About

Nina Joyce is the Manning Assistant Professor of Epidemiology at Brown University and a member of the Center for Gerontology and Health Care Research (CGHCR). Her work focuses on drawing causal inferences from observational data with applications to older adults' health, traffic safety, and pharmacoepidemiology. She is also interested in advancing methods for generalizing findings from cluster randomized trials to new target populations, with a particular focus on applications for nursing home research. Nina Joyce is the founder and director of the Causal Inference Working Group (CIWG), an interdepartmental group at the school of public health dedicated to causal inference. She completed her PhD at Brown University in 2015 and her postdoctoral training at Harvard Medical School, Department of Health Care Policy, from 2015 to 2017. She has been an assistant professor at Brown University School of Public Health since 2018, teaching courses such as Intermediate Methods in Epidemiologic Research, Contemporary History of Epidemiologic Methods, and Modern Foundations of Causal Inference in Epidemiologic Research.

Research topics

• Medicine
• Environmental health
• Gerontology
• Ecology
• Biology
• Environmental science
• Psychiatry
• Psychology
• Geography

Selected publications

• Concordance Between the Minimum Data Set Kidney Impairment I1500 Item and eGFR Records in Nursing Homes

  Journal of the American Medical Directors Association · 2026-04-02

  articleOpen access

  OBJECTIVES: To evaluate the concordance between the Minimum Data Set (MDS) version 3.0 I1500 item and estimated glomerular filtration rate (eGFR) laboratory values for identifying kidney impairment among nursing home (NH) residents. DESIGN: A retrospective cohort study was conducted using data from 8 multistate, multifacility NH chains from January 1, 2018, to July 1, 2022. MDS 3.0 assessments with a complete I1500 item were linked to eGFR values measured on the assessment reference date or within 7 days prior. SETTING AND PARTICIPANTS: The study included NH residents with at least 1 MDS 3.0 assessment linked to a matching eGFR within a 7-day look-back window. METHODS: for the primary analysis and <45 for the secondary analysis. RESULTS: The study included 454,436 MDS assessments from 225,557 NH residents. Using an eGFR threshold of <60, agreement with the MDS I1500 item was 60.3% (kappa = 0.19), with 25.8% sensitivity and 92.3% specificity. At the <45 threshold, agreement rose to 71.8% (kappa = 0.22), with 29.9% sensitivity and 89.2% specificity. CONCLUSIONS AND IMPLICATIONS: The MDS 3.0 I1500 measure demonstrated fair agreement, low sensitivity, and high specificity for identifying kidney impairment compared with eGFR values. Cross-referencing federally mandated MDS data with eGFR data could improve kidney impairment identification and care planning in NHs.

  PublisherDOI

• Linkage of Medicare insurance claims to police-reported motor vehicle crashes: advancing traffic safety research in older adult populations

  Injury Epidemiology · 2026-02-06

  articleOpen accessSenior author

  BACKGROUND: Motor vehicle crashes (MVCs) are a leading cause of injury among adults aged 65 years and older ("older adults"). As the number of older drivers grows, it is increasingly important to understand clinical factors associated with an increased risk of MVC. A major barrier, however, is the lack of data. To address this, we linked two large-scale administrative databases, the New Jersey Safety and Health Outcomes (NJ-SHO) Data Warehouse, which contains information on all police-reported crashes in New Jersey from 2004 to 2019, and Medicare Fee-for-Service (FFS) insurance claims, which contains health care encounters and prescription drug dispensings among older adults in the United States over the same period. This paper explains the linkage process, describes selected work leveraging these data to study MVCs in older drivers, and highlights features and strengths of this linkage for future research. METHODS: The NJ-SHO-Medicare linkage was performed using categories of name (first and last), sex, age (birth and death date), and residence (state and ZIP code). Matches were ranked by quality and overall confidence. RESULTS: After comparing different match strategies, we accepted a match when (1) the name match quality was High or Medium and the age match was High or (2) the name, sex, and residence match categories were all High. Of the 2,722,773 individuals successfully linked, we accepted 2,661,782 matches (97.76% of individuals linked and 91.59% of those submitted for linkage). All accepted matches were Strong or Fair. Among accepted matches who enrolled in Medicare FFS in 2019, 342,422 (28.57%) were 65-69 years old, 619,437 (51.69%) were female, and 955,309 (79.72%) were non-Hispanic White. Only 29,561 (2.47%) experienced an MVC and 25,478 (2.13%) received a citation. The most prevalent clinical conditions ever diagnosed were cataracts (669,044; 55.83%); chronic pain, fatigue, and fibromyalgia (367,165; 30.64%); and glaucoma (287,420; 23.98%). CONCLUSIONS: With extensive temporal and population coverage, the NJ-SHO-Medicare linkage supports studying the relationships between clinical exposures (e.g., medications ), driving events (e.g., crashes, citations) and medical care trajectories, which can help advance the driving safety of older adults and inform future efforts to integrate administrative data.

  PublisherDOI

• Linkage of Medicare insurance claims to police-reported motor vehicle crashes: advancing traffic safety research in older adult populations

  Figshare · 2026-02-06

  otherOpen accessSenior author

  Abstract Background Motor vehicle crashes (MVCs) are a leading cause of injury among adults aged 65 years and older (“older adults”). As the number of older drivers grows, it is increasingly important to understand clinical factors associated with an increased risk of MVC. A major barrier, however, is the lack of data. To address this, we linked two large-scale administrative databases, the New Jersey Safety and Health Outcomes (NJ-SHO) Data Warehouse, which contains information on all police-reported crashes in New Jersey from 2004 to 2019, and Medicare Fee-for-Service (FFS) insurance claims, which contains health care encounters and prescription drug dispensings among older adults in the United States over the same period. This paper explains the linkage process, describes selected work leveraging these data to study MVCs in older drivers, and highlights features and strengths of this linkage for future research. Methods The NJ-SHO–Medicare linkage was performed using categories of name (first and last), sex, age (birth and death date), and residence (state and ZIP code). Matches were ranked by quality and overall confidence. Results After comparing different match strategies, we accepted a match when (1) the name match quality was High or Medium and the age match was High or (2) the name, sex, and residence match categories were all High. Of the 2,722,773 individuals successfully linked, we accepted 2,661,782 matches (97.76% of individuals linked and 91.59% of those submitted for linkage). All accepted matches were Strong or Fair. Among accepted matches who enrolled in Medicare FFS in 2019, 342,422 (28.57%) were 65–69 years old, 619,437 (51.69%) were female, and 955,309 (79.72%) were non-Hispanic White. Only 29,561 (2.47%) experienced an MVC and 25,478 (2.13%) received a citation. The most prevalent clinical conditions ever diagnosed were cataracts (669,044; 55.83%); chronic pain, fatigue, and fibromyalgia (367,165; 30.64%); and glaucoma (287,420; 23.98%). Conclusions With extensive temporal and population coverage, the NJ-SHO–Medicare linkage supports studying the relationships between clinical exposures (e.g., medications ), driving events (e.g., crashes, citations) and medical care trajectories, which can help advance the driving safety of older adults and inform future efforts to integrate administrative data.

  PublisherDOI

• Additional file 1 of Linkage of Medicare insurance claims to police-reported motor vehicle crashes: advancing traffic safety research in older adult populations

  Figshare · 2026-02-06

  articleOpen accessSenior author

  Supplementary Material 1.

  PublisherDOI

• Linkage of Medicare insurance claims to police-reported motor vehicle crashes: advancing traffic safety research in older adult populations

  Figshare · 2026-02-06

  otherOpen accessSenior author

  Abstract Background Motor vehicle crashes (MVCs) are a leading cause of injury among adults aged 65 years and older (“older adults”). As the number of older drivers grows, it is increasingly important to understand clinical factors associated with an increased risk of MVC. A major barrier, however, is the lack of data. To address this, we linked two large-scale administrative databases, the New Jersey Safety and Health Outcomes (NJ-SHO) Data Warehouse, which contains information on all police-reported crashes in New Jersey from 2004 to 2019, and Medicare Fee-for-Service (FFS) insurance claims, which contains health care encounters and prescription drug dispensings among older adults in the United States over the same period. This paper explains the linkage process, describes selected work leveraging these data to study MVCs in older drivers, and highlights features and strengths of this linkage for future research. Methods The NJ-SHO–Medicare linkage was performed using categories of name (first and last), sex, age (birth and death date), and residence (state and ZIP code). Matches were ranked by quality and overall confidence. Results After comparing different match strategies, we accepted a match when (1) the name match quality was High or Medium and the age match was High or (2) the name, sex, and residence match categories were all High. Of the 2,722,773 individuals successfully linked, we accepted 2,661,782 matches (97.76% of individuals linked and 91.59% of those submitted for linkage). All accepted matches were Strong or Fair. Among accepted matches who enrolled in Medicare FFS in 2019, 342,422 (28.57%) were 65–69 years old, 619,437 (51.69%) were female, and 955,309 (79.72%) were non-Hispanic White. Only 29,561 (2.47%) experienced an MVC and 25,478 (2.13%) received a citation. The most prevalent clinical conditions ever diagnosed were cataracts (669,044; 55.83%); chronic pain, fatigue, and fibromyalgia (367,165; 30.64%); and glaucoma (287,420; 23.98%). Conclusions With extensive temporal and population coverage, the NJ-SHO–Medicare linkage supports studying the relationships between clinical exposures (e.g., medications ), driving events (e.g., crashes, citations) and medical care trajectories, which can help advance the driving safety of older adults and inform future efforts to integrate administrative data.

  PublisherDOI

• Additional file 1 of Linkage of Medicare insurance claims to police-reported motor vehicle crashes: advancing traffic safety research in older adult populations

  Figshare · 2026-02-06

  articleOpen accessSenior author

  Supplementary Material 1.

  PublisherDOI

• Center-specific causal inference with multicenter trials—Interpreting trial evidence in the context of each participating center

  Statistical Methods in Medical Research · 2025-10-21

  article

  In multicenter randomized trials, when effect modifiers have a different distribution across centers, comparisons between treatment groups that average (standardize) effects over centers may not apply to any of the populations underlying the individual centers. In the presence of such heterogeneity, interpreting the evidence produced by a multicenter trial in the context of the local population underlying each center may be necessary. Here, we identify center-specific effects under conditions that are largely supported by the study design and are weaker than those underlying popular methods for the analysis of multicenter studies, in the presence of associations between center membership and the outcome ("center-outcome associations" conditional on baseline covariates and treatment). We then consider an additional testable condition of "no center-outcome associations," given baseline covariates and treatment. We propose methods for estimating center-specific average treatment effects, when center-outcome associations are present and when they are absent. When center-outcome associations are absent, we discuss how the proposed methods are often more efficient and make weaker conditions than related transportability methods applied to multicenter trials. We evaluate the performance of the methods in a simulation study and illustrate their implementation using data from the Hepatitis C Antiviral Long-Term Treatment Against Cirrhosis trial.

  PublisherDOI

• Effects of Initiating Different Antidepressant Subclasses on Motor Vehicle Crash Risk Among Older Adults With Depression

  Journal of the American Geriatrics Society · 2025-05-20

  articleOpen access

  BACKGROUND: Antidepressants are prescribed for depression among older adults but might increase the risk of motor vehicle crash (MVC) through adverse effects (AEs) like sedation, dizziness, and blurred vision. Antidepressant subclasses may have different MVC risks since AE risks vary across subclasses. Our objective was to estimate the comparative one-year risks of MVC upon initiating atypical (AA) or tricyclic (TCA) versus selective serotonin reuptake inhibitor (SSRI) antidepressants. METHODS: We emulated 470 sequential target trials each week from January 6, 2008, through January 1, 2017, using Medicare fee-for-service claims linked to New Jersey police-reported MVCs and driver's licensing data. Our sequential target trial emulation included older adults aged ≥ 66 years with a recent diagnosis of depression who initiated AAs, SSRIs, or TCAs. The unit of analysis was the "person-trial" a unique instance of a person in a sequential trial. Using inverse probability of treatment and censoring weighted Kaplan-Meier estimators to account for potential confounding and selection bias, we estimated the intention-to-treat cumulative incidence and risk ratios (RRs) of MVC over 1 year of follow-up. RESULTS: We identified 13,034 person-trials from 11,604 persons (median [first quartile, third quartile] age: 76.0 [71.0, 82.0] years, 69.8% female, 89.4% non-Hispanic White race). There were 31 (37.6 [95% confidence limits {CLs} 20.3, 59.5] per 1000), 65 (37.6 [95% CLs 25.7, 47.7] per 1000), and 380 (38.0 [95% CLs 24.1, 39.7] per 1000) MVCs among 644 TCA-treated, 2130 AA-treated, and 10,260 SSRI-treated person-trials, respectively. The adjusted RRs were 0.99 (95% CLs 0.72, 1.56) comparing AAs versus SSRIs and 0.99 (95% CLs 0.56, 1.86) comparing TCAs versus SSRIs. CONCLUSION: We observed no differences in the one-year risk of MVC between antidepressant subclasses. When selecting among antidepressant subclasses to manage depression in older adults, MVC risk should not guide prescribing decisions, and other considerations should take precedence.

  PublisherOA PDFDOI

• Agreement Between Fingerstick Blood Glucose and Continuous Glucose Monitor Measures Among Long‐Term Care Facility Residents

  Journal of the American Geriatrics Society · 2025-08-02 · 6 citations

  articleOpen access

  BACKGROUND: Long-term care facility (LTCF) residents with diabetes are at high risk of hypoglycemia. Continuous glucose monitoring (CGM), which measures interstitial glucose at 5-min intervals over 10-14 days, and fingerstick blood glucose (FBG) which analyzes glucose from a drop of blood, are both used to monitor glucose levels. Observational studies using electronic health record (EHR) data containing FBG measures could help to identify ways to reduce hypoglycemia risk. We first need to understand the validity of such data. Our objective was to compare EHR-based FBG measures against reference-standard CGM measures of hypoglycemia. METHODS: We studied two cohorts of residents with diabetes in parallel. In Cohort 1, we analyzed linked CGM and Long-Term Care Data Cooperative EHR-based FBG data collected in 2023. In Cohort 2, we analyzed linked CGM and EHR-based FBG data obtained directly from LTCFs between 2022 and 2023. We defined hypoglycemia as glucose < 70 mg/dL and assessed the sensitivity and specificity of FBG versus CGM measures to detect hypoglycemia. The unit of analysis was each pair of contemporaneous FBG-CGM measures. RESULTS: In Cohort 1, two White female residents with a mean (standard deviation [SD]) age of 81 [12.7] years generated 25 daily hypoglycemia measurements. The sensitivity and specificity were 14% and 100%, respectively, for FBG-measured hypoglycemia. Cohort 2 included 40 residents (mean [SD] age 68 [11] years, 45% females, 60% White race) who generated 425 daily measurements of hypoglycemia. The sensitivity and specificity were 13% and 99%, respectively. CONCLUSION: EHR FBG measures of hypoglycemia had high specificity but failed to identify four out of every five hypoglycemic events among LTCF residents. Researchers and healthcare providers should assume hypoglycemia is measured with substantial errors in EHRs and account for this in their research and clinical practice.

  PublisherOA PDFDOI

• Reply to: Comment on: Agreement Between Fingerstick Blood Glucose and Continuous Glucose Monitor Measures Among Long‐Term Care Facility Residents

  Journal of the American Geriatrics Society · 2025-11-19

  articleOpen access

  We thank Dr. Song for their thoughtful and constructive letter, “Comment on: Agreement Between Fingerstick Blood Glucose and Continuous Glucose Monitor Measures Among Long-Term Care Facility Residents” about our article [1]. We appreciate their attention to this critical issue of hypoglycemia under-detection in long-term care facilities (LTCFs) and concur with their call to revise current glucose monitoring practices to advance patient care and safety. Below, we address Dr. Song's comments and reference our prior letter for additional context on the points raised [2]. We agree with Dr. Song's sentiments about the importance of transparent reporting in observational studies, particularly disclosing the basis for inclusion and exclusion criteria, and describing any informative patterns of missing data. In reporting our findings, we adhered to the guidelines on linking data in pharmacoepidemiology and observational research [3, 4]. As delineated in our article, residents in Cohort 1 who generated continuous glucose monitoring (CGM) readings were matched to their corresponding records in the Long-Term Care Data Cooperative (LTCDC) using the date of birth, sex, LTCF name, zip code, state, and calendar day of the glucose measurements [1]. Of the 9 residents, 5 could be connected to the LTCDC electronic health records (EHRs). Among those connected, 3 residents were dropped for lacking CGM and fingerstick blood glucose (FBG) measures on the same days. To facilitate a valid comparison, it was crucial to keep only residents who had both CGM and LTCDC glucose measures. Any absence of same-day CGM-FBG measures reflects logistical and clinical factors such as variability in residents' daily care schedules, instances when FBG tests were delayed because glucose values were recently obtained or residents were asleep, periods when CGM sensors were temporarily removed, or differences in timing between routine CGM readings and FBG testing, rather than investigator exclusion of data. We recognize that the small size of Cohort 1 may have reduced the generalizability of our findings. Accordingly, we analyzed a relatively larger sample of residents in Cohort 2, for whom both CGM and FBG data were directly available, thereby circumventing the data-linkage limitations encountered in Cohort 1, where 4 out of 9 (44%) of the residents without successful linkage to the LTCDC data were necessarily excluded. The findings from both cohorts were largely consistent, lending credibility to the main conclusion: FBG fails to capture a significant proportion of hypoglycemia events compared to CGM. We emphasize that including a second, larger cohort does not eliminate issues around external validity; rather, it mitigates the issue, thereby strengthening the clinical interpretation and relevance of our findings. As we described more fully in our previous correspondence [5], we encourage future researchers to study more heterogeneous LTCF populations to enhance the generalizability of results. Sampling a diverse array of LTCFs is also important since there is facility-level variation in glucose monitoring practices, including the timing of FBG checks and hypoglycemia response protocols, which could influence the detection of hypoglycemia. Finally, we agree with Dr. Song that variation in FBG meter models can affect measurement accuracy and limit generalizability. This further underscores the advantage of CGM, which provides continuous glucose data rather than a single reading and offers a more complete and reliable picture of glycemic control. Based on our findings, reliance on EHR-based FBG data alone can lead to many undetected hypoglycemia events, and consequently, missed opportunities to de-intensify glucose-lowering therapy and prevent harmful outcomes. Although current LTCF glucose monitoring relies almost exclusively on FBG testing, our study adds to a growing body of evidence that CGM is a valuable emerging technology that could improve hypoglycemia detection in LTCFs. At the same time, it is important to acknowledge that CGM use is not without downsides. Among diabetic individuals using glucose monitoring devices, including CGMs, the most frequently reported cutaneous reaction has been allergic contact dermatitis, an inflammatory skin reaction caused by wearing the sensor for long periods and exposure to certain chemicals in the adhesive [6]. Skin tears, bruising, and bleeding might also be of concern in frail or anticoagulated LTCF residents. Although rare, local infection or cellulitis could be of concern if CGMs remain in place for longer than recommended or if hygiene is difficult to maintain. In line with Dr. Song's recommendation, CGM remains warranted for large-scale use in LTCFs, as its benefits outweigh the risks. We deeply appreciate Dr. Song's engagement with our study and the important points raised in their letter. Together, our study [1] and the accompanying letters [2, 5] form a constructive dialogue that advances the evidence base on improving glucose monitoring in LTCFs and helps set priorities for future research and clinical implementation. All authors are responsible for the content of this letter. The Long-Term Care Data Cooperative (LTCDC) is sponsored by the National Institute on Aging (NIA) through a supplemental grant (U54AG063546-S6) to the NIA Imbedded Pragmatic Alzheimer's Disease and AD-Related Dementias Clinical Trials Collaboratory (NIA IMPACT Collaboratory). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health nor the investigators of the IMPACT Collaboratory or the LTCDC. The sponsors had no role in the design, methods, subject recruitment, analysis, and preparation of the paper or any other aspect of the work. Andrew R. Zullo received grant funding from Sanofi paid directly to Brown University for collaborative research on the epidemiology of infections and vaccine use among LTCF residents. He is a U.S. Government employee; the views expressed in this manuscript are those of the author and do not necessarily reflect the position or policy of the Department of Veterans Affairs, or, the United States Government. Medha N. Munshi is a consultant for Sanofi. The other authors declare no conflicts of interest. This publication is linked to a related Letter to the Editor by Xiaohong Song To view this article, visit https://doi.org/10.1111/jgs.70204.

  PublisherOA PDFDOI

Frequent coauthors

• Andrew R. Zullo

  Brown University

  74 shared

• Vincent Mor

  Providence College

  66 shared

• Gregory A. Wellenius

  Universidad de Sevilla

  46 shared

• Seth A. Margolis

  Brown University

  44 shared

• Allison E. Curry

  University of Pennsylvania

  41 shared

• Brian R. Ott

  Brown University

  37 shared

• Stefan Gravenstein

  Brown University

  35 shared

• Susan C. Miller

  Brown University

  26 shared

Education

• Ph.D.

  Brown University

  2015

• Other

  Harvard Medical School, Department of Health Care Policy

  2015