About

Sally L. Baxter is an Associate Professor of Ophthalmology at UC San Diego. She is a comprehensive ophthalmologist and informaticist with research interests that include health information technology integration into clinical workflows, predictive modeling utilizing traditional statistical methods, machine learning, deep learning, and natural language processing. Her work focuses on the implementation and clinical translation of these models, as well as integrating sensors, wearables, and other digital health technologies to support clinical care. Baxter is especially interested in leveraging technology to improve healthcare outcomes for special populations such as older individuals, people with disabilities, and minority groups. Her research activities include developing multi-modal health information technology innovations for precision management of glaucoma and applying artificial intelligence to deliver healthcare from the eye.

Research topics

• Medicine
• Computer Science
• Political Science
• Family medicine
• Psychology
• Ophthalmology
• Medical education
• Data science
• Pedagogy
• Machine Learning
• Internal medicine
• Data Mining
• Telecommunications
• Engineering
• Biology
• Optometry
• Medical physics
• Pathology
• Surgery
• Nursing

Selected publications

• Social Determinants of Health Associated with Orbital and Facial Fractures in the United States using National Institutes of Health All of Us Research Program

  Ophthalmology · 2026-02-01

  articleOpen access
  PublisherOA PDFDOI

• Use of Electronic Health Records to Examine Margin-to-Reflex Distance in Normal Patients

  Research Square · 2026-02-27

  preprintOpen access
  PublisherOA PDFDOI

• Broadband Access and Ophthalmologist Density in the United States: Cross-Sectional Questionnaire Study

  JMIR Public Health and Surveillance · 2026-04-10

  articleOpen accessSenior author

  Background: Rural US communities experience disproportionately high rates of visual disability yet have limited access to ophthalmologists. Teleophthalmology may help address these gaps, but its effectiveness depends on broadband connectivity. The relationship between broadband access and ophthalmologist density has not been well characterized. Objective: The aim of this study is to quantify the association between household broadband access-defined as subscription rates or connection prevalence-and county-level ophthalmologist density and to identify sociodemographic predictors of access. Methods: We conducted an ecological study of all 3141 US counties using 2019 data from the American Community Survey, Area Health Resources File, and National Center for Health Statistics (NCHS). Broadband access was the primary exposure; ophthalmologist count with county population as an offset was the outcome. The primary analysis used negative binomial regression, adjusting for urbanicity, income, education, age, sex, race/ethnicity, unemployment, and insurance status. Sensitivity analyses included population-weighted linear regression and state fixed effects models. County-level heatmaps illustrated geographic patterns. Results: Median household broadband access was 56.6%, ranging from 72.2% in the most urban counties (NCHS category 1) to 49.1% in the most rural (NCHS category 6). In unadjusted negative binomial regression, each 10-percentage-point increase in broadband access was associated with a 68% higher ophthalmologist rate (incidence rate ratio=1.68, 95% CI 1.61-1.76; P<.001). After adjustment, each 10-percentage-point increase was associated with a 46% higher rate (incidence rate ratio=1.46, 95% CI 1.37-1.56; P<.001). Sensitivity analyses were consistent with primary analysis. Regions with both low broadband access and zero ophthalmologist density were concentrated in the South, Mountain West region, and Alaska. Conclusions: Broadband access is strongly associated with ophthalmologist availability across US counties, independent of sociodemographic factors. Areas lacking ophthalmologists also tend to lack broadband adoption, creating compounded barriers to both in-person and teleophthalmic care. Efforts to expand broadband may support more equitable access to vision services in underserved regions.

  PublisherDOI

• Multi-ancestry genome-wide association study in all of Us for primary open-angle glaucoma

  Scientific Reports · 2026-03-17

  articleOpen accessSenior author

  This study aims to identify new genetic loci associated with primary open-angle glaucoma (POAG) and explore shared genetic risk factors across African, European, and Admixed American/Latino populations. Genome-wide Association Study (GWAS) utilizing data from the All of Us Research Program. The study included 374,254 participants, with 4,305 individuals diagnosed with POAG and 369,949 controls. Participants were categorized by ancestry: European, African, and Admixed American/Latino. We used short-read sequencing data and applied strict quality control measures (MAF > 0.01, INFO > 0.8). GWAS were conducted for each ancestry group using a logistic mixed model, adjusting for age, sex, and the top 11 principal components. A fixed-effect meta-analysis combined the results across ancestries. Genome-wide significance was set at p < 5 × 10− 8. The primary outcome measures were the identification of genetic loci associated with POAG, and the analysis of transcription factors linked to these loci in relevant tissues. In the European cohort, we identified four novel loci associated with POAG, linked to the TUT4, RYK, MOXD1, and UBAP2 genes, as well as the previously known TMCO1 locus. In the African cohort, we found five new loci, including TSPAN17, SLC16A7, LOC100506869, LINC02388, and LOC107984606. For the Admixed American/Latino cohort, we identified GATA5, FAM135B, and LINC00871 genes as novel loci. Our analysis identified three novel loci in individuals of European ancestry, mapped to the genes TUT4, RYK, and MOXD1. We identified 56 genome-wide significant variants, including six putative novel loci, and found that most ancestry-specific signals replicated in the cross-ancestry meta-analysis, with the exception of several attenuated associations in the smaller Admixed American/Latino cohort. These findings indicate that the genetic determinants contributing to POAG may differ across populations, underscoring the importance of accounting for population-specific genetic architectures in the study of complex traits. Given the substantial variation in POAG prevalence among ancestries, it is plausible that certain genetic variants exert ancestry-specific effects. Consequently, conducting ancestry-stratified GWAS is essential for elucidating these unique genetic contributions.

  PublisherOA PDFDOI

• A Manual of Procedures for the Generation of the AI-Ready and Exploratory Atlas for Diabetes Insights (AI-READI) Database

  medRxiv · 2026-04-04

  articleOpen access

  The ability to understand and affect the course of complex, multi-system diseases like diabetes has been limited by a lack of well-designed, high-quality and large multimodal datasets. The NIH Bridge2AI AI-READI project (aireadi.org) aims to address this shortfall by generating an AI-ready dataset to support AI discoveries in type 2 diabetes mellitus (T2DM). This manual of procedures provides a detailed description of the AI-READI protocol.

  PublisherOA PDFDOI

• PyOPV: An Open-Source Python Package for Ophthalmic Visual Field Data Management

  Journal of Glaucoma · 2026-02-17

  articleSenior authorCorresponding

  PRÉCIS: PyOPV is a software designed and validated for handling standard visual field DICOM files, enabling multiple functionalities for glaucoma researchers. PURPOSE: To introduce PyOPV, a novel vendor-agnostic Python-based software package we designed for the management and analysis of OPhthalmic Visual field (OPV) DICOM data. PyOPV addresses limitations in interoperability and data accessibility encountered by vision researchers by providing tools that check DICOM compliance, parse, and convert OPV DICOM files into formats easily usable for research and integration with research data systems (eg, Pandas Dataframes, JSON). METHODS: PyOPV was developed using Python 3.8.2. It uses Supplement 146 of the DICOM standard to check compliance, which defines the "ophthalmic-visual-field-static-perimetry-measurements" Composite Information Object Definition. Sample OPV DICOM files from 3 vendors that provide perimetry devices were used to design the package and analyzed for DICOM. The functionalities were then validated at 2 different institutions. RESULTS: PyOPV successfully extracted and converted OPV DICOM data into Pandas DataFrames and JSON formats, facilitating data access, analysis, and visualization. The validation on longitudinal files from different protocols demonstrated excellent agreement between PyOPV outputs and ground truth data extracted using in-place workflows of each institution. Further, it highlighted significant interoperability challenges by demonstrating missing attributes across vendors, with a considerable proportion (range: 17%-51%) of the required tags missing from the files. CONCLUSIONS: PyOPV provides an efficient solution for handling ophthalmic visual field data, bridging a critical gap in data interoperability and research scalability. It can incorporate OPV files from different vendors and distinct protocols in bulk, thereby enhancing the ability to analyze and integrate visual field data into large-scale health data warehouses, supporting ophthalmic informatics and advancing clinical research. PyOPV is limited by the vendors' failure to provide all data elements.

  PublisherDOI

• Fostering Multidisciplinary Collaboration in Artificial Intelligence and Machine Learning Education: Perspectives from the AI-READI Bootcamp (Preprint)

  2025-08-28

  articleOpen accessSenior author

  <sec> <title>BACKGROUND</title> The integration of artificial intelligence (AI) and machine learning (ML) into biomedical research requires a workforce fluent in both computational methods and clinical applications. Structured, interdisciplinary training opportunities remain limited, creating a gap between data scientists and clinicians. The National Institutes of Health’s Bridge2AI initiative launched the Artificial Intelligence–Ready and Exploratory Atlas for Diabetes Insights (AI-READI) Data Generation Project to address this gap. AI-READI is creating a multimodal, FAIR (Findable, Accessible, Interoperable, and Reusable) dataset—including ophthalmic imaging, physiologic measurements, wearable sensor data, and survey responses—from approximately 4,000 participants with or at risk for type 2 diabetes. In parallel, AI-READI established a yearlong mentored research program that begins with a two-week immersive summer bootcamp to provide foundational AI/ML skills grounded in domain-relevant biomedical data. </sec> <sec> <title>OBJECTIVE</title> To describe the design, iterative refinement, and outcomes of the AI-READI Bootcamp, and to share lessons for creating future multidisciplinary AI/ML training programs in biomedical research. </sec> <sec> <title>METHODS</title> Held annually at UC San Diego, the bootcamp combines 80 hours of lectures, coding sessions, and small-group mentorship. Year 1 introduced Python programming, classical ML techniques (e.g., logistic regression, convolutional neural networks), and data science methods such as principal component analysis and clustering, using public datasets. In Year 2, the curriculum was refined based on structured participant feedback—reducing cohort size to increase individualized mentorship, integrating the AI-READI dataset (including retinal images and structured clinical variables), and adding modules on large language models and FAIR data principles. Participant characteristics and satisfaction were assessed through standardized pre- and post-bootcamp surveys, and qualitative feedback was analyzed thematically by independent coders. </sec> <sec> <title>RESULTS</title> Seventeen participants attended Year 1 and seven attended Year 2, with an instructor-to-student ratio of approximately 1:2 in the latter. Across both years, post-bootcamp evaluations indicated high satisfaction, with Year 2 participants reporting improved experiences due to smaller cohorts, earlier integration of the AI-READI dataset, and greater emphasis on applied learning. In Year 2, mean scores for instructor effectiveness, staff support, and overall enjoyment were perfect (5.00/5.00). Qualitative feedback emphasized the value of working with domain-relevant, multimodal datasets; the benefits of peer collaboration; and the applicability of skills to structured research projects during the subsequent internship year. </sec> <sec> <title>CONCLUSIONS</title> The AI-READI Bootcamp illustrates how feedback-driven, multidisciplinary training embedded within a longitudinal mentored research program can bridge technical and clinical expertise in biomedical AI. Core elements—diverse trainee cohorts, applied learning with biomedical datasets, and sustained mentorship—offer a replicable model for preparing health professionals for the evolving AI/ML landscape. Future iterations will incorporate additional pre-bootcamp onboarding modules, objective skill assessments, and long-term tracking of research engagement and productivity. </sec>

  PublisherOA PDFDOI

• A Federated Learning-based Optic Disc and Cup Segmentation Model for Glaucoma Monitoring In Color Fundus Photographs

  medRxiv · 2025-12-04

  articleOpen access

  ABSTRACT Importance Glaucoma, a leading cause of blindness worldwide, depends on accurate optic nerve head assessment, particularly optic disc and cup segmentation, for diagnosis and monitoring. Deep learning (DL) models can automate these measurements, but models trained on smaller, site-specific datasets often fail to generalize. While larger, multi-site datasets help, data privacy concerns limit centralized training. Objective To evaluate a federated learning (FL) framework with site-specific fine-tuning for optic disc and cup segmentation, aiming to match central model performance while preserving privacy and improving generalizability. Design Comparative evaluation of three different approaches: (1) a central model trained on multi-site data, (2) site-specific local model training (3) standard FL models, against an FL with site-specific fine-tuning. Setting Multicenter study incorporating nine publicly available datasets, representing varied clinical environments, populations, and imaging protocols. Participants 5,550 color fundus photographs from at least 917 individuals across nine datasets includingboth routine care and research sources from 7 countries. Exposures Optic disc and cup segmentationin color fundus photographs using training with local model, central model, standard FL, and FL with site-specific fine-tuning.. Main Outcomes and Measures Segmentation accuracy measured by Dice score. Comparisons were labeled as performance “wins” or “losses” based on statistically significant differences via Wilcoxon signed-rank test (P &lt; 0.05). Results Site-specific fine-tuning of FL with site-specific fine tuning matched central model performance for cup segmentation across all sites (9/9) and for disc segmentation in most sites (7/9). Compared with site-specific local models, it preserved within-site performance (cup: 9/9; disc: 5/9) while substantially improving cross-site generalizability, achieving significant gains in 54.2% (39/72) of disc and 25.0% (18/72) of cup external-site evaluations, with no significant losses. Compared to standard FL pipelines, site-specific fine-tuning improved performance by 52% for disc and 26% for cup. Conclusions and Relevance Site-specific fine-tuning within an FL framework effectively personalizes generalized models to local data distributions, achieving central-level performance without data sharing and enhancing cross-site robustness. This approach enables privacy-preserving, scalable AI deployment across heterogeneous clinical settings for reproducible and generalizable glaucoma assessment KEY POINTS Question How can we train an AI model to segment the optic cup and disc across multiple sites without sharing data, yet achieve performance comparable to a central model trained on pooled datasets? Findings In this federated learning (FL) study of 5,550 fundus photographs from nine sites, a site-specific fine-tuning FL strategy matched the central model’s performance and outperformed other standard FL techniques, with notable gains in cross-site generalizability. Meaning Site-specific fine-tuning effectively personalizes FL models to local data distributions, combining data privacy with robust, generalizable performance.

  PublisherOA PDFDOI

• Broadband Access and Ophthalmologist Density in the United States: Ecological Analysis (Preprint)

  2025-11-25

  articleOpen accessSenior author

  <sec> <title>BACKGROUND</title> Rural United States (US) communities experience disproportionately high rates of visual disability yet have limited access to ophthalmologists. Teleophthalmology may help address these gaps, but its effectiveness depends on broadband availability. The relationship between broadband access and ophthalmologist density has not been well characterized. </sec> <sec> <title>OBJECTIVE</title> To quantify the association between household broadband access and county‑level ophthalmologist density and to identify sociodemographic predictors of limited broadband availability. </sec> <sec> <title>METHODS</title> We conducted an ecological study of all 3,141 US counties using 2019 data from the American Community Survey, Area Health Resources File, and National Center for Health Statistics. Broadband access was the primary exposure; ophthalmologist density per 100,000 residents was the outcome. We used simple and multiple linear regression adjusting for urbanicity, income, education, age, sex, race/ethnicity, unemployment, and insurance status. Sub-group analyses compared metropolitan and non-metropolitan counties. County-level heatmaps illustrated geographic patterns. </sec> <sec> <title>RESULTS</title> Median household broadband access was 56.6%, ranging from 72.2% in the most urban counties to 49.1% in the most rural. Each 10-percentage-point increase in broadband access was associated with 1.13 more ophthalmologists per 100,000 residents in unadjusted models and 0.86 more in adjusted models (p &lt; 0.001). Associations persisted in both metropolitan (β = 0.129) and non-metropolitan (β = 0.047) counties. Regions with both low broadband access and low ophthalmologist density were concentrated in the South, Western Mountain region, and Alaska. </sec> <sec> <title>CONCLUSIONS</title> Broadband access is strongly associated with ophthalmologist availability across U.S. counties, independent of sociodemographic factors. Areas lacking ophthalmologists also tend to lack broadband infrastructure, creating compounded barriers to both in-person and teleophthalmic care. Efforts to expand broadband may support more equitable access to vision services in underserved regions. </sec>

  PublisherDOI

• Fostering Multidisciplinary Collaboration in Artificial Intelligence and Machine Learning Education: Tutorial Based on the AI-READI Bootcamp

  JMIR Medical Education · 2025-11-10 · 1 citations

  articleOpen accessSenior author

  Background: The integration of artificial intelligence (AI) and machine learning (ML) into biomedical research requires a workforce fluent in both computational methods and clinical applications. Structured, interdisciplinary training opportunities remain limited, creating a gap between data scientists and clinicians. The National Institutes of Health's Bridge to Artificial Intelligence (Bridge2AI) initiative launched the Artificial Intelligence-Ready and Exploratory Atlas for Diabetes Insights (AI-READI) data generation project to address this gap. AI-READI is creating a multimodal, FAIR (findable, accessible, interoperable, and reusable) dataset-including ophthalmic imaging, physiologic measurements, wearable sensor data, and survey responses-from approximately 4000 participants with or at risk for type 2 diabetes. In parallel, AI-READI established a year-long mentored research program that begins with a 2-week immersive summer bootcamp to provide foundational AI/ML skills grounded in domain-relevant biomedical data. Objective: To describe the design, iterative refinement, and outcomes of the AI-READI Bootcamp, and to share lessons for creating future multidisciplinary AI/ML training programs in biomedical research. Methods: Held annually at the University of California San Diego, the bootcamp combines 80 hours of lectures, coding sessions, and small-group mentorship. Year 1 introduced Python programming, classical ML techniques (eg, logistic regression, convolutional neural networks), and data science methods, such as principal component analysis and clustering, using public datasets. In Year 2, the curriculum was refined based on structured participant feedback-reducing cohort size to increase individualized mentorship, integrating the AI-READI dataset (including retinal images and structured clinical variables), and adding modules on large language models and FAIR data principles. Participant characteristics and satisfaction were assessed through standardized pre- and postbootcamp surveys, and qualitative feedback was analyzed thematically by independent coders. Results: Seventeen participants attended Year 1 and 7 attended Year 2, with an instructor-to-student ratio of approximately 1:2 in the latter. Across both years, postbootcamp evaluations indicated high satisfaction, with Year 2 participants reporting improved experiences due to smaller cohorts, earlier integration of the AI-READI dataset, and greater emphasis on applied learning. In Year 2, mean scores for instructor effectiveness, staff support, and overall enjoyment were perfect (5.00/5.00). Qualitative feedback emphasized the value of working with domain-relevant, multimodal datasets; the benefits of peer collaboration; and the applicability of skills to structured research projects during the subsequent internship year. Conclusions: The AI-READI Bootcamp illustrates how feedback-driven, multidisciplinary training embedded within a longitudinal mentored research program can bridge technical and clinical expertise in biomedical AI. Core elements-diverse trainee cohorts, applied learning with biomedical datasets, and sustained mentorship-offer a replicable model for preparing health professionals for the evolving AI/ML landscape. Future iterations will incorporate additional prebootcamp onboarding modules, objective skill assessments, and long-term tracking of research engagement and productivity.

  PublisherDOI

Recent grants

• PAGE-G: Precision Approach combining Genes and Environment in Glaucoma

  NIH · $316k · 2023–2026

• Multimodal Artificial Intelligence to Predict Glaucomatous Progression and Surgical Intervention

  NIH · $1.8M · 2022–2026

• Short-Term Research training In Vision and Eye health (STRIVE)

  NIH · $144k · 2022–2027

• Bridge2AI: Salutogenesis Data Generation Project

  NIH · $32.7M · 2022–2026

• Multi-modal Health Information Technology Innovations for Precision Management of Glaucoma

  NIH · $2.1M · 2020–2025

Frequent coauthors

• Bharanidharan Radha Saseendrakumar

  Fleet Science Center

  47 shared

• Robert N. Weinreb

  University of California, San Diego

  45 shared

• Lucila Ohno‐Machado

  36 shared

• Eric Nudleman

  31 shared

• Niloofar Radgoudarzi

  UC San Diego Health System

  30 shared

• Don O. Kikkawa

  University of California, San Diego

  29 shared

• Fritz Gerald P. Kalaw

  University of California, San Diego

  28 shared

• Christine Thorne

  UC San Diego Health System

  27 shared

Education

• Ph.D., Ophthalmology

  University of California, San Diego

  2005

• M.D., Medicine

  University of California, San Diego

  2001

• B.A., Biology

  University of California, San Diego

  1997

Awards & honors

• Heed Ophthalmic Foundation 2018 - 2019 Heed Fellowship
• UCSD 2017 - 2018 Lamont Ericson, MD Award for Outstanding Pa…
• California Academy of Eye Physicians and Surgeons 2017 Starr…
• National Eye Institute 2017 Travel Grant for the Association…
• University of Pennsylvania 2014 Charles A. Oliver Memorial P…