KZKai Zheng

Kai Zheng

*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


Kai Zheng has a total of 233 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Recruitment, Engagement, Experiences, and Self-Reported Changes in a Digital Patient Activation Intervention to Improve Hemodialysis Session Stability: Mixed Methods Study

    by Megan Wickens, Jennifer Bragg-Gresham, Rajiv Saran, Meera Tandan, Marissa Argentina, Dinesh Chatoth, Kelli Collins Damron, Michael Heung, Sarah Krein, Vu Le, et al.

    Background: When chronic kidney disease progresses to end-stage kidney disease, many patients receive in-center hemodialysis. Although lifesaving, hemodialysis is associated with intradialytic hypotension. Peer mentoring and patient activation support in hemodialysis have not been well studied. Objective: This Dialysafe study combined patient education and peer mentoring to promote health behaviors supporting prevention of intradialytic hypotension. We aimed to describe recruitment, engagement, completion, and attrition; assess participants’ health goals and self-reported behavior changes; and evaluate satisfaction and recommendations for improvement. Methods: At 10 Dialysafe hemodialysis centers assigned to patient activation, all eligible English-speaking patients were invited. The intervention comprised 5 educational modules and 5 videoconference peer-mentoring sessions. Mixed methods analyses integrated electronic health record and platform data, postintervention surveys, and clinic operations committee notes. An engagement score (range 0-50) summarized intervention use. Descriptive statistics, t tests, chi-square tests, logistic regression, and zero-inflated Poisson regression were used to assess recruitment, completion (all 5 mentoring sessions), and engagement. Qualitative data were inductively coded and integrated through joint displays. Results: Of the 639 eligible patients, 155 (24.3%) were recruited. Recruited patients were younger than nonrecruited patients (mean 57.6, SD 13.8, vs 64.3, SD 14.7 y; P<.001); each additional year was associated with 4% lower odds of recruitment (odds ratio [OR] 0.96, 95% CI 0.94-0.98; P<.001). Recruitment was more likely among college graduates (OR 1.70, 95% CI 1.02-2.83; P=.04) and patients with sleep disorders (OR 2.27, 95% CI 1.06-4.87; P=.03). Attrition was concentrated before mentoring sessions began. Of recruited patients, 72 of 155 (46.5%) met with a peer mentor and 58 of 155 (37.4%) completed all 5 sessions. Completers were 1.9 years younger (56.4, SD 13.8, vs 58.3, SD 13.9 y) and had Charlson Comorbidity Index scores 0.4 points higher (3.4, SD 2.0, vs 3.0, SD 2.3) than noncompleters. The average engagement score was 18.4 of 50 (SD 20.1). Engagement was higher among men (incidence rate ratio [IRR] 1.33, 95% CI 1.22-1.45; P<.001), Black patients (IRR 1.27, 95% CI 1.16-1.39; P<.001), and patients with dialysis vintage <1 year (IRR 1.22, 95% CI 1.05-1.42; P=.01). Higher Charlson Comorbidity Index scores predicted engagement (IRR 1.04, 95% CI 1.01-1.07; P=.002) and completion (OR 1.37, 95% CI 1.05-1.79; P=.02). Of 79 patients who set goals, 64 (81%) selected communication with the care team; 22 of 54 (40.7%) survey respondents identified this as their main behavior change. Both mentees and mentors identified peer-mentoring sessions as the most valuable component. Among mentee respondents, 55 of 55 (100%) respondents rated the program good or excellent, and 53 of 54 (98.1%) respondents would recommend it. Conclusions: Participants constituted a self-selecting group. Patient-clinician communication was a priority, and participants valued peer mentoring. Digital patient activation intervention strategies to increase reach are needed. Trial Registration: ClinicalTrials.gov NCT03171545; https://clinicaltrials.gov/ct2/show/NCT03171545 International Registered Report Identifier (IRRID): RR2-10.2196/46187

  • Open Access

    Recruitment, Engagement, Experiences, and Self-Reported Changes in a Digital Patient Activation Intervention to Improve Hemodialysis Session Stability: Mixed Methods Study (Preprint)

    by Megan Wickens, Jennifer Bragg-Gresham, Rajiv Saran, Meera Tandan, Marissa Argentina, Dinesh Chatoth, Kelli Collins Damron, Michael Heung, Sarah Krein, Vu Le, et al.

    BACKGROUND When chronic kidney disease progresses to end-stage kidney disease, many patients receive in-center hemodialysis. Although lifesaving, hemodialysis is associated with intradialytic hypotension. Peer mentoring and patient activation support in hemodialysis have not been well studied. OBJECTIVE This Dialysafe study combined patient education and peer mentoring to promote health behaviors supporting prevention of intradialytic hypotension. We aimed to describe recruitment, engagement, completion, and attrition; assess participants’ health goals and self-reported behavior changes; and evaluate satisfaction and recommendations for improvement. METHODS At 10 Dialysafe hemodialysis centers assigned to patient activation, all eligible English-speaking patients were invited. The intervention comprised 5 educational modules and 5 videoconference peer-mentoring sessions. Mixed methods analyses integrated electronic health record and platform data, postintervention surveys, and clinic operations committee notes. An engagement score (range 0-50) summarized intervention use. Descriptive statistics, t tests, chi-square tests, logistic regression, and zero-inflated Poisson regression were used to assess recruitment, completion (all 5 mentoring sessions), and engagement. Qualitative data were inductively coded and integrated through joint displays. RESULTS Of the 639 eligible patients, 155 (24.3%) were recruited. Recruited patients were younger than nonrecruited patients (mean 57.6, SD 13.8, vs 64.3, SD 14.7 y; P<.001); each additional year was associated with 4% lower odds of recruitment (odds ratio [OR] 0.96, 95% CI 0.94-0.98; P<.001). Recruitment was more likely among college graduates (OR 1.70, 95% CI 1.02-2.83; P=.04) and patients with sleep disorders (OR 2.27, 95% CI 1.06-4.87; P=.03). Attrition was concentrated before mentoring sessions began. Of recruited patients, 72 of 155 (46.5%) met with a peer mentor and 58 of 155 (37.4%) completed all 5 sessions. Completers were 1.9 years younger (56.4, SD 13.8, vs 58.3, SD 13.9 y) and had Charlson Comorbidity Index scores 0.4 points higher (3.4, SD 2.0, vs 3.0, SD 2.3) than noncompleters. The average engagement score was 18.4 of 50 (SD 20.1). Engagement was higher among men (incidence rate ratio [IRR] 1.33, 95% CI 1.22-1.45; P<.001), Black patients (IRR 1.27, 95% CI 1.16-1.39; P<.001), and patients with dialysis vintage <1 year (IRR 1.22, 95% CI 1.05-1.42; P=.01). Higher Charlson Comorbidity Index scores predicted engagement (IRR 1.04, 95% CI 1.01-1.07; P=.002) and completion (OR 1.37, 95% CI 1.05-1.79; P=.02). Of 79 patients who set goals, 64 (81%) selected communication with the care team; 22 of 54 (40.7%) survey respondents identified this as their main behavior change. Both mentees and mentors identified peer-mentoring sessions as the most valuable component. Among mentee respondents, 55 of 55 (100%) respondents rated the program good or excellent, and 53 of 54 (98.1%) respondents would recommend it. CONCLUSIONS Participants constituted a self-selecting group. Patient-clinician communication was a priority, and participants valued peer mentoring. Digital patient activation intervention strategies to increase reach are needed. CLINICALTRIAL ClinicalTrials.gov NCT03171545; https://clinicaltrials.gov/ct2/show/NCT03171545 INTERNATIONAL REGISTERED REPORT RR2-10.2196/46187

  • Open Access

    Ambient AI Documentation in Mixed-Language Encounters: A Heuristic Evaluation of Reenacted Mandarin–English and Spanish–English Clinical Conversations

    by Di Hu, Daniel Flores, Lidia Flores, Ruby Chien, Kyle Lam, Emilie Chow, Yawen Guo, Steven Tam, Danielle Perret, Deepti Pandita, et al.

    Ambient AI documentation systems rely on automatic speech recognition to transcribe patient–provider conversations before generating clinical notes. However, little evidence exists on how these systems perform in mixed-language clinical encounters. We conducted a mixed-methods heuristic evaluation of an ambient AI documentation tool using 24 reenacted primary care conversations, including 12 Mandarin–English conversations developed from real-world encounter excerpts and 12 Spanish–English adapted counterparts. Quantitative analyses measured mixed error rate (MER) and code-switching. Overall MER was low, with a median of 4% and less variation in Spanish–English conversations, and 9% in Mandarin–English conversations, but with outliers reaching 67%. The system generally detected language switches reliably, although deletions occurred frequently in Mandarin–English transcripts at switch points. Qualitative analysis revealed transcription errors related to phonetic similarity, automatic translation, clinical terminology recognition, and language-specific challenges. These findings highlight considerations for improving ambient AI tools to support multilingual providers in delivering care for linguistically diverse populations.

  • Open Access

    We Need Granular Sharing of De-Identified Data—But Will Patients Engage? Investigating Health System Leaders' and Patients' Perspectives on A Patient-Controlled Data-Sharing Platform CSCW043

    by Xi Lu, Di Hu, An T. Nguyen, Brad Morse, Lisa M. Schilling, Kai Zheng, Michelle S. Keller, Lucila Ohno-Machado, Yunan Chen

    Patient-controlled data-sharing systems are increasingly promoted as a way to empower patients with greater autonomy over their health data. Yet it remains unclear how different stakeholders, especially patients and health system leaders, perceive the benefits and challenges of enabling granular control over the sharing of de-identified medical data for research. To address this gap, we developed a high-fidelity prototype of a patient-controlled, web-based consent platform and conducted a two-phase mixed-methods study: semi-structured interviews with 16 health system leaders and a survey with 523 patient participants. While both groups appreciated the potential of such a platform to enhance transparency and autonomy, their views diverged in meaningful ways. Leaders viewed transparency and granular control through the lens of informed consent and institutional ethics, whereas patients interpreted these factors as safeguards against potential risks and uncertainties. Our findings underscore critical tensions such as individual control and research integrity. We offer design implications for building trustworthy, context-aware systems that support flexible granularity, provide ongoing benefit‑centered transparency, and adapt to diverse literacy and privacy needs.

  • Top Cited Papers
    Open Access

    Large Language Model–Based Chatbots and Agentic AI for Mental Health Counseling: Systematic Review of Methodologies, Evaluation Frameworks, and Ethical Safeguards

    by Ha Na Cho, Jiayuan Wang, Di Hu, Kai Zheng

    Background: Large language model (LLM)–based chatbots have rapidly emerged as tools for digital mental health (MH) counseling. However, evidence on their methodological quality, evaluation rigor, and ethical safeguards remains fragmented, limiting interpretation of clinical readiness and deployment safety. Objective: This systematic review aimed to synthesize the methodologies, evaluation practices, and ethical or governance frameworks of LLM-based chatbots developed for MH counseling and to identify gaps affecting validity, reproducibility, and translation. Methods: We searched Google Scholar, PubMed, IEEE Xplore, and ACM Digital Library for studies published between January 2020 and May 2025. Eligible studies reported original development or empirical evaluation of LLM-driven MH counseling chatbots. We excluded studies that did not involve LLM-based conversational agents, were not focused on counseling or supportive MH communication, or lacked evaluable system outputs or outcomes. Screening and data extraction were conducted in Covidence (Veritas Health Innovation) following PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) 2020 guidelines. Study quality was appraised using a structured traffic light framework across 5 methodological domains (design, dataset reporting, evaluation metrics, external validation, and ethics), with an overall judgment derived across domains. We used narrative synthesis with descriptive aggregation to summarize methodological trends, evaluation metrics, and governance considerations. Results: Twenty studies met the inclusion criteria. GPT-based models (GPT-2/3/4) were used in 45% (9/20) of studies, while 90% (18/20) used fine-tuned or domain-adaptation models such as LLaMa, ChatGLM, or Qwen. Reported deployment types were not mutually exclusive; standalone apps were most common (18/20, 90%), and some systems were also implemented as virtual agents (4/20, 20%) or delivered via existing platforms (2/20, 10%). Evaluation approaches were frequently mixed, with qualitative assessment (13/20, 65%), such as thematic analysis or rubric-based scoring, often complemented by quantitative language metrics (18/20, 90%), including BLEU (Bilingual Evaluation Understudy), ROUGE (Recall-Oriented Understudy for Gisting Evaluation), or perplexity. Quality appraisal indicated consistently low risk for dataset reporting and evaluation metrics, but recurring limitations were observed in external validation and reporting on ethics and safety, including incomplete documentation of safety safeguards and governance practices. No included study reported registered randomized controlled trials or independent clinical validation in real-world care settings. Conclusions: LLM-based MH counseling chatbots show promise for scalable and personalized support, but current evidence is limited by heterogeneous study designs, minimal external validation, and inconsistent reporting of safety and governance practices. Future work should prioritize clinically grounded evaluation frameworks, transparent reporting of model and prompt configurations, and stronger validation using standardized outcomes to support safe, reliable, and regulatory-ready deployment.


  • Open Access

    Impact of Adverse Childhood Experiences (ACEs) on Mental Health Help-Seeking Among Asian American Adults: Findings from the 2021 California Health Interview Survey

    by Cindy Chwa, Biblia S. Cha, Gloria J. Kim, Judith Borghouts, Elizabeth Eikey, Dana B. Mukamel, Stephen M. Schueller, Margaret Schneider, Nicole A. Stadnick, Kai Zheng, et al.

    Cumulative adverse childhood experiences (ACEs) and psychological distress have been shown to negatively affect mental health across the lifespan. Less is known, however, about how ACEs might impact mental health help-seeking behavior in adulthood, especially among Asian Americans, a population with high exposure to trauma who also face cultural barriers to mental health care. The study’s aim was to evaluate the prevalence of ACEs within a sample of Asian American respondents, the relationship of ACEs and psychological distress with help-seeking, and the association of ACEs with different types of help-seeking behaviors. Data from the 2021 California Health Interview Survey (N = 4,345) were analyzed. Pairwise comparisons examined differences in ACEs and covariates across the seven Asian American subgroups (Chinese, Filipino, Japanese, Korean, Vietnamese, South Asian, and Other Asian American). Multivariable logistic regression analyses evaluated the relationship between ACEs and covariates with professional and emerging digital mental health help-seeking behaviors. The joint effect between ACEs and Asian American subgroup was evaluated for each type of mental health help-seeking. Covariates included psychological distress, gender, age, marital status, insurance, education, English proficiency, self-rated health, and being U.S.-born. Asian American adults with 4 + ACEs were more likely to seek mental health help from primary care practitioners, mental health professionals, and social media/blogs/online forums than respondents with ≤ 3 ACEs. Moderate/severe psychological distress increased likelihood to seek mental health help. No significant interaction between ACEs and Asian American subgroup was found. Findings indicate that Asian American respondents with elevated ACEs and distress are more likely to seek mental health help from professional and emerging digital resources. This suggests a demand for these resources among those with higher needs. Differences across ACEs and distress levels in help-seeking behaviors emphasize the need for culturally tailored interventions and accessible mental health resources to better support the diverse help-seeking preferences within the Asian American community.


  • Open Access

    Adult Digital Mental Health Tool Use From 2019-2022: Findings from the California Health Interview Survey

    by Biblia S. Cha, Jeongmi Kim, Judith Borghouts, Elizabeth V. Eikey, Margaret L. Schneider, Stephen M. Schueller, Nicole A. Stadnick, Kai Zheng, Dana B. Mukamel, Dara H. Sorkin

    Digital mental health interventions (DMHIs) provide tools to seek mental health resources, providers, and facilitate and/or complement in-person treatment. Limited research has examined what factors are associated with DMHI uptake. We used California Health Interview Survey data to examine DMHI use among California adults (2019–2022), estimating three multi-variable logistic regression models to assess if DMHI use to seek mental health support (Model 1), connect with mental health professionals (Model 2), and connect with others with similar concerns (Model 3) varied by psychological distress or sociodemographic variables. We used Wald Chi-square statistics tests to examine reasons for not using DMHIs by the same variables. DMHI use to seek mental health support (OR = 1.6) and connect with professionals (OR = 1.4) increased between 2019–2022. High psychological distress individuals used DMHIs for all three outcomes significantly more than low/no distress individuals (Model 1: OR = 14.9; Model 2: OR = 11.9; Model 3: OR = 13.0). The top reason for not using online tools regardless of distress was in-person treatment. The second reasons were low perceived treatment utility (high/medium distress individuals), and low perceived need (low/no distress individuals). Overall, younger, female, more educated, insured, unmarried, and non-Hispanic White participants were more likely to use DMHIs than older, male, less educated, uninsured, married, and Asian counterparts. Adult DMHI use to seek mental health support and professional treatment increased between pre-pandemic and pandemic years. Many respondents who did not use DMHIs sought in-person support. Future research can examine how to increase perceived DMHI efficacy among people with high/medium distress.

  • Open Access

    Clinical Information Extraction From Notes of Veterans With Lymphoid Malignancies: Natural Language Processing Study

    by Lu He, Matthew R Moldenhauer, Kai Zheng, Helen Ma

    Background: Clinical natural language processing (cNLP) techniques are commonly developed and used to extract information from clinical notes to facilitate clinical decision-making and research. However, they are less established for rare diseases such as lymphoid malignancies due to the lack of annotated data as well as the heterogeneity and complexity of how clinical information is documented. In addition, there is increasing evidence that cNLP techniques may be prone to biases embedded in clinical documentation or model development. These biases can result in disparities in performance when extracting clinical information or predicting patient outcomes. Objective: This study aims to report the development and validation of a cNLP pipeline that extracts clinical information such as performance status, staging, and diagnosis, as well as less common information such as substance use and military environmental exposures, from the clinical notes of veterans with lymphoid malignancies. Methods: We developed a rule-based cNLP pipeline that integrates domain expertise. We tested and compared the performance of the cNLP pipeline on notes from 2 veteran patient cohorts: one from non-Hispanic White veterans and the other from non-Hispanic Black veterans. Results: Overall, our pipeline achieved promising performance on our study data, especially for extracting entities that have standard clinical documentation, such as performance status. We also found that while the pipeline has robust performance across the two patient groups, the false-positive and false-negative rates were significantly associated with race for detecting the primary diagnosis (P=.001 for both); the false-negative rate was significantly associated with race for identifying substance use (P=.02). Conclusions: The system exhibits satisfying and comparable performance for most clinical entities of interest except for (1) the primary diagnosis and (2) substance use. Future work will address the challenges encountered in developing and deploying the cNLP pipeline on the Department of Veterans Affairs data for rare cancers and enhance the performance of cNLP systems to avoid biases.

  • Open Access

    Characteristics Associated With the Use of the Mindfulness Meditation App Headspace in a Large Public Health Deployment: Cross-Sectional Survey Study

    by Judith Borghouts, Elizabeth V Eikey, Cinthia De Leon, Stephen M Schueller, Margaret Schneider, Nicole A Stadnick, Kai Zheng, Dana B Mukamel, Dara H Sorkin

    Background: Mindfulness-based apps can be an effective and accessible resource for mental health support. However, little is known about their use outside of research settings and what user characteristics relate to app use. Objective: This study aimed to examine the characteristics of people who decided to use, not use, or stop using Headspace within the context of a large-scale public deployment, which offered the mindfulness meditation app Headspace as a free mental health resource to community members. Methods: Nearly 100,000 community members received Headspace. All members (N=92,311) received an email inviting them to complete a voluntary and uncompensated survey. In total, 2725 participants completed the survey. The 20-minute survey asked about the use of Headspace, user experience, mental health problems, mental health stigma, and mental health use. Logistic regression models were used to examine relationships between predictors and nonuse, past use, or current use of Headspace. Results: Participants who were still using Headspace at the time of completing the survey (2076/2725, 76.18%) were more likely to experience mental health challenges and distress and make more use of other digital mental health resources (ie, online tools and connecting with people online) than people who were not using Headspace. In addition, current users of Headspace rated the app higher on user experience compared with past users. The most common reasons for abandoning Headspace were that people were already using other strategies to support their mental health (198/570, 34.7%), no longer needed Headspace (73/570, 12.8%), or did not think Headspace was useful (46/570, 8.1%). Conclusions: Results indicate that a person’s mental health challenges, a perceived need for support, and familiarity with digital resources were associated with continued use of Headspace. While the most common reason for not using Headspace was that people were already using other resources, it is important to consider the continuity of mental health support beyond these free programs for those who may not have easy access to other resources. We discuss potential implications of our findings for offering and using apps such as Headspace as a mental health resource, along with factors that influence engagement with this app.


  • Top Cited Papers
    Open Access

    Ambient Listening in Clinical Practice: Evaluating EPIC Signal Data Before and After Implementation and Its Impact on Physician Workload

    by Yawen Guo, Di Hu, Jiayuan Wang, Kai Zheng, Danielle Perret, Deepti Pandita, Steven Tam

    The widespread adoption of EHRs following the HITECH Act has increased the clinician documentation burden, contributing to burnout. Emerging technologies, such as ambient listening tools powered by generative AI, offer real-time, scribe-like documentation capabilities to reduce physician workload. This study evaluates the impact of ambient listening tools implemented at UCI Health by analyzing EPIC Signal data to assess changes in note length and time spent on notes. Results show significant reductions in note-taking time and an increase in note length, particularly during the first-month post-implementation. Findings highlight the potential of AI-powered documentation tools to improve clinical efficiency. Future research should explore adoption barriers, long-term trends, and user experiences to enhance the scalability and sustainability of ambient listening technology in clinical practice.