Spring quarter 2022 Tier 1 pilot grant awardees report final impact of funded projects

Patient acceptability of clinical suggestion of social needs automatically identified from clinical notes in the electronic health record

Investigators
Andrea Hartzler, Department of Biomedical Informatics & Medical Education
Angad Singh, Department of Family Medicine
Patrick Wedgeworth, Department of Biomedical Informatics & Medical Education
Kevin Lybarger, Department of Biomedical Informatics & Medical Education
Serena Xie, Department of Biomedical Informatics & Medical Education
Herbie Duber, Department of Emergency Medicine
Brian Wood, Department of Medicine

Project summary
Identifying patients’ social needs is a first critical step for health systems to address social determinants of health (SDoH). Yet, SDoH screening is underused since it relies on arduous clinician entry and patient-entered questionnaires in the electronic health record (EHR). While these methods identify a subset of patients with social needs, they primarily capture patients who are already engaged in healthcare and don’t scale for enterprise-wide population health management, especially when considering patient populations who may need the greatest level of outreach. Emerging CMS mandates for health systems to report SDoH data as a quality metric for federal level tracking will only amplify the need to address barriers to screening. Innovative strategies could bolster screening efforts by filling data gaps for a fuller picture of upstream social factors impacting the health of the population.

We investigated autosuggestion as a novel strategy that surfaces social needs information previously documented in clinical notes in the EHR. The free text of clinical notes contains rich detail about patients’ social needs, but this SDoH data is scattered across patient encounters, difficult to find, and thus remains underutilized. Natural language processing (NLP) makes it technically feasible to automatically extract SDoH from clinical notes with high accuracy. With this automated approach “SDoH autosuggestion” in mind, our team envisions a future in which the EHR surfaces SDoH automatically extracted from clinical notes, and pre-fills those social needs as auto-suggestions for users to accept/reject in clinical tools, such as Epic flowsheets, MyChart, health equity dashboards. Although SDoH autosuggestion could help facilitate and scale up social needs screening, SDoH data is sensitive. This study addresses how patients would feel about having their EHR mined for data to identify social needs.

The purpose of this project was to illuminate the voice of patients regarding potential use of SDoH auto-suggestion in health care. Below we describe key findings and achievements for the two aims that address this objective. This project enabled multidisciplinary collaboration among domain experts, community advisers with lived experience, two PhD students, and a clinical informatics fellow.

Aim 1. Co–produce inclusive study materials with clinical and community champions

• Convened a five-member board of community advisors with professional and lived experience for 3 workshops to: 1) produce recruitment materials and storyboards that explain SDoH auto suggest to patients with varied digital literacy, 2) analyze data together, and 3) create dissemination plans.
• Conducted key informant interviews with 8 interdisciplinary domain experts to vet and refine the interview study design and recruitment plan – the expertise of these “clinical champions” spans primary care, emergency medicine, health equity, NLP, and human-computer interaction.
• Convened meetings with subgroup of five clinical champions for data analysis and manuscripts.
• Collaborated with clinical and community champions on a publication in the Journal of the American Medical Informatics Association published in the special issue on NLP and SDoH.
• Demonstrated an effective model of “multi stakeholder co-production” for future research.

Aim 2. Interview patients about their acceptability of SDoH auto-suggestion

• Partnered with four clinics serving underrepresented and hard-to-reach patient populations.
• Interviewed 16 patients who experience social needs about their acceptability, concerns and recommendations for future implementation of SDoH auto suggestion.
• We used storyboard scenarios to explain how SDoH autosuggestion could work in future care in which NLP-extracted SDoH data extracted from a patient’s clinical notes could be presented in the EHR to healthcare providers for employment, housing stability, food insecurity and substance use domains.
• Most participants found SDoH autosuggestion acceptable, with greater acceptability of SDoH domains perceived as more health-related (i.e., substance use) compared to others (i.e., employment).
• Concerns that participants raised focused on tradeoffs between potential risks of data inaccuracies, privacy/security threats, promotion of stigma and bias and impaired patient-provider relationships.
• Participants recommended that future implementation of automated SDoH systems foster transparency, patient autonomy and empathetic clinical communication about social needs.
• These findings add critical perspective of patients on careful consideration of how automated SDoH systems are implemented and used in clinical care and strategies to safeguard risks.
• Dissemination for Aim 2 to date has included an oral presentation and submission of a manuscript to the American Medical Informatics Association, which is under review.

Understanding the Impacts of Washington Paid Family Leave Policy on American Indian and Alaska Native Birthing Parents and Newborns

Investigators
Teresa Abrahamson-Richards, School of Social Work
Jennifer Romich, School of Social Work
Rose James, Urban Indian Health Institute

Project summary
This project intended to lay the foundation for future research into Paid Family and Medical Leave (PFML) access among American Indian and Alaska Native (AIAN) Washingtonians. Both paid and unpaid parental leave have been consistently associated with positive perinatal outcomes in a variety of studies. However, little is known about access, uptake, and outcomes among AIAN people. Furthermore, intersections between state level policy-making and Tribal sovereignty are also unexplored in this area. As stated in our original project abstract: “Washington’s PFML went into effect on January 1, 2020 and is available to eligible workers in the state who have a qualifying family or medical event such as welcoming a new child. This project will undertake a community-engaged study design process, conduct preliminary analyses, and enable the researchers to acquire the necessary quantitative data to conduct a future study examining birth outcomes among first time AIAN birthing parents.”

We began meeting with community partners at the beginning of the grant period. The partners in these groups included nurses, home visitors, representatives from community social services organizations/nonprofits, public health professionals, and others who work directly with AIAN families during the perinatal period. These partners expressed enthusiasm for the potential to learn about PFML, and overall parental leave access, among AIAN families. However, they also talked about a broader framing to contextualize this work. In addition to these meetings, we also met with researchers to gain insights into appropriate study designs and with state partners to learn more about PFML policy development vis a vis Tribes and AIAN populations.

The initial analysis on pre-PFML employment trends and likely PFML eligibility was completed. The analysis focuses on the proportion of AIAN birthing parents in 2013- 2016 whose work hours would have made them eligible for PFML under the future policy’s rules as well as the employment and earnings of these families pre- and post-birth. The analysis also examines how these trends differ by urban versus rural locality and education status. Preliminary results were presented at one conference in April 2023 and to community partners, as approved by the state of Washington.

Another goal of this work was to design a follow-on quantitative study of PFML access and birth outcomes among AIAN Washingtonians using merged Employment Security Department PFML data and Department of Health birth record data. Based on community partner feedback, we determined that more initial research was needed into the holistic perinatal experiences and needs of working AIAN parents in the state welcoming a new child. In line with this, and with community partner involvement, we designed a qualitative study with the aim to learn about (1) individuals’ in-depth experiences of parental leave type and duration, (2) positive aspects and challenges related to parental leave access, (3) AIAN mothers’ situated perspectives on parental leave design and adequacy, and (4) visions and recommendations for future parental leave policymaking. Preliminary results from these interviews have been presented to community partners, and will be included alongside our initial quantitative findings funded through this Tier 1 population health grant in the manuscript that is under development.

This pilot grant ultimately enabled us to do deep work with our community for the benefit of family health and well-being into the future. By far the most exciting outcome is the community partner enthusiasm for the work, the generation of preliminary results and policy context to share with them and the interest in continued partnership moving forward.

Examining share pantry/refrigerator safety and potential as a hyper-local alternate means of food assistance in the Puget Sound

Investigators
Emily Hovis, Department of Environmental & Occupational Health Sciences
Joe Graham, Washington State Department of Health
Tania Busch Isaksen, Department of Environmental & Occupational Health Sciences

Project summary
As a result of the COVID-19 pandemic, many communities and individuals began providing food in neighborhood pantries/refrigerators to help feed their neighbors. These “share pantries” are generally small, unattended, open‐access food pantries that may or may not include refrigerated foods. Donations are provided by members of a community and generally use a “take what you need, leave what you can” model. Anecdotal evidence suggests that some shared pantries and community refrigerators may provide unsafe food (such as home-prepared foods, unpackaged foods, and food held at improper temperatures). Access to safe and wholesome food, regardless of whether donated or purchased, is an expectation for all Washingtonians. For this reason, our primary project focus was to evaluate the safety of shared pantries/refrigerators within the context of existing food rules. Our secondary focus was to analyze existing locations and gather user data to better understand where they are, who is using them, and why.

We mapped 286 pantries and 12 community refrigerators within the project area of Puget Sound using publicly available information from websites such as littefreepantry.com and freedge.com. A random subset of these sites was visited for in-person safety assessments, including 118 pantries and 12 community refrigerators. The majority of sites were located at private residences in King County, with non-profit organizations being the second most common site. Also, with few exceptions, the sites are located within 1 mile of public transportation.

Interestingly, we identified that a large proportion of sites (33% of fridges and 18% of pantries) were no longer operating at the time of the field visit. The characteristics that contribute to the long-term viability of shared pantries as a community resource is something that we are interested in exploring in the future. In particular, if certain operational models (private residence vs. community organization) are more successful for long-term success.

We conducted safety assessments during the hottest period of the year, from August – October 2022. Inadequate temperature control was identified as the most significant food safety risk at community refrigerator sites, with 75% of refrigerator ambient temperatures greater than 41°F (excluding data from one unplugged refrigerator). Other food safety risks included moldy/decaying food present (55%), food in torn/open containers (40%), presence of leftovers/partially eaten food (25%), potential cross-contamination due to improper animal protein storage (15%), and the presence of home prepared food (10%). Dented/rusty cans (33%) and non-commercially packaged/unlabeled foods (32%) were identified as the most common food safety risks observed at pantry sites. Insects and rodents were identified to be another common risk observed in 29% of pantries. Improper door construction seems to be a common contributing factor, which may lead to pest entry. Our findings suggest there are safety concerns associated with this type of food sharing, but opportunities exist for community education to help improve safety at these sites.

Development and Evaluation of a Culturally-informed Food Insecurity Screening Protocol with American Indian Adults: Evidence from Northern Navajo Medical Center

Investigators
Amanda M. Fretts, Department of Epidemiology
Terra Yabeny, Navajo Department of Health
Preyanka Makadia, Northern Navajo Medical Center
Cassandra Nguyen, Washington State University Partnerships for Native Health

Project summary
Food insecurity (i.e., insufficient access to nutritious food) is a pressing public health concern among American Indian people. Recent reports indicate that more than 46% of American Indian households have experienced food insecurity during the past year. To address this, Indian Health Services (IHS) encourages providers to screen patients for food insecurity to identify households that can be referred to federal or Tribal food assistance programs. The IHS currently utilizes a 2-question food insecurity screener that asks if the food patients buy does not last, or if they worry about food running out before they have money to buy more. Though these questions are approved and recommended for use in healthcare settings in the USA, providers at the IHS Northern Navajo Medical Center found that the 2-question screener is stigmatizing for many patients given the deficit-oriented wording of the questions; the questions make some patients feel uncomfortable and they refuse to complete the screener. Providers at the IHS Northern Navajo Medical Center are interested in identifying an alternate set of strengths-based screener questions that focus on patients’ capacity to maintain food security, while opening a dialogue about opportunities to access food through non-market activities (e.g., farming, foraging, hunting), community resources, and/or food assistance programs.

This pilot project comprised two phases. The first phase was a qualitative phase that included brief 20-minute interviews with patients at the IHS Northern Navajo Medical Center. Patients who enrolled in the study provided feedback on the existing two question food insecurity screener currently being used at IHS, as well as an alternate set of (strengths-based) food security screening questions. In total, we enrolled 25 participants in Phase 1. Interviews were recorded. An external service was used to transcribe all interviews. We analyzed the transcripts using a basic content analysis approach to gather insights and common sentiments about the existing and alternate food insecurity screening questions. Findings from the interviews informed the second phase of the study.

During Phase 2, 97 participants completed both food insecurity questions (i.e., the two question food insecurity screener currently being used at IHS and alternate set of food insecurity questions developed/refined during Phase 1). Agreement between responses on the screeners were compared to assess whether the new screener could validly replace the existing screener.

Most participants in the study (both Phase 1 and Phase 2) were less than 65 years old, identified as female, had no or some college experience, and lived in households with children and incomes below $50,000 per year. Over 60% of participants were classified as food insecure based on responses to the 2 question food insecurity screener currently being used at IHS. During the interviews, many respondents reported a high prevalence of food insecurity in the community and felt that discussing food access with providers at the IHS Northern Navajo Medical Center was valuable. However, interviewees felt the questions on the IHS-approved Food Insecurity Screener are confusing and may produce shame or embarrassment for some respondents. Most respondents felt that screening would be best if providers took the time to discuss food security with sensitivity and with the support of visual aids. Participants felt that discussing food choices with a focus on composition and portions was more comfortable than the format of the current IHS screener.

The alternate screener that we developed included three questions with visual aids. High risk of food insecurity was classified affirmatively if participants indicated that the amount of food usually available was either “Always not enough” or “sometimes not enough” or if they indicated they did not eat “Almost only healthy and traditional Southwest foods” and felt average or unhappy about the kinds of foods they ate. When this classification criteria were used, the Pearson correlation coefficient between the two screeners (i.e., the 2 question food insecurity screener currently being used at IHS and the alternate set of questions developed during Phase 1) was only moderate. More work is needed to better understand the discordant findings.

We are currently focusing on dissemination activities. In September 2023, we gave a one hour learning session (for CME credit) to providers and staff at the Northern Navajo Medical Center. We have additional presentations scheduled in November 2023 for Navajo Department of Health and Navajo Nation Human Research Review Board. We are also in the process of scheduling a data work session with relevant Tribal programs.

A Framework for Prioritizing Urban Green Space Protection and Restoration to Improve Health

Investigators
Kristin Hayman, School of Marine & Environmental Affairs
Jeremy Hess, Department of Environmental & Occupational Health Sciences
Phillip Levin, School of Marine & Environmental Affairs
Josh Lawler, School of Environmental & Forest Sciences
Jamie Robertson, The Nature Conservancy, Washington
Julie Fox, Washington State Department of Health
Buffi LaDue, Washington State Department of Health
Allison Ertl, Washington State Department of Health
Christopher Ahmed, Washington State Department of Health
Marnie Boardman, Washington State Department of Health

Project summary
The purpose of our project is to work with public health and conservation practitioners to develop and implement a framework for identifying locations across the Puget Sound region where protection and restoration of urban green space have the greatest potential to promote health and wellbeing, abate environmental health disparities, and further environmental resilience.

The team conducted an extensive literature review of over 100 peer-reviewed journal articles. Via this literature review, they investigated the dose-responses of a suite of health outcomes to green space, with a major take-away that very few dose-responses between green space and health outcomes have been identified by the literature to date.

Via the literature review, the team also selected three focal health metrics that are affected by green space and conveyed via direct contact with green space: all-cause mortality, depression and self-reported health status. They pulled data from the Environmental Health Disparities Index and the Center for Disease Control and Prevention (CDC) as proxies for the sensitivity of populations in each census tract to the focal health metrics.

The team also pulled data from the Stormwater Heat Map to estimate green space density in each census tract. Further, they used green space density per census tract as a proxy for the average exposure of populations in each census tract to green space. They pulled data from the US Census Bureau to estimate population in each census tract, and then prioritized census tracts and created heat maps of the Puget Sound region that depict the relative priority of each census tract.

Establishing a rural Food Equity Collaborative to increase healthy food retail options for Latinos in WA

Investigators
Barbara Baquero, Department of Health Systems and Population Health
Elizabeth Torres, Northwest Communities Education Center/Radio KDNA
Ria Francisco, Department of Health Systems and Population Health

Project summary
Through this community-based participatory research project, the University of Washington Health Promotion Research Center and Northwest Communities Education Center (NCEC)/ Radio KDNA, conducted formative research to: 1) establish an academic-community Collaborative to advance Latino food equity in rural WA; and 2) examine the contextual factors (geographical, epidemiological, socio-cultural, socio-economic, ethical, legal, political) that would influence healthy food retail promotions for Latinos in rural WA.

Both objectives were accomplished through the following tasks:

• Completed a literature review on academic and gray literature and publicly available sources to draft a summary report on contextual factors (geographical, epidemiological, cultural, economic, legal, ethical, political) that would influence healthy food retail promotions for Latinos in Yakima, Benton and Franklin counties.
• Completed 14 informational interviews with 18 individuals and utilized a combination of inductive and deductive methods to conduct thematic analysis.
• Established a Food Research and Equity Collaborative (FrEsCo) of nine community leaders from Yakima Health District, Yakima Valley Partners in Education (Heritage University), 211, People for People, Yakima School District Fresh Fruit and Vegetable Program, Yakima Valley Farmworkers Clinic Snap Ed, Toppenish Community Chest, and Benton-Franklin Health District representing a multi-sectoral team addressing intersecting community priorities with thoughtful, inclusive solutions.
• Completed training with health educators at Radio KDNA who conducted food retail audits at a sample of 10% of all tiendas across Yakima, Benton, and Franklin (n=9)

Eighteen interviews were completed across 14 organizations. The themes that emerged:

• Transportation is a barrier to accessing services.
• Food is not affordable with current living wages.
• Tiendas (Latino small retail stores) are a comfortable way to purchase culturally relevant foods.
• Communities share wealth of knowledge and resources through informal networks.
• Intersecting structural and social factors emerged as the most important, persistent and complex to address. For example, limited transportation services in these areas affect access to foods, combined with lower wages in rural areas, limit the ability to afford healthier foods.
• Immigration status prevents undocumented residents from accessing government services such as SNAP, despite the fact they participate in the agricultural workforce of the region.
• The strong community ties within the Latino community and access to tiendas, where culturally relevant foods were available for purchase, were leveraging resources to support Latinos.
• Tiendas served as a safe, familiar, affordable setting where Latinos buy healthier food while feeling connected socially and culturally and could be an effective setting for a nutrition intervention.

Understanding mechanisms of social determinants of health and symptoms in adults with subarachnoid hemorrhage

Investigators
Eeeseung Byun, Department of Biobehavioral Nursing and Health Informatics
Michael Levitt, Department of Neurological Surgery
Maria Bleil, Department of Child, Family & Population Health Nursing

Project summary
The goals of our study were 1) Aim 1: to explore the relationship between individual-level social determinants of health (e.g., demographics and socio-economic status) and symptoms (sleep disturbance, fatigue, and impaired cognition) in subarachnoid hemorrhage (SAH) survivors as well as the potential mechanistic role of inflammation in explaining any observed associations and 2) Aim 2: to examine demographics and socio-economic status in SAH patients admitted to Harborview Medical Center in 2021, a potential site for a large grant application, as well as to develop relationships with other potential recruitment sites.

For Aim 1, we conducted a secondary data analysis based on a previous study that examined symptoms (sleep disturbance, fatigue, and impaired cognition) and inflammation (Toll-Like Receptor 4 [TLR4] messenger RNA [mRNA] expression and plasma concentrations of Tumor Necrosis Factor-alpha [TNFα] and Interleukins [IL1β and IL6]) in the first 6 months after SAH in 43 SAH survivors. The findings indicate that demographics and socio-economic status (e.g., age, gender, race/ethnicity, education, employment, and income) were associated with symptoms (e.g., sleep disturbance, fatigue, and impaired cognition) in SAH survivors over 6 months post-SAH. In addition, inflammatory biomarkers (e.g., TLR4 mRNA, TNFα, IL1β, and IL6) were associated with symptoms (e.g., sleep disturbance, fatigue, and impaired cognition). Inflammatory biomarkers also moderated the relationship between social determinants of health (e.g., age, gender, race/ethnicity, education, employment, and income) and symptoms, which indicates that the effect of social determinants of health on symptoms may differ based on the degree of inflammation after SAH. Further studies, with a large sample size, an extended follow- up period, and rigorous measures of social determinants of health are warranted to confirm the associations of social determinants of health, symptoms, and inflammation in people with SAH.

For Aim 2, we reviewed 80 medical charts of aneurysmal SAH patients admitted to Harborview Medical Center in 2021 and examined their demographics and socio-economic status (e.g., race/ethnicity, gender, marital status, education, employment, insurance, areas of living, and social support) and clinical characteristics (e.g., severity of SAH and comorbidity). Our findings indicate that SAH patients at Harborview had diverse demographics and socio-economic status, indicating that Harborview is an excellent recruitment site for a future large grant application. In addition, we have been disseminating research findings related to social determinants of health in SAH at international and national scientific meetings and initiated a network of potential collaborators. We will further foster interdisciplinary research collaboration and initiatives with a high potential for further research related to social determinants of health in people with SAH.

PrEParing for the future: preferences for HIV pre-exposure prophylaxis products and delivery models among cisgender men, transgender, and non-binary persons in Seattle, Washington

Investigators
David Katz, Department of Global Health
Lauren Violette, Department of Epidemiology
Jodi Greathouse, Department of Global Health

Project summary
he primary aims of this project were to 1) design and pilot a discrete choice experiment (DCE) to identify HIV pre-exposure prophylaxis (PrEP) product and delivery preferences of cisgender men who have sex with men, transgender persons, and non-binary people, and 2) describe and identify characteristics associated with preferences for PrEP product attributes and delivery models.

The DCE design process spanned from April to August 2022. We used a combination of current literature, expert opinion, and previously conducted qualitative research to guide attribute and level development. Ngene and SAS were used to program the experiment. The final DCE was piloted by the study team, translated, and programmed into REDCap in both English and Spanish. We enrolled 325 participants from the Public Health – Seattle & King County (PHSKC) Sexual Health Clinic between August 2022 and December 2023. The majority (87%) of participants identified as cisgender men, while 10% identified as non-binary or genderqueer, and 2% were transgender women. Over half (67%) had previously used (20%) or were currently using (47%) PrEP, while 30% were PrEP-naive.

Final analyses for the primary outcomes of this project will be completed and summarized in a manuscript by Summer 2024. Depending on availability, we may also present these data at a “Work in Progress” meeting with a UW Center for AIDS Research (CFAR) working group or core. These data will be disseminated to our project collaborators at the PHSKC Sexual Health Clinic to enhance their existing PrEP services and plan for implementation of new PrEP products. Our DCE can also serve as a tool for local community organizations and partners to tailor PrEP delivery to their clients. As such, we plan to create a one-page summary with key takeaways of our study to disseminate to other local stakeholders including, but not limited to, the UW CFAR, community-based organizations in King County, and the Washington State Department of Health. Finally, Lauren Violette, the project co-investigator, will use this DCE design to explore PrEP delivery preferences among similar populations in Boston during a post-doctoral fellowship within the Department of Population Medicine at Harvard Medical School.

Residential Segregation and Pediatric Injury and Violence in Seattle, Spokane, and Tacoma

Investigators
Chelsea Hicks, Department of Pediatrics
Frederick Rivara, Department of Pediatrics
Abril Harris, School of Social Work
Rebecca Walter, Department of Real Estate
Monica Vavilala, Department of Anesthesiology & Pain Medicine
Qian Qiu, Department of Pediatrics
Tony Escobar, Mary Bridge Children’s Hospital
Timothy Bax, Sacred Heart Medical Center

Project summary
The original project proposal defined three aims: (1) Create a model of the relationship between residential segregation, the built environment and pediatric injury exposure through a scoping review. (2) Quantify present day association between residential segregation and pediatric injury and violence outcomes in three Washington state cities (Seattle, Tacoma, Spokane). (3) Identify potential interventions through moderating and mediating mechanisms in the relationship between residential segregation and pediatric injury and violence.

Progress toward each aim is as follows:

1. The research team conducted a systematic literature review that included approximately 25 journals and databases. Over the course of several months over 3,000 articles were screened for inclusion. We dual abstracted data from 34 eligible studies and conducted quality assessments of the studies. Our team is currently in the process of finishing a manuscript to be submitted to a peer reviewed journal on the associations other studies have found between geographic location, built environment and pediatric injury or violence outcomes.
2. In partnership with Harborview Medical Center, Sacred Heart Medical Center (Spokane) and Mary Bridge Children’s Hospital (Tacoma), we received pediatric injury data from their trauma registries for patients from years 2012-202l (Seattle) and 2015-2021 (Spokane and Tacoma). The team used trauma registry data from Harborview Medical Center, Mary Bridge Children’s Hospital, and Sacred Heart Medical Center to identify residential zip codes in these cities of youth treated at these hospitals for injuries and violence. While it was initially believed we could acquire access to the residential addresses to geocode into census tracts, there were difficulties in receiving this data due to personal health information concerns. We thus were forced to rely on zip code data. Using the available hospital data and the American Community Survey (ACS) data, we did not find any association of sociodemographic characteristics of the zip code and the injury rates. This was due to the fact that injury rates were relatively low per zip code and there was a lack of variation in injury rates at the zip code level. However, this research project and analysis did lend to methodological insights to improve measuring injury and violence outcomes at the environment level while noting limitations in existing data formats.
3. Based on the null findings at the zip code level, we identified that the methods that are typically used to measure mediating and moderating mechanisms were unfeasible for our variables of residential segregation and pediatric injury and violence at the zip code level and a more granular unit of analysis is required for future research.

Piloting a Strategy to Improve Pediatric Inpatient Guideline Adherence in Migori, Kenya

Investigators
Megan Coe, Department of Child, Family, and Population Health Nursing
Sarah Gimbel, Departments of Global Health and Child, Family, and Population Health Nursing
Arianna Rubin Means, Department of Global Health
Beth Kolko, Department of Human Centered Design & Engineering
Benson Singa, Kenya Medical Research Institute
Ann Van Haney, Department of Global Health

Project summary
This pilot study used participatory approaches to identify and tailor an implementation strategy aimed at improving pediatric hospital care in Migori, Kenya. In Aim 1, a nominal group technique approach was used to match potential interventions to identified barriers and select a strategy to pilot. This strategy was then tailored to the needs of the health care workers using an iterative design process (Aim 2). After piloting, we evaluated the usability and feasibility of the piloted strategy for ongoing use through mixed methods data collection and analysis (Aim 3).

Aim 1 of the project was completed in August 2022. A targeted implementation strategy was selected using participatory techniques with seven staff members from Migori County Referral Hospital. First, preliminary results of an earlier qualitative study conducted at the site were presented to the participants. This study had identified barriers and facilitators of use of pediatric treatment guidelines. Four challenges that came out strongly in the qualitative results and were modifiable at the facility-level were identified and participants ranked them on their likelihood of change and potential for impact. There was consensus among participant that the focus of this pilot should be the challenge: Health care workers rely on nutritionists to handle many tasks related to nutrition. If the nutritionist is not available this can delay guideline adherent care (especially at night/weekends).

The group was then presented with examples of strategies that were mapped to the specific barriers that had been identified in the qualitative study. They brainstormed to identify specific strategies that would be relevant to their setting. Six ideas were generated by the group. Participants ranked these options on three criteria (acceptability, feasibility, and impact), had a discussion about these options, and then reached consensus on an idea to pilot: Develop a summary of key information from the Integrated Management of Acute Malnutrition and other guidelines, merge them, and agree how they will be delivered by staff. Our team decided to call this idea the Team Approach to Malnutrition Services (TeAMS) Toolkit.

Based on the outcome of Aim 1, the study team refined plans for the pilot study and ethics approval for the remainder of the study was granted in February 2023. In Aim 2, an iterative design process was used to identify the content and features that health care workers at the facility preferred in the TeAMS Toolkit. Twenty participants completed pre-pilot surveys describing their experience and competence in caring for children with malnutrition. Between April and July 20203, 18 one-on-one design sessions were held with participants, consisting of interactive activities to guide refinement of the TeAMS Toolkit. The toolkit was designed to support staff by: (1) offering guidance on teamwork and clarifying team member roles and (2) consolidating recommendations from national guidelines so that health workers have easier access to the information they need to fulfil their roles. This is accomplished in four sections: Teamwork Skills, Team Member Roles and Responsibilities, Key Information from IMAM, Monitoring and Documentation. The TeAMS Toolkit was launched at the facility in September 2023 and a 10-week pilot period commenced.

At the conclusion of the TeAMS Toolkit pilot in November 2023, 29 participants completed a post-pilot survey. In addition, one focus group discussion and six in-depth interviews were conducted with participants. Among 19 participants with both pre- and post- data, mean scores for frequency of completing seven of eight tasks related to malnutrition increased, however changes in perceived competence scores on these tasks were variable (some increasing and some decreasing). While the survey data did not show increases in competence, during qualitative interviews several participants described additional staff being able to initiate feedings for children with malnutrition by utilizing the information on Toolkit posters. Responses to five out of six items assessing teamwork perceptions increased between the pre- to post-pilot surveys. This aligned with sentiments shared in the interviews describing improved teamwork and information sharing among staff. One participant stated: “I used to think it is a nutritionist duty, but currently I can also initiate [feeding]. …I have learnt that for success then, it has to be teamwork.”

Among all participants who completed the post-pilot survey, the mean System Usability Score for the TeAMS Toolkit was 77.4 on a 100-point scale, which can be interpreted as good system usability. The SUS score has two dimensions – usability and learnability. The mean score on the 8 usability questions was 3.3 out of 4, while the mean score on the two learnability questions was 2.3 out of 4. In interviews, participants noted the ease of use of most of the content in the TeAMS Toolkit, but often requested additional training opportunities to support development of knowledge and skills related to the TeAMS Toolkit. On three questions related to the feasibility of continuing implementation of the TeAMS Toolkit within the facility, mean scores among staff within the Paediatric Ward ranged from 4.29 to 4.71; where 4 indicates agreement and 5 indicates strong agreement that it would be possible to continue implementation of the Toolkit.

Exploring Multicultural and Multilingual Methods in Developing Dialog-Based Health Technologies

Investigators
Weichao Yuwen, School of Nursing & Healthcare Leadership, UW Tacoma
Trevor Cohen, Department of Biomedical Informatics and Medical Education
Magaly Ramirez, Department of Health Systems and Population Health
Serena Jinchen Xie, Department of Biomedical Informatics and Medical Education

Project summary
In the United States, over 20 million family caregivers self-identified as having a racial or ethnic minority background spend on average more than 20 hours per week providing unpaid care to a care receiver. In addition, at least one-third of caregivers speak a language other than English at home. Existing barriers that prevent these marginalized caregivers from accessing services include a lack of culturally sensitive and linguistically appropriate programs of care. On-demand digital health technologies delivering interventions through text-based dialogs could be cost-effective solutions to support family caregivers. Recent innovations in machine learning and artificial intelligence (AI) have the potential to automate labor-intensive work such as language translation. However, these technologies, if used without careful intention, can increase biases and existing imbalances for marginalized groups. To our knowledge, there has not been the development of efficient multilingual AI language models in the health domain. We proposed to conduct a literature review of multicultural and multilingual model development methods in developing health dialogs (aim 1) and perform pilot training using an existing caregiver-domain-specific dataset in English to translate to and train in Spanish (aim 2).

We established an interdisciplinary research team consisting of faculty members, undergraduate students, pre-doctoral and post-doctoral trainees from nursing, biomedical informatics, and public health. The team conducted a narrative review on the topic of multilingual model for health dialog development and exploring the capabilities and limitations of large multilingual language models like mBERT and XLM-R. We submitted an abstract of the preliminary results to the American Medical Informatics Association Annual Symposium (AMIA 2023), and plan to submit the review for publication this summer. In addition, we conducted preliminary exploration of our existing dataset for machine translation and training, and used the results in a grant application to the National Institute of Health (NIH), which is currently under review. The team was recently awarded an Emergent Innovation Grant from the Rita & Alex Hillman Foundation for a pilot project building on this project.