Investigators
Anna Ratzliff, Department of Psychiatry & Behavioral Sciences
Elizabeth Austin, Department of Health Systems and Population Health
Brittany Blanchard, Department of Psychiatry & Behavioral Sciences

Project summary
It is unknown whether the Collaborative Care Model (CoCM) reaches underserved populations equitably, but current data suggests that variable CoCM fidelity likely contributes to care disparities. Assessing CoCM fidelity is critical for equitable care, quality improvement and research, yet no measure of CoCM fidelity exists. For this Population Health Initiative Tier 2 project, our goal was to refine and evaluate the performance of a Collaborative Care fidelity tool. Initial item content was based on model components developed by the AIMS Center, which was then refined via a modified Delphi process with clinicians and academic experts. Our specific aims were to 1) develop item(s) to assess healthcare equity in CoCM delivery, 2) evaluate the psychometric properties of the CoCM Checklist and 3) identify CoCM components associated with patient improvement using clinical registry data.

Aim 1. Qualitative interviews were conducted to understand what equity in CoCM meant to clinicians, and responses indicated equitable access and individualized care. Several clinicians indicated that CoCM was a form of equitable care. Seven equity items were drafted by a multidisciplinary team, followed by cognitive interviews with clinicians. Items were perceived as interpretable and relevant. Items spanned equity in screening, access, treatment tailoring, patient barriers to care, ability to assess program equity and implementation efforts to address equity gaps.

Aim 2. The 24 fidelity items and 7 equity items were administered via REDCap survey to clinicians (combined n = 68), and retest administration occurred within 1 week (n = 56). Results indicated range restriction, with most participants endorsing 鈥渁lmost always鈥� for many items. Most variation was observed for items that asked about team members鈥� behaviors (e.g., psychiatric consultant, leadership). The intraclass correlation coefficient (ICC) was good (ICC > .75) for only 5 of 24 fidelity items and 2 of the 7 equity items. However, ICC may be artificially low due to range restriction, so we will conduct sensitivity analyses using appropriate tests (e.g., weighted kappa; spearman鈥檚 rho). Further, over 10% endorsement of improbable responses (e.g., daily caseload reviews with psychiatric consultants) indicates inaccurate responding. This could be due to multiple factors (e.g., social desirability bias, response set bias, item misinterpretation) which we could not discern with our design. However, results indicate that self-report alone is likely insufficient to assess many components of CoCM fidelity. Data from qualitative interviews were further analyzed using Rapid Assessment Process to better understand variation in practice fidelity and opportunities for fidelity checklist enhancement.

Aim 3. We are currently merging registry and electronic health record data from both samples to maximize power for Aim 3 analyses to evaluate whether specific CoCM components predict patient improvement and evaluate whether specific equity gaps exist in current CoCM programs. This large clinical dataset will allow us to explore whether specific components predict patient improvement and inform optimization of CoCM delivery. Quantitative results will be triangulated with qualitative interview data exploring CoCM clinician perspectives on essential CoCM components.

Additional deliverables. As part of our semi-structured interviews, we asked participants to describe impacts of recent healthcare funding changes on their clinics. Qualitative analysis of 19 interview transcripts representing 24 unique clinics indicated less accessible and affordable mental health services for patients, increased acuity of patient needs for mental health and social services, increased distress, burnout and staffing shortages and increased fragility of organizational financing and support structures necessary to sustain integrated care.

Investigators
Barclay Stewart, Department of Surgery
Megan Moore, School of Social Work
Tony Machacha, King County Regional Homelessness Authority
Caitlin Orton, Department of Surgery

Project summary
The aims for the project Piloting the Equity in Injury Prevention Toolkit with People Who are Unhoused:

Aim 1: Pilot the EQUIP Toolkit with people experiencing homelessness (PEH) living within high burn injury risk areas of Seattle; and

Aim 2: Assess the EQUIP Toolkit implementation using the RE-AIM Framework (e.g., reach, effectiveness, adoption, implementation, maintenance) to support the future large-scale dissemination within King County.

We partnered with four outreach specialists from a large, local outreach organization to implement EQUIP (consumer-tested injury prevention education, PEH preferred fire extinguishing equipment and dissemination strategies). These outreach specialists integrated EQUIP into their routine outreach activities within four high fire and burn and cold injury risk areas of Seattle. They each completed online surveys every three weeks along with a post-implementation interview to capture reach, adoption and implementation outcomes (i.e., EQUIP acceptability, feasibility of EQUIP integration, fidelity to protocols, consistency of delivery, required adaptations). Data were also collected from PEH through the completion of key informant interviews at the start and end of project implementation (total of 62 interviews). These interviews also captured reach and implementation outcomes (i.e., required adaptations, use of EQUIP, confidence with using EQUIP) as well as provided additional information to inform the future scaling of EQUIP (i.e., gaps in education, future collaborators). Researcher notes tracked the cost of implementation, number and type of supplies provided and any feedback shared outside of the interviews and surveys.

Overall we found disseminating EQUIP during routine outreach activities to be acceptable and feasible. Outreach specialists reported a desire to continue disseminating EQUIP after the pilot had concluded. Their barriers to adoption were level of comfort with talking about burn and cold injury prevention and time limitations as other responsibilities and tasks arose. Outreach specialists were able to reach people living outside, in tents, vehicles and RVs, however reach was limited to those who spoke English. PEH confirmed the acceptability of EQUIP and highlighted that distribution through outreach greatly reduced the barriers to prevention education and equipment. PEH stored the fire extinguishing equipment in accessible areas in their tents or in their campsites so that they and those around them could easily access if needed. The results from this pilot will inform the scaling of EQUIP: 1) integrating EQUIP into King County Regional Homelessness Authority鈥檚 tiered severe weather programing, 2) collaborating with additional community partners to include additional outreach teams, street medicine, food banks, shelters, day centers, clinics and needle exchanges and 3) scaling EQUIP across King County.

Investigators
Brittany St John, Department of Rehabilitation Medicine
Tanya St. John, Department of Speech & Hearing Sciences
Yev Veverka, Institute on Human Development & Disability
Sara Kover, Department of Speech & Hearing Sciences
Jessica Greenson, Institute on Human Development & Disability
Nicole Bergstrom, Institute on Human Development & Disability
Sarah Lemke, Institute on Human Development & Disability
Kelleen Dunley, Institute on Human Development & Disability

Project summary
The purpose of this project was to pilot the co-developed First Bite program for families of autistic children awaiting feeding intervention services at the 天美影视传媒 Autism Center (UWAC). First Bite is a 10-week caregiver education and consultation program to address feeding challenges for autistic children. Primary outcomes measured in this study include achievement of child goals and reduction of need for direct feeding services. Secondary outcomes evaluated were feasibility, acceptability and changes in caregiver feeding knowledge and efficacy post program.

First Bite was piloted with 14 families on the waitlist for feeding intervention at the UW Autism Center. Results of the pilot study suggest high rates of acceptability and high feasibility for delivery to families awaiting services. In addition, families demonstrated significant progress toward feeding goals and improvements in caregiver knowledge and feeding efficacy, suggesting the program provides clinically impactful feeding supports.

Investigators
Avanti Adhia, ScD, Department of Child, Family, and Population Health Nursing
Heather D. Hill, Daniel J. Evans School of Public Policy & Governance
Krista Neumann, Department of Pediatrics
Ann Richey, Department of Epidemiology

Project summary
Domestic violence (DV) is prevalent and has substantial individual, community and societal costs and consequences. State “safe leave鈥� policies (including paid leave), which provide survivors with time off to address legal, medical or safety issues related to DV, serve as one potential strategy to address DV at the outer layers of the social ecology. The overall goal of this PHI Tier 2 project was to provide preliminary evidence about the potential effectiveness and implementation of paid leave programs for DV. Specifically, we completed three aims:

Aim 1: We conducted an in-depth policy landscape analysis of state safe leave programs in the context of DV. We systematically reviewed summary reports, policy documents and state legislature websites for all 50 states and the District of Columbia to examine three state policy tools that provide safe leave鈥攑aid family and medical leave, sick leave and victim leave鈥攁nd their key dimensions. We organized these data to descriptively analyze policy features. We published a peer-reviewed Current Issues article in the American Journal of Preventive Medicine in March 2026 titled 鈥淪tate Safe Leave Policies to Address Domestic Violence鈥� based on these findings.

Aim 2: We used data from CDC鈥檚 Pregnancy Risk Assessment Monitoring System (PRAMS) to examine the association between state paid family and medical leave programs and DV during pregnancy. Using data from 2000 to 2022, we employed a quasi-experimental difference-in-differences design. We found that state paid family and medical leave program adoption was associated with modest reductions in self-reported DV during pregnancy, with larger estimated effects observed five to eight years after implementation. A manuscript with these findings is under review. Co-I Neumann also presented findings at the Society for Epidemiologic Research Annual Meeting in June 2026 in a session titled 鈥淰iolence: Context, Disparities, and Consequences.鈥� Findings will also be presented at the Association for Public Policy Analysis and Management Fall Research Conference in November 2026.

Aim 3: We explored the use of paid leave for DV in two states with different paid leave program structures鈥擶ashington and Oregon. We interviewed a total of 25 state paid leave staff and service providers (e.g., advocates, legal providers) about their knowledge regarding paid leave being used for reasons related to DV, benefits or barriers to using paid leave for these reasons and recommendations to improve access for workers. Qualitative analyses showed that paid leave can be a mechanism for survivor safety, recovery and employment stability; however, there are some misalignments between the paid leave program design and survivors鈥� complex realities. Our interviews highlighted barriers to accessing and using paid leave in the context of DV, which included limited awareness, structural eligibility constraints, administrative burdens and challenges with program delivery, confidentiality and navigation. To increase support for survivors, state paid leave staff and service providers suggested adding DV as a qualifying event for Washington鈥檚 program, broadening the range of individuals authorized to certify leave, strengthening confidentiality protections and expanding training and outreach. This qualitative data collection provided rich insights into paid leave that complement analyses in Aims 1 and 2. A manuscript is in preparation, and findings will be presented at the Association for Public Policy Analysis and Management Fall Research Conference in November 2026.

A portion of the funding for this award came via a partnership with the , which seeks to enhance the UW鈥檚 global engagement and reach.

Investigators
Mehmet Kurt, Department of Mechanical Engineering (and by courtesy, Bioengineering and Radiology)
Jacob Ruzevick, Department of Neurosurgery
Udunna Anazodo, McGill University and Crestview Radiology Ltd., Lagos, Nigeria
Maruf Adewole, Medical Artificial Intelligence Laboratory (MAI Lab), Crestview Radiology Ltd., Lagos, Nigeria

Project summary
The goal of this project was to develop and validate equitable artificial intelligence methods for ischemic stroke diagnosis that can remain useful in settings where MRI is limited and CT-based imaging is the most practical acute-care option. The work focused on improving stroke lesion segmentation from noncontrast CT, CT angiography and CT perfusion by incorporating clinically informed image preprocessing, validating the approach against MRI-derived lesion labels and building international training and data partnerships that can support future evaluation in underrepresented populations and resource-constrained clinical environments.

A major technical outcome was a clinically informed preprocessing pipeline that uses brain extraction, targeted intensity windowing and vessel-aware CTA information before nnU-Net segmentation. In the attached study, the approach increased mean Dice score from 0.162 to 0.224 across ten folds, a 38.4% relative improvement over baseline nnU-Net preprocessing. Replacing raw CTA with binary vessel segmentations produced a further 21% mean improvement across five folds. The best model was submitted to the ISLES鈥�24 challenge and placed first overall, providing blinded external validation of the core approach.

The project also established a stronger translational dataset base. We curated 160 paired CT-MRI stroke datasets from Harborview Medical Center in collaboration with the Stroke and Applied Neuroscience Center. These paired studies will allow direct MRI-referenced evaluation of CT-based lesion prediction in a real clinical population. In parallel, ethics and regulatory submissions were initiated to support acquisition of stroke lesion datasets from multiple Sub-Saharan African countries, including Uganda and Nigeria. Ethics/IRB application fees associated with establishing these international cohorts were supported by the grant, and data acquisition and curation are now being organized with partner sites.

Capacity building was an integral part of the project. PHI funds helped sponsor three SPARK Academy trainees to travel to MICCAI 2025 and participate in the broader MIRASOL community focused on medical image computing in resource-constrained settings. The project鈥檚 stroke-related SPARK mentorship also included T. Musah, P. E. Adjei and K. O. Otoo, whose work addressed automated ischemic stroke lesion segmentation from non-contrast CT. Juampablo Heras Rivera and Dr. Kurt provided mentorship around stroke segmentation, clinically meaningful preprocessing, model evaluation and the challenges of building robust AI for settings with limited imaging and compute resources.

Together, these activities moved the project from proof-of-concept algorithm development toward multisite validation and sustainable global collaboration. The technical results show that relatively simple, clinically motivated processing can materially improve CT-based stroke segmentation, while the training and international partnership activities have created a pathway to test whether these gains generalize to populations and clinical environments that remain underrepresented in current medical AI datasets.

Investigators

Lesley Steinman Department of Health Systems & Population Health
Najma Mohamed, Neighborhood House
Jacob Bentley, Department of Rehabilitation Medicine
Tricia Aung, Department of Human Centered Design & Engineering
Roberto Orellana, School of Social Work
KeliAnne Hara-Hubbard, Department of Health Systems & Population Health
Farhiya Osman, Neighborhood House

Project summary
Our goal was to evaluate whether and how adapted Problem Management Plus (PM+) worked for social service providers and Somali clients. We trained CBO providers in PM+ and assessed PM+ impact and delivery. We also honed methods for CBO-engaged mental health research.

The project trained five social service providers in PM+ and delivered the five-week intervention to 11 Somali clients. Participants demonstrated improvements in well-being, functioning and PTSD symptoms, with 10 of 11 participants improving their PSYCHLOPS scores by an average of 35%, WHO-5 well-being scores increasing by 24.6% and PCL-5 PTSD scores decreasing by 48%. Interview and diary data suggested that adaptations integrating Islamic prayers and prophet stories were acceptable, appropriate, usable and feasible. Participants reported using PM+ skills, especially slow breathing and connecting with faith and family, alongside Neighborhood House social services to cope with health challenges, social needs and immigration-related stressors. The project suggests it is possible to re-imagine mental health care that meets people where they are mentally, spiritually and physically.

Investigators
Lisa Manzer, Center for Women鈥檚 Welfare, School of Social Work
Sarah Brolliar, Center for Women鈥檚 Welfare, School of Social Work
Marisol Tapia Hopper, Workforce Development Council of Seattle-King County

Project update
This project expands the Self-Sufficiency Standard (SSS) to incorporate individual marketplace premiums and premium tax credits (PTCs), producing a more accurate measure of income adequacy and estimating the household impact of PTC expiration.

To date, work has focused on building the infrastructure required to support this modeling. The SSS production system has been fully transitioned from Excel to a reproducible, R-based workflow, with all components version-controlled in GitHub. A comprehensive federal and state tax calculation framework has been developed, enabling accurate modeling of tax liabilities and credits, including the structure needed to incorporate PTCs. The framework for integrating marketplace premiums into the SSS is established, and parameters for modeling PTCs under both current law and post-expiration scenarios are being finalized. The population modeling system is also complete: the R-based workflow for applying the SSS to American Community Survey (ACS) microdata is ready to generate household estimates under different policy scenarios.

The urgency of this work has increased: with Enhanced PTCs having expired at the end of 2025, the project is now positioned to document real-world impacts rather than projected ones, strengthening the policy relevance of findings.

Final policy outputs have not yet been produced, but the core systems required to generate them are in place. Remaining work is focused on completing integration, running simulations, and producing results.

Investigators
Lupita Santillan, Department of Psychiatry and Behavioral Sciences
Julia Mattson, Department of Pediatrics
Ana C. Hernandez, Seattle Children鈥檚 Hospital
Weili Yuan, Elson S. Floyd College of Medicine, Washington State University
Gulaiim Almatkyzy, Department of Oral Health Sciences
Brent Collett, Department of Pediatrics

Project update
Our project aims to design and pilot a mobile health intervention to improve oral health behaviors among young children through parent engagement. To date, we have obtained IRB approval from the 天美影视传媒, as well as site approval from Seattle Children鈥檚. We are currently awaiting Washington State University approval for one study team member who will support recruitment at our rural partner site (Joya Child and Family Development). Spanish translations of IRB-approved materials are in progress.

We have also expanded our recruitment network to include Seattle Children鈥檚 Early Childhood Clinic and the 天美影视传媒 Institute on Human Development and Disability鈥檚 Infant Development Follow-up Clinic. All English-language consent forms and questionnaires have been completed and programmed in REDCap. Additionally, our BOOST mobile application has been fully developed and tailored for the study in English. We are preparing to begin recruitment of English-speaking families within the next 1-2 weeks. Participant materials (e.g., tripods, toothbrushing kits) are in the process of being ordered.

Investigators
Sam Bernecker, Department of Psychiatry and Behavioral Sciences
Jennifer Turns, Department of Human Centered Design & Engineering
Patrick Raue, Department of Psychiatry and Behavioral Sciences

Project update
Psydkick is a digitally guided, reciprocal coaching tool designed to overcome common barriers to evidence-based psychosocial interventions (EBPIs). In Psydkick, pairs of users take turns delivering Problem-Solving Therapy (PST) to one another, guided by real-time prompts on a laptop, tablet, or phone.

The reciprocal, peer-delivered structure addresses both structural barriers (cost, provider scarcity, wait times) and attitudinal barriers (stigma, desire for self-reliance) that keep many people from receiving evidence-based mental health care. With Initiative funding, the team is conducting the discovery work needed to inform prototype development and support a follow-on pilot trial.

Progress to date includes the following:

• Refocused the project鈥檚 target population from rural-dwelling adults to community college (CC) students. CC students share the high unmet treatment need that motivated the original focus on rural adults鈥攐ne large NIMH-funded study found that more than two-thirds of CC students screened positive for depression, anxiety, or PTSD, while only 11% had received psychotherapy in the past year鈥攂ut they are more demographically diverse, more efficiently reached through institutional partners, and represent a large population (roughly half of working U.S. adults take at least one CC course). This refocus is expected to enhance the generalizability of findings.
• Established community college partnerships. The team secured letters of support from two Washington State community colleges (Highline College and Lake Washington Institute of Technology, with a combined enrollment of more than 20,000), including building relationships with one institution鈥檚 Executive Committee. Outreach is ongoing with several additional colleges.
• Developed study materials and submitted an IRB protocol for a mixed-methods user-needs study of community college students experiencing elevated depression and/or anxiety symptoms.
• Submitted an NIH K23 career development award proposal (PI: Bernecker) that builds directly on this work and proposes a pilot trial of Psydkick at community colleges. Initial review feedback is expected in the coming months.

Investigators
Dario Cvencek, Institute for Learning & Brain Sciences
Jessica C. Jones-Smith, Health Systems and Population Health

Project update
This project aims to investigate the development of children鈥檚 food attitudes and their associations with dietary habits and physical health outcomes. The study is guided by four primary objectives, and progress to date is summarized below.

Development of Assessment Measures. We have successfully developed and implemented a comprehensive battery of child- and parent-report measures within our data collection platform. These measures assess a broad range of factors related to children鈥檚 physical health, including home food availability (healthy and unhealthy), physical activity levels, presence and use of electronic devices and children鈥檚 food attitudes.

Demonstration of Project Feasibility. Data collection is well underway, with 78% of the target sample obtained to date. Recruitment and participation rates achieved within a relatively short timeframe indicate strong feasibility for a larger-scale longitudinal study. Both children and parents have demonstrated willingness and ability to complete the assessment battery in a remote (online) format.

Identification of Factors Associated with Food Attitudes. Although data collection is still in progress and analyses have not yet been conducted, successful administration of the full assessment battery supports our ability to carry out the planned analyses upon completion of data collection.

Examination of Links to Physical Health. Similarly, planned analyses examining associations between children鈥檚 food attitudes and physical health outcomes will be conducted once data collection is complete.

Investigators
Sebastian Tong, Department of Family Medicine
Peter Kahn, Department of Psychology & School of Environmental and Forest Sciences
Ashley Park, Department of Family Medicine
Hongfei Li, College of Built Environments

Project update
The Nature Engagement Project aims to translate Interaction Pattern Theory into clinical practice through the development of the Interaction Pattern Preference Inventory (IPPI), a patient- and provider-facing tool to support more specific, feasible and equity-informed nature prescriptions. Guided by the Health Equity Implementation Framework, the project focuses on understanding embodied interactions with nature and identifying barriers shaped by social and environmental contexts.

To date, the team has collaboratively developed core Tier 1 study materials, including a draft IPPI prototype, a survey instrument and three semi-structured interview guides tailored for patients, health care providers and community planners/practitioners. A recruitment plan has been implemented, healthcare providers have been successfully recruited, and qualitative interviews with providers are actively underway. An early conceptual version of the IPPI is being used within interviews to gather formative feedback on usability, feasibility and clinical relevance to inform future refinement. In addition, the project has supported the mentorship of an undergraduate student who is actively engaged in research activities and gaining applied experience in health equity鈥揻ocused, practice-based research.

Investigators
Thanh P. Nguyen, Department of Pediatrics
Gabi Sepulveda, Department of Psychiatry & Behavioral Sciences
Matilda Sampson, Excelsior Wellness
Shayla Collins, UW Center for Child & Family Well-Being
Jim Mancini, Department of Pediatrics

Project update
This project works toward three interrelated aims to improve coordinated care for autistic youth with mental health needs and their families in Washington State.

For Aim 1, the team has initiated a multi-method data collection process to identify gaps, barriers and ideal practices in care coordination. A needs assessment survey has been developed and distributed to care coordinators and families of autistic youth, and data collection is currently underway. In parallel, the team is preparing for regional research World Caf茅 sessions鈥攁 qualitative approach designed to efficiently gather rich, practice-informed insights鈥攊ncluding finalizing procedures, booking venues, and purchasing materials. Planning is also in progress for virtual focus groups that will engage diverse stakeholders such as caregivers, autistic individuals, coordinated care professionals and administrators.

Progress has also been made toward Aim 2, which focuses on developing a community-informed and vetted Care Coordination Training Curriculum (CCTC) manual. The team has drafted the training content and identified key topic areas for the manual. This draft will undergo iterative refinement through at least three rounds of stakeholder input gathered during the World Caf茅 sessions and the virtual focus groups.

For Aim 3, the team has engaged with several agencies across Washington State to explore alignment and potential interest in future collaboration related to a pilot study of the CCTC program. Ongoing data collection through the needs assessment survey, World Caf茅 sessions and virtual focus groups will continue to inform manual development and partnership building.

Investigators
Patience Binambiba Jaman, Department of Health Systems and Population Health
Kaboni Whitney Gondwe, Department of Child, Family, and Population Health Nursing
Rabi Yunusa, Department of Global Health
Mpiima Mugambe, Immigration Guide
Francis Abugbilla, Ghanaians in Seattle Association-GHASEA

Project update
This project seeks to understand the barriers and opportunities to equitable gestational hypertension care among Black immigrant women in Washington State through a community-partnered needs assessment. Using a mixed-methods and community-based participatory approach, the project aims to generate culturally informed insights to guide future interventions and policy-relevant recommendations. The project timeline includes preparation and recruitment (January鈥揊ebruary 2026); data collection, including interviews (March鈥揗ay 2026); data analysis and community validation (June鈥揓uly 2026); and dissemination activities, including a stakeholder meeting (August 2026).

Progress to date reflects meaningful advancement toward these milestones, with several components completed as planned and others actively underway. The project has since focused on foundational activities essential to successful implementation. Institutional Review Board (IRB) approval for the study was obtained, and data collection tools, including surveys and interview guides, have been developed and refined with the help of our community partners and community advisory board.

Key partnerships have been established with community organizations serving Black immigrant populations, including Immigrant Guide and the Ghana Association of Greater Seattle (GHASEA). We have had five meetings with our community partners since the beginning of the study, where the research team has consistently provided progress updates and gathered feedback for improvement. They have actively engaged in shaping the study’s priorities and outreach strategies. The research team has also recruited and formed a Community Advisory Board (CAB) consisting of five community members to ensure cultural relevance and shared ownership of the research process. This board has met once to date, with a primary focus on addressing recruitment challenges.

Recruitment has started and is ongoing; flyers were shared at two Black community events, put at the Bantaba (African) restaurant, the Masjid, and the African Market in Lynnwood. Flyers have also been shared via email and on community platforms through the community partners. Despite these efforts, there are no confirmed participants among the targeted women at this stage. Participation has been slower than anticipated due to heightened fear and hesitancy among immigrant communities related to current immigration enforcement concerns, which has affected willingness to engage. Adaptive recruitment strategies are being implemented to address these challenges. Several strategies were discussed during the first CAB meeting to strengthen recruitment, including engaging a community recruiter to actively identify and approach potential participants. Additionally, we plan to advertise this study on social media at a cost.

This recruitment phase, scheduled for March to May 2026, will depend on the success of recruitment efforts over the next several weeks. If the recruitment strategies above prove effective, the project is expected to proceed as planned. Overall, the project is progressing as planned and is well-positioned to continue data collection and move into data analysis, with a strong foundation of trust, collaboration and community partnerships.

Investigators
Wanda Pratt, Information School
Yue Wu, Department of Ophthalmology
Mike Teodorescu, Information School
Kennedy Orwa, Information School

Project update
This project aims to lay the foundation for equitable, AI-enabled diabetic retinopathy (DR) screening by addressing limitations in generalizability of diagnostic AI systems across diverse populations and healthcare settings. The project is guided by two core objectives: (1) generating a rigorous, peer-review-ready evidence base on bias and generalizability in DR diagnostic AI, and (2) engaging multidisciplinary stakeholders to co-design pathways toward AI-ready health systems.

At mid-term, substantial progress has been achieved across both aims. The systematic literature review is complete, with a manuscript prepared for peer review. This work is synthesized through the Biomedical Informatics Generalizability (BIG) Framework, which conceptualizes generalizability as a systems-level property emerging from interactions across four interdependent layers: model, data, workflow and governance. The framework highlights how variability at any layer propagates across the system, shaping model performance and clinical outcomes.

In parallel, stakeholder engagement has progressed from initial mapping to active, context-grounded partnership development. A field mission to western Kenya was successfully conducted, involving site visits to multiple healthcare institutions and consultations with clinical and administrative stakeholders. This mission generated critical insights into infrastructure, workflow, and data ecosystem constraints, while also establishing strong institutional relationships and identifying partners for ongoing collaboration.