Awardees of spring 2024 Tier 1 pilot research grants report final project outcomes

Integrating end-user needs and perspectives in the measurement of young adult climate change distress

Investigators
Jessica Acolin, Department of Psychiatry & Behavioral Sciences
Sonya Jampel, Public Health – Seattle & King County
Jennifer Atkinson, Interdisciplinary Arts & Sciences (UW Bothell)
McKenna Parnes, Department of Psychiatry & Behavioral Sciences
Jason Kilmer, Department of Psychiatry & Behavioral Sciences
Brittney Hultgren, Department of Psychiatry & Behavioral Sciences

Project summary
This project performed pilot work to support the development of a brief Climate Distress Questionnaire (CDQ) suitable for measuring climate change distress in epidemiologic surveys. This work represents a foundational step towards facilitating future research on the public health burden and health impacts of climate distress which are currently understudied.

The project occurred over three stages. First, we conducted post-hoc qualitative analysis of a series of meetings with key informants from local public health jurisdictions to identify practitioner demands of climate distress measurement to support public health. Second, we conducted qualitative focus groups with end-users of the proposed CDQ. This work consolidated and reconciled differing disciplinary perspectives on climate distress, identified multi-sectoral demands for climate distress measurement and explored priorities for future work across fields. Participants included environmental epidemiologists, mental health clinicians, public health professionals and community-based key informants. Additional interviews with survey administrators explored the logistic demands regarding the development of epidemiologic surveys and the inclusion of novel measures.

Across disciplinary perspectives, overall consensus highlighted the complex nature of climate distress. Consistent with prior work, respondents described the contributions of individual exposure to extreme weather, social norms, behavioral change and sociopolitical climate to climate distress. Novel findings regarding climate distress measurement included the need for brevity, alignment with existing programs and acceptability by respondents. Respondents also noted the need for future work to address and overcome barriers in order to advance public health, including polarization regarding the term “climate change,” competing demands for limited funds and institutional inertia. Innovative findings regarding future directions for climate distress work also emerged.

This work resulted in a preliminary list of approximately 30 items to measure climate distress among young adults, selected to capture conceptually relevant aspects of climate distress while meeting the demands of cross-sectoral end-users. These items provided the foundation for the third stage: qualitative pre-testing of items with young adults to ensure that young adults perceived items were relevant, clear and easy to understand. A series of focus groups with 20 young adults recruited nationally reduced the preliminary list to 15 items.

Using remaining pilot funds, we have extended project aims to include quantitative psychometric testing of the remaining 15 items. Data collection is ongoing and will include a national sample of young adult college students. Preliminary findings were presented at the Oregon Public Health Association Annual Conference, the Washington State Public Health Association Conference, the American Public Health Association Annual Conference and the American Meteorological Society Annual Meeting. Two research manuscripts summarizing these findings are being prepared for submission to peer reviewed journals.

Identity safety in medicine: a qualitative study to elicit patient perspectives and develop a framework to apply in clinical settings and research

Investigators
Brian Wood, Department of Medicine
Justin Bullock, Department of Medicine
Crystal Brown, Department of Medicine
Reggie Casanova-Perez, Department of Biomedical Informatics and Medical Education

Project summary
The principal aim of our project is to gain a better understanding of patient perspectives on identity safety in healthcare, which we hope will contribute to development of strategies to help individuals feel safe during medical encounters, feel that they can be their true authentic selves with medical staff and feel that they can discuss difficult issues or share sensitive information which will improve the quality of their healthcare. We are accomplishing this through a qualitative study in which we enrolled people with HIV at clinical sites with varied geopolitical environments (sites from a major metropolitan area in Washington, a more mid-size city in Washington, a mid-size city in Montana and a lower population area of Idaho).

We enrolled individuals with HIV because of a relatively high rate of stigma around HIV plus relatively frequent occurrence of intersectionality with other stigmatized conditions such as mental health issues, substance use, incarceration and others. A secondary aim of the study is to create a framework for understanding and studying identity safety in healthcare which will benefit future research efforts.

We feel that we have had a high degree of success with this study. During the grant period we were able to accomplish the following:

• Completed the IRB review and approval process
• Completed a successful community advisory board meeting during which we gained insights and input from 10 community members who gave feedback on our study goals and methods (including recruitment strategy and interview plans)
• Our study team reviewed, revised and finalized the interview guide and focused on integrating advice and feedback from the community advisory board meeting
• We recruited at all planned clinical sites via study flyers and assistance of staff at those clinics who helped distribute flyers and disseminate information about the study
• We successfully completed a total of 17 qualitative interviews (12 participants from Washington, 4 participants from Idaho and 1 participant from Montana) and feel that we reached a relatively diverse group of participants and gathered rich and valuable input from the interviews
• Volunteers who completed interviews were given Tango gift cards as a thank you for their time and participation
• We have uploaded all interview audio recordings to Rev, developed a code book for analysis and are actively working on coding the transcripts

Although the grant period has ended, the study team will continue to code the transcripts and then will move to analyzing the data, developing a manuscript and developing other tools that can be informative to clinics and providers to help promote feelings of safety for patients during medical encounters.

Exploring the Impact of Communication Styles in Health Chatbot using Large Language Models to Support Family Caregivers from Multicultural backgrounds

Investigators
Elina Hwang, Foster School of Business
Stephanie Lee, Foster School of Business
Serena Jinchen Xie, Department of Biomedical Informatics and Medical Education
Rebekah Baik, Foster School of Business
Weichao Yuwen, School of Nursing & Healthcare Leadership (UW Tacoma)

Project summary
Recent advancements in generative large language models (LLMs) have opened up significant opportunities for AI-based chatbots in the healthcare sector. However, there are ongoing concerns regarding their cultural sensitivity. It remains uncertain whether these chatbots can grasp the nuanced cultural differences that are essential for effective communication in healthcare, where building rapport is crucial for patient compliance, satisfaction and overall outcomes. Our project aims to establish a foundation for developing health chatbots that can adapt to various cultural communication styles.

Our study had two aims. First, we aimed to conduct a comprehensive literature review on cultural preferences in communication styles, specifically within healthcare. This review will help identify factors that can assist in adapting AI-based chatbots to be more culturally sensitive (Aim 1). Second, we proposed to develop prototypes of AI-based therapy bots that incorporate a variety of communication styles and gather feedback from participants of diverse racial and ethnic backgrounds (Aim 2). We expect these findings to enhance the cultural sensitivity and effectiveness of therapy chatbots for patients from various backgrounds. To achieve our goals, we have assembled an interdisciplinary team of faculty and doctoral students from the business school, the school of medicine, the college of arts and sciences (department of communication) and the school of nursing and healthcare leadership.

Our team has satisfactorily completed both aims. For Aim 1, we conducted a thorough literature review on health communication styles, drawing from various fields, including communication, psychology, anthropology, linguistics, human-computer interaction (HCI) and computer science. From this review, we identified four communication style dimensions that could influence the effectiveness of AI-based chatbots designed for caregivers from diverse cultural backgrounds. These dimensions are authoritative, sentimental, informal and concise.

For Aim 2, we developed prototypes of therapy chatbots and evaluated their effectiveness. Based on the communication styles identified in Aim 1, our team created several chatbot prototypes that incorporate these different styles. We utilized a multi-agent system with large language models (LLMs) and Langchain to engineer prompts tailored for each communication style. We validated these prototypes through a controlled experiment, where we assessed human subjects’ evaluations of the chatbots’ communication styles. The experiment demonstrated that the prototypes successfully produced noticeable differences in communication styles, with the adapted responses receiving significantly higher ratings for style-specific characteristics compared to the unadapted responses. Furthermore, we validated these prototypes using dictionary-based measures of linguistic markers, specifically the Linguistic Inquiry and Word Count (LIWC 2024). These results indicate that LLM-based therapeutic chatbots can be effectively tailored to exhibit distinct communication styles, establishing an essential foundation for the current proposed study. Our paper summarizing the findings from this Tier 1 project has been accepted to CHI 2025 as Late-Breaking Work.

Clinical and structural approaches to addressing substance use disorder in HIV care in Zimbabwe: Formative work

Investigators
Helen Jack, Department of Medicine
Malinda Kaiyo-Utete, University of Zimbabwe
Theresa Matson, Kaiser Permanente Washington Health Research Institute
Kudakwashe Takarinda, OPHID
Emily Williams, Department of Health Systems and Population health

Project summary
The aims of this study are to validate 1) a single-item substance use disorder screening tool (1) and 2) the AUDIT-C, a 3-item alcohol use disorder screening tool (2) against a clinical interview (CIDI diagnostic interview) (3) in a sample of 1064 patients in HIV care in Zimbabwe. This study will also provide the first prevalence estimate of alcohol and substance use in HIV care and one of the first prevalence estimates in the country overall. This is a collaborative project between the Organization for Public Health and Development (OPHID), a Zimbabwean non-profit organization and University of Washington (UW). Data collection for this study took place in Bulawayo and Gwanda districts in southern Zimbabwe. The overall study was supported by the UW Behavioral Research Center for HIV (BIRCH); the UW Addictions, Drug and Alcohol Institute (ADAI); and the UW Population Health Initiative (PHI). PHI specifically supported expansion of the study to include the single-item SUD screening tool.

The study took place in 8 HIV clinics in Bulawayo (urban) and Gwanda (rural) districts. People living with HIV presenting for care were randomly selected to participate in the study. A trained data collector administered a sociodemographic questionnaire, the full AUDIT (of which the AUDIT-C is the first 3 questions) and the single-item substance use screening tool to people who consented to participate. The participants then met with a nurse who administered a structured clinical diagnostic interview (the CIDI) for alcohol and substance use disorders then completed timeline followback, a standardized method for eliciting the number of standard drinks that someone has consumed in the past 30 days. In preparation for future work, we asked participants with alcohol or substance use disorder what types of treatment they would find most helpful. During analysis, receiver operator curves and areas under the curve were created for the alcohol and substance use screening tools, comparing their performance to the diagnostic interview and the sensitivity and specificity at each threshold was calculated.

We enrolled 1064 people living with HIV in the study, the a priori target sample size. The sample was 62% female, 36.1% were between the ages of 18 and 40 and 64.0% had a secondary school level of education. 40.7% of the sample stated that they “often” or “always” lacked resources for basic necessities such as housing or food in the past 12 months. We found a high prevalence of alcohol use disorder based on the diagnostic instrument with prevalence twice as high among males compared to females. We found a markedly lower prevalence of substance use disorder and the most common substance used was cannabis. The AUDIT-C had excellent performance for predicting alcohol use disorder and the single-item substance use screen had good performance. The vast majority of people with alcohol or substance use disorder wanted to reduce their use and the treatment that they thought would help the most was having a job or other income-generating activity. Of more standard substance use treatments, counseling from a lay health worker or psychologist were most commonly endorsed and there was less interest in counseling from a peer.

A community co-led epidemiologic study of methamphetamine use patterns and associated factors among cisgender men and transgender people who have sex with men

Investigators
Matthew R. Golden, Department of Medicine
Christopher Archiopoli, Strength Over Speed
Mike Barry, Department of Epidemiology
Jack Harlan, Peer Seattle
Barbra A. Richardson, Department of Biostatistics

Project summary
In this research, we aimed to recruit cisgender men and transgender people who have sex with males (CMTSM, N >2,500) for a web-based biobehavioral survey previously developed by our research team. Our scientific aims were to (1) identify methamphetamine (meth) use classes (e.g. experience types) in this population and (2) compare characteristics of these groups to one another. The public health significance and expected impact of this research was to understand the diversity of meth use in this population and discern which characteristics are associated with the most morbid manifestations of meth use. Additionally, this work was guided by Community-Based Participatory Research principals and included a formal partnership with the Strength Over Speed (SOS) program of Peer Seattle. We proposed hosting two public meetings with the SOS group for the research team to receive input at various stages of the study from CMTSM who use meth in the Seattle Area.

Results and findings for Aim 1: From June-November 2024, the survey was attempted 2,579 times. Among these, 1,720 complete deduplicated surveys comprised the analytic sample. In this sample, 428 (25%) reported ever using meth. Among those who ever used meth, 403 (94%) provided sufficient data to be included in a stimulant use disorder (StUD) and meth-related adverse outcomes (MRAO) analysis (i.e. the meth use class analysis). In this sample, 269 (67%) reported evidence of developing StUD/MRAO. On average, CMTSM developed the first evidence of StUD/MRAO 6 years following first meth use. The most common manifestations of StUD/MRAO among those who developed it were disruptions in relationship/school/work (50%) and requiring medical attention due to meth use (20%). We estimated a 13% lifetime prevalence of StUD/MRAO in the CMTSM population.

Results and findings for Aim 2: CMTSM who ever used meth were more likely than those who did not to (1) be living with HIV (2) have experienced housing instability/homelessness and (3) have ever exchanged sex for money, drugs or a place to stay. CMTSM who developed evidence of StUD/MRAO were more likely than other CMTSM who used meth to (1) have received an HIV diagnosis (2) have experienced >1 of housing instability/homelessness, sex exchange or inability to meet basic needs. We held two community meetings at SOS; these were well-attended and supported the development of analyses and the interpretation of our findings.

Paid Family Leave and Parent Mental Health: Evidence from Administrative Data

Investigators
Tom Lindman, Evans School of Public Policy & Governance
Anirban Basu, School of Pharmacy
Anjum Hajat, Department of Epidemiology
Heather D. Hill, Evans School of Public Policy & Governance
Jessica Jones-Smith, Department of Health Systems and Population Health

Project summary
Research indicates that when workers have paid time off to care for themselves and others, they have more stable economic circumstances and improved physical health (Bartel et al. 2023). However, less is known about the effect of access to paid leave on mental health particularly among parents using leave around a birth. Although recent studies on U.S. state-level paid leave programs find reductions in mental distress for new parents, these studies compare mothers in states with and without paid leave programs, an approach that can be biased by unobserved differences between states (Bullinger 2019; Doran et al. 2020; Irish et al. 2021). In addition, all studies in the U.S. use self-reported mental health outcomes that are not closely linked to specific conditions or mental health care. Using data from the Washington Paid Family and Medical Leave Program (WA PFML) linked with insurance claims from the Washington All Payer Claims Database (APCD), we evaluated the impact of paid leave on the likelihood of birthing parents submitting insurance claims related to depression and anxiety following the birth of a child. Our study leverages a plausibly random feature of eligibility rules—the program work-hour eligibility requirement—using regression discontinuity designs. We separately estimate the impact of paid leave on parents submitting insurance claims related to (1) depression and (2) anxiety in the calendar quarter following paid leave use.

Results suggest that access to Washington’s Paid Family and Medical Leave program reduces the likelihood of birthing parents submitting medical claims for depression by approximately 26.5% (-3.18 percentage points, SE 1.30) with much larger impacts in the treatment-on-the-treated analysis focused on leave users. We do not find statistically significant impacts on anxiety diagnoses. Our sample of working parents has low earnings and high rates of Medicaid use, suggesting that state paid leave programs positively affect the mental health among working parents with low incomes and/or unstable employment. This study is the first in the U.S. to evaluate paid leave’s impact on measures of mental health care or diagnosis and speaks directly to the efficacy of state paid leave programs at supporting the mental health of working parents.

Ensuring Equality: Language Access in Civil Protection Orders

Investigators
N. Jeanie Santaularia, Department of Epidemiology,
Alice M. Ellyson, Department of Pediatrics
Avanti Adhia, Department of Child, Family, and Population Health Nursing
Leticia Figueroa, King County Prosecuting Attorney’s Office

Project summary
Over 50% of women experience some form of interpersonal violence (e.g. domestic violence, sexual assault, harassment) in their lifetime. A civil protection order (CPO) is one critical intervention to support victim-survivors. CPOs are legal remedies meant to reduce the risk of threat and harm to a victim-survivor; however, equitable access to CPOs is limited. Individuals can encounter language barriers inhibiting effective communication at many points in their help-seeking and CPO process. The overall goal of our Tier 1 project was to describe language access challenges in CPO processes across counties in Washington State. The aims were to: (1) characterize language access of written materials (e.g. webpages) required to file and receive a CPO for interpersonal violence and (2) assess language access in verbal settings (e.g. interpretation services in hearings) using court observations.

We worked closely with our community partners at the King County Prosecuting Attorney’s Office and Washington Administrative Office of the Courts to receive input and feedback on all aspects of the project. We had two graduate students assisting on the project: an MPH student in Epidemiology and a PhD student in Health Services. We met weekly as a research team to check in and make progress.

Aim 1: We created a data abstraction tool, piloted the tool with a small number of counties, double coded a subset of counties with the tool (i.e. had two independent coders code the same webpages using the tool and comparing coding) and collected data from all county webpages. We systematically collected data to evaluate search results and webpage resource availability across the 38 counties with CPO webpages for the six most commonly spoken languages in the state other than English (Spanish, Russian, Chinese, Korean, Vietnamese and Tagalog/Filipino) to better understand the resources available to English language learners. We conducted web searches to assess whether two search terms (“protection order” and “restraining order”) directed individuals to a county CPO webpage and assessed the translation availability and features of these CPO webpages. Across the six languages, the search term “protection order” returned the county’s CPO webpage for 3–79% of counties and “restraining order” for 0–76% of counties. Between 53%–63% of county CPO webpages had a translation option available where Spanish was the most common and Chinese the least common. Most webpages used Google Translate and while over half of county CPO webpages offered a translation option, many elements (e.g. instructional PDFs, links, images) did not translate. We identified several possible solutions to these language access gaps, such as linking to translated versions of documents from the Washington State Administrative Office of the Courts website, embedding non-English search terms into webpages’ HTML code and using human translations or verifying quality of automatic translations.

Aim 2: We gained access to CPO hearings in King County that had language access needs and created a standardized data abstraction tool. We conducted court observations (n=50) by joining hearings via Zoom or accessing digital recordings of hearings held between December 2024 and March 2025. Hearings were eligible if they involved first-time cases and an interpreter was requested for at least one party. Our data abstraction tool included a mix of closed- and open-ended fields and had measures on hearing and CPO type, petitioner and respondent presence, interpretation requests and presence, hearing outcome, victim advocate presence, lawyer presence, judicial officer type, behavior and actions. Interpretation quality was assessed for a subset of Spanish interpreter cases observed by a native Spanish speaker on our team. Among the 50 CPO hearings, 38% resulted in a granted CPO, 36% in a continuance, 12% in a dismissal, 8% in a denial. An interpreter was requested for the petitioner in 89% of cases and present when requested in 92% of cases. A total of thirteen different languages were requested with Spanish being the most common. While 68% of hearings included an opening statement by the commissioner, not all opening statements were interpreted (83%). In open text fields, observers noted that there were hearings where evidence was not being considered due to lack of interpretation. Approximately 40% of hearings had an advocate present for the petitioner, 11% had a lawyer present for the petitioner and 23% had a lawyer present for the respondent. Overall, these exploratory analyses of observations of CPO hearings in King County, the most well-resourced county in the state, highlighted opportunities for improving language access in CPO hearings to ensure survivors of interpersonal violence are able to receive adequate legal protection.

Dissemination: We have disseminated our findings in several ways. First, we submitted a manuscript summarizing the Aim 1 findings on language access in CPO webpages to a peer-reviewed journal. We presented to the Washington Domestic & Sexual Violence Committee and the broader Washington State Supreme Court Gender and Justice Commission where members of the Interpreter and Language Access Commission were also invited. These commissions include relevant partners from across the state (e.g. judicial officers, court administrators) who were interested in our empirical findings.

Exploring stakeholder perspectives on the use of controversial datasets in population health research

Investigators
Stephanie M. Fullerton, Department of Bioethics & Humanities
Sarah C. Nelson, Department of Biostatistics
Jacklyn Dahlquist, Department of Bioethics & Humanities
Stephanie M. Gogarten, Department of Biostatistics

Project summary
The aims of this pilot project were to (1) perform a scoping literature review to identify relevant scientific and scholarly discussions of use of a controversial dataset (the Human Genome Diversity Project, or HGDP) in population health research and (2) conduct qualitative interviews with key informants knowledgeable about the historic and/or current relevance of HGDP data for population health research and which communities may be impacted by its ongoing use.

For Aim 1, we completed the literature review and are now preparing a manuscript for publication. Our review was based in reverse citation searches of six key HGDP resource papers, identifying 1,848 manuscripts published since 2010 (as publications prior to 2010 were analyzed in a previous publication). We manually screened abstracts then full text to exclude 631 publications that were either not primary research reports (e.g. commentaries and reviews) or did not directly use the HGDP data. The final collection of 1,217 articles was then assigned to one of four categories of HGDP research use: health-related study, non-health related phenotype, no phenotype (e.g. population genetics study) or methods development. Relevant metadata, including dates of publication, geographic locations of study authors and funding source, has also been extracted and is being analyzed for the planned manuscript. A primary finding from this analysis is that the HGDP data continue to be widely used largely by investigators located in the United States or Europe and despite early controversies surrounding the project. Most published uses were in the realm of population genetics and evolutionary biology, in line with the original goals of the project. We also observed both health-related and commercial uses of the data, which were not anticipated at the time the sample collection was established. Importantly, because HGDP data are also embedded in several prominent databases and other resources without full attribution, we believe this review captures only a fraction of all secondary uses of those data.

For Aim 2, we completed Zoom-based interviews with a sample of bioinformaticians, geneticists, Indigenous scientists, social scientists and historians about the current utility of and potential concerns regarding ongoing use of the HGDP data. To identify key informants for recruitment, we began with social scientists and researchers who had written about the HGDP or used HGDP in their work (including from our Aim 1 literature review); we then used snowball sampling to identify others with related interests and experiences. In total, we successfully recruited 20 and spoke with 18 different scholars based in the United States, United Kingdom or Latin America. These interviews, which were primed by a brief introduction to the findings of the Aim 1 literature review, focused on key informants’ understandings of the early history of the project, their thoughts on the benefits and risks involved in ongoing uses and ended by soliciting responses to the currently hypothetical possibility of researcher engagement with the Indigenous and other communities whose members contributed data to the project. All interviews were recorded with permission, transcribed and redacted. We are now in the process of qualitatively analyzing the transcripts and plan to summarize our findings in peer reviewed publications and at professional meetings. A preliminary finding is the striking heterogeneity of views among the different stakeholders we interviewed and varying familiarity with the history and early controversies of the HGDP collection. While some regard the data as fundamentally unique and of high scientific utility, others suggested that donor communities could hold concerns about the ongoing use of their members’ information particularly so many decades after sample collection. Most regarded the possibility of community engagement as practically infeasible or undesirable.

Building and evaluating AI-augmented treatment support for individuals with tuberculosis

Investigators
Martine De Cock, School of Engineering and Technology (UW Tacoma)
Sarah Iribarren, Department of Biobehavioral Nursing and Health Informatics
Fernando Rubinstein, Institute for Clinical Effectiveness and Health Policy
Daniil Filienko, School of Engineering & Technology (UW Tacoma)
Weichao Yuwen, School of Nursing & Healthcare Leadership (UW Tacoma)

Project summary
Tuberculosis (TB) remains one of the world’s deadliest diseases, killing three people per minute. Each year an estimated 10.6 million people develop TB and 1.5 million die from this curable disease. A critical challenge to meeting the World Health Organization’s targets to end TB is ensuring patients’ adherence to the 6 to 9-month treatment regimen. Innovative patient-centered approaches are needed to empower individuals with TB and frontline healthcare workers to meet the end TB goals. The TB Treatment Support Tools (TB-TST) intervention, iteratively developed by a UW team with patients and clinical partners, combines a Spanish-language mobile application for patient-centered support, monitoring and communication with a drug metabolite test. While helpful, following up on patients through Digital Adherence Technologies (DATs) still requires significant human involvement from healthcare providers. A pragmatic clinical trial across four public hospitals in Argentina demonstrated TB-TST’s effectiveness but challenges such as 24/7 support needs and workforce burden highlighted the need for further innovation.

Large Language Models (LLMs), a new kind of powerful AI model such as the GPT series from OpenAI, offer a promising advancement to scale patient support and reduce provider workload. They can answer medical questions, provide treatment guidance and enhance patient engagement, potentially transforming TB care delivery. Yet, no Spanish-language chatbots have been developed specifically for TB treatment support. In this pilot project (1) we developed an LLM-powered TB treatment support AI agent based on pragmatic clinical trial data of the TB-TST intervention tool and patient needs and (2) we evaluated the AI agent in terms of linguistic appropriateness, empathy, medical accuracy and privacy.

AI agent development. We iteratively developed a series of GPT-based conversational models designed to be deployed as human-supervised treatment supporters for Spanish-speaking individuals with TB. These models can answer questions such as “Do I take all of the pills together or some in the morning and some in the evening?” or “Is it normal to continue having night sweats?” The models were created using different prompt engineering techniques, few-shot learning and Retrieval Augmented Generation (RAG). To enhance domain-specific responses, we integrated publicly available TB guidelines and medication suggestions including from the WHO and the CDC; previous TB trial messages; and manually crafted dialogue samples mimicking real conversations to be used by the model. To safeguard patient privacy, we applied differentially private text sanitization to trial messages used in few-shot prompts. We also developed a multi-agent classification sequence in which the first LLM agent uses a classification prompt to identify whether a user query is empathy-seeking or information-heavy based on examples curated by clinical experts. Queries classified as empathy-seeking are directed to an empathy-optimized agent while information-heavy queries are routed to a fact-focused RAG agent. This modular setup enables the system to provide context-appropriate responses while leveraging the strengths of each specialized agent.

AI agent evaluation. A considerable amount of effort went into assessing the quality of the developed models. We created rubrics to evaluate the conversational agents in terms of linguistic appropriateness, empathy and medical accuracy. After multiple rounds of internal evaluation with our bilingual research team (physician, Argentinian researcher, nurses) we selected the top-performing models for external evaluation by ten Spanish-speaking TB experts. Using predefined patient personas and common queries identified in the clinical trial, evaluators engaged in structured chatbot interactions and assessed performance across nine dimensions based on the QUEST framework for human evaluations of LLMs (Quality of Information, Understanding and Reasoning, Expression Style and Persona, Safety and Harm and Trust and Confidence) (rated 1–5 with 5 = highest). Empathy was rated separately (0–2 scale). Qualitative feedback was analyzed thematically.

Findings demonstrated the feasibility of LLM-powered chatbots for TB treatment support, with all three top-performing models scoring ≥4 across evaluation dimensions. The FS model had the highest overall rating (4.81, SD 0.47) and empathy score (1.82, SD 0.49). Key challenges included overly technical language, difficulty distinguishing between latent and active TB and occasional incomplete responses.

Assessing readiness for newborn pulse oximetry screening in Northern Ghana

Investigators
Nelangi Pinto, Department of Pediatrics
Kristin Beima-Sofie, Department of Global Health
Alhassan Abdul-Mumin, Tamale Teaching Hospital
Donna Denno, Departments of Pediatrics and Global Health
Abdulai Abubakari, University for Development Studies
Rafiuk Cosmos Yakubu, University for Development Studies

Project summary
The goal of this project was to conduct a mixed methods assessment grounded to inform community co-designed implementation of newborn pulse oximetry screening (POS) in the Northern Region of Ghana. Our specific aims were: 1) to assess the current healthcare organizational capacity in the Northern Region for newborn POS and cascade care for positive screens and 2) to explore POS acceptability, appropriateness and feasibility across facility levels and use our findings to co-develop/adapt interview guides and site assessments for use across the Northern Region.

We collected abstracted high level health care data and completed detailed facility surveys at 17 health facilities in Northern Ghana (Table 1). This included 1 teaching/referral hospital, 1 regional hospital, 14 district hospitals and 2 primary health centres. We elicited important data regarding capacity for POS implementation including the availability and use of neonatal pulse oximetry in almost all health facilities except for one maternity home. Pulse oximetry for neonates was not universal but used as needed and all but 2 facilities had oxygen available for neonates if required. In general results showed a high level of resource capacity for POS screening but not for downstream testing for positive results. Most facilities also had the capacity for specialty consultation with pediatric cardiologists and transfers though transportation was a constrained resource with most relying on a combination of ambulance and private car transfers.

We performed focus groups and key informant interviews with providers and staff from 4 health facilities representing all levels of care in Northern Ghana (teaching hospital, regional hospital, district hospital and primary health center/maternity home). Between 3/25 and 5/25 we conducted 13 individual interviews and 4 focus group discussions (FGDs) with healthcare providers (N=42: RNs=17, midwives=15, doctors=8, administrators=2). Interviews and FGDs lasted an average of 47 minutes, were conducted in English and transcribed for analysis. Transcripts were analyzed using a thematic analysis approach to characterize current newborn assessment processes and identify key influences on acceptability and feasibility of future POS implementation. Preliminary analysis results identified that providers are highly motivated to implement interventions that improve neonatal mortality but work in busy clinics, lack access to clinical specialists, rely on personal communication networks to facilitate referrals of high-risk patients and face challenges navigating patients’ needs and preferences when providing neonatal care. Overall, providers found POS acceptable and were enthusiastic about the potential for POS to improve infant outcomes, believed the intervention was effective and that there would be motivation to adopt POS if training was well conducted and providers saw a positive impact on infant morbidity and mortality. Providers also believed POS was feasible to implement because it was non-invasive, quick to perform and would be easy to train providers on results interpretation. Providers identified challenges affecting their beliefs about feasibility including lack of access to pulse oximeters, overcrowding in labor and delivery settings, staff resistance to change and new procedures and high rates of staff turnover. Participants specifically noted concerns about future pulse oximeter availability and functionality if initial access were contingent on research study investment rather than broader national policies or guidelines. Providers also expressed concerns that limited POS acceptability included how to deal with positive screening results when the recommended treatment or follow-up diagnostics were not available or were not acceptable because of high costs. To ensure successful POS implementation, providers recommended engagement of leadership at national and local levels, repeated sensitization and training of all staff and integration of POS screening into already existing neonatal care guidelines and standardized neonatal assessment forms.

Sembrando Salud: Designing Healthy Eating Futures for Latinx Adolescents

Investigators
Carla Castillo, Department of Human Centered Design & Engineering
Sean Munson, Department of Human Centered Design & Engineering
Julie Kientz, Professor, Department of Human Centered Design & Engineering
Pia Chaparro, Department of Health Systems and Population Health

Project summary
Adolescence is a pivotal period when teens develop eating agency—the ability to make and carry out independent food choices—and eating patterns formed during this stage can carry into adulthood. Teens’ food choices are shaped by their environments (e.g. home and school), social influences (e.g. peers and family), cultural context, food availability and family routines. This transition period is especially important to understand among Latinx adolescents as Latinx communities experience higher risk of obesity and diet-related conditions (e.g. type 2 diabetes). Despite this importance, limited research examines how Latinx adolescents experience food decision-making as they gain autonomy, particularly in rural contexts. This project aimed to (1) understand how Latinx adolescents in rural Washington, specifically Lower Yakima Valley, experience food choices as they gain autonomy (Aim 1) and (2) engage teens, caregivers and community partners in co-design to identify challenges and opportunities that could inform community-informed interventions that reflect their needs and contexts (Aim 2). We engaged 7 caregiver–teen dyads and 5 teens participating independently, recruited through schools and community organizations. We also engaged 5 community partners from community- and school-based organizations, including family-serving, healthcare, youth and school district settings in roles such as registered dietitian and program coordinator.

For Aim 1, we conducted food diaries and in-depth interviews with teens and caregivers. Teens shared their eating experiences, goals and influences on food choices and reflected on their diary entries. Some teens described goals related to healthier eating while others focused on weight-related goals; a few used tracking apps to support these goals. Mapping teens’ responses to the social ecological model, we found that at the individual level, teens’ eating was influenced by food preferences, cravings, perceptions and relationships with food (e.g. body image concerns or disordered eating), dietary sensitivities (e.g. lactose intolerance) and physical reactions (e.g. nausea). At the family level, influences included specific family members, home food availability, social gatherings (e.g. parties) and growing their own fruits and vegetables. At the community level, influences included school food and nearby grocery stores as well as broader factors such as money, friends and social media. While influences varied by teen, social media and school food were most often described as barriers to eating in desired ways.

For Aim 2, we held co-design sessions with teens, caregivers and community partners. In teen sessions, participants generated individual ideas for addressing challenges they experience, including meal planning, portion control and improving access to healthier foods. Teens with eating challenges (e.g. eating disorders or difficult relationships with food) described solutions such as supportive eating settings, food diaries, meal planning and counseling. In these same sessions, teens emphasized parental support that felt helpful—autonomy, validation, emotional support and privacy—and expressed resistance to control, judgment, comparison and sharing their experiences without consent. We also held individual caregiver co-design sessions in which caregivers discussed their teens’ eating patterns and food preferences and ways they support them. Caregivers additionally discussed school-based supports, including creating space for food gardens and communicating concerns to schools about food quality. Community partners reviewed ideas from these sessions and shared insights from their experience in the area, including existing family programs and implementation challenges. A registered dietitian in a school district talked about the challenge of meeting nutrition standards while still serving food that students want to eat. In a final survey, participants rated the helpfulness of generated ideas and selected among five program concepts: (1) food education, (2) cooking and meal skills, (3) strengthening caregiver–teen relationships around food, (4) a teen food support group and (5) a food-support app. Teens most often selected the app and cooking/meal skills while caregivers most often selected cooking/meal skills and food education; across both groups, some participants also selected the caregiver–teen relationships program. These preferences clarify where teens and caregivers see the most useful support and can guide future intervention design.

While this work focuses on improving teen eating, it is also important to highlight teens’ experiences with school food in Lower Yakima Valley. Teens frequently described concerns about school food quality, particularly food that seemed expired. Survey responses echoed these experiences, with over half of teens reporting encountering expired-seeming school food in the past three months, including some who reported it happening more than once. These experiences may influence eating behavior as many teens reported avoiding school food at least sometimes due to worry about expiration. When faced with questionable food, teens most often threw it away or stopped eating rather than asking staff for a replacement, suggesting limited perceived options for addressing the issue. One teen shared, “Sometimes the milk is sour or slightly chunky, but I still drink it anyways; however, I usually eat around fruits and other food that seems rotten or expired.” These findings point to a concrete opportunity for schools to address food quality concerns and support safe eating environments for teens.

Building Reparative Connections and Community Health Strategies across Sites Impacted by Nuclear Weapons Development

Investigators
Sasha Su-Ling Welland, Department of Gender, Women & Sexuality Studies
Holly M. Barker, Department of Anthropology

Project summary
This project advanced collaborative research about the environmental and health impacts of nuclearism on frontline communities created through weapons development and testing, mining and manufacturing and waste (mis)management. From March 5–7, 2025, we held a workshop for current and future leaders from impacted communities to discuss their experiences, develop trust and gauge interest in potential collaboration. Workshop participants traveled to UW from Majuro (Marshall Islands), Yakama Nation, Livermore CA, Portland OR, Minneapolis MN, Lacey WA and Seattle to connect, learn, think and brainstorm together.

In addition to the project PIs, the group members included:

• Community leaders addressing environmental health and restoration: Laurene Contreras (Yakama Nation Environmental Restoration & Waste Management), Moriana Phillips (Republic of Marshall Islands Environmental Protection Agency) and Lyla Lemari (Marshall Islands researcher at Lawrence Livermore National Laboratory)
• Graduate students and public health leaders: Johnny Buck (Wanapum/Yakama, Human Centered Design & Engineering) and Leimamo Wase (Marshall Islands, Anthropology)
• Undergraduate researchers and leaders: Torin Burns and Anika Palep (UW Students for Nuclear Justice)
• Artists: Lauren Iida (Japanese American public artist, WA State Arts Commission) and Johnny Buck (Wanapum/Yakama)
• Non-profit organization managers and strategists: Liz Mattson and Miya Burke (Hanford Challenge)
• Published scholar and community leader: Michael Buck (Wanapum/Yakama)

A loosely structured agenda gave people time for discussions within and across smaller groups with an emphasis on establishing trust, gaining local knowledge about sites and communities siloed from one another by government secrecy and identifying opportunities for future collaboration. The group’s intergenerational composition reflects the importance of connecting the next generation of community leaders, especially students enrolled at UW, with knowledge holders from their communities as essential aspects of their learning. The proposal for this project stressed the importance of relationship-building as a critical component of growing and extending conditions for community-directed change; and the workshop achieved that goal by bringing impacted Marshallese, Yakama and Wanapum community members together for the first time to reflect upon a shared history of nuclear harm and dispossession from land, water and cultural resources. Our findings align with Indigenous research methods that emphasize the importance of relationship building rather than extractive knowledge as an essential foundation for learning and possibility.

Without question, the major result to emerge from this workshop was the creation of a new and extremely close “nuclear family,” a term we coined to identify ourselves in a familial way. Our group hypothesizes that we were able to build a sense of family in just a few short days because of familiarity with and similarity between the issues experienced in “kin” communities, including removal from the land and waters of home, high incidences of cancer, long-term environmental contamination and complexity and intractability of interactions with governmental agencies. We arrived as individuals committed to community-led efforts to address nuclear justice but we left feeling and behaving like a family. The result, which exceeded every expectation for this workshop, was seeing that a group of people with vastly different ages, languages, knowledges and geographies could unite quickly and deeply over shared interests. This result supports our understanding about the nuclear realm: that the strongest possible defense system to nuclear violence is building community that centers care and concern for one another, for our relations to land, plants and animals and to the future.

Another key finding is awareness about the many similarities among frontline communities and how cathartic and healing it felt to share the pain with people who understood and could relate to it, including the profound grief of losing beloved family members and community leaders. The project’s model for sharing and relationality revealed the disproportionate impacts of the nuclear military industrial complex on Indigenous communities, creating conditions to expand allyship, enter collaboration and center possibilities for a healthy future that draw on the strengths of culture and community. Immediate outcomes include:

• Creation of a bi-monthly video call to keep the nuclear family connected. All workshop participants expressed enthusiasm to not only continue our conversations but to expand “family” and health strategies for the future.
• New artistic collaboration between Lauren Iida and Johnny Buck interweaving nuclear impacts on Japanese, Japanese American, Yakama and Wanapum communities, initiated in the Burke Museum’s Artist Studio in May 2025 and exhibited at Seattle Center and the annual From Hope to Hiroshima ceremony at Green Lake honoring the victims of the atomic bombings of Hiroshima and Nagasaki (2025 marked the 80th anniversary).
• A poem, “Love Letter to Our Mother Earth,” by Johnny Buck selected by independent juror Sierra Nelson as winner of the Hanford Challenge 2025 Hot Poetry Contest.
• A new archival collaboration between Torin Burns, who will begin the UW MLIS program in the fall, and Lauren Contreras to identify declassified documents from the UW Radiation Ecology Laboratory in the Libraries Special Collections of interest to Yakama Nation Environmental Restoration & Waste Management; Torin will scan and upload these documents to the digital archive “Archiving for Community Use: Digital Collections of the United States Nuclear Legacy.”
• A commitment by workshop participants to show up for one another at conferences, events (such as the August 2025 From Hope to Hiroshima event), student organizations and classes and to share food and laughter as part of healing, resistance and survivance.

Individual and community-level effects from climate change driven heat and wildfire smoke co-exposures among a Washington state agricultural community

Investigators
John Flunker, Department of Environmental & Occupational Health Sciences
Coralynn Sack, Departments of Environmental & Occupational Health Sciences and Pulmonology
June Spector, Department of Environmental & Occupational Health Sciences
Pablo Palmández, Department of Environmental & Occupational Health Sciences;

Project summary
Our project seeks to examine the impact of extreme heat and wildfire smoke (WFS) co-exposures on the respiratory health of an at-risk agricultural population in Yakima County, Washington. Yakima County is the state’s top apple producing county and also supports the state’s largest population of tree fruit workers. Residents of Yakima County and many Eastern WA agricultural regions experience WFS and extreme heat co-exposure events that are increasing in frequency and duration. Tree fruit workers often must endure these exposures while engaging in demanding physical labor yet with limited access to cool clean air. Little is known about the impacts of such exposures on the respiratory health of tree fruit workers and the communities in which they reside. We hypothesize that worker respiratory health will decline following acute heat (≥90°F) and/or WFS exposure events (PM2.5 ≥100 µg/m3) with the most pronounced reductions in respiratory health resulting from dual heat and WFS exposure events in comparison to no/low exposure conditions.

During the award period, we met our proposed goals including establishing a cohort of tree fruit workers and assessing WFS and heat exposure associated respiratory health. Specifically, we recruited 35 workers from three apple farms in Yakima County, Washington, obtaining sequential measures of respiratory health following heat and WFS exposure events from workers at their worksite. Measures of respiratory health were taken at baseline and within two days of the onset of an exposure event by a bilingual multi-disciplinary team of UW researchers. Measures included self-reported heat-related illness and respiratory symptoms, spirometry and exhaled Nitric Oxide (i.e. airway inflammation). All measures of worker lung health were obtained in person at the worksite. There were no acute exposure wildfire smoke days in August and September 2024 with relatively few acute heat exposure days. Furthermore, no dual acute heat and wildfire smoke events occurred during the study period. As such, we were only able to measure respiratory health following either singular high heat or lower intensity WFS exposure events.

We successfully retained orchard worker participants for the two-month study duration. However, one farm (approximately ten workers) was not able to participate in our final (third) set of measurements due to apple harvesting activities. Measures of health effects followed WFS exposure events with PM2.5 concentrations of approximately 60 µg/m3 and heat exposure events with a dry air temperature of 90°F. Results indicate that concern over WFS and heat exposure is highly prevalent among our sample of orchard workers. Most workers reported receiving WFS and heat training from their employer during the 2024 work season. Self-reported respiratory symptoms and heat related illness symptoms were common following WFS and heat exposure events. Although the majority of workers reported that employers made N95 respirators available during WFS events, workers reported infrequent usage of N95 respirators and frequent usage of facial coverings such as a bandana. Changes in pace, workload, shift duration and shift start time during WFS and heat exposure events were reported frequently. Most workers reported living in housing or apartments with access to clean and cool air. No changes in lung function were observed following WFS and heat exposure events. Overall, our results point to an outdoor worker group at risk of experiencing adverse respiratory health effects related to WFS and heat exposures. Access to and usage of available exposure reduction tools both in the workplace and at home may help to mitigate potential adverse health effects experienced by workers and members of the regional community following WFS and heat exposure events.