the public’s ideas to address them. %M 39207842 %R 10.2196/59641 %U https://infodemiology.jmir.org/2024/1/e59641 %U https://doi.org/10.2196/59641 %U http://www.ncbi.nlm.nih.gov/pubmed/39207842 %0 Journal Article %@ 2564-1891 %I JMIR Publications %V 4 %N %P e51328 %T Descriptions of Scientific Evidence and Uncertainty of Unproven COVID-19 Therapies in US News: Content Analysis Study %A Watson,Sara %A Benning,Tyler J %A Marcon,Alessandro R %A Zhu,Xuan %A Caulfield,Timothy %A Sharp,Richard R %A Master,Zubin %+ Department of Social Sciences and Health Policy, Wake Forest University School of Medicine, Medical Center Boulevard, Suite 310, Winston-Salem, NC, 27157, United States, 1 3367164289, zmaster@wakehealth.edu %K COVID-19 %K COVID-19 drug treatment %K information dissemination %K health communication %K uncertainty %K content analysis %K information sources %K therapy %K misinformation %K communication %K scientific evidence %K media analysis %K news report %K COVID-19 therapy %K treatment %K public awareness %K public trepidation %K therapeutic %K therapeutics %K vaccine %K vaccines %K pandemic %K United States %K media analysis %K safety %K efficacy %K evidence %K news %K report %K reports %D 2024 %7 29.8.2024 %9 Original Paper %J JMIR Infodemiology %G English %X Background: Politicization and misinformation or disinformation of unproven COVID-19 therapies have resulted in communication challenges in presenting science to the public, especially in times of heightened public trepidation and uncertainty. Objective: This study aims to examine how scientific evidence and uncertainty were portrayed in US news on 3 unproven COVID-19 therapeutics, prior to the development of proven therapeutics and vaccines. Methods: We conducted a media analysis of unproven COVID-19 therapeutics in early 2020. A total of 479 discussions of unproven COVID-19 therapeutics (hydroxychloroquine, remdesivir, and convalescent plasma) in traditional and online US news reports from January 1, 2020, to July 30, 2020, were systematically analyzed for theme, scientific evidence, evidence details and limitations, safety, efficacy, and sources of authority. Results: The majority of discussions included scientific evidence (n=322, 67%) although only 24% (n=116) of them mentioned publications. “Government” was the most frequently named source of authority for safety and efficacy claims on remdesivir (n=43, 35%) while “expert” claims were mostly mentioned for convalescent plasma (n=22, 38%). Most claims on hydroxychloroquine (n=236, 79%) were offered by a “prominent person,” of which 97% (n=230) were from former US President Trump. Despite the inclusion of scientific evidence, many claims of the safety and efficacy were made by nonexperts. Few news reports expressed scientific uncertainty in discussions of unproven COVID-19 therapeutics as limitations of evidence were infrequently included in the body of news reports (n=125, 26%) and rarely found in headlines (n=2, 2%) or lead paragraphs (n=9, 9%; P<.001). Conclusions: These results highlight that while scientific evidence is discussed relatively frequently in news reports, scientific uncertainty is infrequently reported and rarely found in prominent headlines and lead paragraphs. %M 39207825 %R 10.2196/51328 %U https://infodemiology.jmir.org/2024/1/e51328 %U https://doi.org/10.2196/51328 %U http://www.ncbi.nlm.nih.gov/pubmed/39207825 %0 Journal Article %@ 1438-8871 %I JMIR Publications %V 26 %N %P e55927 %T Benchmarking State-of-the-Art Large Language Models for Migraine Patient Education: Performance Comparison of Responses to Common Queries %A Li,Linger %A Li,Pengfei %A Wang,Kun %A Zhang,Liang %A Ji,Hongwei %A Zhao,Hongqin %+ Department of Neurology, The Affiliated Hospital of Qingdao University, No. 59 Haier Road, Qingdao, 266035, China, 86 13864873935, zhaohongq@qdu.edu.cn %K migraine %K large language models %K patient education %K ChatGPT %K Google Bard %K language model %K patient education %K education %K headache %K accuracy %K OpenAI %K AI %K artificial intelligence %K AI-assisted %K holistic %K migraine management %K management %D 2024 %7 23.7.2024 %9 Research Letter %J J Med Internet Res %G English %X This study assessed the potential of large language models (OpenAI’s ChatGPT 3.5 and 4.0, Google Bard, Meta Llama2, and Anthropic Claude2) in addressing 30 common migraine-related queries, providing a foundation to advance artificial intelligence–assisted patient education and insights for a holistic approach to migraine management. %M 38828692 %R 10.2196/55927 %U https://www.jmir.org/2024/1/e55927 %U https://doi.org/10.2196/55927 %U http://www.ncbi.nlm.nih.gov/pubmed/38828692