RLRobert Patrick Lennon

Robert Patrick Lennon

*This is an algorithmically generated author record based on publicly available information.

Robert Patrick Lennon has a total of 265 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Vaccine Hesitancy Driven by Social Media Misinformation in COVID-19 and HPV: A Large Language Model-Based Content Analysis (Preprint)

    by Samira Malek, Christopher Griffin, Robert D Fraleigh, Robert Lennon, Lijiang Shen

    BACKGROUND The rapid growth of the internet and social media platforms has transformed them into major sources of information for the public. However, the widespread dissemination of misinformation on social media—particularly regarding vaccines—poses a critical threat to public health by undermining public trust, reducing vaccine uptake, and increasing vaccine hesitancy. As the volume of online data continues to expand, there is a growing need for an end-to-end automated and reliable content analysis of large-scale data to identify prevalent misinformation narratives and public concerns. OBJECTIVE The goal of this paper is to analyze misinformation related to COVID-19 and HPV vaccines, both separately and jointly, to identify and compare the dominant themes contributing to vaccine hesitancy. To do this, we apply the Misinformation Detection and Inoculation Process (MDIP) method—an end-to-end automated framework for health misinformation analysis introduced in (1). Besides, by deploying MDIP across multiple vaccine misinformation datasets and obtaining consistent, interpretable thematic structures, we demonstrate its effectiveness as a scalable framework for analyzing health misinformation. METHODS We utilized MDIP (1) to analyze unstructured social media data. Accordingly, first, a fine-tuned BERT model was employed to detect misinformation posts. Second, we applied BERTopic with an assignment algorithm to cluster documents. Finally, unlike traditional word-level models, we used structured prompt to automatically generate sentence-level topic descriptions and hierarchical themes, enabling granular analysis of narratives. RESULTS The analysis revealed COVID-19 misinformation was predominantly fueled by Conspiracy Theories (45.90%), including narratives such as microchips and 5G. In contrast, HPV misinformation was overwhelmingly driven by Safety and Concerns (58.98%), including infertility and adverse effects. Collectively, the combined analysis reveals a fundamental convergence in the determinants of vaccine hesitancy, demonstrating that public apprehension is universally rooted in safety concerns and skepticism regarding institutional motives. Temporal analysis showed that misinformation intensity is reactive, consistently spiking in response to specific regulatory approvals and public health milestones. CONCLUSIONS The proposed content analysis of HPV and COVID-19 includes four types of analysis: describing topics, categorizing topics into themes, examining theme distribution, and tracking changes over time. This approach helps extract the major concerns and dynamics surrounding the HPV and COVID-19 vaccines.

  • Top Cited Papers
    Open Access

    Intervention in Health Misinformation Using Large Language Models for Automated Detection, Thematic Analysis, and Inoculation: Case Study on COVID-19

    by Samira Malek, Christopher Griffin, Robert D Fraleigh, Robert Lennon, Vishal Monga, Lijiang Shen

    Background: The rapid growth of social media as an information channel has enabled the swift spread of inaccurate or false health information, significantly impacting public health. This widespread dissemination of misinformation has caused confusion, eroded trust in health authorities, led to noncompliance with health guidelines, and encouraged risky health behaviors. Understanding the dynamics of misinformation on social media is essential for devising effective public health communication strategies. Objective: This study aims to present a comprehensive and automated approach that leverages large language models (LLMs) and machine learning techniques to detect misinformation on social media, uncover the underlying causes and themes, and generate refutation arguments, facilitating control of its spread and promoting public health outcomes by inoculating people against health misinformation. Methods: We use 2 datasets to train 3 LLMs, namely, BERT, T5, and GPT-2, to classify documents into 2 categories: misinformation and nonmisinformation. In addition, we use a separate dataset to identify misinformation topics. To analyze these topics, we applied 3 topic modeling algorithms—Latent Dirichlet Allocation, Top2Vec, and BERTopic—and selected the optimal model based on performance evaluated across 3 metrics. Using a prompting approach, we extract sentence-level representations for the topics to uncover their underlying themes. Finally, we design a prompt text capable of identifying misinformation themes effectively. Results: The trained BERT model demonstrated exceptional performance, achieving 98% accuracy in classifying misinformation and nonmisinformation, with a 44% reduction in false-positive rates for artificial intelligence–generated misinformation. Among the 3 topic modeling approaches used, BERTopic outperformed the others, achieving the highest metrics with a Coherence Value of 0.41, Normalized Pointwise Mutual Information of −0.086, and Inverse Rank-Biased Overlap of 0.99. To address the issue of unclassified documents, we developed an algorithm to assign each document to its closest topic. In addition, we proposed a novel method using prompt engineering to generate sentence-level representations for each topic, achieving a 99.6% approval rate as “appropriate” or “somewhat appropriate” by 3 independent raters. We further designed a prompt text to identify themes of misinformation topics and developed another prompt capable of detecting misinformation themes with 82% accuracy. Conclusions: This study presents a comprehensive and automated approach to addressing health misinformation on social media using advanced machine learning and natural language processing techniques. By leveraging LLMs and prompt engineering, the system effectively detects misinformation, identifies underlying themes, and provides explanatory responses to combat its spread. The proposed method was tested on an English language COVID-19–related dataset and has not been evaluated on real-world online social media data; the experiments were conducted offline.


  • Open Access

    A Methodology Framework for Analyzing Health Misinformation to Develop Inoculation Intervention Using Large Language Models: A case study on covid-19

    by Samira Malek, Christopher Griffin, Robert Fraleigh, Robert P. Lennon, Vishal Monga, Lijiang Shen

    Background The rapid growth of social media as an information channel has enabled the swift spread of inaccurate or false health information, significantly impacting public health. This widespread dissemination of misinformation has caused confusion, eroded trust in health authorities, led to noncompliance with health guidelines, and encouraged risky health behaviors. Understanding the dynamics of misinformation on social media is essential for devising effective public health communication strategies. Objective This study aims to present a comprehensive and automated approach that leverages Large Language Models (LLMs) and Machine Learning (ML) techniques to detect misinformation on social media, uncover the underlying causes and themes, and generate refutation arguments, facilitating control of its spread and promoting public health outcomes by inoculating people against health misinformation. Methods We use two datasets to train three LLMs, namely BERT, T5, and GPT-2, to classify documents into two categories: misinformation and non-misinformation. Additionally, we employ a separate dataset to identify misinformation topics. To analyze these topics, we apply three topic modeling algorithms—Latent Dirichlet Allocation (LDA), Top2Vec, and BERTopic—and selected the optimal model based on performance evaluated across three metrics. Using a prompting approach, we extract sentence-level representations for the topics to uncover their underlying themes. Finally, we design a prompt text capable of identifying misinformation themes effectively. Results The trained BERT model demonstrated exceptional performance, achieving 98% accuracy in classifying misinformation and non-misinformation, with a 44% reduction in false positive rates for AI-generated misinformation. Among the three topic modeling approaches employed, BERTopic outperformed the others, achieving the highest metrics with a Coherence Value (CV) of 0.41, Normalized Pointwise Mutual Information (NPMI) of -0.086, and Inverted RBO (IRBO) of 0.99. To address the issue of unclassified documents, we developed an algorithm to assign each document to its closest topic. Additionally, we proposed a novel method using prompt engineering to generate sentence-level representations for each topic, achieving a 99.6% approval rate as “appropriate” or “somewhat appropriate” by three independent raters. We further designed a prompt text to identify themes of misinformation topics and developed another prompt capable of detecting misinformation themes with 80% accuracy. Conclusions This study presents a comprehensive and automated approach to addressing health misinformation on social media using advanced machine learning and natural language processing techniques. By leveraging large language models (LLMs) and prompt engineering, the system effectively detects misinformation, identifies underlying themes, and provides explanatory responses to combat its spread. (Journal of Medical Internet Research) doi:


  • Open Access

    A Methodology Framework for Analyzing Health Misinformation to Develop Inoculation Intervention Using Large Language Models: A case study on covid-19 (Preprint)

    by Samira Malek, Christopher Griffin, Robert D Fraleigh, Robert Lennon, Vishal Monga, Lijiang Shen

    BACKGROUND The rapid growth of social media as an information channel has enabled the swift spread of inaccurate or false health information, significantly impacting public health. This widespread dissemination of misinformation has caused confusion, eroded trust in health authorities, led to noncompliance with health guidelines, and encouraged risky health behaviors. Understanding the dynamics of misinformation on social media is essential for devising effective public health communication strategies. OBJECTIVE This study aims to present a comprehensive and automated approach that leverages Large Language Models (LLMs) and Machine Learning (ML) techniques to detect misinformation on social media, uncover the underlying causes and themes, and generate refutation arguments, facilitating control of its spread and promoting public health outcomes by inoculating people against health misinformation. METHODS We use two datasets to train three LLMs, namely BERT, T5, and GPT-2, to classify documents into two categories: misinformation and non-misinformation. Additionally, we employ a separate dataset to identify misinformation topics. To analyze these topics, we apply three topic modeling algorithms—Latent Dirichlet Allocation (LDA), Top2Vec, and BERTopic—and selected the optimal model based on performance evaluated across three metrics. Using a prompting approach, we extract sentence-level representations for the topics to uncover their underlying themes. Finally, we design a prompt text capable of identifying misinformation themes effectively. RESULTS The trained BERT model demonstrated exceptional performance, achieving 98% accuracy in classifying misinformation and non-misinformation, with a 44% reduction in false positive rates for AI-generated misinformation. Among the three topic modeling approaches employed, BERTopic outperformed the others, achieving the highest metrics with a Coherence Value (CV) of 0.41, Normalized Pointwise Mutual Information (NPMI) of -0.086, and Inverted RBO (IRBO) of 0.99. To address the issue of unclassified documents, we developed an algorithm to assign each document to its closest topic. Additionally, we proposed a novel method using prompt engineering to generate sentence-level representations for each topic, achieving a 99.6% approval rate as "appropriate" or "somewhat appropriate" by three independent raters. We further designed a prompt text to identify themes of misinformation topics and developed another prompt capable of detecting misinformation themes with 80% accuracy. CONCLUSIONS This study presents a comprehensive and automated approach to addressing health misinformation on social media using advanced machine learning and natural language processing techniques. By leveraging large language models (LLMs) and prompt engineering, the system effectively detects misinformation, identifies underlying themes, and provides explanatory responses to combat its spread.

  • Open Access

    The educational impact of medical students’ participation in a short-term health expedition: The Iguape one health experience

    by Marcelo Arruda Candido, Vinicius Venturini, Matheus Polly, Matheus Belloni Torsani, Alexandre Sizilio, João Mitsuji Sakô, Wen-Jan Tuan, Robert Patrick Lennon, Anna Sara Shafferman Levin, Iolanda de Fátima Lopes Calvo Tibério


  • Open Access

    Perceived barriers to research and scholarship among physicians

    by Makayla B Lagerman, Lauren Dennis, Traci Deaner, Adam Sigal, David Rabago, Huamei Dong, Joseph P. Wiedemer, Alexis Reedy-Cooper, Jessica Parascando, Karl T. Clebak, et al.

    Introduction: Inadequate scholarship is a common concern in graduate medical education. Many barriers to scholarship have been identified, but there are limited data on which are most important. A rank-order of barriers from learners’ perspectives would better enable programs to address perceived barriers. Methods: Given that learners are experts in their own perceptions, we applied the Delphi method of generating consensus expert opinion to construct ranked lists of physician-perceived barriers to scholarly activity at various sites. The survey was conducted within three separate health systems. Respondents were asked to identify their perceived barriers via free text and the listed barriers were consolidated by the research team. In the second round, respondents were presented with the consolidated lists and asked to rank them. In the third and final round, each respondent was shown a comparison of their own rankings to that of their peers and given an opportunity to make changes. Ranking differences between programs were compared using Rank Biased Overlap (RBO). Results: The Delphi method is a straightforward method to obtain a ranked list of perceived learner barriers to scholarly activity; its primary limitation is learner engagement and, of note in our study, high dropout rates. RBO is an effective method of ranking differences between programs and specialties. Top barriers across programs included time, overwhelm with the research process, and lack of interest or energy. Discussion: Several of the identified barriers may be addressed with enthusiastic mentors, streamlined administrative processes, and education. This could be done within a hospital system or on a national level through specialty organizations.

  • Open Access

    Assessing the risk of COVID-19 reinfection and severe outcomes among individuals with substance use disorders: a retrospective study using real-world electronic health records

    by Wen-Jan Tuan, Hailey M Kindt, Robert P Lennon

    Objective Despite advancement in vaccines and treatments for COVID-19 over the past 2 years, many concerns remain about reinfection and waning immunity against COVID-19 and its variants, especially among people with substance use disorder (SUD). The study assessed the risk of COVID-19 reinfection and severe illness among adults with SUD and their vaccination status to inform management in this vulnerable population as the pandemic continues. Design Retrospective cohort study. Setting Nationwide electronic health records (TriNetX database) in the USA among adults with COVID-19 infection from January 2020 to June 2022. Participants Adults (age ≥18 years) who were infected by COVID-19, excluding those who had cancer or lived in nursing homes or palliative care facilities. Outcome measures COVID-19 reinfection was defined as a new diagnosis after 45 days of the initial infection. Logistic regression was applied to assess the OR of COVID-19 reinfection and severe outcomes within 30 day of reinfection for adults with alcohol (AUD), opioid (OUD), cocaine (CUD), stimulant (STUD), cannabis (CAUD) and other use disorders, controlled for demographic and comorbid conditions. Results The SUD cohort was 13%–29% more likely to be reinfected by COVID-19 and had significantly higher 30-day mortality. Adults with AUD, STUD and OUD were at greater risks (adjusted ORs, AORs=1.69–1.86) of emergency department, hospital and intensive care admissions after 30 days of reinfection. Individuals with SUD and multiple vaccines doses were associated with decreased risks of worse COVID-19 outcomes. Lower COVID-19 reinfection rates (AORs=0.67–0.84) were only found among individuals with AUD, CUD or CAUD who had COVID-19 vaccination. Conclusions Individuals with SUD had greater risks of COVID-19 reinfection and poor outcomes, especially those with OUD, STUD and AUD. Multiple vaccinations are recommended to reduce severe illness after COVID-19 reinfection in the SUD population.


  • Open Access

    Non-inferiority comparison of on-site ophthalmic diagnosis by residents versus remote, asynchronous diagnosis by faculty

    by Elen Deng, Carolina Martines, Matheus Enrico Dias Vaz Monteiro, Flavio Moura Travassos de Medeiros, Augusto Cesar Villar de Almeida, Davi Martins, Amanda Latuffe Soares Damião, Kunal Koka, Matheus Polly, Pedro Carlos Carricondo, et al.

    Context: Teleophthalmology via portable retina cameras usable by non-ophthalmologists may increase access to ophthalmology care in underserved populations, but there is limited data on their accuracy. Objective: To evaluate the diagnostic accuracy of two novel optic devices between on-site residents and remote, asynchronous faculty. Study Design and Analysis: Single-blinded prospective comparative trial. Ophthalmology residents from the University of São Paulo (USP) provided care to a remote population in Iguape, Brazil through a joint effort between USP and Unidade Mista de Saúde. Residents recorded a brief history and visual analysis using two novel digital cameras. Image(s) and/or video(s) of the anterior anatomy of the eye were captured using a prototype camera with 2 settings: diffuse light and slight lamp. Images of the posterior chamber were captured using Eyer®. Residents documented their provisional diagnosis. USP ophthalmology faculty completed a blinded review of the history and images and documented a diagnosis. On-site versus remote diagnoses were compared, and considered the same if they were either identical or if the difference was not clinically significant, as determined by independent faculty review. Concordance was measured by Cohen’s kappa for inter-rater reliability to 95% confidence. Setting: Rural outpatient clinic. Population Studied: Adults living in the greater Iguape, Brazil catchment area. Intervention: Ophthalmic assessment. Outcome Measures: Similarity of ophthalmologic diagnoses. Results: Inter-rater reliability for 120 anterior chamber images examined from the prototype camera and 130 retinal images from Eyer® were 0.98 (95%CI 0.96-1.00) and 0.97 (95%CI 0.94-0.99) respectively. Expected Outcomes: On-site resident diagnoses of optic diseases almost perfectly agrees to that of remote, asynchronous faculty diagnoses. Asynchronous teleophthalmology consultation using a brief history and detailed image(s) and/or video(s) may be used to expand ophthalmology capacity in remote areas.

  • Adoption and Discontinuance of Innovation Packages: A Longitudinal Study of Transitions in COVID-19 Mitigation

    by Rachel A. Smith, Jessica Gall Myrick, Molly A. Martin, Robert P. Lennon, Lauren J. Van Scoy, Meg L. Small

    The past decades have seen growing interest and application of interventions targeting the change of multiple behaviors at once. We advance this work by using the diffusion of innovations theory (DOI) to consider constellations of behaviors as innovation packages: multiple innovations that are logically related, interdependent in their use or effects, and often promoted as a set (Rogers, 2003). In addition, we embraced DOI's focus on behavioral decisions as a continual process that can include adoption and discontinuance over time, especially as new innovations (e.g., COVID-19 vaccine) appear. To that end, we conducted a latent transition analysis of COVID-19 mitigation behaviors (N = 697; 97% received a COVID-19 vaccine) across three time points in the pandemic: initial outbreak; a secondary, record-breaking rise in cases; and after the CDC recommended that fully vaccinated adults could discontinue wearing masks. This analysis allowed us to identify latent classes based on shared behavioral patterns and transitions between classes over time. The results showed evidence of three possible packages: (a) a package of traditional, symptom-management behaviors (covering coughs and sneezes, staying home if ill, and seeking medical care), (b) a package of just-novel COVID-19 behaviors (wearing masks, keeping six feet apart, and avoiding mass gatherings), and (c) a package of all COVID-19 mitigation behaviors. Movement between classes exemplified adoption and discontinuance of different packages, as well as widespread discontinuance with the replacement innovation: COVID-19 vaccines. Additional analyses showed that increases in hope were associated with sustained and delayed adoption; decreases in social approval were associated with discontinuance. Future directions in theorizing around innovation packages are discussed.


  • Open Access

    Experiences of adults with opioid-treated chronic low back pain during the COVID-19 pandemic: A cross-sectional survey study

    by Aleksandra E. Zgierska, Cindy A. Burzinski, Eric L. PhD Garland, Bruce Barrett, Robert P. Md Lennon, Roger L. Brown, Anthony R. Bs Schiefelbein, Yoshio Nakamura, Barbara Stahlman, Robert N. Jamison, et al.

    This study aimed to evaluate the impact of the COVID-19 pandemic on adults with opioid-treated chronic low back pain (CLBP), an understudied area. Participants in a “parent” clinical trial of non-pharmacologic treatments for CLBP were invited to complete a one-time survey on the perceived pandemic impact across several CLBP- and opioid therapy-related domains. Participant clinical and other characteristics were derived from the parent study’s data. Descriptive statistics and latent class analysis analyzed quantitative data; qualitative thematic analysis was applied to qualitative data. The survey was completed by 480 respondents from June 2020 to August 2021. The majority reported a negative pandemic impact on their life (84.8%), with worsened enjoyment of life (74.6%), mental health (74.4%), pain (53.8%), pain-coping skills (49.7%), and finances (45.3%). One-fifth (19.4%) of respondents noted increased use of prescribed opioids; at the same time, decreased access to medication and overall healthcare was reported by 11.3% and 61.6% of respondents, respectively. Latent class analysis of the COVID-19 survey responses revealed 2 patterns of pandemic-related impact; those with worse pandemic-associated harms (n = 106) had an overall worse health profile compared to those with a lesser pandemic impact. The pandemic substantially affected all domains of relevant health-related outcomes as well as healthcare access, general wellbeing, and financial stability among adults with opioid-treated CLBP. A more nuanced evaluation revealed a heterogeneity of experiences, underscoring the need for both increased overall support for this population and for an individualized approach to mitigate harms induced by pandemic or similar crises.