Examining the Contribution of Diabetes and Obesity to Alzheimer's Disease

Brian Gordon, Ph.D.

 Project Summary: 

The number of adults aged 65 and older is expected to increase to nearly 1 billion by 2030. With this dramatic increase in elderly individuals, there is a rising concern over the implications of a corresponding increase in the number of individuals suffering from dementia. The objective of this project was to examine the influence that obesity and diabetes have on Alzheimer’s disease (AD) risk and pathology. Epidemiological studies suggest that both obesity and diabetes increase the risk of developing dementia. It is unclear if these two conditions act directly on AD pathology, or rather represent comorbid pathology that act in parallel to AD to increase dementia risk. Studying and unraveling this distinction is critical in understanding in what ways obesity and diabetes impact AD. Using longitudinal cognitive and neuroimaging data from the Knight Alzheimer’s Disease Research Center we examined the influence that obesity and diabetes had on cognition and AD biomarkers. We quantified obesity using body mass index (BMI) and diabetes using self-reported medication and glycated hemoglobin (HbA1c). Our primary cognitive outcome was whether individuals would ever develop dementia as measured using the clinical dementia rating (CDR). We used a neuroimaging technique called positron emission tomography (PET) to measure levels of beta-amyloid (Aβ), the primary pathological protein associated with AD pathogenesis. In addition to these primary measures, we examined interactions with the apolipoprotein genotype (APOE) as well as baseline age group (mid-life or late-life).

We found that there was no relationship between mid-life or late-life BMI at risk of developing dementia. In the subset with HbA1c we found no relationship between insulin resistance and risk of developing dementia. When examining Aβ levels we found a negative association, such that greater levels of BMI, and also greater insulin resistance, was associated with lower levels of beta-amyloid pathology measured with PET. We found that there were no interactions with APOE genotype (presence or absence of e4 allele). With did find significant interactions with baseline age group, with the significant effect of BMI only being present in late-life. This work suggests that if BMI and diabetes increase the risk for dementia the effect is subtle and not detectable in our current cohorts. Although we found a significant relationship between health measures and levels of Aβ, the direction of the relationship was such that worse health profiles were associated with less pathology. This result is consistent with two other recent publications examining the link between AD, obesity, and diabetes, but contrary to expectations from the animal literature.


With this dramatic increase in elderly individuals in the population, there is a rising concern over age-related diseases such as dementia and diabetes. The number of adults aged 65 and older is expected to increase to nearly 1 billion by 2030. Already the worldwide costs of dementia are estimated to be over 156 billion, with this cost only increasing in the future. Obesity is another age-related public health concern tied to an increased risk of diabetes, heart disease, and cancer 1. Together these health concerns represent a significant public health crisis.

Epidemiological studies have shown that being overweight or having diabetes increases the risk of developing dementia2–4. While this is the general pattern, some research has shown no increased risk of dementia tied to these factors 5,6 or even a reduced risk of dementia 7. When examining AD dementia specifically, epidemiological work has shown that being overweight both increases 8 and decreases 9,10 the risk of dementia. These inconsistent findings may be driven by the heterogeneous nature of the cohorts being studied. For example, there appear to be differential outcomes when comparing obesity at mid-life to obesity later in life. Mid-life obesity has shown to increase the risk of later dementia 11–13 while in older adults a lower body-mass (BMI) index is actually related to an increased risk of dementia 2. This reversal may be driven by weight loss near the onset of cognitive decline 14,15. The relationship between obesity, diabetes, and AD may be modified by the apolipoprotein (APOE) e4 allele 16–18.

Prior work has focused on how obesity and diabetes impact the overall risk of developing dementia. As a result, it is unclear if these factors directly interact with AD pathological process, or instead represent parallel comorbidities that lower the brain’s reserve against other insults. Work with animal models suggest that diabetes and obesity directly act through the accumulation of AD pathology 19–21. In human work, only a modest number of studies have examined the relationship between obesity, diabetes, and AD pathology 17,18,22,23. The two hallmark pathologies in AD are extracellular plaques composed of beta-amyloid (Aβ) and neurofibrillary tangles composed of hyperphosphorylated tau 24. Abnormal levels of Aβ can be detected decades before the onset of dementia 25 and abnormal Aβ levels in cognitively normal individuals is a marker of “preclinical” AD 26. It is possible to measure levels of Aβ in the cerebrospinal fluid (CSF) or using positron emission tomography (PET) ligands that bind to Aβ plaques. Only a modest number of studies in the field have tried to relate such measures of Aβ pathology to measures of obesity and diabetes to see if these health factors directly impact AD pathology 17,18.

We sought to better understand what relationship, if any, is present between obesity, diabetes, AD pathology, and the risk of becoming demented. To do this we examined data from longitudinal studies of aging and AD collected by the Knight Alzheimer’s Disease Research Center (ADRC) at Washington University in St. Louis. The Knight ADRC has been continuously funded since the late 1980’s to study the transition from cognitive normality to impairment. Work from the ADRC pioneered the concept of “preclinical” AD, and as part of its ongoing work collects cognitive, neuroimaging, and genetic data on its participants. Although not a primary focus, the ADRC also collects a number of lifestyle factors (e.g. height and weight) that we can leverage to explore the role between such health measures and AD.

The focus of our research was organized around a number of key questions. The first was 1) using the longitudinal cognitive data obtained by the ADRC since the 1980s, could we see any relationship between obesity and the longitudinal risk of developing dementia. The second question was 2) can we see any relationship between diabetes and obesity and AD pathology as measured using Aβ PET imaging. Finally, we were interested in whether 3) the relationship between health factors and AD biomarkers interact with the APOE genotype and age (midlife vs. late life).


Participants: The first task our project needed to complete was to identify potential data for our analyses. For our project we analyzed data collected by the Knight ADRC. Our focus was on individuals that were cognitively normal at their baseline visit, that had longitudinal cognitive data, had measures of AD pathology, and had some information pertaining to obesity and diabetes. Not all subjects had all types of data, leading to a number of sub-analyses devoted to each question. The data we were interested in is described in more detail below. In addition to the measures below, we collected demographic information such as age, sex, and years of education. For some analyses we stratified individuals based upon whether their baseline was in mid-life (45-60 years) or late-life (61+),

 Cognitive Measures: Clinical status was determined using the Clinical Dementia Rating (CDR) 27 This is a commonly used, standardized, clinical rating scale where a value of 0 represents someone who is non-demented, with values greater than 0 representing progressively greater impairment (CDR’s 0.5, 1, 2 and 3). Neuropsychological cognitive testing has also been acquired by the ADRC, but over the years the test battery has changed. To maximize our sample, we focused only on the CDR. All individuals in our analyses had to be cognitively normal at baseline (CDR=0). Longitudinal observations were used to determine if someone stayed cognitively normal (CDR=0 at all visits) or ever demented (CDR>0 at any future visit). We identified 664 individuals that were cognitively normal at baseline, had at least one longitudinal cognitive assessment, and had BMI data available. There were a total of 3080 cognitive assessments on this group.

 Measure of Obesity: To quantify obesity we calculated a body mass index 28. BMI is calculated from the ratio of weight(kilograms) to height2 (meters). We primarily examined BMI as a continuous measure, but additionally classified all individuals using current NIH guidelines: underweight (BMI < 18.5), normal (BMI 18.5-24.9), overweight (BMI 25-29.9), and obese (BMI > 30). The underweight subdivision was not used for additional analyses as there were too few individuals in this group. Height and weight were available for all subjects.

 Measures of Diabetes and Metabolic Health: As the primary focus of the knight ADRC is on the preclinical and clinical phases of AD, only a modest amount of effort has been paid to characterize additional health factors. As part of the biomarker and genetic studies performed fasting blood is stored. On a subset of individuals (~n=200) this blood has been processed to measure of glycated hemoglobin (HbA1c). This allowed us to classify individuals as having normal (below 5.7%), prediabetes (5.7 to 6.5%) or diabetic levels (<6.5%). If preliminary results indicated a relationship between HbA1c and cognition and AD pathology funds from the grant would be utilized to obtain HbA1c values on further blood samples stored by the ADRC.

Additionally, all participants self-report their medications. The self-reported medication is entered into a stored database, but the medications are not coded in any way. Early medication data acquired by the ADRC was also typed into the database rather than selecting specific medications within an electronics record system. This led to many transcription errors. To address this problem, we developed a string-matching algorithm. This computer code compared the entered medication text against a list of potential canonical medication names to identify the most likely true medication. Once the medication names had been cleaned up in this way, we determined if individuals were on any of a series of medications typically prescribed for diabetes. In this way we can denote whether an individual was on any diabetes medication at that visit. The majority of our cohort had multiple visits spanning years to decades of time in the studies. If someone was on a diabetes medication at any visit, they were considered to be in the “diabetes medication” category. We recognize that this is only a rough staging of potential disease mechanism, but given the rough nature of the health data it represented a starting point for our inquiries.

 Measures of AD Pathology: Aggregated plaques composed of beta-amyloid (Aβ) are the most common neuropathological hallmark of AD 24. Positron emission tomography (PET) is a neuroimaging technique commonly used to measure levels of Aβ protein in the brain. This technique uses a radiolabeled ligand that selectively binds to Aβ plaques. As a result the tracer collects in areas with high concentrations of plaques. As the label decays, it emits positrons which are detected by the scanner. This allows us to quantify the level of Aβ deposition in the brain. For the current project we identified all individuals from the ADRC cohort that had undergone imaging with either [18F] florbetapir (AV-45) or [11C] Pittsburgh Compouond B (PiB). The two tracers were combined using the Centiloid Scale 29 which is a linear scale where 0 represents levels in healthy younger controls and 100 levels in individuals with AD dementia. We identified 375 individuals who were cognitively normal at baseline and had BMI. A subset of 137 individuals also had Hb1Ac levels measured from fasting blood. These data are presented in figures 1-5.

Genotype: The e4 allele of the apolipoprotein (APOE) gene confers an increased risk of developing AD. There is prior evidence that the effects of obesity and diabetes are modulated by the APOE genotype 30. As previously described 31 , all participants underwent genotyping for APOE.

 Relating health factors to cognition: Our first question was whether our measured health factors of interest were related to an increased risk of every developing dementia. We performed these analyses using logistic regressions. Our main predictive variables were 1) continuous BMI at baseline or 2) BMI group at baseline. Only a modest subset of individuals had HbA1c at baseline (~200) which is why we focused on BMI as a predictor.

Relating health factors to AD biomarkers: Our second focus was relating our health factors to levels of AD pathology as measured by PET. General linear regression models (GLMs) related continuous measures of BMI and cortical burden (measured in Centiloid units). Models predicated levels of Aβ using continuous BMI, APOE e4 carrier status, sex, age, and education as predictors. As prior work examining BMI has shown different relationships based upon measurements in mid-life (45-60 years) or late-life (61+), we also included a BMI by age group (mid or late-life) interaction. As prior work suggests an interaction between BMI and APOE genotype, we also investigated the association of continuous measures of BMI and cortical Aβ burden by including an interaction term for continuous BMI and APOE e4 carrier status. 

For additional analyses, we investigated the differences in mean cortical Aβ burden between BMI groups (normal, overweight, obese) by computing a one-way analysis of variance (ANOVA) testing whether there was a main effect of group on Aβ level followed by a multiple pairwise-comparison t-test between the means of the groups. Demographics for this cohort is presented in Table 1. Finally, in the subset with HbA1c, we used linear regression to examine the continuous relationships between insulin resistance and Aβ burden and ANCOVA to examine the effect of groups.


Obesity Predicting Risk of Dementia: Using logistic regression in our sample of 664 individuals who were cognitively normal at baseline (CDR=0) we examined whether BMI as a either a continuous predictor, or categorical variable was associated with an increased risk of developing dementia (CDR <0). We found that neither continuous nor categorical measures of BMI were related to an increased risk of developing dementia, and that there were not significant interactions between BMI and APOE genotype or on age group in predicting dementia (all p’s>0.05).

Obesity Predicting Cortical Burden in Mid- and Late-Life: The relationship between a continuous measure of BMI and cortical Aβ burden is presented in Figure 1. In the late-life cohort, higher BMI was associated with lower cortical Aβ burden (β = -0.81, p=0.0066) in the linear regression model after including age, sex, years of education, and APOE e4 carrier status. BMI in the mid-life cohort did not show any significant association with Aβ burden (β=0.02, p > 0.5) after adjusting for the same covariates (Figure 2). When modeling an interaction between BMI and age group for the whole cohort, the BMI x age-group interaction term was significant (p = 0.0229). Thus, there is a differential effect of BMI in mid-life versus late-life (Figure 2). There was no significant interaction between BMI and APOE genotype in either the mid-life (β=0.28, p = 0.41) or the late-life (β=0.17, p > 0.5) groups (Figure 3).

Instead of examining BMI as a continuous measure we also looked at three BMI groups (normal, overweight, and obese) in mid- or late-life, we computed a one-way ANOVA test. In late-life there were significant differences in Aβ levels across the three groups (F2,270= 6.45, p = 0.002), with the highest levels in the normal group, and lowest levels in the obese group. In the mid-life cohort there was no significant effect of BMI group (F2,92=0.452, p > 0.5).

The BMI analyses has the large sample size of any of our health factors. As a follow-up we also examined those individuals that also had HbA1c measures. Not surprisingly we found that HbA1c levels were related to BMI (Figure 4). As with BMI, we saw that individuals with insulin resistance actually had lower levels of Aβ measured with PET compared to those with normal insulin function (Figure 5). Individuals with controlled diabetes (normal HbA1c but on diabetes medication) also had low levels of Aβ measured with PET.


Prior work has suggested a link between obesity and diabetes and an increased risk of developing AD dementia. In our analyses we did not find any relationship between increasing BMI in mid-life or late-life and longitudinal risk of dementia or that being on a diabetes medication or having abnormal levels HbA1c increases dementia risk. We found no evidence of an increased risk of dementia even when stratifying our cohort by mid-life and later-life as well as by APOE genotype. Even though our cognitive analyses examining BMI had a large cohort (n=664) this number is lower than the sizes of typical epidemiological studies and would not have power to detect very modest increases in the risk of developing dementia. It is also possible that midlife obesity is the key factor in determining longitudinal risk of dementia. The vast majority of our middle-aged cohort has less than ten years of longitudinal follow-up. With only a decade of follow-up we are only on the edge of the period of life (70+) where indices of dementia drastically increase. As we gain more cognitive follow-up we will be able to see if an effect emerges in the middle-aged cohort.

We were also interested in whether there would be any relationship between obesity and diabetes and AD pathology as measured through Aβ PET imaging. We found a modest association; such that having greater BMI and greater insulin resistance was associated with less AD pathology. This is counter to a prediction that obesity and diabetes directly exacerbate AD pathology, but consistent with prior work in the field similarly showing negative rather than positive associations between these health factors and AD pathology in the Harvard Aging Brain Study 17 and Alzheimer’s Disease Neuroimaging Initiative 18 cohorts. Finding similar results across three independent cohorts suggests that this is a real finding, but one that does not have an immediately clear interpretation.

In the Washington University cohort we required individuals to be cognitively normal at baseline. We found that individuals that were obese or that had insulin resistance at baseline also had lower levels of Aβ. There is a possibility that this is due to a selection bias. That is individuals with both high levels of pathology and poor health would no longer be cognitively normal, but would already be demented. This would remove the most diseased individuals from our cohort, leading to the negative relationship we reserved. However, work from other centers (34, 35) did include impaired individuals, suggesting that some sort of selection bias is not driving the results. We also had a high concordance between BMI and HbA1c, suggesting that in our cohort BMI was serving as a good predictor of metabolic health. Obesity and insulin resistance are also complicated by the fact that individuals often lose weight immediately preceding the onset of dementia and as dementia worsens. As a result the one time observation of BMI and HbA1c may not tell the full story of an individual’s health. We are continuing to examine the data and now are incorporating longitudinal analyses.

Future Plans:

Our work relating BMI and HbA1c to Aβ PET imaging is currently being written up by graduate student Austin McCullough and undergraduate Vineeth Thirunavu. The paper will be finalized by the end of the calendar year and submitted in the spring. The original funds for the proposal were primarily to support Austin’s stipend and supper pay for Vineeth to work in the lab. In addition, Austin was named a Willman Scholar to support his work and Vineeth was awarded a scholarship through the BioSURF program.

The current cohort also allows us to do additional analyses. Vineeth has made a computer program that will identify and classify the medications that our participants are on. While we have measures of HbA1c this is just on a small cohort. By analyzing the medication data we will be able to separate individuals into controlled and uncontrolled diabetes, and also to identify individuals on diabetes medication in the larger cohort. We are also interested in looking at whether change in weight (i.e. weight loss or weight gain) is associated with the risk of developing dementia and AD biomarkers. Finally, we also hope to still use the remaining funds on the gran to run samples for HbA1c.

Our current analyses have been a first step towards focusing on such health-related factors such as obesity and diabetes in the context of AD. We recognize that the available measures in our cohort are only rough approximations of the true measures we would like to get. Rather than looking at BMI, central adiposity or body composition from a DEXA scan would be a better measure. We need to get fasting HbA1c and lipid panels as a routine measure on all ADRC participants, not just a subsample. Ideally, we would even get an oral glucose tolerance test on participants. To pursue these aims we currently are preparing two R01s. One R01 we are collaborating on is primarily looking at the effects diabetes and obesity has on white matter integrity in the brain. This study will specifically recruit individuals with diabetes and is proposing to get DEXA and OGTT data on its participants in addition to MRI imaging. A second R01 in the works would provide funds to add additional measures to the ADRC cohort. This proposal would add dietary surveys, blood tests (HbA1c), body composition measures of central adiposity, DEXA, a five minute walk test, and a peak exercise test for aerobic capacity (VO2). Getting this data would enrich the ADRC cohort, and make it possible to ask more in-depth questions. The data funded by the LLF grant will serve as preliminary data in this R01 submission.


1.        Agha, M. & Agha, R. The rising prevalence of obesity. Int. J. Surg. Oncol. 2, e17 (2017).

2.        Xu, W. L. et al. Midlife overweight and obesity increase late-life dementia risk: a population-based twin study. Neurology 76, 1568–74 (2011).

3.        Fitzpatrick, A. L. et al. Midlife and Late-Life Obesity and the Risk of Dementia. Arch. Neurol. 66, 336–42 (2009).

4.        Xu, W. et al. Mid- and late-life diabetes in relation to the risk of dementia: a population-based twin study. Diabetes 58, 71–7 (2009).

5.        Singh-Manoux, A. et al. Obesity trajectories and risk of dementia: 28 years of follow-up in the Whitehall II Study. Alzheimers. Dement. 14, 178–186 (2018).

6.        Østergaard, S. D. et al. Associations between Potentially Modifiable Risk Factors and Alzheimer Disease: A Mendelian Randomization Study. PLOS Med. 12, e1001841 (2015).

7.        Qizilbash, N. et al. BMI and risk of dementia in two million people over two decades: a retrospective cohort study. lancet. Diabetes Endocrinol. 3, 431–436 (2015).

8.        Luchsinger, J. a, Patel, B., Tang, M.-X., Schupf, N. & Mayeux, R. Measures of adiposity and dementia risk in elderly persons. Arch. Neurol. 64, 392–8 (2007).

9.        Lainscak, M., von Haehling, S., Doehner, W. & Anker, S. D. The obesity paradox in chronic disease: facts and numbers. J. Cachexia. Sarcopenia Muscle 3, 1–4 (2012).

10.      Atti, A. R. et al. Late-life body mass index and dementia incidence: nine-year follow-up data from the Kungsholmen Project. J. Am. Geriatr. Soc. 56, 111–6 (2008).

11.      Whitmer, R. A., Gunderson, E. P., Barrett-Connor, E., Quesenberry, C. P. & Yaffe, K. Obesity in middle age and future risk of dementia: a 27 year longitudinal population based study. BMJ 330, 1360 (2005).

12.      Gorospe, E. C. & Dave, J. K. The risk of dementia with increased body mass index. Age Ageing 36, 23–29 (2006).

13.      Chuang, Y.-F. et al. Midlife adiposity predicts earlier onset of Alzheimer’s dementia, neuropathology and presymptomatic cerebral amyloid accumulation. Mol. Psychiatry 21, 910–5 (2016).

14.      Albanese, E. et al. Dementia severity and weight loss: a comparison across eight cohorts. The 10/66 study. Alzheimers. Dement. 9, 649–56 (2013).

15.      White, H., Pieper, C. & Schmader, K. The association of weight change in Alzheimer’s disease with severity of disease and mortality: a longitudinal analysis. J. Am. Geriatr. Soc. 46, 1223–7 (1998).

16.      Martins, I. J. et al. Apolipoprotein E, cholesterol metabolism, diabetes and the convergence of risk factors for Alzheimer’s disease and cardiovascular disease. Mol. Psychiatry 11, 721–736 (2006).

17.      Hsu, D. C. et al. Lower Late-Life Body-Mass Index is Associated with Higher Cortical Amyloid Burden in Clinically Normal Elderly. J. Alzheimer’s Dis. 53, 1097–1105 (2016).

18.      Vidoni, E. D., Townley, R. A., Honea, R. A., Burns, J. M. & Alzheimer’s Disease Neuroimaging Initiative. Alzheimer disease biomarkers are associated with body mass index. Neurology 77, 1913–1920 (2011).

19.      Zhao, L. et al. Insulin-degrading enzyme as a downstream target of insulin receptor signaling cascade: implications for Alzheimer’s disease intervention. J. Neurosci. 24, 11120–6 (2004).

20.      Ho, L. et al. Diet-induced insulin resistance promotes amyloidosis in a transgenic mouse model of Alzheimer’s disease. FASEB J. 18, 902–4 (2004).

21.      Farris, W. et al. Insulin-degrading enzyme regulates the levels of insulin, amyloid beta-protein, and the beta-amyloid precursor protein intracellular domain in vivo. Proc. Natl. Acad. Sci. U. S. A. 100, 4162–7 (2003).

22.      Buchman, A. S., Schneider, J. A., Wilson, R. S., Bienias, J. L. & Bennett, D. A. Body mass index in older persons is associated with Alzheimer disease pathology. Neurology 67, 1949–54 (2006).

23.      Grundman, M., Corey-Bloom, J., Jernigan, T., Archibald, S. & Thal, L. J. Low body weight in Alzheimer’s disease is associated with mesial temporal cortex atrophy. Neurology 46, 1585–91 (1996).

24.      Braak, H. & Braak, E. Neuropathological stageing of Alzheimer-related changes. Acta Neuropathol. 82, 239–259 (1991).

25.      Bateman, R. J. et al. Clinical and biomarker changes in dominantly inherited Alzheimer’s disease. N. Engl. J. Med. 367, 795–804 (2012).

26.      Jack, C. R. et al. Tracking pathophysiological processes in Alzheimer’s disease: an updated hypothetical model of dynamic biomarkers. Lancet Neurol. 12, 207–216 (2013).

27.      Morris, J. C. Clinical dementia rating: a reliable and valid diagnostic and staging measure for dementia of the Alzheimer type. Int. psychogeriatrics 9 Suppl 1, 173-176; discussion 177-178 (1997).

28.      Jensen, M. D. et al. Executive summary: Guidelines (2013) for the management of overweight and obesity in adults. Obesity 22, S5–S39 (2014).

29.      Klunk, W. E. et al. The Centiloid project: Standardizing quantitative amyloid plaque estimation by PET. Alzheimer’s Dement. 11, 1–15.e4 (2015).

30.      Craft, S. Insulin Metabolism in Alzheimer ’ s Disease Differs According to Apolipoprotein E Genotype and Gender. 98108, 146–152 (1999).

31.      Morris, J. C. et al. APOE predicts amyloid-beta but not tau Alzheimer pathology in cognitively normal aging. Ann. Neurol. 67, 122–31 (2010).

Tables and Figures:

Table 1. Demographics of Mid-Life and Late-Life Knight ADRC cohorts with PET







Age: mean (range)

53.5 (45-60)

71.4 (61 – 100)

Sex (% female)



Years of Education: mean (range)

16.3 (6-24)

15.8 (8-24)

APOE4 Status (% carriers)



Body Mass Index

- Underweight (BMI < 18.5)

- Normal (BMI 18.5 – 24.9)

- Overweight (BMI 25 – 29.9)

- Obese (BMI > 30)

28.4 (18.1 – 42.3)

1.0 % (n=1)


30.2% (n=29)


35.5% (n=34)


33.3% (n=32)

27.5 (14.1 – 58.7)

1.4% (n=4)


35.4% (n=98)


37.9% (n=105)


25.3% (n=70)

Centiloid value: mean(range)

-0.96 (-8.5 – 47.1)

13.5 (-16.6 – 131.05)

Table 2. Summary of additional analyses of those with BMI and PET. All models adjusted for age (or age-group), sex, years of education, and APOE4 status.




P – value


BMI × Age-Group Model

BMI x Age-Group



BMI Subdivision Model (Late-Life)

Normal BMI



Overweight BMI



Obese BMI



BMI × APOE4 Models

BMI ×APOE4 (late-life)


> 0.5

BMI ×APOE4 (mid-life)




Figure 1. Scatter plot of BMI and cortical burden by age group. The scatter plot reflects raw values for ease of interpretation. Late-life participants are in orange, and mid-life participants are in teal. Lines of best fit are also shown.

Figure 2. Boxplots of the distributions of cortical burden in each of the three BMI groups in late-life (3A) and mid-life (3B) individuals. Pairwise comparisons are denoted as * p < 0.05 and *** p < 0.001

Figure 3. Scatterplot of BMI and cortical burden within APOE4 subdivisions in mid-life (4A) and late-life (4B). The scatter plots reflect raw values for ease of interpretation. E4 non-carriers are in orange, and E4 carriers are in teal. Lines of best fit are also shown. The BMI x APOE4 interaction term was not significant in the mid-life (β=0.28, p = 0.41) or the late-life (β=0.17, p > 0.5) groups.

Figure 4. Boxplot of BMI across normal, prediabetes, and diabetic groups as determined by HbA1c. Unsurprisingly there was a significant effect of HbA1c group, with more impaired individuals having higher BMI values on average.

Figure 5. Boxplot of across normal, prediabetes, and diabetic groups. Mirroring what was seen with BMI, individuals with insulin resistance had lower levels of Aβ.