CCCheryl Renee Clark

Cheryl Renee Clark

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Cheryl Renee Clark has a total of 349 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Health system–community partnerships in a Medicaid nutrition and housing program, 2020-2023: a mixed-methods evaluation

    by Jessica L McCurley, Vicki Fung, Douglas E Levy, Sydney McGovern, Christine Vogeli, Cheryl R Clark, Anne N Thorndike

    Introduction: Health systems are partnering with social service organizations (SSOs) to establish health-related social needs (HRSN) programs, yet best practices have not been established. Methods: This study assessed the implementation of a partnership between one accountable care organization (ACO) and SSOs in Massachusetts' Medicaid Flexible Services Program (FSP) from 2020 to 2023 to address food and housing insecurity. To be FSP-eligible, individuals needed to be in the ACO, have food or housing insecurity, and meet specific health criteria. Quantitative measures of reach, enrollment, and services were integrated with findings from interviews with ACO (n = 12) and SSO (n = 11) staff. Results: Overall, 15.1% of 16 289 FSP-eligible adults and 11.8% of 5476 FSP-eligible children were enrolled. Nutrition enrollments increased over time, but housing enrollments remained low. ACO and SSO staff described positive impacts, including tailored support for patients' needs, SSO organizational growth, and understanding of how SSOs can partner with health care. Implementation challenges included complex contracts, administrative burden, and housing shortages. State funding for SSO infrastructure, cross-sector data sharing, workflow simplification, and ACO–SSO communication improved efficiency and reduced burden. Conclusion: The ACO–SSO partnerships expanded resources to address beneficiaries' HRSN. Understanding factors that contribute to success on both sides of the partnership can inform future HRSN programs.

  • Open Access

    Extracting Social Determinants of Health From Electronic Health Records: Development and Comparison of Rule-Based and Large Language Model Methods

    by Bo Wang, Dia Kabir, Cheryl Renee Clark, Karmel W Choi, Jordan W Smoller

    Background: Social determinants of health (SDoH) are critical drivers of health outcomes but are often underdocumented in structured electronic health record (EHR) data. Instead, SDoH are more commonly recorded in unstructured clinical notes, and unlocking this information could have far-reaching implications for advancing population health research and informing clinical decision-making. Objective: This study develops and systematically evaluates cost-efficient methods for extracting SDoH information from unstructured clinical text using rule-based natural language processing (NLP) and large language model (LLM)–based approaches. Methods: We constructed a gold-standard annotated corpus comprising clinical text segments from 171 patients in the Mass General Brigham Research Patient Data Registry, covering 7 SDoH domain categories and 23 subcategories. A rule-based system (RBS) was developed and evaluated alongside 7 OpenAI GPT models (GPT-4o, 4.1, 4.1-mini, o4-mini, GPT-5, GPT-5-mini, and o3) under zero-shot and few-shot settings using multiple prompting strategies. We additionally implemented late-fusion ensemble approaches that combined outputs from rule- and LLM-based methods. Performance was assessed using precision, recall, and F1-score, alongside qualitative error analysis. Results: The RBS achieved high precision for SDoH domain categories (0.96) but substantially lower recall (0.68). GPT-based models consistently outperformed the RBS in overall recall and F1-scores. The best domain-level performance was observed for GPT-5 and GPT-5-mini in few-shot settings (F1-score=0.89), while o4-mini achieved the highest subcategory-level performance (F1-score=0.88). A late-fusion ensemble integrating RBS and GPT outputs further improved domain-level performance (F1-score=0.92), with balanced precision (0.93) and recall (0.90), but did not improve subcategory-level performance. Conclusions: Recent GPT models with advanced reasoning capabilities, including the newly released mini models (eg, o4-mini and GPT-5-mini), demonstrated strong performance for SDoH extraction without task-specific fine-tuning and consistently outperformed the rule-based NLP system. Integrating rule- and LLM-based methods via late fusion further enhanced domain-level extraction performance. Our results demonstrate a cost-efficient framework for the accurate identification of SDoH from clinical text, facilitating downstream population health research and clinical informatics applications.


  • Open Access

    Retention strategies and predictors of retention for longitudinal observational cohort studies: A scoping review

    by Jeonghyun Shin, Alexa M. Bond, Marlika Marceau, Alexandra Businger, Marcelo G. Binker, Colin A. Settle, Callie A. Gu, Jordan W. Smoller, Cheryl Clark

    ObjectivesParticipant retention is key to generating meaningful findings in longitudinal research studies. Previous systematic reviews have identified strategies for increasing participant retention. This scoping review examines recent innovations in methodology used for participant retention and representation to inform retention practices in the All of Us Research Program (AoU).MethodsThree databases (PubMed, Embase, and CINAHL) were searched for recent studies to include in the literature review. Landmark longitudinal cohort studies were manually selected and searched on study websites and PubMed. AoU archive materials were identified on the AoU Confluence platform that stores internal study resources. Data on retention strategies were extracted, categorized according to retention themes identified in the systematic reviews by Robinson et al. (2015) and Teague et al. (2018), and summarized. Additional data on predictors of retention, digital retention methods, and new retention strategies were also extracted and narratively synthesized.ResultsTwenty-two longitudinal observational studies, 10 landmark cohort studies, and materials from 28 sites in the 14 AoU consortia met inclusion criteria. Contact and scheduling methods, visit characteristics, and special tracking methods were the most common retention themes used among the three data sources. Disability and lack of healthcare access were associated with lower odds of retention, whereas higher education and older age were associated with higher odds of retention. Digital strategies mostly involved the use of study websites, social media platforms, and digital tools to conduct follow-up, contact participants, or track participant information. The most common new strategies that were not included in previous reviews were retention committees, staff meetings on retention, and check-in calls. Segmentation approaches were identified as a new retention theme, including retention call lists and participant segmentation to individualize retention.ConclusionsThis scoping review reports on current methods used to retain participants in recent literature, landmark longitudinal cohort studies, and AoU archive materials. The review identified recent progress in the development and implementation of participant retention practices, as well as gaps in research assessing the effectiveness of specific retention strategies on increasing retention rates among different demographics, especially those who are underrepresented in biomedical research (UBR).

  • Open Access

    Extracting Social Determinants of Health From Electronic Health Records: Development and Comparison of Rule-Based and Large Language Model Methods (Preprint)

    by Bo Wang, Dia Kabir, Cheryl Renee Clark, Karmel W Choi, Jordan W Smoller

    BACKGROUND Social determinants of health (SDoH) are critical drivers of health outcomes but are often underdocumented in structured electronic health record (EHR) data. Instead, SDoH are more commonly recorded in unstructured clinical notes, and unlocking this information could have far-reaching implications for advancing population health research and informing clinical decision-making. OBJECTIVE This study develops and systematically evaluates cost-efficient methods for extracting SDoH information from unstructured clinical text using rule-based natural language processing (NLP) and large language model (LLM)–based approaches. METHODS We constructed a gold-standard annotated corpus comprising clinical text segments from 171 patients in the Mass General Brigham Research Patient Data Registry, covering 7 SDoH domain categories and 23 subcategories. A rule-based system (RBS) was developed and evaluated alongside 7 OpenAI GPT models (GPT-4o, 4.1, 4.1-mini, o4-mini, GPT-5, GPT-5-mini, and o3) under zero-shot and few-shot settings using multiple prompting strategies. We additionally implemented late-fusion ensemble approaches that combined outputs from rule- and LLM-based methods. Performance was assessed using precision, recall, and F1-score, alongside qualitative error analysis. RESULTS The RBS achieved high precision for SDoH domain categories (0.96) but substantially lower recall (0.68). GPT-based models consistently outperformed the RBS in overall recall and F1-scores. The best domain-level performance was observed for GPT-5 and GPT-5-mini in few-shot settings (F1-score=0.89), while o4-mini achieved the highest subcategory-level performance (F1-score=0.88). A late-fusion ensemble integrating RBS and GPT outputs further improved domain-level performance (F1-score=0.92), with balanced precision (0.93) and recall (0.90), but did not improve subcategory-level performance. CONCLUSIONS Recent GPT models with advanced reasoning capabilities, including the newly released mini models (eg, o4-mini and GPT-5-mini), demonstrated strong performance for SDoH extraction without task-specific fine-tuning and consistently outperformed the rule-based NLP system. Integrating rule- and LLM-based methods via late fusion further enhanced domain-level extraction performance. Our results demonstrate a cost-efficient framework for the accurate identification of SDoH from clinical text, facilitating downstream population health research and clinical informatics applications.

  • Barriers to the Equitable Implementation of Risk-Based Cervical Cancer Management Guidelines

    by Cheryl R. Clark, Jacquelyn M. Lykken, Patricia M. Chen, Jennifer S. Haas, Sarah Feldman, Chul Ahn, Callie A. Gu, Sarah Kobrin, Michelle I. Silver, Steven J. Atlas, et al.

    Background The 2019 American Society for Colposcopy and Cervical Pathology provides risk-based management guidelines for abnormal cervical cancer screening test result management. It is not known whether populations facing inequities in primary care delivery have the prior screening history needed to take advantage of risk-based management of abnormal cervical cancer screening results. Objective Determine the prevalence and associations of unknown screening history for people with an abnormal cervical cancer screening result. Design Retrospective multi-center population-based study of three large, structurally diverse healthcare systems in Washington state, Massachusetts, and Dallas County, Texas. Participants Females ages 25–65 at index abnormal Pap cytology or positive Human Papillomavirus (HPV) result from 2010 to 2019 (n = 63,739). Main Measure(s) Unknown cervical cancer screening history, assessed as missing or insufficient documentation of cervical screening, diagnostic tests, and procedures extracted from administrative data and electronic health records of health systems. Key Results Over a third of patients had unknown cervical cancer screening history prior to the index abnormal test during the study period (38.3% overall, 27.6 to 52.3% across healthcare systems). Patients empaneled to a primary care provider at the time of their abnormal test had a lower prevalence of an unknown screening history than those who were not (prevalence ratio 0.74, 95% CL 0.72–0.75). Those who were uninsured or had multiple insurance transitions had a higher prevalence of unknown screening history than those who were commercially insured (PR 1.13, 95% CL 1.09–1.16). There were health system-specific associations with having an unknown screening history. Conclusions In three large health systems, historical screening data needed to implement ASCCP risk-based guidelines were frequently missing. Our findings call for greater attention to potential management inequities for patients with missing data. Importantly, more attention is needed to address potential primary care access barriers to equitable delivery of risk-based cervical cancer surveillance and care.


  • Top Cited Papers
    Open Access

    Measuring social determinants of health in the All of Us Research Program

    by Samantha Tesfaye, Robert M. Cronin, Maria Lopez-Class, Qingxia Chen, Christopher S. Foster, Callie A. Gu, Andrew Guide, Robert A. Hiatt, Angelica S. Johnson, Christine L. M. Joseph, et al.

    To accelerate medical breakthroughs, the All of Us Research Program aims to collect data from over one million participants. This report outlines processes used to construct the All of Us Social Determinants of Health (SDOH) survey and presents the psychometric characteristics of SDOH survey measures in All of Us. A consensus process was used to select SDOH measures, prioritizing concepts validated in diverse populations and other national cohort surveys. Survey item non-response was calculated, and Cronbach’s alpha was used to analyze psychometric properties of scales. Multivariable logistic regression models were used to examine associations between demographic categories and item non-response. Twenty-nine percent (N = 117,783) of eligible All of Us participants submitted SDOH survey data for these analyses. Most scales had less than 5% incalculable scores due to item non-response. Patterns of item non-response were seen by racial identity, educational attainment, income level, survey language, and age. Internal consistency reliability was greater than 0.80 for almost all scales and most demographic groups. The SDOH survey demonstrated good to excellent reliability across several measures and within multiple populations underrepresented in biomedical research. Bias due to survey non-response and item non-response will be monitored and addressed as the survey is fielded more completely.


  • Open Access

    Measuring Social Determinants of Health in the All of Us Research Program: Technical Document

    by Samantha Tesfaye, Robert Cronin, Maria Lopez-Class, Qingxia Chen, Christopher S. Foster, Callie Gu, Andrew Guide, Robert A. Hiatt, Angelica Johnson, Christine Lm Joseph, et al.

    Background: To advance precision medicine and improve population health, the All of Us Research Program aims to collect data, including a survey of social determinants of health (SDOH), from over 1 million participants. This report (1) outlines the process used to construct the All of Us SDOH survey and (2) presents the psychometric characteristics and scoring recommendations for SDOH survey measures. Methods: A consensus process was used to select a definition of SDOH and conceptual frameworks to develop inclusion and exclusion criteria and priorities for construct consideration. Cognitive interviewing was used to provide an assessment of SDOH survey item performance in English and Spanish. Scales and scored items were constructed in alignment with validated literature. Item non-response was calculated, and Cronbach ‘s alpha was used to analyze the psychometric properties of each scale, overall and by demographic characteristics. Multivariable logistic regression models were used to examine if demographic categories were associated with item non-response. Results: Twenty-nine percent (N=117,783) of All of Us participants submitted SDOH survey data by June 30, 2022. Among those who provided any SDOH survey data, item non-response was infrequent, with most scales having less than 5% incalculable scores due to item non-response. Item non-response varied most along the lines of racial identity, educational attainment, and the language in which the survey was administered (Spanish or English). In our regression models, for most scales, patterns of missing data due to item non-response were seen by racial identity, educational attainment, income level, and age. Internal consistency reliability was greater than 0.80 for almost all scales, with variability by racial identity, educational attainment, and the language of survey administration. Conclusion: The SDOH survey demonstrated good to excellent reliability across several measures of SDOHs and within multiple population groups that are underrepresented in biomedical research. Bias due to survey non-response and item non-response should be monitored and addressed as the survey is fielded more completely.

  • Top Cited Papers

    Racial and Ethnic Disparities in Opioid Access and Urine Drug Screening Among Older Patients With Poor-Prognosis Cancer Near the End of Life

    by Andrea C. Enzinger, Kaushik Ghosh, Nancy L. Keating, David M. Cutler, Cheryl R. Clark, Narjust Florez, Mary Beth Landrum, Alexi A. Wright

    PURPOSE: To characterize racial and ethnic disparities and trends in opioid access and urine drug screening (UDS) among patients dying of cancer, and to explore potential mechanisms. METHODS: Among 318,549 non-Hispanic White (White), Black, and Hispanic Medicare decedents older than 65 years with poor-prognosis cancers, we examined 2007-2019 trends in opioid prescription fills and potency (morphine milligram equivalents [MMEs] per day [MMEDs]) near the end of life (EOL), defined as 30 days before death or hospice enrollment. We estimated the effects of race and ethnicity on opioid access, controlling for demographic and clinical factors. Models were further adjusted for socioeconomic factors including dual-eligibility status, community-level deprivation, and rurality. We similarly explored disparities in UDS. RESULTS: Between 2007 and 2019, White, Black, and Hispanic decedents experienced steady declines in EOL opioid access and rapid expansion of UDS. Compared with White patients, Black and Hispanic patients were less likely to receive any opioid (Black, –4.3 percentage points, 95% CI, –4.8 to –3.6; Hispanic, –3.6 percentage points, 95% CI, –4.4 to –2.9) and long-acting opioids (Black, –3.1 percentage points, 95% CI, –3.6 to –2.8; Hispanic, –2.2 percentage points, 95% CI, –2.7 to –1.7). They also received lower daily doses (Black, –10.5 MMED, 95% CI, –12.8 to –8.2; Hispanic, –9.1 MMED, 95% CI, –12.1 to –6.1) and lower total doses (Black, –210 MMEs, 95% CI, –293 to –207; Hispanic, –179 MMEs, 95% CI, –217 to –142); Black patients were also more likely to undergo UDS (0.5 percentage points; 95% CI, 0.3 to 0.8). Disparities in EOL opioid access and UDS disproportionately affected Black men. Adjustment for socioeconomic factors did not attenuate the EOL opioid access disparities. CONCLUSION: There are substantial and persistent racial and ethnic inequities in opioid access among older patients dying of cancer, which are not mediated by socioeconomic variables. [Media: see text]


  • Contributors to Early Mortality in African Americans, the Jackson Heart Study

    by Katherine C. Brooks, Mark J. Ommerborn, Lara I. Brewer, Mario Sims, Adolfo Correa, Gabriel S. Tajeu, Cheryl R. Clark

    Introduction: In recent years, premature “deaths of despair” (ie, due to alcohol, drug use, and suicide) among middle-aged White Americans have received increased attention in the popular press, yet there has been less discussion on what explains premature deaths among young African Americans. In this study, we examined factors related to deaths of despair (alcohol use, drug use, smoking) and contextual factors (perceived discrimination, socioeconomic status, neighborhood conditions) as predictors of premature deaths before the age of 65 years among African Americans. Methods: The Jackson Heart Study (JHS) is a longitudinal cohort study of African Americans in the Jackson, Mississippi, metropolitan statistical area. We included participants younger than 65 years at baseline (n=4000). Participant enrollment began in 2000 and data for these analyses were collected through 2019. To examine predictors of mortality, we calculated multivariable adjusted hazard ratios (HRs; 95% CI), using Cox proportional hazard models adjusted for age, sex, ideal cardiovascular health metrics, drug use, alcohol intake, functional status, cancer, chronic kidney disease, asthma, waist circumference, depression, income, education, health insurance status, perceived neighborhood safety, and exposure to lifetime discrimination. Results: There were 230 deaths in our cohort, which spanned from 2001-2019. After adjusting for all covariates, males (HR, 1.50; 95% CI, 1.11-2.03), participants who used drugs (HR, 1.53; 95% CI, 1.13-2.08), had a heavy alcohol drinking episode (HR, 1.71; 95% CI, 1.22-2.41), reported 0-1 ideal cardiovascular health metrics (HR, 1.78; 95% CI, 1.06-3.02), had cancer (HR, 2.38; 95% CI, 1.41-4.01), had poor functional status (HR, 1.68; 95% CI, 1.19-2.37), or with annual family income less than $25,000 (HR, 1.63; 95% CI, 1.02-2.62) were more likely to die before 65 years of age. Conclusions: In our large cohort of African American men and women, clinical predictors of premature death included poor cardiovascular health and cancer, and social predictors included low income, drug use, heavy alcohol use, and being a current smoker. Clinical and social interventions are warranted to prevent premature mortality in African Americans.


  • Achieving a Representative Sample of Asian Americans in Biomedical Research Through Community-Based Approaches: Comparing Demographic Data in the All of Us Research Program With the American Community Survey

    by Fornessa T. Randal, Paula Lozano, Siya Qi, Chieko Maene, Sameep Shah, Yicklun Mo, Francis Ratsimbazafy, Eric Boerwinkle, Mine Cicek, Cheryl R. Clark, et al.

    Background: Underrepresented persons are often not included in biomedical research. It is unknown if the general Asian American population is being represented in All of Us . The purpose of this study was to compare the Asian demographic data in the All of Us cohort with the Asian nationally representative data from the American Community Survey. Method: Demographic characteristics and health literacy of Asians in All of Us were examined. Findings were qualitatively compared with the Asian data in the 2019 American Community Survey 1-year estimate. Results: Compared with the national composition of Asians, less All of Us participants were born outside the United States (64% vs 79%), were younger, and had higher levels of education (76% vs 52%). Over 60% of All of Us participants reported high levels of health literacy. Conclusion: This study had implications for the development of strategies that ensure diverse populations are represented in biomedical research.