1. Epidemiology and Global Health
  2. Medicine

Type 2 diabetes mellitus accelerates brain aging and cognitive decline: Complementary findings from UK Biobank and meta-analyses

  1. Botond Antal
  2. Liam P McMahon
  3. Syed Fahad Sultan
  4. Andrew Lithen
  5. Deborah J Wexler
  6. Bradford Dickerson
  7. Eva-Maria Ratai
  8. Lilianne R Mujica-Parodi  Is a corresponding author
  1. Department of Biomedical Engineering, Stony Brook University, United States
  2. Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital and Harvard Medical School, United States
  3. Department of Computer Science, Stony Brook University, United States
  4. Diabetes Center, Massachusetts General Hospital and Harvard Medical School, United States
  5. Department of Neurology, Massachusetts General Hospital and Harvard Medical School, United States
  6. Department of Neurology, Stony Brook University School of Medicine, United States
https://doi.org/10.7554/eLife.73138
Share this article
Cite this article
  1. Botond Antal
  2. Liam P McMahon
  3. Syed Fahad Sultan
  4. Andrew Lithen
  5. Deborah J Wexler
  6. Bradford Dickerson
  7. Eva-Maria Ratai
  8. Lilianne R Mujica-Parodi
(2022)
Type 2 diabetes mellitus accelerates brain aging and cognitive decline: Complementary findings from UK Biobank and meta-analyses
eLife 11:e73138.
https://doi.org/10.7554/eLife.73138
  2. Download BibTeX
  3. Download .RIS

Abstract

Background:

Type 2 diabetes mellitus (T2DM) is known to be associated with neurobiological and cognitive deficits; however, their extent, overlap with aging effects, and the effectiveness of existing treatments in the context of the brain are currently unknown.

Methods:

We characterized neurocognitive effects independently associated with T2DM and age in a large cohort of human subjects from the UK Biobank with cross-sectional neuroimaging and cognitive data. We then proceeded to evaluate the extent of overlap between the effects related to T2DM and age by applying correlation measures to the separately characterized neurocognitive changes. Our findings were complemented by meta-analyses of published reports with cognitive or neuroimaging measures for T2DM and healthy controls (HCs). We also evaluated in a cohort of T2DM-diagnosed individuals using UK Biobank how disease chronicity and metformin treatment interact with the identified neurocognitive effects.

Results:

The UK Biobank dataset included cognitive and neuroimaging data (N = 20,314), including 1012 T2DM and 19,302 HCs, aged between 50 and 80 years. Duration of T2DM ranged from 0 to 31 years (mean 8.5 ± 6.1 years); 498 were treated with metformin alone, while 352 were unmedicated. Our meta-analysis evaluated 34 cognitive studies (N = 22,231) and 60 neuroimaging studies: 30 of T2DM (N = 866) and 30 of aging (N = 1088). Compared to age, sex, education, and hypertension-matched HC, T2DM was associated with marked cognitive deficits, particularly in executive functioning and processing speed. Likewise, we found that the diagnosis of T2DM was significantly associated with gray matter atrophy, primarily within the ventral striatum, cerebellum, and putamen, with reorganization of brain activity (decreased in the caudate and premotor cortex and increased in the subgenual area, orbitofrontal cortex, brainstem, and posterior cingulate cortex). The structural and functional changes associated with T2DM show marked overlap with the effects correlating with age but appear earlier, with disease duration linked to more severe neurodegeneration. Metformin treatment status was not associated with improved neurocognitive outcomes.

Conclusions:

The neurocognitive impact of T2DM suggests marked acceleration of normal brain aging. T2DM gray matter atrophy occurred approximately 26% ± 14% faster than seen with normal aging; disease duration was associated with increased neurodegeneration. Mechanistically, our results suggest a neurometabolic component to brain aging. Clinically, neuroimaging-based biomarkers may provide a valuable adjunctive measure of T2DM progression and treatment efficacy based on neurological effects.

Funding:

The research described in this article was funded by the W. M. Keck Foundation (to LRMP), the White House Brain Research Through Advancing Innovative Technologies (BRAIN) Initiative (NSFNCS-FR 1926781 to LRMP), and the Baszucki Brain Research Fund (to LRMP). None of the funding sources played any role in the design of the experiments, data collection, analysis, interpretation of the results, the decision to publish, or any aspect relevant to the study. DJW reports serving on data monitoring committees for Novo Nordisk. None of the authors received funding or in-kind support from pharmaceutical and/or other companies to write this article.

Editor's evaluation

This work emphasizes the role of diabetes in brain aging and cognitive functions that is a critical gap that needs to be filled due to the increasing trend in the prevalence of diabetes around the world. This article provides valuable information about specific brain regions altered during aging and diabetes. Further, this article reports how T2DM accelerates the aging-associated decline in cognition and brain function. Extensive analysis of human datasets and comparison with published data from other researchers support the conclusion of this study; however, certain diabetic interventions that do not rescue brain damage need further validation.

https://doi.org/10.7554/eLife.73138.sa0

Introduction

Approximately 6.6% of the global population carries a diagnosis of type 2 diabetes mellitus (T2DM) (Kaiser et al., 2018). Patients with T2DM are at greater risk for developing dementia and Alzheimer’s disease (AD) and have been reported to exhibit inferior cognitive performance when compared to age-matched healthy controls (HCs) (Moheet et al., 2015). Several human neuroimaging studies have linked T2DM with brain atrophy and cognition (Moheet et al., 2015; Callisaya et al., 2019; Gold et al., 2007; Mankovsky et al., 2018; Manschot et al., 2006); recent research suggested that T2DM resulted in a more rapid rate of cognitive decline than typically associated with natural aging (Rawlings et al., 2014; Zilliox et al., 2016; Pelimanni and Jehkonen, 2018).

Despite strong preliminary evidence linking T2DM to neurological and cognitive decline, few patients with T2DM undergo a comprehensive neurocognitive evaluation as part of their clinical care (Zilliox et al., 2016; Biessels et al., 2006; Ninomiya, 2014). This may reflect the fact that T2DM diagnosis often occurs in middle age, hindering dissociation of patients’ cognitive changes from normal aging. Several studies published to date focused on the neurocognitive effects of T2DM including age-matched participants. However, because none has compared lifespan neurological changes to those experienced by equivalently aged patients with T2DM, it is currently unknown whether neurocognitive effects represent a T2DM-specific neurodegenerative pathway or the exacerbation of typical brain aging. Moreover, there remain limited data (Tuligenga, 2015) evaluating the impact of chronicity or role of effective treatment in the progression of cognitive and neurological decline.

Routine clinical protocols typically focus on peripheral biomarkers (e.g., blood glucose and insulin levels, body fat percentage) as diagnostic modalities for T2DM. However, the neurological effects of T2DM may reveal themselves many years before they can be detected by peripheral markers (Callisaya et al., 2019; Zilliox et al., 2016). As such, by the time T2DM is diagnosed and treated by standard measures, patients may have already sustained irreversible brain damage. Thus, there are direct clinical implications with respect to defining the neurocognitive impact of T2DM and to determine how these negative sequelae might be prevented or treated (Kaiser et al., 2018).

Given these unknowns and their clinical importance, here we focus on addressing three questions. First, we establish T2DM neurocognitive effects compared to age, sex, education, and hypertension-matched HCs. To do so, we leverage the robust statistical power made possible by UK Biobank (Sudlow et al., 2015), the largest (N = 20,314, ages 50–80 years) neurocognitive lifespan dataset to date. The UK Biobank results are then compared to a meta-analysis of the published literature (34 cognitive studies, 60 neuroimaging studies) to assess convergence. Second, we ask whether changes in the brain observed in T2DM represent accelerated aging or a nonaging-related degenerative pathway specific to T2DM. Third, we test whether T2DM chronicity exacerbates, and medication status ameliorates, the progression of neurocognitive effects.

Methods

Analysis of UK Biobank dataset (N = 20,314)

General overview

UK Biobank data were analyzed for both cognitive and neuroimaging data. Datafield identifiers for all utilized features are shown in Supplementary file 1. The primary factor of interest was T2DM, which we dissociated from age-related effects by age matching T2DM and HC. To permit comparison of T2DM-specific effects to age-specific effects, we also assessed the same neurocognitive variables with age as a factor of interest from samples that excluded patients diagnosed with T2DM. To control for potential neurocognitive confounds, T2DM and HC were exact pairwise matched for not only age, but also sex, education, and hypertension status. T2DM status was assessed based on self-reported diagnosis by doctor. We considered education as a binary variable based on possession of a college degree. Hypertension was quantified using measured blood pressure values: all individuals with systolic blood pressure >140 mmHg or diastolic blood pressure >90 mmHg were labeled with hypertension (Chobanian et al., 2003). To exclude potential confounding effects due to menopausal transition, particularly relevant to the youngest age group in our sample (age cutoff >50 years), we excluded all female subjects who did not report menopause or if they reported ongoing hormone therapy (Maki and Henderson, 2016). To minimize the number of individuals with type 1 diabetes in our sample, rather than type 2 diabetes, we only included diabetic individuals with a self-reported age of onset ≥ 40 years (Thomas et al., 2018).

We fitted linear regression models to neurocognitive variables and quantified associated effects as the maximum likelihood point estimates and confidence intervals (95% CIs) of the corresponding regression coefficients. T2DM was accounted for as a binary factor with two states corresponding to HCs and individuals diagnosed with T2DM, whereas age was considered as a linear continuous factor with increments in years. The latter was justified given the linear trends we observed across age in all modalities (Appendix 1—figures 4–6). Regression models were fit using the Statsmodels Python library (Seabold and Perktold, 2010). To account for multiple comparisons, we applied Bonferroni correction to cognitive and structural results, and adjusted for false discovery rate (FDR) in our brain activation results; we report adjusted p-values exclusively.

To determine whether T2DM neurocognitive effects suggested an acceleration of typical aging trajectories, versus nonaging-related degenerative pathways specific to T2DM, we compared the progression of neurodegeneration seen in T2DM to that seen in relation to age across brain regions and cognitive domains using bivariate Pearson correlations.

Cognition

Data on five cognitive domains for 18,829 participants (T2DM: N = 914, HC: N = 17,915) were extracted from the UK Biobank dataset, including abstract reasoning, executive function, processing speed, reaction time, and numeric memory (~2–3 s). Exact sample sizes varied across cognitive domains based on data availability, and therefore are noted separately for each result (Appendix 1—figures 1C and 2C). We employed linear regression and considered the maximum likelihood estimates of coefficients belonging to age and T2DM to estimate their associations with performance in each of the five domains. Effect sizes in cognition were quantified as percentages by dividing the estimated β coefficient and 95% CIs of the factor of interest with the average performance of HC.

Brain structure

Using structural MRI data from the UK Biobank dataset, we assessed the effects associated with T2DM (T2DM: N = 821, HC: N = 821, Appendix 1—figure 1A) compared to non-T2DM-specific age-related effects (N = 4775, Appendix 1—figure 2A) on atrophy of gray matter volume , calculated as mm3 (Callisaya et al., 2019) for the whole brain as well as for 139 anatomical regions. For region-specific analyses, we coarse-grained the default unilateral parcellation of 139 regions provided by UK Biobank into 45 bilateral regions and normalized gray matter volumes for head size. We applied linear regression and quantified atrophy in each anatomical region as a relative percentage change in average gray matter volume by dividing the estimated regression coefficients and 95% CIs of the factor of interest by the average gray matter volume of HC.

Brain function

Using functional MRI data from the UK Biobank dataset, we assessed the effects occurring with T2DM (T2DM: N = 646; HC: N = 646, Appendix 1—figure 1B) compared to non-T2DM-specific age-related effects (N = 2250, Appendix 1—figure 2B) on resting-state brain activity. Data were accessed already preprocessed by UK Biobank according to their standard pipelines (Alfaro-Almagro et al., 2018). After transforming functional images to Montreal Neurological Institute (MNI) space, we performed spatial smoothing with a full width at half maximum (FWHM) of 5 mm, then quantified brain activation by calculating the amplitude of low-frequency fluctuation (ALFF) (Zang et al., 2007). We used the program 3dRSFC, which is a component of Analysis of Functional NeuroImages (AFNI) (Cox, 1996; Cox and Hyde, 1997), to compute ALFF in voxel space. ALFF was computed from the 0.01–0.08 Hz frequency band, within a gray matter-only brain mask. Computed voxel space ALFF values were normalized to the global mean of each individual subject. Statistical analyses were performed in voxel space using the Nistats Python library. We used a significance threshold of p<0.05 and a minimum cluster size of 12 voxels (~100 mm3) and controlled for multiple comparisons using FDR.

Implications of T2DM duration

To evaluate the implications of T2DM chronicity, we analyzed whole-brain gray matter volume with time since T2DM diagnosis as a regressor. Time since diagnosis was derived from self-reported age at T2DM diagnosis. To improve the accuracy of self-reported values, we averaged the reported age of onset values from three visits, separated by multiple years. To estimate the degree to which T2DM progression was associated with gray matter loss relative to age, we calculated the ratio of regression coefficients corresponding to T2DM duration and age, and expressed it as a percentage, using Fieller’s theorem (Fieller, 1954) to quantify the confidence interval (95%) for this ratio.

Implications of metformin treatment

For patients with T2DM, we evaluated whether metformin, a first-line medication for the treatment of T2DM, was associated with improved outcomes in terms of cognition, atrophy, and/or brain activation. To isolate medication effects specific to metformin, we compared subjects who reported not taking any medications to treat T2DM to subjects who reported taking metformin but no other medications. For these comparisons, we exact-matched for sex, education, and hypertension, and coarse-matched for age (bin size of 5 years) and disease duration (bin size of 3 years). Since UK Biobank did not measure HbA1c levels, we also included body mass index (BMI) as a regressor since it was the only available proxy measure for disease severity (Bower et al., 2017; Bae et al., 2016).

Implications of sex

To determine whether results detected in the sample might be driven by sex-specific factors, we additionally performed analyses separately in females and males. For these, we evaluated our cognitive and neurobiological measures in association with age and T2DM and quantified the overlap separately for the two subsamples.

Meta-analysis of published literature (N = 24,185)

Search strategy and selection criteria (cognition)

We conducted a literature search for peer-reviewed articles published up to 28 August 2020 from PubMed/Medline using the following search terms: “type-2-diabetes,” “diabetes mellitus, type 2,” “insulin-resistance,” < AND > “cognition,” “cognitive-function,” “cognitive-dysfunction,” “cognitive-performance,” and “neuropsychological tests.” Search results were filtered to include manuscripts that had undergone peer-review, were published in English with full-text availability, and reported relevant results. Our cognitive meta-analysis adhered to PRISMA guidelines (Page et al., 2021).

We included studies that compared cognitive performance between people diagnosed with T2DM and HCs. We excluded studies that (1) included participants with neurological or psychiatric diagnoses, (2) utilized treatment interventions without first obtaining baseline cognitive measurements, (3) included only diagnostic threshold instruments for dementia (e.g., the Mini-Mental State Examination [MMSE]), (4) included a novel cognitive test without adequate explanation of the scoring procedures, (5) did not perform age- and education-matching of the participants diagnosed with T2DM to their HC, or (6) failed to provide summary statistics needed to calculate effect sizes. In the latter case, the authors were contacted to obtain relevant data.

Our literature search yielded 219 articles; relevant reviews were also screened for eligible studies. Seventy-five articles were identified for full-text evaluation; 34 studies were eligible for inclusion. Among the studies that were excluded (Supplementary file 2), 8 featured inadequate testing or scoring procedures, 14 included secondary analyses of the same patient sample that was used in previous publications, and 5 failed to perform appropriate education-matching of the study groups. Furthermore, one longitudinal study did not report baseline scores and another reported inconsistent sample sizes. Fifteen authors were contacted to obtain data not provided in the text; three authors provided the data requested, and the remaining 12 studies were excluded. Eligible studies included a total of 4735 subjects diagnosed with T2DM and 17,496 HC (Supplementary file 3).

Data analysis (cognition)

We extracted data, including publication year, authors, sample demographics, and cognition, from all included studies. We extracted baseline data only from longitudinal studies to avoid practice effects. We sorted individual cognitive tests into several domains, including abstract reasoning, executive function, processing speed, numeric memory, visual memory, verbal memory, verbal fluency, visuospatial reasoning, and working memory (Supplementary file 4).

Statistical analyses were performed using R version 3.6.1 (R Development Core Team, 2018) and the Metafor package version 2.4-0 (Viechtbauer, 2010). Cognitive differences between participants diagnosed with T2DM and HC were determined by calculating standardized mean difference (SMD) effect sizes and 95% CIs for all cognitive domains. Effect size analyses were chosen to account for within-domain variability in the type and sensitivity of cognitive tests across different reports. We calculated effect sizes as Cohen’s d by dividing the mean difference in group scores by the pooled standard deviation of individual domains (Cohen, 1988); an SMD (Cohen’s d) of –1.0 was interpreted as a difference of 1 SD in the negative direction. We used random-effects models to account for variability between samples not due to sampling error with significance at p<0.05, and effect size heterogeneity was evaluated using values for Cochran’s Q and I (Moheet et al., 2015; Higgins et al., 2003). Publication bias was evaluated with funnel plots. We applied Bonferroni correction to account for multiple comparisons across cognitive domains.

Search strategy and selection criteria (brain)

We used NeuroQuery (Dockès et al., 2020) to conduct a meta-analysis of published neurobiological results associated with T2DM and age. NeuroQuery is an automated Coordinate-Based Meta-Analysis (Fox et al., 1998; Van Horn et al., 2004; Yarkoni et al., 2010) tool based on a database of z-scores collected by crawling through texts and tables of published research articles by an automated algorithm (Yarkoni et al., 2011). We utilized NeuroQuery to address limitations of standard approaches to meta-analyses of neuroimaging results, which rely on summary statistics and thus risk overfitting to what typically comprise a relatively small number of in-sample studies (i.e., they fail to generalize to out-of-sample studies). NeuroQuery optimizes for rigor and reproducibility by utilizing predictive modeling, a higher threshold for results than statistical significance. In a quantitative evaluation of its generalization performance with 16-fold cross-validation and 10:90 test-train splits, NeuroQuery was found to accurately produce brain maps for out-of-sample neuroimaging studies (Yarkoni et al., 2011). We note that because of NeuroQuery’s criteria for neuroimaging data quality and completeness in reporting, the algorithm draws only from journals that focus on functional neuroimaging results, and thus can exclude some general interest and non-neurological medical journals. This exclusion criterion is important to reduce false positives and ensure the quality and relevance of the compiled results. The database in total contains 149,000 neuroimaging papers and represents the single largest database of neuroimaging foci to date. By the law of large numbers, NeuroQuery therefore provides the most unbiased approach to choosing representative papers, even at the risk of excluding relevant and well-cited articles specific to any one field.

Using the collected database of articles, NeuroQuery applies a multivariate model to predict the spatial distribution of voxel activations corresponding to a search term. The search terms we used to obtain the meta-analytic maps were “diabetic” and “age.” These terms identified the 30 most relevant neuroimaging studies for T2DM and 30 most relevant neuroimaging studies for age (Supplementary file 5). To account for any errors in the automated search results, the identified set of studies were cross-validated by an independent manual search using the same search terms for Google Scholar and PubMed to verify their relevance, as well as to confirm that they included T2DM age-matched HC and T2DM (not type 1 diabetes). In the T2DM datasets, 23 were fMRI (ALFF), 2 were structural (T1), 3 were FDG positron emission tomography (PET), and 2 were tractography (diffusion tensor imaging [DTI]). In the age datasets, 22 were fMRI (ALFF), 3 were structural (T1), and 5 were tractography (DTI, diffusion weighted imaging).

Data analysis (brain)

For region- and voxel-level comparisons of the meta-analytic T2DM and age maps from NeuroQuery with their structural and functional counterparts from UK Biobank, the meta-analytic statistical maps were transformed onto comparable coordinate space and spatial resolution. At the voxel level, the meta-analytic maps were resampled to the standard MNI affine (the transformation matrix that maps from voxel indices of the data array to actual real-world locations of the brain; no registration was required as images were already aligned). For region-level comparisons, the transformed voxel maps were coarse-grained to the 45 regions of interest from UK Biobank by masking with each individual region and computing the mean activation of the masked voxels as the representative region value.

Results

Cognitive correlates with age and T2DM

Individuals without T2DM showed age-based cognitive effects across all domains in the UK Biobank (Figure 1A). The strongest effects were observed in executive function, which showed 1.9% ± 0.1% decrease in performance per year (N = 2450, T = –17.2, p<1e–10), and processing speed, which showed 1.5% ± 0.2% decrease in performance per year (N = 2525, T = –22.8, p<1e–10). Our analyses identified further cognitive deficits associated with T2DM that were consistent with accelerated age-related cognitive decline (Figure 1B). As with aging, the strongest T2DM effects were also observed in executive function, which showed a further 13.1% ± 6.9% decrease in performance, beyond age-related effects (T2DM: N = 446; HC: N = 446; T = –3.7, p=0.001), and processing speed, which showed a further 6.7% ± 3.2% decrease in performance, beyond age-related effects (T2DM: N = 454; HC: N = 454; T = –4.1, p=0.0002). A more modest decline (3.7% ± 2.3%) was observed in numeric memory (~2–3 s) (T2DM: N = 483; HC: N = 483; T = –3.2, p=0.007), whereas abstract reasoning (T2DM: N = 886; HC: N = 886; T = –2.4, p=0.08) and reaction time (T2DM: N = 914; HC: N = 914; T = –1.0, p=0.32) were not statistically significant. Our meta-analysis confirmed that individuals with T2DM exhibited markedly lower performance when compared to age- and education-matched controls, over an even broader set of domains (Figure 1C). These again included executive function (K = 18, d = –0.40, p=0.009), processing speed (K = 31, d = –0.34, P = 5e–8), and numeric memory (~2–3 s) (K = 16, d = –0.21, p=0.05), as well as abstract reasoning (K = 8, d = –0.36, p=1e–7), immediate (~30 s) verbal memory (K = 23, d = –0.39, p=0.001), delayed (~20 min) verbal memory (K = 21, d = –0.21, p=0.005), verbal fluency (K = 25, d = –0.37, p=2e–8), visuospatial reasoning (K = 13, d = –0.32, p=4e–7), and working memory (K = 12, d = –0.36, p=0.002) (Supplementary file 6).

Figure 1 with 3 supplements see all
Download asset Open asset
Cognitive deficits are apparent with respect to both age and type 2 diabetes mellitus (T2DM) diagnosis.

(A) Using the UK Biobank dataset, we performed a quantitative analysis of the effects related to age on cognitive performance across five cognitive domains. Associated changes were derived from estimated regression coefficients as percentages and are shown on the y-axis. Age was associated with significant deficits in all five domains, with the strongest effects observed in executive function and processing speed. (B) Using the same dataset, we also analyzed cognitive performance in T2DM, with negative values on the y-axis representing performance below that of age-, sex-, and education-matched healthy controls (HCs). As per age effects, executive function and processing speed showed the highest magnitude changes. (C) Cognitive deficits identified in UK Biobank data were confirmed by our meta-analysis, which included 11 domains from 34 studies. Average effect sizes (Cohen’s d) corresponding to T2DM are shown on the y-axis. Values below the cutoff line (y = 0) indicate cases in which subjects with T2DM performed less well than age- and education-matched HC. Numbers next to labels identify domains common across panels. Marker sizes represent sample sizes scaled (per area) as indicated in the bottom-right corner of each panel. On panel (C), sample size indicates the number of individual studies. Underlying sample size distributions can be found in Appendix 1—figures 1C and 2C. Error bars are 95% CI. *p≤0.05; **p≤0.01; ***p≤0.001, Bonferroni corrected.

Neurobiological correlates with age and T2DM

Brain atrophy

HCs (N = 4,775) showed a linear decrease in brain gray matter with age. This was most pronounced in the ventral striatum, which showed a 1.0% ± 0.06% decrease per year (T = –30.4, p<1e–10), and Heschl’s gyrus, which also showed a 0.9% ± 0.06% decrease per year (T = –30.4, p<1e–10) (Figure 2A). Compared to their age-matched HC, T2DM patients showed further decreases in gray matter beyond typical age-related effects (T2DM: N = 821; HC: N = 821). These included both cortical and subcortical regions, with the most severe atrophy observed in the ventral striatum, which showed on average a 6.2% ± 1.6% further decrease in volume, beyond age-related effects (T = –7.5, p<1e-10), in the cerebellum with an additional 4.9% ± 1.1% decrease in volume, beyond age-related effects (T = –8.8, p<1e-10), and in the putamen, which showed a 4.7% ± 2.3% further decrease, beyond age-related effects (T = –4.1, p=0.002) (Figure 2B).

Figure 2 with 8 supplements see all
Download asset Open asset
Widespread gray matter atrophy can be observed with respect to both age and type 2 diabetes mellitus (T2DM) diagnosis status.

Using the UK Biobank dataset, we measured gray matter atrophy across 45 anatomical regions. Associated changes were derived from estimated regression coefficients as percentages and are shown on the x-axes. (A) We observed significantly decreased gray matter volume in both cortical and subcortical brain regions with respect to age in healthy controls (HCs). Age was associated with an average of ~0.5% brain-wide decrease in gray matter volume per year, most prominently for the ventral striatum and Heschl’s gyrus. (B) Gray matter atrophy was also seen in patients diagnosed with T2DM compared to age-matched HC, most prominently for the ventral striatum, cerebellum, and putamen. The distribution of T2DM-related effects overlapped with those associated with age, with degeneration of the ventral striatum and preservation of the thalamus and caudate. Underlying sample size distributions can be found in Appendix 1—figures 1A and 2A. Error bars are 95% CI. *p≤0.05; **p≤0.01; ***p≤0.001, Bonferroni corrected.

Brain activity

Age was associated with functional reorganization of brain activation (ALFF), rather than global decrease or increase. Brain activation in T2DM showed similar reorganization. Normalized to whole-brain activation, both age (HC: N = 2250) and T2DM (T2DM: N = 646, HC: N = 646) were associated with decreased activation in the caudate and premotor cortex, and with increased brain activity in the subgenual area, orbitofrontal cortex, posterior cingulate cortex, and brainstem (Figure 3A).

Figure 3 with 6 supplements see all
Download asset Open asset
Age- and type 2 diabetes mellitus (T2DM)-associated reorganization patterns in brain activity are significant and overlap.

(A) For functional MRI data obtained from the UK Biobank dataset, we used the amplitude of low-frequency fluctuation (ALFF) to quantify brain activation. Effects linked to age are shown in the form of an unthresholded z-map represented by the pink-green color gradient, with pink indicating increased activation and green showing decreased. T2DM-related effects were thresholded (minimum cluster size ~100 mm3, false discovery rate [FDR] p<0.05) to result in significant clusters. The outlines of these significant clusters are overlaid on the age-related z-map to demonstrate overlapping effects. The largest significant clusters with respect to T2DM were in the subgenual area (increased) orbitofrontal cortex (increased), premotor cortex (decreased), and caudate (decreased). The highlighted regions were similarly impacted across age, indicating substantial overlap between the two contrasts. Underlying sample size distributions can be found in Appendix 1—figures 1B and 2B. (B) Using multimodal neuroimaging data, we performed a meta-analysis for the same contrasts using NeuroQuery. We extracted contrast maps for age and T2DM with NeuroQuery and overlaid the outlines of thresholded (minimum cluster size ~100 mm3, FDR p<0.05) z-maps from T2DM on unthresholded z-maps belonging to age. The overlapping effects were present in several regions, most importantly in the posterior cingulate gyrus, thalamus, caudate, and premotor cortex. These results support the hypothesis that neurodegeneration in both T2DM and aging may share common mechanistic pathways.

NeuroQuery

Our meta-analysis of 60 multimodal neuroimaging studies (30 age-specific, 30 T2DM-specific) independently identified the same regions as UK Biobank (premotor cortex, caudate, posterior cingulate gyrus), but additionally identified clusters of decreased activity in Broca area and the frontal eye fields and increased activity in the thalamus and inferior temporal gyrus (Figure 3B).

Neurocognitive changes associated with T2DM and age overlap, consistent with common pathways

Together, these analyses confirm that T2DM patients show evidence of neurocognitive deficits, with the most consistent and profound effects observed in structural atrophy. Even after controlling for education, cognitive deficits remained statistically significant. Both age and T2DM implicated the same areas of greatest vulnerability: for brain atrophy, these were the ventral striatum, Heschl’s gyrus, and cerebellum; for cognition, these were executive function and processing speed. When assessed across all brain regions, T2DM-related patterns in brain atrophy exhibited strong overlap with those associated with age (r = 0.60, p=0.0002) (Figure 2—figure supplements 1 and 2A). Similarly, T2DM-related changes in brain activation (ALFF) also exhibited significant overlap with those associated with age (r = 0.64, p=0.00004) (Figure 3—figure supplements 1 and 2B). The meta-analysis, which included multimodal neuroimaging measures (not only atrophy and brain activity, but also glucose uptake via FDG-PET), also yielded equivalent results in terms of the overlap between neurobiological effects of T2DM and age (r = 0.58, p=0.0005) (Figure 2—figure supplements 1 and 2C).

T2DM chronicity exacerbates neurocognitive symptoms

Neurocognitive effects were more severe with increased disease duration, particularly for structural changes (T = –3.8, p=0.0002) (Figure 4). T2DM progression was associated with 26% ± 14% acceleration of typical neurogenerative age-related effects as per the linear shift along the horizontal time axis shown in Figure 4.

Figure 4
Download asset Open asset
Progression of type 2 diabetes mellitus (T2DM) disease is significantly associated with gray matter atrophy, accelerating neurodegenerative effects seen in brain aging.

For a quantitative evaluation of the impact of T2DM progression on gray matter volume, we considered time since T2DM diagnosis as the main factor of interest from the UK Biobank dataset. The T2DM+ cohort was divided into two groups based on disease duration (separated at 10 years) with a healthy control (HC) cohort also included for visualization purposes. We matched age, sex, education, and hypertension across these three groups and performed linear regression within T2DM+ subjects focusing on disease duration. Evaluation of our sample suggested that time since diagnosis was a significant factor, with each year after diagnosis of T2DM associated with an additional ~0.26 ± 0.14 years of brain aging beyond that of age-matched HC. Underlying sample size distributions can be found in Appendix 1—figure 3. Error bars are standard error of the mean. *p≤0.05, **p≤0.01, ***p≤ 0.001.

T2DM patients treated with metformin do not demonstrate improved neurocognitive symptoms

Even after matching groups for disease duration and controlling for BMI, T2DM patients who were treated with metformin alone (N = 498) did not differ with respect to cognition or brain atrophy compared to T2DM patients who were unmedicated (N = 352) (Figure 1—figure supplement 3, Figure 2—figure supplement 8). Likewise, treatment status showed no significant impact on resting-state brain activity.

Age- and T2DM-associated effects were consistent in females and males

For the UK Biobank dataset, we identified marked sex-related differences both in neurobiological and cognitive measures, consistent with the literature (Figure 1—figure supplement 1, Figure 2—figure supplement 3, Figure 3—figure supplement 3; Sanchis-Segura et al., 2019; Jain et al., 2015). All reported age and T2DM effects were seen for both males and females, and strongly correlated between them (Figure 2—figure supplement 6, Figure 3—figure supplement 5). However, age and T2DM effects were stronger in males (Figure 1—figure supplement 2, Figure 2—figure supplement 5, Figure 3—figure supplement 4) and neurodegeneration (brain atrophy, in particular) was more similar across sex for age (rage = 0.93, Bonferroni-corrected p≤0.001) than for T2DM (rT2DM = 0.51, Bonferroni-corrected p≤0.01). Consistent with the latter, the overlap between effects associated with age and T2DM was statistically significant in males (gray matter volume: r = 0.77, p=4e–9; brain activation [ALFF]: r = 0.51, p=0.002) but not in the female-only subsample (Figure 2—figure supplement 6, Figure 2—figure supplement 7, Figure 3—figure supplement 5, Figure 3—figure supplement 6). All females in our UK Biobank sample were menopausal and not on hormone replacement therapy. For our meta-analyses, we were unable to perform a sex-based comparison as the underlying articles from which our data were derived did not always perform sex-matching across their subject pools and did not provide access to the individual subject-level data that would be required to control for sex in our analyses.

Discussion

The UK Biobank dataset confirms that T2DM patients show deficits in cognitive performance compared to HC, even after controlling for age, sex, education, and hypertension. These findings were supported by meta-analysis of the published literature. Deficits in cognitive performance were accompanied by marked brain atrophy in the T2DM sample compared to age-matched HC. The atrophy was most severe (6.2% gray matter loss compared to HC) in the ventral striatum, a region critical to learning, decision-making, goal-directed behavior, and cognitive control. These cognitive functions, collectively known as executive functioning, were (with processing speed) also those most affected by T2DM. Neurodegeneration severity for all regions increased with longer disease duration. We detected qualitatively consistent results in females and males; however, males exhibited stronger effects in relation to T2DM (Figure 2—figure supplement 4). This result is consistent with the well-established neuroprotective effects of female hormones such as estrogen (Behl, 2002). This result also suggests that the T2DM neurological effects observed result from chronic degenerative processes, which, for our female participants, may have been at least partially ameliorated prior to menopause.

Our findings indicate that structural brain imaging, in particular, can provide a clinically valuable metric for identifying and monitoring neurocognitive effects associated with T2DM. Normalizing across sample sizes to compare the measures of neurocognitive effects – structural MRI, functional MRI, and cognitive testing – structural atrophy showed global effects that were far more statistically robust (p<2e–10) than either global cognitive measures (p=0.0001) or global brain activation (p=0.002). One additional advantage of structural MRI over cognitive testing is that the former avoids confounding associated with education and practice effects, and therefore may be more interpretable in real-world clinical settings in which matching for education and practice effects is not feasible. Structural MRI also showed advantages as a biomarker over a functional MRI-derived measure of brain activation (ALFF). The reorganization of brain activity seen with T2DM may reflect the brain’s switch to less metabolically expensive networks to conserve energy in the face of diminishing access to glucose, a pattern previously documented in aging (Tomasi et al., 2013; Weistuch et al., 2021; Tomasi and Volkow, 2012; Mujica-Parodi et al., 2020). Yet activation patterns that are spatially reorganized, rather than globally increased or decreased, are less straightforward to quantify. Moreover, functional MRI is an inherently more complex measure than structural MRI, reflecting both neuronal and hemodynamic influences. Each of these influences may be differentially affected by T2DM, further complicating its interpretation in a clinical setting.

The localization of brain atrophy in T2DM to the ventral striatum, followed by the cerebellum, may reflect the fact that these two brain regions contain the densest concentrations of insulin-dependent GLUT-4 (El Messari et al., 2002; Kobayashi et al., 1996; El Messari et al., 1998; Vannucci et al., 1998) compared to non-insulin-dependent isoforms GLUT-1 and GLUT-3. The ventral striatum functions as a critical hub within the reward circuit, integrating inputs (including external stimuli) from both cortical and subcortical regions, and therefore is a key structure required for all learning. Rat studies have shown modulation of nitric oxide within the ventral striatum to control release of acetylcholine (Prast et al., 1995), a neurotransmitter severely reduced in dementia (Kurotani et al., 2003) and a target for its pharmaceutical treatment (Ferreira-Vieira et al., 2016; Muramatsu et al., 2019). Release of nitric oxide is insulin dependent and reduced in T2DM (Tessari et al., 2010). Together, these suggest a potential mechanistic pathway between insulin resistance, atrophy of the ventral striatum, and widespread deficits with respect to learning. In this context, memory deficits may be primarily driven by failure to encode rather than failure to retrieve, which would be consistent with our results that did not identify the hippocampus to be one of the regions most affected. Importantly, the structural and functional changes associated with T2DM show marked overlap with age-related effects but appear earlier. This, on the one hand, suggests that neurocognitive changes seen in T2DM may progress via a common mechanistic trajectory as normal brain aging, but which is accelerated. From another perspective, the overlap implies that brain aging itself may be a metabolic syndrome driven by impaired brain insulin signaling and glucose metabolism, the same processes that, outside the brain, are well-established with respect to T2DM.

Our analyses had two limitations, inherent in the datasets analyzed, which represent important directions for future research. First, our use of a lifespan dataset permitted tracking how variables change with age, but not for the same subjects. A more rigorous assessment of phase shift between trajectories of neurodegeneration for patients with T2DM and HC would be made possible only with a longitudinal study. Second, while we had access to disease duration and BMI, we did not have HbA1c measures, which would have provided a more direct measure of disease severity. While metformin was not found to be associated with better neurocognitive measures, even when matched to unmedicated patients with equivalent disease duration and after controlling for BMI (a proxy measure for disease severity; Bower et al., 2017; Bae et al., 2016), it was not possible to determine other diabetes-related characteristics. As such, our medication findings should be considered suggestive but not conclusive.

Consistent with findings from earlier studies that focused on the brain and energy metabolism (Sokoloff, 1955; Clark, 1999), our results suggest that T2DM and its progression may accelerate pathways associated with typical brain aging. As T2DM diminishes s glucose availability within the brain, this chronic loss of energy may compromise the brain’s structure and function . We consider the possibility that, by the time T2DM is formally diagnosed, neuronal insulin resistance may have already resulted in significant damage. As such, our findings underscore the need for additional research into brain-based biomarkers for T2DM and treatment strategies that specifically target its neurocognitive effects (Kaiser et al., 2018).

Appendix 1

Appendix 1—figure 1
Download asset Open asset
Sample size distributions in type 2 diabetes mellitus (T2DM) vs. healthy control (HC) analyses.

The values shown represent T2DM+ only, T2DM+ were matched with an equivalent number of HCs, resulting in twice the number of samples in every analysis. The bars are stacked. (A) Gray matter volume analyses, (B) brain activation (amplitude of low-frequency fluctuation [ALFF]) analyses, and (C) cognition analyses (five domains).

Appendix 1—figure 2
Download asset Open asset
Sample size distributions of the analyses that investigated age as the factor of interest.

Samples included healthy control (HC) only and were matched for sex, education, and hypertension across age. The matching method resulted in ‘sample chains’ across age: within each chain, samples were equivalent in terms of sex, education, and hypertension. The bars are stacked. (A) Gray matter volume analyses, (B) brain activation (amplitude of low-frequency fluctuation [ALFF]) analyses, and (C) cognition analyses (five domains).

Appendix 1—figure 3
Download asset Open asset
Sample size distribution in the analysis that investigated type 2 diabetes mellitus (T2DM) duration as the main factor.

The values shown represent one out of the three duration groups (healthy controls [HCs], T2DM+ for 0–9 years, and T2DM+ for 10+ years). Samples across the three duration groups were matched for age, sex, education, and hypertension. Given the three groups, the total sample size was three times what is shown here. The statistical analysis included T2DM+ samples only (for which duration was defined), HCs were included for visual representation only. The bars are stacked.

Appendix 1—figure 4
Download asset Open asset
Plots representing trends across age and type 2 diabetes mellitus (T2DM) for each of the five investigated cognitive domains.

For executive function and reaction time, lower scores represented better performance. The linear trends across age were robust and justified modeling age as a linear factor. The apparent deviation from linear trends in the youngest and oldest age groups is explained by markedly smaller sample sizes in those age groups (see underlying sample sizes in Appendix 1—figure 1C).

Appendix 1—figure 5
Download asset Open asset
Plots representing trends across age and type 2 diabetes mellitus (T2DM) for six regions that exhibited the strongest trends in gray matter volume changes.

The linear trends across age were robust and justified modeling age as a linear factor. The apparent deviation from linear trends in the youngest and oldest age groups is explained by markedly smaller sample sizes in those age groups (see underlying sample sizes in Appendix 1—figure 1A).

Appendix 1—figure 6
Download asset Open asset
Lineplots representing trends across age and type 2 diabetes mellitus (T2DM) for six regions that exhibited the strongest trends in brain activation (amplitude of low-frequency fluctuation [ALFF]).

The linear trends across age were robust and justified modeling age as a linear factor. The apparent deviation from linear trends in the youngest and oldest age groups is explained by markedly sample sizes in those age groups (see underlying sample sizes in Appendix 1—figure 1B).

Appendix 1—figure 7
Download asset Open asset
Sample size distributions in metformin medication status-related analyses.

The values shown represent metformin medicated type 2 diabetes mellitus (T2DM)+ only, these samples were matched with an equivalent number of unmedicated T2DM+ samples, resulting in twice the number of samples in every analysis. Samples were matched for age, sex, education, and hypertension. The bars are stacked. (A) Gray matter volume analyses, (B) brain activation analyses, and (C) cognition analyses (five domains).

Data availability

All analyses reported in this manuscript were on either publicly available data (UK Biobank) or meta-analyses of listed published articles. Source code for all analyses conducted for this manuscript is uploaded as Source Code (https://github.com/lcneuro/pub_t2dm_age_meta, copy archived at swh:1:rev:81c32d4b6c25191eef4028a7dd2337b71af59628).

The following previously published data sets were used

References

  8. Book
    1. Clark DD
    (1999)
    Basic Neurochemistry: Molecular, Cellular and Medical Aspects
    Lippincott.
  9. Book
    1. Cohen J
    (1988)
    Statistical Power Analysis for the Behavioral Sciences
    Erlbaum Associates.
  34. Software
    1. R Development Core Team
    (2018) R: A language and environment for statistical computing
    R Foundation for Statistical Computing, Vienna, Austria.
    http://www.r-project.org

Decision letter

  1. Pankaj Kapahi
    Reviewing Editor; Buck Institute for Research on Aging, United States
  2. Carlos Isales
    Senior Editor; Medical College of Georgia at Augusta University, United States

Our editorial process produces two outputs: i) public reviews designed to be posted alongside the preprint for the benefit of readers; ii) feedback on the manuscript for the authors, including requests for revisions, shown below. We also include an acceptance summary that explains what the editors found interesting or important about the work.

Decision letter after peer review:

Thank you for submitting your article "Type 2 diabetes mellitus accelerates brain aging and cognitive decline: complementary findings from UK Biobank and meta-analyses." for consideration by eLife. Your article has been reviewed by 2 peer reviewers, and the evaluation has been overseen by a Reviewing Editor and Carlos Isales as the Senior Editor. The reviewers have opted to remain anonymous.

The reviewers have discussed their reviews with one another, and the Reviewing Editor has drafted this to help you prepare a revised submission.

In this timely study the authors explore the impact of age and diabetes status on cognition. The study leverages the UK Biobank data set and validates findings using a meta-analysis of the relevant clinical literature. The major finding is that Type II diabetes is associated with changes in brain structure and cognition similarly to those identified in aging but occurring earlier in individuals with Type II diabetes.

Please respond to all the issues raised by the reviewers.

Essential Revisions (for the authors):

1. Sexual dimporphism.

2. Control group- is it free of diabtes or pre- diabetics as well.

3. Exclusion of T1DM based on age cutoff is problematic as well because people can develop T1DM even after 20 years of age.

4. It takes many years before people develop T2DM, and transition from a normal glucose metabolism, to insulin resistance, prediabetes, diabetes; associated with these changes there are many other associated inflammatory, metabolic and molecular changes of systemic and cerebral blood vessels. Authors should discuss this point in their paper.

5. The duration of diabetes and metformin treatment.

Reviewer #1 (Recommendations for the authors):

1. Please provide more information regarding how the mid-life or late-life onset of T2DM influences cognition and neurodegeneration in the analyzed dataset. The duration of diabetes must be considered along with age, sex, and education while performing paired analysis. Provide the average age of subjects selected in T2DM and HC groups? Also, include the minimum and maximum age of subjects included in each group for the assays.

2. Since sex hormones play an essential role in T2DM and neurodegeneration, I would recommend the authors separate the cognition data based on sex difference and present it as a figure to show how sex differences and their association with T2DM influence cognitive performance.

3. Does early diabetic intervention limit neurodegeneration and cognition decline in the analyzed dataset? What is the disease severity between the metformin-treated and the untreated groups?

4. To compare the meta-analysis with UK Biobank data analysis, the parameters used for analysis should be similar. Is the meta-analysis for cognitive performance include sex-based comparison? There is no information regarding this in the methods section. Please clarify

5. Please check the dataset, for other disease conditions or co-morbidities that could potentially affect the studied cognition domain or brain structure and activity. Because conditions like hypertension that are usual in aged and diabetic conditions can influence the observed data.

Reviewer #2 (Recommendations for the authors):

Strengths of the study include

a) the use of the UK Biobank,

b) identification of overlap between the age-sensitive brain structural and functional changes with those in the T2DM group,

c) demonstration that the duration of T2DM impacts the extent of change,

d) interesting outcome of the ALFF pointing to reorganization rather than simple decline in "activity" by brain region,

e) the validation using the meta-analysis and neuroquery combination,

f) the regression analysis showing congruence for structural outcomes and for cognitive outcomes between the UK Biobank and Meta-analysis approaches.

Moderate concerns include

a) The primary focus here is on T2DM, for which (according to the methods) the comparisons to healthy controls were pairwise matched by age, sex, education. The actual number of subjects used is unclear: The UK Biobank dataset included cognitive and neuroimaging data (N=26,125) including 1,270 T2DM and 24,855 HC. T2DM cohort is 5% the size of the HC. What was the actual reference analysis? Were 1270 matched HC used as the reference or were all HC included – how did this impact variance in the measures?

b) Given the cross-sectional nature of the work differences in composition of the cohort at the various times points and the numbers contributing to each of the increments would be helpful. Were the numbers of data points equivalent across the 50-75 years of age?

c) Can the authors provide an explanation as to why males and females were combined?

d) It would be helpful to understand how other potentially important variables such as BMI contributed in the analysis.

e) The declines in the various parameters with age are quite modest, making those identified in the T2DM all the more striking. In the narrative declines per year are described but actual age is not mentioned and does not appear at all until later in the manuscript (Figure 4) and (Figure S1). The data are shown as 5-year increments from 50-75 years. It is unclear what happened before this time point – is there a period during adulthood when volume remains steady? Appreciating the ~8.3 to ~7.6 GMV voxel count over the 25 year time frame – it would be helpful to understand if the declines are detected from 50 years or earlier.

f) Was sex or BMI a contributing factor in either the T2DM or the HC outcomes?

g) The final conclusion of the abstract "As such, neuroimaging-based biomarkers may provide a valuable adjunctive measure of T2DM progression and treatment efficacy based on neurological effects" is not supported. Actual comparisons to established clinical assessments of glucoregulatory impairment were not conducted (effect size and sensitivity), and apart from the obvious considerations of cost and accessibility, this comment seems unnecessary and detracts from the work.

Comments:

a) The authors point to "accelerated" cognitive aging but it seems the slopes of decline are rather similar for aging and for T2DM – what differs is the starting value at age 50. It is semantics I suppose but accelerated seems a better fit if the pace of aging is increased, i.e. a different slope. It is appreciated that premature onset of aging phenotypes has in the field been described as accelerated but often it is also accompanied by a quicker rate of decline. What seems to be happening here is premature age-related decline.

b) Speculation about "switch to less metabolically expensive networks" is unwarranted. In the absence of glucose uptake measures "diminished access to glucose" is not tested. Even with those data, inferences about the metabolism of glucose cannot be made as PET tells nothing about the fate of glucose. This argument ignores the possibility of a role of insulin, lipid signaling, or even inflammatory processes that might equally be speculated upon.

c) It is somewhat alarming that while 559 of the T2MD were treated with metformin for their condition, 473 were unmedicated. Is this usual?

https://doi.org/10.7554/eLife.73138.sa1

Author response

Essential revisions:

1. Sexual dimporphism.

In our original submission, we had already excluded the possibility that our results were solely due to sexual dimorphism, as we matched for sex in all analyses in the UK Biobank (every female with T2DM was matched with a female HC control, and equivalently for males). When investigating age, effects were derived from a combination of equally distributed female only and male only subsamples. However, we agree with the reviewers that differences due to sexual dimorphism in the UK Biobank are relevant and should be reported. Edits can be found on lines 192-195, 358-375, and 385-390.

To eliminate interaction effects stemming from the menopausal transition process in females, we excluded all female subjects who did not report menopause or reported hormone replacement therapy. As the distribution of the raw sample was strongly skewed towards ages >60 years, these exclusions did not result in significant loss of sample size.

To address sexual dimorphism, we compared females and males across all three investigated sets of features: cognitive performance, atrophy (gray matter volume), and brain activity (ALFF). For all three, we found significant differences between males and females, which are now reported in Figure 1—figure supplement 1, Figure 2—figure supplement 3 and Figure 3—figure supplement 3.

In addition, we investigated the interaction between sex and our main factors of interest: T2DM and age, by performing all main analyses independently for females only and for males only. We found that, although age-related effects were equivalent in females and males, T2DM effects were more pronounced in males, even if matched for disease duration (see Figure 1—figure supplement 2, Figure 2—figure supplement 4, Figure 2—figure supplement 5, Figure 3—figure supplement 4). This manifested not only in reduced statistical power in females due to smaller sample sizes, but also in smaller mean differences between HC and T2DM in females. We discuss these results and a potential explanation in the discussion on lines 385-390. Our second level analyses correlating effects associated with T2DM and age also showed stronger results in the male only subsample (Figure 2—figure supplements 6, 7 and Figure 3—figure supplements 5, 6). However, given that both males and females show the same effects (the differences are of degree, but not of kind), we also present our results from the combined sample.

2. Control group- is it free of diabtes or pre- diabetics as well.

The groups were defined as those that had vs. had not been diagnosed with diabetes by a doctor (“Diabetes diagnosed by doctor”, UK Biobank datafield #2443). While not perfect, this was the best available such feature in the dataset. We acknowledge this limitation in the discussion on lines 430-435. However, it should reassure the reviewer to note that, if there were prediabetics in the control group, this would have weakened the effect size for the comparison, rather than strengthened it. Therefore, the fact that the control group may have contained prediabetics, and yet we still found marked statistical differences between that group and those diagnosed with diabetes, suggests that the results are very robust and might even be more robust than reported.

3. Exclusion of T1DM based on age cutoff is problematic as well because people can develop T1DM even after 20 years of age.

According to a recent study using the same UK Biobank dataset, investigating the distribution of age at diagnosis of T1DM based on genetic markers, only ~1.8% of diagnoses above the age of 40 involved T1DM [1]. Thus, we re-analyzed the data, choosing 40 years as the new cutoff in terms of age at diagnosis.

Reviewer #1 (Recommendations for the authors):

1. Please provide more information regarding how the mid-life or lateLife onset of T2DM influences cognition and neurodegeneration in the analyzed dataset. The duration of diabetes must be considered along with age, sex, and education while performing paired analysis. Provide the average age of subjects selected in T2DM and HC groups? Also, include the minimum and maximum age of subjects included in each group for the assays.

We have made edits to the manuscript (line 106) and have added several new figures to clarify. Exact sample sizes and corresponding distributions across age are shown on Appendix figures 1, 2, 3, and 7 for all analyses. Regarding the actual age distributions, all samples were aged 50-80 years, the average age typically being ~65 years across analyses. Given this distribution and the T1 diabetes-related age of onset cutoff at 40 years, our analyses were specific to cases with mid-life and late-life onset of T2DM. Our analysis of disease duration with respect to gray matter volume did not reveal different outcomes for cases of mid-life vs late-life onset of T2DM, while matching for duration (see Figure 4). The significant association between disease duration and neurodegeneration appears to be consistent in its extent across the 50-80y age range.

2. Since sex hormones play an essential role in T2DM and neurodegeneration, I would recommend the authors separate the cognition data based on sex difference and present it as a figure to show how sex differences and their association with T2DM influence cognitive performance.

As described above (Essential Comment #1), we re-analyzed all UK Biobank data to investigate the interaction between sexual dimorphism and T2DM. We now present these results in the manuscript (lines 358-375) and show relevant results on Figure 1—figure supplement 2.

3. Does early diabetic intervention limit neurodegeneration and cognition decline in the analyzed dataset? What is the disease severity between the metformin-treated and the untreated groups?

Unfortunately, the dataset did not include sufficient information to evaluate this question. Accordingly, we indicate in the Discussion section (lines 430-435), that we did not have access to appropriate biomarkers such as blood-based measures to assess disease severity between metformin treated and untreated groups. We did control for BMI as the best proxy available, but we did not observe marked differences in this measure: the average BMI was 29.1 and 30.6 for unmedicated and medicated samples, respectively.

4. To compare the meta-analysis with UK Biobank data analysis, the parameters used for analysis should be similar. Is the meta-analysis for cognitive performance include sex-based comparison? There is no information regarding this in the methods section. Please clarify

Our meta-analysis of cognitive performance was unable to perform a sex-based comparison, as the underlying papers that our data were derived from did not perform sex-matching across their subject pools and did not provide access to the individual subject-level data that would be required to control for sex in our analyses.

5. Please check the dataset, for other disease conditions or co-morbidities that could potentially affect the studied cognition domain or brain structure and activity. Because conditions like hypertension that are usual in aged and diabetic conditions can influence the observed data.

To address this issue, we now have quantified hypertension based on peripheral blood pressure measurements for all samples, as described in the Methods section (lines 121-123). Consistent with literature, we observed hypertension to influence gray matter volume in several brain regions [2] as well as cognition (short-term memory and abstract reasoning). Since hypertension moderately correlated with T2DM+ status, we repeated all analyses in the UK Biobank by matching for hypertension (along with age, sex and education), to mitigate the variable as a confound.

Reviewer #2 (Recommendations for the authors):

a) The primary focus here is on T2DM, for which (according to the methods) the comparisons to healthy controls were pairwise matched by age, sex, education. The actual number of subjects used is unclear: The UK Biobank dataset included cognitive and neuroimaging data (N=26,125) including 1,270 T2DM and 24,855 HC. T2DM cohort is 5% the size of the HC. What was the actual reference analysis? Were 1270 matched HC used as the reference or were all HC included – how did this impact variance in the measures?

This information is now clarified in the manuscript, primarily in Appendix figures 1, 2, and 3. In our T2DM vs HC comparisons, we used all eligible T2DM+ samples and matched each one of them with a HC from the total sample, in a 1:1 ratio. For instance, in the case of gray matter volume, we compared 821 T2DM+ to 821 HC which were matched for age, sex, education and hypertension, but otherwise randomly picked from the total sample. Consequently, this resulted in numerous HC samples being discarded. When addressing effects associated with age, we investigated within HC only, resulting in significantly larger sample sizes and statistical power.

b) Given the cross-sectional nature of the work differences in composition of the cohort at the various times points and the numbers contributing to each of the increments would be helpful. Were the numbers of data points equivalent across the 50-75 years of age?

Please see Appendix figures 1, 2, 3, and 7, which now clearly indicate the distribution of samples across the age span, separately for every analysis we performed in the UK Biobank.

c) Can the authors provide an explanation as to why males and females were combined?

We have now re-analyzed all UK Biobank data to investigate the interaction between sex and T2DM status. We refer the reviewer to our response given to an earlier comment (Essential Comment #1) where we explain our findings in detail. Please note that all our UK Biobank analyses matched for sex, to exclude it as a confounder. With respect to sex specific T2DM related effects, we did not detect qualitatively different results when performing analyses in females only vs in males only, which justified our decision to present the main results combining sexes. Sex-specific results are now presented in Supplemental Figures.

d) It would be helpful to understand how other potentially important variables such as BMI contributed in the analysis.

In response to this comment, we re-performed all our analyses with BMI as an additional covariate and found BMI to be a significant predictor for gray matter volume, but not for cognition or brain function. As for structure, we were unable to dissociate BMI and T2DM associated effects given the strong collinearity between increased BMI and T2DM+ status. The collinearity was so strong that even matching was deemed impractical due to significant sample size loss (lack of T2DM+ with low BMI).

e) The declines in the various parameters with age are quite modest, making those identified in the T2DM all the more striking. In the narrative declines per year are described but actual age is not mentioned and does not appear at all until later in the manuscript (Figure 4) and (Figure S1). The data are shown as 5-year increments from 50-75 years. It is unclear what happened before this time point – is there a period during adulthood when volume remains steady? Appreciating the ~8.3 to ~7.6 GMV voxel count over the 25 year time frame – it would be helpful to understand if the declines are detected from 50 years or earlier.

The ~0.5-1% decline per year we identified with respect to age (as shown in Figure 1 for cognition and Figure 2 for gray matter volume) were indeed specific to 50 years and above. Less than 1% of our sample was younger than 48 years and none was younger than 44 years. This precludes the evaluation of aging-related neurocognitive deterioration prior to age 50.

f) Was sex or BMI a contributing factor in either the T2DM or the HC outcomes?

Sex was a statistically significant contributing factor across all our features; consequently, we matched for it. Please once again refer to our earlier response to Essential Comment #1. BMI, as described in our response above to another comment above, was also significant but only in the context of gray matter volume.

g) The final conclusion of the abstract "As such, neuroimaging-based biomarkers may provide a valuable adjunctive measure of T2DM progression and treatment efficacy based on neurological effects" is not supported. Actual comparisons to established clinical assessments of glucoregulatory impairment were not conducted (effect size and sensitivity), and apart from the obvious considerations of cost and accessibility, this comment seems unnecessary and detracts from the work.

We respect the reviewer’s perspective. However, our remark occurs at the end of the Discussion section in the context of a direction for future clinical research. We believe that this research is warranted based on the widespread neurodegenerative effects that we identified in patients with T2DM.

Comments:

a) The authors point to "accelerated" cognitive aging but it seems the slopes of decline are rather similar for aging and for T2DM – what differs is the starting value at age 50. It is semantics I suppose but accelerated seems a better fit if the pace of aging is increased, i.e. a different slope. It is appreciated that premature onset of aging phenotypes has in the field been described as accelerated but often it is also accompanied by a quicker rate of decline. What seems to be happening here is premature age-related decline.

This is more clearly represented in the analyses where we break up the T2DM+ cohort with respect to disease chronicity. Specifically, in Figure 4 we show that as disease duration increases, the deviation between T2DM+ and HC increases (acceleration of 0.26% ± 0.14%).

b) Speculation about "switch to less metabolically expensive networks" is unwarranted. In the absence of glucose uptake measures "diminished access to glucose" is not tested. Even with those data, inferences about the metabolism of glucose cannot be made as PET tells nothing about the fate of glucose. This argument ignores the possibility of a role of insulin, lipid signaling, or even inflammatory processes that might equally be speculated upon.

This claim is based on existing literature [3, 4], providing evidence not only for inhomogeneity among brain regions in terms of metabolic requirements and efficiency. Subcortical regions and the cerebellum were found to be metabolically cheaper, which is relevant to the brain activity patterns we observed in relation to T2DM and age: brain activity manifested a general shift from cortical to subcortical regions (Figure 3) [3].

c) It is somewhat alarming that while 559 of the T2MD were treated with metformin for their condition, 473 were unmedicated. Is this usual?

Half of unmedicated subjects were diagnosed less than five years earlier (25% within two years), whereas for medicated subjects, average duration was twice as high. This suggests that a significant number of subjects might not have been diagnosed for sufficient time to start medication. In addition, our sample of 498 subjects who reported taking metformin included those only who took metformin exclusively (we excluded those taking multiple medications to treat their diabetes). There were an additional ~150 subjects who were excluded due to polypharmacy, further raising the ratio.

References:

[1]: Thomas, Nicholas J., et al. "Frequency and phenotype of type 1 diabetes in the first six decades of life: a cross-sectional, genetically stratified survival analysis from UK Biobank." The Lancet Diabetes and Endocrinology 6.2 (2018): 122-129.

[2]: Gianaros, Peter J., et al. "Higher blood pressure predicts lower regional grey matter volume: Consequences on short-term information processing." Neuroimage 31.2 (2006): 754-765.

[3]: Tomasi, Dardo, Gene-Jack Wang, and Nora D. Volkow. "Energetic cost of brain functional connectivity." Proceedings of the National Academy of Sciences 110.33 (2013): 13642-13647.

[4]: Shokri-Kojori, Ehsan, et al. "Correspondence between cerebral glucose metabolism and BOLD reveals relative power and cost in human brain." Nature communications 10.1 (2019): 1-12.

https://doi.org/10.7554/eLife.73138.sa2

Article and author information

Author details

  1. Botond Antal

    1. Department of Biomedical Engineering, Stony Brook University, Stony Brook, United States
    2. Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital and Harvard Medical School, Charlestown, United States
    Contribution
    is the sole first author, Formal analysis, Investigation, Methodology, Visualization, Writing – original draft, Writing – review and editing
    Competing interests
    No competing interests declared
    Additional information
    is the sole first author
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-0775-6033

  2. Liam P McMahon

    1. Department of Biomedical Engineering, Stony Brook University, Stony Brook, United States
    2. Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital and Harvard Medical School, Charlestown, United States
    Contribution
    Data curation, Formal analysis, Investigation, Methodology, Writing – original draft, Writing – review and editing
    Competing interests
    No competing interests declared

  3. Syed Fahad Sultan

    Department of Computer Science, Stony Brook University, Stony Brook, United States
    Contribution
    Formal analysis, Investigation, Methodology, Writing – original draft, Writing – review and editing
    Competing interests
    No competing interests declared

  4. Andrew Lithen

    1. Department of Biomedical Engineering, Stony Brook University, Stony Brook, United States
    2. Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital and Harvard Medical School, Charlestown, United States
    Contribution
    Investigation
    Competing interests
    No competing interests declared

  5. Deborah J Wexler

    Diabetes Center, Massachusetts General Hospital and Harvard Medical School, Boston, United States
    Contribution
    Writing – review and editing
    Competing interests
    is part of a Novo Nordisk data monitoring committee service for semaglutide in SOUL and FLOW trials. The author has no other competing interests to declare

  6. Bradford Dickerson

    1. Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital and Harvard Medical School, Charlestown, United States
    2. Department of Neurology, Massachusetts General Hospital and Harvard Medical School, Boston, United States
    Contribution
    Writing – review and editing
    Competing interests
    received royalties from Cambridge University Press and Oxford University Press, and consulting fees from Acadia, Alector, Arkuda, Biogen, Denali, Lilly, Merck, Novartis, Takeda, Wave LifeSciences (unrelated to the present work). Also participates in a Lilly Data Safety Monitoring Board (unrelated to the present work) and participates in leadership roles in Alzheimer's Association and Association for Frontotemporal Degeneration. The author has no other competing interests to declare

  7. Eva-Maria Ratai

    Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital and Harvard Medical School, Charlestown, United States
    Contribution
    Writing – review and editing
    Competing interests
    received honoraria from Harvard Catalyst. The author has no other competing interests to declare

  8. Lilianne R Mujica-Parodi

    1. Department of Biomedical Engineering, Stony Brook University, Stony Brook, United States
    2. Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital and Harvard Medical School, Charlestown, United States
    3. Department of Neurology, Stony Brook University School of Medicine, Stony Brook, United States
    Contribution
    Conceptualization, Data curation, Funding acquisition, Methodology, Project administration, Resources, Supervision, Visualization, Writing – original draft, Writing – review and editing
    For correspondence
    lilianne.strey@stonybrook.edu
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-3752-5519

Funding

W. M. Keck Foundation

  • Lilianne R Mujica-Parodi

National Science Foundation (NSFNCS-FR 1926781)

  • Lilianne R Mujica-Parodi

Baszucki Brain Research Fund

  • Lilianne R Mujica-Parodi

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Senior Editor

  1. Carlos Isales, Medical College of Georgia at Augusta University, United States

Reviewing Editor

  1. Pankaj Kapahi, Buck Institute for Research on Aging, United States

Publication history

  1. Preprint posted: May 26, 2021 (view preprint)
  2. Received: August 17, 2021
  3. Accepted: April 26, 2022
  4. Accepted Manuscript published: May 24, 2022 (version 1)
  5. Version of Record published: May 25, 2022 (version 2)

Copyright

© 2022, Antal et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 8,421
    Page views
  • 1,795
    Downloads
  • 12
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

  1. Botond Antal
  2. Liam P McMahon
  3. Syed Fahad Sultan
  4. Andrew Lithen
  5. Deborah J Wexler
  6. Bradford Dickerson
  7. Eva-Maria Ratai
  8. Lilianne R Mujica-Parodi
(2022)
Type 2 diabetes mellitus accelerates brain aging and cognitive decline: Complementary findings from UK Biobank and meta-analyses
eLife 11:e73138.
https://doi.org/10.7554/eLife.73138

Categories and tags

Research organism

Further reading

    1. Epidemiology and Global Health

    Predictive performance of multi-model ensemble forecasts of COVID-19 across European nations

    Katharine Sherratt, Hugo Gruson ... Sebastian Funk
    Research Article Updated

    Background:

    Short-term forecasts of infectious disease burden can contribute to situational awareness and aid capacity planning. Based on best practice in other fields and recent insights in infectious disease epidemiology, one can maximise the predictive performance of such forecasts if multiple models are combined into an ensemble. Here, we report on the performance of ensembles in predicting COVID-19 cases and deaths across Europe between 08 March 2021 and 07 March 2022.

    Methods:

    We used open-source tools to develop a public European COVID-19 Forecast Hub. We invited groups globally to contribute weekly forecasts for COVID-19 cases and deaths reported by a standardised source for 32 countries over the next 1–4 weeks. Teams submitted forecasts from March 2021 using standardised quantiles of the predictive distribution. Each week we created an ensemble forecast, where each predictive quantile was calculated as the equally-weighted average (initially the mean and then from 26th July the median) of all individual models’ predictive quantiles. We measured the performance of each model using the relative Weighted Interval Score (WIS), comparing models’ forecast accuracy relative to all other models. We retrospectively explored alternative methods for ensemble forecasts, including weighted averages based on models’ past predictive performance.

    Results:

    Over 52 weeks, we collected forecasts from 48 unique models. We evaluated 29 models’ forecast scores in comparison to the ensemble model. We found a weekly ensemble had a consistently strong performance across countries over time. Across all horizons and locations, the ensemble performed better on relative WIS than 83% of participating models’ forecasts of incident cases (with a total N=886 predictions from 23 unique models), and 91% of participating models’ forecasts of deaths (N=763 predictions from 20 models). Across a 1–4 week time horizon, ensemble performance declined with longer forecast periods when forecasting cases, but remained stable over 4 weeks for incident death forecasts. In every forecast across 32 countries, the ensemble outperformed most contributing models when forecasting either cases or deaths, frequently outperforming all of its individual component models. Among several choices of ensemble methods we found that the most influential and best choice was to use a median average of models instead of using the mean, regardless of methods of weighting component forecast models.

    Conclusions:

    Our results support the use of combining forecasts from individual models into an ensemble in order to improve predictive performance across epidemiological targets and populations during infectious disease epidemics. Our findings further suggest that median ensemble methods yield better predictive performance more than ones based on means. Our findings also highlight that forecast consumers should place more weight on incident death forecasts than incident case forecasts at forecast horizons greater than 2 weeks.

    Funding:

    AA, BH, BL, LWa, MMa, PP, SV funded by National Institutes of Health (NIH) Grant 1R01GM109718, NSF BIG DATA Grant IIS-1633028, NSF Grant No.: OAC-1916805, NSF Expeditions in Computing Grant CCF-1918656, CCF-1917819, NSF RAPID CNS-2028004, NSF RAPID OAC-2027541, US Centers for Disease Control and Prevention 75D30119C05935, a grant from Google, University of Virginia Strategic Investment Fund award number SIF160, Defense Threat Reduction Agency (DTRA) under Contract No. HDTRA1-19-D-0007, and respectively Virginia Dept of Health Grant VDH-21-501-0141, VDH-21-501-0143, VDH-21-501-0147, VDH-21-501-0145, VDH-21-501-0146, VDH-21-501-0142, VDH-21-501-0148. AF, AMa, GL funded by SMIGE - Modelli statistici inferenziali per governare l'epidemia, FISR 2020-Covid-19 I Fase, FISR2020IP-00156, Codice Progetto: PRJ-0695. AM, BK, FD, FR, JK, JN, JZ, KN, MG, MR, MS, RB funded by Ministry of Science and Higher Education of Poland with grant 28/WFSN/2021 to the University of Warsaw. BRe, CPe, JLAz funded by Ministerio de Sanidad/ISCIII. BT, PG funded by PERISCOPE European H2020 project, contract number 101016233. CP, DL, EA, MC, SA funded by European Commission - Directorate-General for Communications Networks, Content and Technology through the contract LC-01485746, and Ministerio de Ciencia, Innovacion y Universidades and FEDER, with the project PGC2018-095456-B-I00. DE., MGu funded by Spanish Ministry of Health / REACT-UE (FEDER). DO, GF, IMi, LC funded by Laboratory Directed Research and Development program of Los Alamos National Laboratory (LANL) under project number 20200700ER. DS, ELR, GG, NGR, NW, YW funded by National Institutes of General Medical Sciences (R35GM119582; the content is solely the responsibility of the authors and does not necessarily represent the official views of NIGMS or the National Institutes of Health). FB, FP funded by InPresa, Lombardy Region, Italy. HG, KS funded by European Centre for Disease Prevention and Control. IV funded by Agencia de Qualitat i Avaluacio Sanitaries de Catalunya (AQuAS) through contract 2021-021OE. JDe, SMo, VP funded by Netzwerk Universitatsmedizin (NUM) project egePan (01KX2021). JPB, SH, TH funded by Federal Ministry of Education and Research (BMBF; grant 05M18SIA). KH, MSc, YKh funded by Project SaxoCOV, funded by the German Free State of Saxony. Presentation of data, model results and simulations also funded by the NFDI4Health Task Force COVID-19 (https://www.nfdi4health.de/task-force-covid-19-2) within the framework of a DFG-project (LO-342/17-1). LP, VE funded by Mathematical and Statistical modelling project (MUNI/A/1615/2020), Online platform for real-time monitoring, analysis and management of epidemic situations (MUNI/11/02202001/2020); VE also supported by RECETOX research infrastructure (Ministry of Education, Youth and Sports of the Czech Republic: LM2018121), the CETOCOEN EXCELLENCE (CZ.02.1.01/0.0/0.0/17-043/0009632), RECETOX RI project (CZ.02.1.01/0.0/0.0/16-013/0001761). NIB funded by Health Protection Research Unit (grant code NIHR200908). SAb, SF funded by Wellcome Trust (210758/Z/18/Z).

    1. Epidemiology and Global Health

    Affectionate touch and diurnal oxytocin levels: An ecological momentary assessment study

    Ekaterina Schneider, Dora Hopf ... Beate Ditzen
    Research Article

    Background:

    Affectionate touch, which is vital for mental and physical health, was restricted during the Covid-19 pandemic. This study investigated the association between momentary affectionate touch and subjective well-being, as well as salivary oxytocin and cortisol in everyday life during the pandemic.

    Methods:

    In the first step, we measured anxiety and depression symptoms, loneliness and attitudes toward social touch in a large cross-sectional online survey (N = 1050). From this sample, N = 247 participants completed ecological momentary assessments over 2 days with six daily assessments by answering smartphone-based questions on affectionate touch and momentary mental state, and providing concomitant saliva samples for cortisol and oxytocin assessment.

    Results:

    Multilevel models showed that on a within-person level, affectionate touch was associated with decreased self-reported anxiety, general burden, stress, and increased oxytocin levels. On a between-person level, affectionate touch was associated with decreased cortisol levels and higher happiness. Moreover, individuals with a positive attitude toward social touch experiencing loneliness reported more mental health problems.

    Conclusions:

    Our results suggest that affectionate touch is linked to higher endogenous oxytocin in times of pandemic and lockdown and might buffer stress on a subjective and hormonal level. These findings might have implications for preventing mental burden during social contact restrictions.

    Funding:

    The study was funded by the German Research Foundation, the German Psychological Society, and German Academic Exchange Service.

    1. Epidemiology and Global Health

    Perceived barriers to cervical cancer screening and motivators for at-home human papillomavirus self-sampling during the COVID-19 pandemic: Results from a telephone survey

    Susan L Parker, Ashish A Deshmukh ... Jane R Montealegre
    Research Article

    Background: Home-based self-sampling for human papillomavirus (HPV) testing may be an alternative for women not attending clinic-based cervical cancer screening.

    Methods: We assessed barriers to care and motivators to use at-home HPV self-sampling kits during the COVID-19 pandemic as part of a randomized controlled trial evaluating kit effectiveness. Participants were women aged 30-65 and under-screened for cervical cancer in a safety-net healthcare system. We conducted telephone surveys in English/Spanish among a subgroup of trial participants, assessed differences between groups, and determined statistical significance at p<0.05.

    Results: Over half of 233 survey participants reported that clinic-based screening (Pap) is uncomfortable (67.8%), embarrassing (52.4%), and discomfort seeing male providers (63.1%). The last two factors were significantly more prevalent among Spanish versus English speakers (66.4% vs. 30% (p=0.000) and 69.9 vs. 52.2% (p=0.006), respectively). Most women who completed the kit found Pap more embarrassing (69.3%), stressful (55.6%), and less convenient (55.6%) than the kit. The first factor was more prevalent among Spanish versus English speakers (79.6% vs. 53.38%, p=0.001) and among patients with elementary education or below.

    Conclusions: The COVID-19 pandemic influenced most (59.5%) to participate in the trial due to fear of COVID, difficulty making appointments, and ease of using kits. HPV self-sampling kits may reduce barriers among under-screened women in a safety-net system.

    Funding: This study is supported by a grant from the National Institute for Minority Health and Health Disparities (NIMHD, R01MD013715, PI: JR Montealegre).

    Clinical trial number: NCT03898167.