About

Kelly M. Bakulski, PhD, is an Associate Professor in the Department of Epidemiology at the University of Michigan School of Public Health. Her research team aims to understand the environmental and genetic etiologies of neurological disorders, promoting rigorous science and the development of scientific expertise. Dr. Bakulski is an environmental health scientist with expertise in life course heavy metals exposure testing related to neurodevelopment and neurodegeneration. She is also a molecular epidemiologist with experience analyzing across multiple levels of the genome, including the epigenome and transcriptome. Her research incorporates population approaches and laboratory experiments to develop biomarkers and cell type tools that inform molecular epidemiology inferences. Her research projects focus on the risk of Alzheimer's disease and related dementias from perinatal lead exposure, evaluating the effects on brain regions and cell types, and understanding how early life lead exposures impact lifelong brain health. She is involved in the Study of the Environment and Alzheimer's Disease and Related Dementias (SEAD), which evaluates modifiable environmental exposures such as lead and cadmium to advance understanding of dementia etiology and inform prevention strategies. Additionally, her work includes studying DNA methylation, genetics, and modifiable risk factors of dementia in diverse populations, as well as assessing toxicant-induced disruptions in gestational tissues with implications for adverse pregnancy outcomes. Dr. Bakulski is actively engaged with the Michigan Alzheimer's Disease Research Center, promoting research, diagnosis, and treatment of dementias through collaborative efforts, and providing training and research opportunities for healthcare professionals, scientists, and students. She is passionate about teaching and mentoring in her field, contributing to the advancement of knowledge in environmental health, neurogenetics, and epidemiology.

Research topics

• Genetics
• Biology
• Medicine
• Bioinformatics
• Computational biology
• Internal medicine
• Immunology
• Psychology
• Cell biology
• Chemistry
• Endocrinology
• Pathology
• Neuroscience
• Psychiatry
• Biochemistry
• Environmental chemistry

Selected publications

• Linking sleep apnea and arthritis in the National Alzheimer Coordinating Center Cohort: A cross-sectional analysis

  Medicine · 2026-02-20

  articleOpen access

  Sleep apnea-related intermittent hypoxia and the chronic inflammation of arthritis share oxidative-stress pathways, yet their epidemiologic overlap remains under-described. The prevalence of both conditions increases with age and presents unique challenges for patient management. To quantify the association between clinician-suspected arthritis and self-reported sleep apnea and to explore whether demographic or cognitive factors modify that link. We analyzed 17,013 adults enrolled in the referral-based National Alzheimer Coordinating Center Uniform Data Set, version 3. Complete-case binary logistic regression modeled obstructive sleep apnea (OSA) (yes/no) on arthritis (yes/no) with adjustment for age, sex, race (White vs non-White), years of education, cognitive status (normal, mild cognitive impairment, Alzheimer disease), body mass index, and cardiometabolic comorbidities. A pre-specified interaction term tested whether cognition modified the arthritis-OSA association. Multiple imputation was used to address missing data. Arthritis was associated with 60% higher odds of OSA (adjusted odds ratio = 1.60, 95% confidence interval: 1.46-1.76, P < .001). The effect was attenuated in Alzheimer disease. Male sex, atrial fibrillation, stroke, diabetes, and higher body mass index were additional correlates (all P < .001); age was not independently significant. Imputation yielded similar estimates. Clinician-suspected arthritis was robustly associated with self-reported OSA even after extensive adjustment, although unmeasured confounding and exposure misclassification cannot be excluded. Both OSA and arthritis were ascertained by self-report or single-clinician designation without polysomnography, actigraphy, imaging, or serology, raising non-differential misclassification potential. The cross-sectional design prevents causal interpretation, and the predominantly White, highly educated volunteer cohort limits generalizability. Prospective, objectively phenotyped studies, ideally with arthritis sub-typing, are needed to verify directionality and clarify mechanisms. We used records from more than 17,000 volunteers at U.S. Alzheimer Disease Research Centers to ask whether people who say they have arthritis are also more likely to report OSA. After controlling for age, sex, education, cognitive status, weight, and common medical conditions, arthritis still raised the odds of OSA by about 60%. Joint pain and poor sleep can feed off 1 another, so recognizing both problems may help doctors treat them earlier. Neither arthritis nor OSA was confirmed with X-rays, lab tests, or sleep studies, we relied on what participants or clinicians reported. Furthermore, the study looked at 1 point in time, so we cannot tell which problem came 1st; and most volunteers were White and highly educated, so the findings may not reflect every community. Future research that tracks patients over time and uses overnight sleep tests and detailed arthritis subtypes will be crucial.

  PublisherDOI

• Exposure to lead and incidence of Alzheimer's disease and all‐cause dementia in the United States

  Alzheimer s & Dementia · 2026-02-01 · 1 citations

  articleOpen access

  Abstract INTRODUCTION Growing evidence suggests lead exposure may increase dementia risk, but evidence from human studies is limited. We investigated prospective associations between lead exposure and incident Alzheimer's disease (AD) and all‐cause dementia in nationally‐representative US populations. METHODS Baseline measured blood lead and estimated patella and tibia lead from the National Health and Nutrition Examination Survey (NHANES)‐III (1988‐1994, blood n = 6,217, bone n = 5,865) and continuous NHANES (1999‐2016, blood n = 8,038, bone n = 4,824) were linked to Medicare and the National Death Index for incident AD and all‐cause dementia, with up to 30 years of follow‐up. Survey‐weighted Cox regressions estimated hazard ratios (HRs) and 95% confidence intervals (CIs). RESULTS In continuous NHANES, estimated patella lead was associated with AD (HR = 2.96, 95% CI:1.37‐6.39) and all‐cause dementia (HR = 2.15, 95% CI:1.33‐3.46), comparing quartile‐4 vs. quartile‐1. We observed weaker associations in NHANES‐III. Blood lead showed no association. DISCUSSION These findings suggest cumulative lead as a potential dementia risk factor. Highlights We examined Medicare‐linked National Health and Nutrition Examination Survey (NHANES) on lead exposure and incident Alzheimer's disease (AD) and dementia. Lead exposure was assessed by blood lead and algorithm‐estimated bone lead levels. High estimated patella lead was linked to increased all‐cause dementia incidence. Dementia cases drop by 18% if all patella lead levels reduce to the 25 th percentile. Cumulative lead exposure may raise dementia risk, highlighting its potential impact.

  PublisherOA PDFDOI

• MethylModes: computationally efficient detection of multimodal distributions in DNA methylation data

  Bioinformatics · 2026-01-03

  articleOpen access

  SUMMARY: MethylModes is an R package and Shiny application to identify multimodal distributions in human DNA methylation at individual CpG sites. Multimodal distributions, which can be the result of nearby genetic variation, environmental exposures, or assay artifacts, are susceptible to confounding and important to identify for methylation analysis. MethylModes is easily incorporated into existing quality control pipelines of array-based DNA methylation data. The underlying algorithm uses kernel smoothing of probe-level data to locate the number and location of peaks. The algorithm can be parallelized across probes for efficient implementation at genome-scale. We provide a case study implementation of MethylModes in the Health and Retirement Study as well as the Airwave Health Monitoring Study. AVAILABILITY AND IMPLEMENTATION: MethylModes is available on GitHub at https://github.com/lutiffan/methylModes as an R package wrapping an R Shiny application. We include a toy dataset to validate installation. The codebase is also published on Zenodo at https://doi.org/10.5281/zenodo.17448517.

  PublisherDOI

• Particulate air pollution and domain‐specific cognition among Black adults

  Alzheimer s & Dementia Behavior & Socioeconomics of Aging · 2026-05-18

  article

  Abstract INTRODUCTION Long‐term exposure to fine particulate air pollution (PM 2.5 ) may affect cognitive function, yet evidence in diverse cohorts with long‐term exposure averages is lacking. METHODS We analyzed data from 740 adults aged 53 to 94 in the Study of Healthy Aging in African Americans (STAR). Cognitive performance was assessed in three domains (semantic memory, verbal episodic memory, and executive function) and standardized to Z ‐scores. Using linear regression, we evaluated associations with 5‐, 10‐, and 17‐year average PM 2.5 exposure. Fully adjusted models included age, sex, marital status, education, and neighborhood income. RESULTS Greater long‐term PM 2.5 exposure over 17 years was associated with lower semantic memory ( β = −0.61 SD, 95% confidence interval: [−1.03, −0.19] per 5 µg/m 3 ). Associations were similar but slightly attenuated for 5‐ and 10‐year exposures. No associations were observed with executive function or verbal episodic memory. DISCUSSION Long‐term PM 2.5 exposure may contribute to lower semantic memory in midlife and later‐life Black adults.

  PublisherDOI

• Cadmium Exposure and Incidence of All-Cause Dementia and Alzheimer’s Disease in US Adults

  medRxiv · 2026-01-19

  articleSenior author

  Introduction: While longitudinal studies aid in understanding and preventing long-latency disorders like dementia, evidence for cadmium's role in these conditions is still limited. We evaluated associations between cadmium exposure and incident Alzheimer's disease (AD) and all-cause dementia in US adults. Methods: National Health and Nutrition Examination Survey (NHANES) III (1988-1994) and continuous NHANES (1999-2016) data were linked with Medicare claims to identify incident AD and dementia cases through 2018. Urinary and/or blood cadmium were measured during NHANES. We used covariate-adjusted, survey-weighted Cox proportional hazard models to evaluate the associations between cadmium exposure biomarkers and AD/dementia over follow-up. Results: In NHANES III (N=6,122), baseline age was 53.9±0.5 years and urinary cadmium was 0.8±0.02 ug/L. Over a follow-up of 20.4±0.3 years, 743 AD and 1,508 all-cause dementia cases occurred. Urinary cadmium was not associated with AD (HR: 1.01, 95% CI: 0.9-1.0) nor all-cause dementia incidence (HR: 1.02, 95% CI: 0.96-1.08). In continuous NHANES (urinary cadmium N=2,833; blood cadmium N=8,038), baseline age was 64.1±0.2 years, urinary cadmium was 0.5±0.03 ug/L, and blood cadmium was 0.6±0.01 ug/L. Over 9.5±0.1 years, 587 AD and 1,260 all-cause dementia cases occurred. Urinary and blood cadmium showed no associations with AD (HR [95% CI]: 1.09 [0.9, 1.4]; 1.06 [0.9, 1.2]) nor all-cause dementia (HR [95% CI]: 1.07 [0.9, 1.3]; 1.06 [0.95, 1.2]). Conclusion: No association between cadmium exposure and dementia incidence was observed. Our null findings should be interpreted while considering potential methodological issues and verified by subsequent studies.

  PublisherDOI

• Air Pollution and the Progression of Physical Function Limitations and Disability in Aging Adults

  JAMA Network Open · 2026-02-11

  articleOpen access

  Importance: Physical disability reflects the cumulative burden of chronic conditions. Although generally progressive, episodes of disability can be followed by periods of recovery; therefore, there is a need to identify modifiable risk factors that contribute to the dynamic development of disability. Objective: To investigate air pollution as a modifiable risk factor of transitions between states of no physical function limitation, physical function limitations, and activities of daily living (ADL) disability. Design, Setting, and Participants: This cohort study included respondents older than 50 years from the nationally representative Health and Retirement Study (HRS) who participated in at least 2 interviews between 2000 and 2016. Data analysis was conducted from July 2023 to August 2025. Exposures: Ten-year average ambient concentrations of particulate matter with a diameter of 2.5 µm or less (PM2.5), PM with a diameter between 10 and 2.5 µm (PM10-2.5), nitrogen dioxide (NO2), and ozone (O3) were estimated at respondent residential addresses preceding each survey using spatiotemporal models. Main Outcomes and Measures: Physical disability states were assessed using self-reported mobility and Activities of Daily Living (ADL). To examine associations of exposure to air pollution with transitioning between states of physical disability, multistate models, adjusted for individual- and area-level covariates, were used. Results: The sample included 29 790 respondents (mean [SD] age, 63 [11] years; 16 878 [57%] women; 3371 [11%] Hispanic, 5240 [18%] non-Hispanic Black, and 20 314 [68%[ non-Hispanic White), who were followed up for a mean (SD) of 8 (6) years. IQR-increments in PM2.5, PM10-2.5, and NO2 concentrations were mostly associated with greater hazards of transitioning from a state of no physical function limitation toward disability; a 1-IQR increment for PM2.5 was associated with lower odds of a reverse transition. For example, in the single-pollutant model, a 1-IQR higher PM2.5 concentration was associated with a hazard ratio (HR) of 1.06 (95% CI, 1.03-1.09) for transitioning from no physical function limitations to physical function limitations and an HR of 0.96 (95% CI, 0.93-0.99) for reverting back to healthy physical function from physical function limitations. By contrast, a 1-IQR higher O3 concentration was associated with lower hazards of transitioning from no physical function limitations to physical function limitations (HR, 0.92; 95% CI, 0.86-0.98) and ADL disability (HR, 0.89; 95% CI, 0.81-0.97). Conclusions and Relevance: These findings suggest that air pollution may affect the progression of physical disability and hinder recovery in later life.

  PublisherOA PDFDOI

• Blood Essential Trace Elements and Alzheimer's Disease Biomarkers in Midlife

  Alzheimer s & Dementia · 2025-12-01

  articleOpen access

  BACKGROUND: Alzheimer's Disease (AD) is a progressive neurodegenerative disorder characterized by cognitive decline, impacting millions globally. Essential trace elements are implicated in key age-related physiologic processes but have not been fully examined with respect to AD etiology. This study investigates associations between serum levels of essential trace elements (manganese, iron, cobalt, copper, zinc, selenium, and molybdenum) and AD biomarkers (Aβ42, Aβ42/Aβ40 ratio, p-tau181, and total tau) in midlife women. METHOD: This cross-sectional study included 194 midlife women (median age=53.3 years) from the Study of Women's Health Across the Nation, Michigan site. Serum levels of trace elements were measured using inductively coupled plasma-mass spectrometry, and AD biomarkers were quantified using single molecule array assays. Multivariable linear regression models assessed potential associations and Bayesian kernel machine regression (BKMR) was used to account for complex co-exposures and non-linear relationships. RESULT: In the multivariable linear regression models, a doubling of serum molybdenum level was associated with 9.4% higher Aβ42/40 ratio (95% CI: 0.8%, 18.6%; p = 0.03), and a doubling of serum cobalt level with 17.5% higher p-tau181 level (95% CI: 3.1%, 33.8%; p = 0.02). Copper showed an inverse association with the Aβ42/40 ratio, while zinc was positively associated with the Aβ42/40 ratio, though these associations were marginally significant. BKMR analysis confirmed these associations. CONCLUSION: This study identified statistically significant associations of serum molybdenum and cobalt levels with AD biomarkers, suggesting a potential protective effect of molybdenum against Aβ aggregation and exacerbation of pathologic tau phosphorylation by cobalt. These findings underscore the need for further longitudinal studies to explore the role of essential trace elements in AD pathogenesis.

  PublisherOA PDFDOI

• Effects of self-rated mental and physical work demands on cognition are dependent in a cross-sectional sample of the Health and Retirement Study

  Journal of Occupational and Environmental Hygiene · 2025-11-19 · 1 citations

  articleSenior authorCorresponding

  This study assessed whether self-rated physical and mental work demands were associated with cognition among older working adults and whether their effects were dependent. The cross-sectional sample consisted of 6,377 working older adults using the Health and Retirement Study in 2004. Self-rated work demands were summarized from four questions about the frequency of mental or physical demands in the respondent's current job. Cognition was assessed using a subset of the Telephone Interview for Cognitive Status. Multivariable linear regression assessed the associations and additive interaction between physical and mental work demands and cognition, adjusted for age, sex, race, education, and practice effect. Independently, higher physical work demands were associated with poorer cognition (change per one unit increase: 0.50, 95% CI: 0.36, 0.65), and higher mental work demands were associated with better cognition (change per one unit increase: -0.31, 95% CI: -0.44, -0.19). The effect of one work demand measure became more negative as the level of the other increased (Beta for interaction = -0.23, 95% CI: -0.43, -0.03). A one-point increase in mental work demands was associated with 0.79 (95% CI: 0.51, 1.08) points higher cognition score when physical work demands were lowest, but was not associated with cognition when physical work demands were highest (0.11, 95% CI: -0.26, 0.48). The highest predicted cognition score was for the highest mental and lowest physical work demands. Results were robust to additional adjustment for health and behavior covariates. The associations of self-rated mental and physical work demands on cognition are dependent. Future studies should strongly consider examining interactions to capture the range of work demand effects.

  PublisherDOI

• Evaluation of a panel of plasma biomarkers for Alzheimer's disease in a diverse research cohort

  Journal of Alzheimer s Disease · 2025-11-10

  article

  Background Plasma biomarkers show significant promise for Alzheimer's disease (AD) diagnostics and risk prediction, however, much less is known about how these assays perform in a diverse research cohort of older adults. Objective To compare plasma biomarkers with clinical diagnoses and assess variability by demographic factors in a diverse research cohort. Methods Among 331 University of Michigan Memory and Aging Project (UM-MAP) participants, plasma biomarkers (pTau-217, pTau-181, GFAP, NfL, Aβ 42 , Aβ 40 , t-Tau) were measured. Demographic information (age, sex, education, race) was self-reported. Clinical consensus phenotypes (dementia of the Alzheimer Type (DAT), mild cognitive impairment (MCI), cognitively unimpaired (CU) were based on neuropsychological assessments. Logistic regression with machine learning for model variable selection was used to compare participants by clinical phenotypes. Results Comparing CU and DAT participants, areas under the curve (AUCs) from receiver operator characteristic curves of single biomarker models ranged from 0.74–0.89. Optimal performance (AUC 99.7) was observed from stepwise regression with backward selection, which identified pTau-217, GFAP, sex, education, APOE ε4 allele, and race as model variables. When comparing MCI and DAT participants, only pTau-217 differed significantly (AUC 0.80). pTau-181 and pTau-217 levels were higher in white participants than Black/African American participants across all clinical phenotypes. Conclusions Plasma biomarkers demonstrate promise for improving diagnostic accuracy in diverse research cohorts. Incorporating demographic variables facilitates enhanced interpretability of biomarker levels and the development of reference ranges.

  PublisherDOI

• DNA methylation age acceleration and cognitive status among older adults in the US, 1999‐2002

  Alzheimer s & Dementia · 2025-12-01

  articleOpen accessSenior author

  BACKGROUND: Biological aging, measured by DNA methylation, is a potential biomarker for cognitive health outcomes. The purpose of this analysis was to evaluate associations between measures of aging using DNA methylation and cognition in a nationally representative sample of adults aged 60+ in the National Health and Nutrition Examination Survey (NHANES). METHOD: Annual cross-sectional NHANES surveys (1999-2002) with DNA methylation measures and cognitive testing were combined. Genome-wide DNA methylation data were used to create 13 measures of biological aging trained on different aging phenotypes. Cognition was assessed with the Digit Symbol Substitution Test (DSST). Mild cognitive impairment was categorized as DSST score below the weighted 25th percentile. To evaluate the associations between each DNA methylation measure and continuous DSST score, survey weighted linear regression models adjusted for age, sex, race/ethnicity, education, smoking, serum cotinine, and BMI were run. Sensitivity analyses included assessing effect modification by sex, education, and race/ethnicity and running modified Poisson models with the binary mild cognitive impairment outcome. p-values were adjusted using the false discovery rate method. RESULT: Included participants (N = 1,463) were an average of 70.5 years old, 82.7% non-Hispanic White, and 71.6% high school educated or higher. The average DSST score was 46.9 points (maximum 117). 10 of 13 DNA methylation measures were associated with cognition measured by DSST (adjusted p <0.05). One year of GrimAge2 accelerated aging was associated with -0.41 points lower DSST score (95% CI: -0.59, -0.24; adjusted p = 0.001). One standard-deviation decrease in estimated telomere length was associated with -2.19 points lower DSST score (95% CI: -1.16, -3.22; adjusted p = 0.001). In stratified analyses, higher magnitudes of association were observed among male and non-Hispanic White participants across multiple aging measures. Mild cognitive impairment was observed for 39.7% of participants (DSST < 35). Associations between DNA methylation measures and mild cognitive impairment were not significant after multiple comparison adjustment. CONCLUSION: DNA methylation may be a useful biomarker of cognitive status among older adults.

  PublisherOA PDFDOI

Recent grants

• Characterizing disparities in late-onset Alzheimer's disease risk through polygenic risk and epidemiologic factors in the Health and Retirement Survey

  NIH · $1.7M · 2017–2024

• DNA Methylation,Genetics, and Modifiable Risk Factors of Dementia in a Nationally Representative, Multi-Ethnic Cohort

  NIH · $3.8M · 2020–2027

Frequent coauthors

• M. Daniele Fallin

  Emory University

  134 shared

• John Dou

  University of Michigan–Ann Arbor

  130 shared

• Kelly S. Benke

  Johns Hopkins University

  93 shared

• Erin B. Ware

  University of Michigan–Ann Arbor

  89 shared

• Irva Hertz‐Picciotto

  University of California, Davis

  81 shared

• Rebecca J. Schmidt

  Cohort (United Kingdom)

  75 shared

• Jason I. Feinberg

  Center for Autism and Related Disorders

  68 shared

• Heather E. Volk

  Johns Hopkins University

  60 shared