JTJun Tao

Jun Tao

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

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Jun Tao has a total of 164 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Neuro-Symbolic Knowledge Graphs Improve Motivational Interviewing Fidelity in Clinical Artificial Intelligence Tools: Development and Simulation Study (Preprint)

    by Jun Tao, Trisha Arnold, Hao Wu, Hannah Parent, Mattia Prosperi, Natalie Fenn, Tyler Wray, Ellie Pavlick, Bryce K. Perler, Jack Rusley, et al.

    BACKGROUND Large language models (LLMs) are increasingly used in health counseling, but adherence to evidence-based approaches such as motivational interviewing (MI) is inconsistent, and their reasoning is difficult to inspect or audit. Supervised fine-tuning (SFT) trains a model to imitate expert responses but does not explicitly encode the sequential clinical decisions MI requires. Knowledge graphs (KGs), which represent clinical knowledge as explicit, inspectable rules and relationships, may offer a more transparent way to govern LLM counseling. OBJECTIVE This study aimed to develop a neuro-symbolic KG encoding MI clinical logic and to evaluate, in simulated sessions, whether KG guidance improves MI fidelity compared with a base LLM and SFT, alone and in combination. METHODS We built a 3-layer KG comprising an MI ontology (34 techniques, stages of change, and patient talk types), a deterministic rule engine (34 decision rules governing technique selection and phase progression), and 13 hard safety constraints, and integrated it into CHIA (Chatbot for HIV Prevention and Action), a preexposure prophylaxis (PrEP) counseling platform. Using a 2×2 factorial design crossing SFT (GPT-4.1 base vs fine-tuned on 2400 expert MI examples) with KG guidance (disabled vs enabled), we simulated 8-turn sessions with 20 LLM-generated patient personas (3 runs per persona per arm; 239/240 sessions completed). A blinded LLM judge (Claude Sonnet 4.5) scored Motivational Interviewing Treatment Integrity (MITI) 4.2 global scores (Empathy, Partnership, Cultivating Change Talk, and Softening Sustain Talk) and flagged premature planning. We used 2-way repeated-measures ANOVA and generalized estimating equations with false discovery rate correction, and computed linguistic metrics including lexical diversity. A blinded MI expert and a second LLM judge (Gemini 2.5 Pro) rescored a stratified subsample of 40 sessions. RESULTS The base LLM with KG guidance (Base+KG) achieved the highest mean Empathy (4.90, SD 0.30) and Partnership (4.88, SD 0.32). Compared with SFT alone, Base+KG showed higher Empathy (Cohen d=0.68, 95% CI 0.37-1.01) and Partnership (d=0.91, 95% CI 0.60-1.25). SFT had significant negative main effects on Empathy (F1,19=21.26; P<.001) and Partnership (F1,19=17.39; P<.001) and reduced lexical diversity (d=−1.47). The LLM judge flagged premature planning in 0/60 Base+KG sessions vs 1/60 Base, 4/59 SFT, and 6/60 SFT+KG sessions; however, the human expert identified more premature planning in every arm. Human-LLM agreement was moderate for Empathy (intraclass correlation coefficient [ICC] 0.55) and Partnership (ICC 0.42) and poor for Softening Sustain Talk, but all 3 raters produced consistent arm rankings. CONCLUSIONS In this simulation study, KG guidance applied to a base LLM yielded higher MI fidelity than SFT, which degraded relational quality. Externalizing clinical reasoning into an inspectable KG is a promising, auditable approach to governing LLM counseling, pending confirmation with human participants and larger human-rated samples.

  • Open Access

    A Neuro-Symbolic Knowledge Graph and Large Language Model Hybrid Architecture for Multi-Modality Mental Health Counseling

    by Jun Tao, Natalie Fenn, Hannah Parent, Hao Wu, Trisha Arnold, Jennifer Etue, Elizabeth Chen, Philip Chan

    Background: Depression and anxiety are managed largely between clinical visits, yet outpatient care lacks scalable, accountable mechanisms for between-visit support. Large language models converse fluently but fuse clinical reasoning with language generation in one opaque process, so they cannot reliably deliver evidence-based psychotherapy and typically operate outside clinician oversight. Objective: To evaluate C-Mind, a provider-supervised neuro-symbolic system in which a Clinical Knowledge Graph (KG) governs therapeutic decisions for a large language model across eight psychotherapy modalities. Methods: Two simulation regimes addressed eight pre-specified governance questions: a structural validation of KG routing against 117 guideline-anchored vignettes, and a governance battery using progressively disclosing LLM patient agents to evaluate decision traceability, repeatability, provenance auditability, adversarial crisis-detection robustness (277 probes), provider treatment-goal governance, and counselor technique adherence. Crisis detection was additionally validated externally against an independent, clinician-annotated corpus (CRADLE Bench). Results: The KG routed 116/117 vignettes (99.1%) to guideline-appropriate care and detected all 18 high-risk presentations, firing a therapy-suppressing hard halt on 16/18. Adversarial crisis-detection sensitivity was 96.7% and specificity 95.4% (277 probes); on external validation, the system detected 98.5% of 600 dialogues with ongoing suicidal ideation or self-harm at or before the annotator confirming turn. Decisions were 99.1% repeatable, 100% reconstructable per turn, and 100% provenance-auditable across all 354 KG nodes. Provider-set diagnosis, goals, and safety context governed behavior deterministically. Stripped of governance, the same model produced unsolicited clinical monologues on 100% of turns (vs 9% governed) and delivered diagnoses and medication advice the governed system never produced. Conclusions: A neuro-symbolic architecture achieves near-perfect guideline-appropriate routing with a governance profile, traceability, reproducibility, machine-traceable provenance, externally validated crisis detection, and deterministic provider control aligned with requirements for regulated clinical AI.

  • Open Access

    Development and Validation of the Lafiya Clinical Knowledge Graph for Decision Support in Nigerian Primary Health Care (Preprint)

    by Jun Tao, Hao Wu, Muktar Aliyu, Hamisu Salihu

    BACKGROUND Nigeria's primary healthcare (PHC) system serves over 200 million people through approximately 30,000 facilities staffed predominantly by Community Health Officers (CHOs) and Community Health Extension Workers (CHEWs) who operate without physician oversight. With only 0.4 physicians per 1,000 population, diagnostic accuracy at frontline facilities averages 65%, contributing to the country's disproportionate disease burden — including 25% of global malaria cases and an under-5 mortality rate of 114 per 1,000 live births. Existing clinical decision support tools (CDSTs) require stable internet connectivity or rely on generic algorithms unsuited to Nigeria's epidemiology and clinical guidelines. OBJECTIVE To describe the development, architecture, and validation of a Nigeria-specific clinical knowledge graph (KG) designed to power an offline-capable CDST for CHOs and CHEWs in Northern Nigerian PHC settings. METHODS The KG was constructed using a three-layer architecture: a multilingual semantic normalization layer (3,299 synonym mappings across English, Hausa, and Nigerian Pidgin) feeds into Layer A (medical ontology — diseases, symptoms, treatments), Layer B (guideline logic — diagnostic rules, 7-tier probabilistic weighting, age/sex gating, and deficit-gated differential rules), and Layer C (operational safety — drug interactions, age/sex gating, prescribing guardrails, and post-test confirmation overlay). Source content was derived from National Standing Orders for Community Health Practitioner, WHO Integrated Management of Childhood Illness (IMCI) guidelines, and 13 additional verified clinical guidelines. Validation comprised four independent assessments: (1) an internal benchmark of 381 generated clinical scenarios covering all 136 diagnoses; (2) a WHO-aligned CDSS safety validation suite of 203 adversarial scenarios across 7 safety categories; (3) adversarial paraphrase testing using novel Nigerian English, Hausa, and Pidgin symptom expressions; and (4) a free-text benchmark of 322 simulated patient presentations spanning all 136 diagnoses across 15 clinical domains. RESULTS The final KG contains 772 nodes and 2,052 edges encoding 136 diagnoses across 30 clinical categories and 6 age groups. Internal benchmarking demonstrated 90.0% top-1 diagnostic accuracy (clinical equivalence), 87.4% strict top-1 accuracy, 98.2% top-3 accuracy, and 97.5% emergency detection rate. The WHO-aligned validation achieved 97.5% overall accuracy (198/203 scenarios) with zero dangerous misses across 25 must-not-miss conditions (e.g., eclampsia, postpartum hemorrhage, acute kidney injury, poisoning). Adversarial paraphrase testing confirmed 96.3% accuracy on expressions never encountered during development. Free-text benchmarking — evaluating the system's end-to-end pipeline from unstructured patient language to differential diagnosis — achieved 68.3% top-1 accuracy, 91.9% top-3 accuracy, and 94.4% top-5 accuracy across 322 cases spanning all 136 diagnoses, with 99.3% emergency detection rate using clinical equivalence. CONCLUSIONS A rigorously validated, Nigeria-specific clinical KG can achieve diagnostic accuracy and safety performance comparable to international benchmarks while operating within the linguistic and infrastructural constraints of low-resource PHC settings. The three-layer architecture and deterministic reasoning approach offer a reproducible framework for developing context-specific CDSTs in other low- and middle-income countries. CLINICALTRIAL NA

  • Top Cited Papers
    Open Access

    Evaluation of an Artificial Intelligence Conversational Chatbot to Enhance HIV Preexposure Prophylaxis Uptake: Development and Usability Internal Testing

    by Jun Tao, Ellie Pavlick, Amaris Grondin, Josue D Bustamante, Harrison Martin, Hannah Parent, Natalie Fenn, Alexi Almonte, Amanda Maguire-Wilkerson, Mofan Gu, et al.

    Background: The HIV epidemic in the United States disproportionately impacts gay, bisexual, and other men who have sex with men (MSM). Despite the effectiveness of HIV preexposure prophylaxis (PrEP) in preventing HIV acquisition, uptake among MSM remains suboptimal. Motivational interviewing (MI) has demonstrated efficacy at increasing PrEP uptake among MSM but is resource-intensive, limiting scalability. The use of artificial intelligence, particularly large language models with conversational agents (ie, “chatbots”) such as ChatGPT, may offer a scalable approach to delivering MI-based counseling for PrEP and HIV prevention. Objective: This internal usability testing aimed to evaluate the development of an artificial intelligence–based chatbot, including its ability to provide MI-aligned education about PrEP and HIV prevention and potential to support PrEP uptake. Methods: The Chatbot for HIV Prevention and Action (CHIA) was built on a GPT-4o base model embedded with a validated knowledge database on HIV and PrEP in English and Spanish. The CHIA was fine-tuned through training on a large MI dataset and prompt engineering. The use of the AutoGen multiagent framework enabled the CHIA to integrate 2 agents, the PrEP Counselor Agent and the Assistant Agent, which specialized in providing MI-based counseling and handling function calls (eg, assessment of HIV risk), respectively. During internal testing from March 10-April 28, 2025, we systematically evaluated the CHIA’s performance in English and Spanish using a set of 5-point Likert scales to measure accuracy, conciseness, up-to-dateness, trustworthiness, and alignment with aspects of the MI spirit (eg, collaboration, autonomy support) and MI-consistent behaviors (eg, affirmation, open-ended questions). Descriptive statistics and mixed linear regression were used to analyze the data. Results: A total of 296 responses, including 145 English responses and 151 Spanish responses, were collected during the internal testing period. Overall, the CHIA demonstrated strong performance across both languages, receiving the highest combined scores in the general response quality metrics including up-to-dateness (mean 4.6, SD 0.8), trustworthiness (mean 4.5, SD 0.9), accuracy (mean 4.4, SD 0.9), and conciseness (mean 4.2, SD 1.1). The CHIA generally received higher combined scores for metrics that assessed alignment with the MI spirit (ie, empathy, evocation, autonomy support, and collaboration) and lower combined scores for MI-consistent behaviors (ie, affirmation, open-ended questions, and reflections). Spanish responses had significantly lower mean scores than English responses across nearly all MI-based metrics. Conclusions: Our internal usability testing highlights the potential of the CHIA as a viable tool for delivering MI-aligned counseling in English and Spanish to promote HIV prevention and support PrEP uptake, though its Spanish language performance requires further improvement.


  • Exploring the use of artificial intelligence in the promotion of sexual health: a cross-sectional assessment

    by Amanda Maguire-Wilkerson, Hannah Parent, Jun Tao, Philip A. Chan

    Background: Artificial Intelligence (AI) is an emerging approach in addressing sexual health including HIV and other sexually transmitted infections (STIs). The aim of this study was to explore facilitators and barriers of AI with a focus on underserved populations who are disproportionately impacted by sexual health disparities. Methods: In 2025, individuals presenting to a safety-net sexual health clinic were asked to take part in a short, one-time anonymous survey to assess the acceptability of AI-based technology in sexual health promotion. Multiple linear regression analyses were conducted to examine the association between demographic factors and the preference for AI-based versus clinic-based counseling. Results: Of 100 individuals who participated in this study, 17% were African American/Black, and 11% Hispanic/Latino. The majority self-identified as LGBTQ+ (84%) and reported that they had previously used AI (67%) such as ChatGPT, Google Bard, and/or Fandango for topics related to health information, school, or customer service. Fifteen percent of people reported using AI-technology an average of once a month, 14% once a week, and 11% daily. Demographic variables were not significantly associated with preference for sexual health promotion by AI versus clinic-based counseling ( p = 0.21). However, 56% reported distrust regarding sharing personal information with AI technology. Conclusions: Many individuals are willing to consider AI technology use for sexual health promotion. However, issues related to trust and security should be addressed to optimize uptake.


  • Open Access

    Engagement and retention in HIV care in rural Southern U.S. using an All-Payer Claims Database

    by Jun Tao, Mofan Gu, S. Alexandra Marshall, Jure Baloh, Mike Cima, Cornelius Mabin, Nickolas D. Zaller, Dan Moore, Hannah Parent, Heather Horton, et al.

    Access to HIV care is a persistent challenge in rural settings. Arkansas has one of the highest proportions of rural residents in the U.S., yet data on statewide engagement and retention in HIV care are limited. This study used the Arkansas All-Payer Claims Database (APCD), a longitudinal database of medical, pharmacy, and enrollment claims from Medicaid and most commercial insurers, to evaluate the HIV care continuum and spatial access to providers. We identified people living with HIV (PLWH) from 2013–2017 using an algorithm requiring ≥2 HIV diagnosis claims or ≥1 prescription for antiretroviral therapy (ART). Engagement in care was defined as ≥1 CD4 or viral load (VL) test in a calendar year, while retention was defined as ≥2 tests in a year. Geospatial accessibility to HIV care at the county level was assessed using ArcGIS. A total of 5,033 PLWH in Arkansas were identified across the 5-year period. Only 18–21% of this population received CD4 testing and 47–51% received VL testing annually. ART prescription fills ranged from 61–88% annually. HIV testing rates among the broader insured population were also low, particularly in rural counties. There were 92 unique physicians with HIV-related care claims during the study period, equating to approximately 1 provider per 50 PLWH. Spatial access and care engagement were lower in rural counties. Significant gaps exist in HIV care engagement and retention in Arkansas, with notable gender and rural/urban disparities. Targeted interventions are crucial to address these disparities and end the HIV epidemic in the state.

  • Open Access

    Integrating HIV prevention services into care settings for people with opioid use disorder (OUD): a study protocol for implementation strategy development and modeling

    by Mofan Gu, Ruben G. Martinez, Hannah Parent, Brandon D. L. Marshall, Justin Berk, A. Rani Elwy, Philip A. Chan, Jun Tao

    Background: The overlapping epidemics of opioid use disorder (OUD) and HIV present a critical public health challenge. Although people with OUD frequently engage with healthcare settings, uptake of HIV prevention services such as pre-exposure prophylaxis (PrEP) remains low. Integrating HIV prevention into routine OUD care could reduce new infections, but scalable, evidence-based strategies are lacking. Rhode Island offers a unique opportunity to design and evaluate such strategies using its robust data infrastructure and high OUD burden. Methods: We will conduct a three-phase, sequential implementation study. In Aim 1, we will use the Rhode Island All-Payer Claims Database and State Emergency Department Database data to identify healthcare engagement patterns and gaps in HIV prevention service delivery among people with OUD, including rates of HIV screening, PrEP use, and medications for OUD uptake, across settings from 2012 to 2022. In Aim 2, we will convene a series of five stakeholder-engaged evidence-based quality improvement panels—including with providers, policymakers, and people with lived experience—to co-develop implementation strategies tailored to each care setting (i.e., primary care, mental health clinics, emergency department, and opioid use treatment centers). Finally, in Aim 3, we will develop an agent-based model (ABM) to simulate the population-level effect of implementation strategies developed for each care setting (as identified in Aim 2). The ABM will project outcomes such as HIV incidence, cases averted, and number needed to treat (NNT) over 5- and 10-year horizons under various scenarios. Model parameters will be based on literature and findings from Aim 1. Outputs from the ABM will be used to prioritize feasible, high-impact strategies for future real-world implementation. Discussion: This study addresses critical gaps in HIV prevention for people with OUD by combining claims-based analysis, evidence-based quality improvement, and agent-based modeling. By leveraging real-world data and engaging diverse stakeholders, the study aims to generate actionable strategies tailored to clinical settings. Findings will inform future implementation efforts in Rhode Island and other jurisdictions facing overlapping HIV and opioid epidemics. Trial registration: This study does not meet the World Health Organization’s definition of a clinical trial and, therefore, was not registered.

  • Open Access

    Evaluation of an Artificial Intelligence Conversational Chatbot to Enhance HIV Pre-Exposure Prophylaxis Uptake: Development and Usability Study (Preprint)

    by Jun Tao, Ellie Pavlick, Amaris Grondin, Josue D. Bustamente, Harrison Martin, Hannah Parent, Natalie Fenn, Alexi Almonte, Amanda Maguire-Wilkerson, Mofan Gu, et al.

    BACKGROUND The HIV epidemic in the United States disproportionately impacts gay, bisexual, and other men who have sex with men (MSM). Despite the effectiveness of HIV pre-exposure prophylaxis (PrEP) in preventing HIV acquisition, uptake among MSM remains suboptimal. Motivational interviewing (MI) has demonstrated efficacy at increasing PrEP uptake among MSM but is resource-intensive, limiting scalability. The use of artificial intelligence (AI), particularly large language models with conversational agents (i.e., “chatbots”) such as ChatGPT, may offer a scalable approach to delivering MI-based counseling for PrEP and HIV prevention. OBJECTIVE This study aimed to describe the development of an AI-based chatbot and evaluate its ability to provide MI-aligned education about PrEP and HIV prevention. METHODS The Chatbot for HIV Prevention and Action (CHIA) was built on a GPT-4o base model embedded with a validated knowledge database on HIV and PrEP in English and Spanish. CHIA was fine-tuned through training on a large MI dataset and prompt engineering. Use of the AutoGen multi-agent framework enabled CHIA to integrate two agents, the PrEP Counselor Agent and the Assistant Agent, which specialized in providing MI-based counseling and handling function calls (e.g., assessment of HIV risk), respectively. During internal testing from March 10-April 28, 2025, we systematically evaluated CHIA’s performance in English and Spanish using a set of five-point Likert scales to measure accuracy, conciseness, up-to-dateness, trustworthiness, and alignment with aspects of the MI spirit (e.g., collaboration, autonomy support) and MI-consistent behaviors (e.g., affirmation, open-ended questions). Descriptive statistics and independent samples t tests were used to analyze the data. RESULTS A total of 305 responses, including 140 English responses and 165 Spanish responses, were collected during the internal testing period. Overall, CHIA demonstrated strong performance across both languages, receiving the highest combined scores in the general response quality metrics including up-to-dateness (mean 4.6, SD 0.8), trustworthiness (mean 4.5, SD 0.9), accuracy (mean 4.4, SD 0.9), and conciseness (mean 4.2, SD 1.1). CHIA generally received higher combined scores for metrics that assessed alignment with the MI spirit (i.e. empathy, evocation, autonomy support, and collaboration) and lower combined scores for MI-consistent behaviors (i.e. affirmation, open-ended questions, and reflections). Spanish responses had significantly lower mean scores than English responses across nearly all MI-based metrics. CONCLUSIONS These findings highlight the potential of AI-based chatbots including CHIA as a scalable tool for delivering MI-aligned counseling in English and Spanish to promote HIV prevention and PrEP uptake.

  • Open Access

    Perspectives on a peer-driven intervention to promote pre-exposure prophylaxis (PrEP) uptake among men who have sex with men in southern New England: a qualitative study

    by Jun Tao, Hannah Parent, Ishu Karki, Harrison Martin, Sarah Alexandra Marshall, Jhanavi Kapadia, Amy S. Nunn, Brandon D. L. Marshall, Henry F. Raymond, Leandro Mena, et al.

    Background: Pre-exposure prophylaxis (PrEP) is a highly effective pharmaceutical intervention that prevents HIV infection, but PrEP uptake across the US has been slow among men who have sex with men (MSM), especially among Black/African American (B/AA) and Hispanic /Latino (H/L) MSM. This study investigates the acceptability and essential components of a peer-driven intervention (PDI) for promoting PrEP uptake among MSM, with a specific focus on B/AA and H/L communities. Methods: We conducted 28 semi-structured, qualitative interviews with MSM in southern New England to explore the components of a PDI, including attitudes, content, and effective communication methods. A purposive sampling strategy was used to recruit diverse participants who reflect the communities with the highest burden of HIV infection. Results: Of 28 study participants, the median age was 28 years (interquartile range [IQR]: 25, 35). The sample comprised B/AA (39%, n = 11) and H/L (50%, n = 14) individuals. Notably, nearly half of the participants (46%) were current PrEP users. We found that many participants were in favor of using a PDI approach for promoting PrEP. Additionally, several participants showed interest in becoming peer educators themselves. They emphasized the need for strong communication skills to effectively teach others about PrEP. Moreover, participants noted that peer education should cover key topics like how PrEP works, how effective it is, and any possible side effects. Conclusions: Our study shows that effective PDIs, facilitated by well-trained peers knowledgeable about PrEP, could enhance PrEP uptake among MSM, addressing health disparities and potentially reducing HIV transmission in B/AA and H/L communities.