Mid- to Late-Life Body Mass Index and Dementia Risk: 38 Years of Follow-up of the Framingham Study

Abstract

Growing evidence relates body mass index (BMI) to poorer health outcomes; however, results across studies associating BMI and dementia are conflicting. A total of 3,632 Framingham Offspring participants aged 20 to 60 years at their second health examination (1979–1983) were included in this study, with 190 cases of incident dementia identified by 2017. Cox proportional hazards regression models were fitted to investigate the association of BMI at each of their 8 exams as a baseline for dementia risk and the associations between obesity and dementia across age groups. Spline models were fitted to investigate nonlinear associations between BMI and dementia. Each 1-unit increase in BMI at ages 40–49 years was associated with higher risk of dementia, but with lower risk after age 70 years. Obesity at ages 40–49 years was associated with higher risk of dementia. Overall, the relationship between BMI and dementia risk was heterogeneous across the adult age range. Monitoring BMI at different ages might mediate risk for dementia across an individual’s lifetime.

Abbreviations

 
• BMI

  body mass index

 
• FHS

  Framingham Heart Study

Editor’s note:  An invited commentary on this article appears on page 2511, and the authors’ response appears on page 2515.

With the rapid increase in the aging population, there has been an escalation in the number of people living with dementia worldwide. The number of people living with dementia worldwide is currently estimated at 50 million and will almost triple by 2050 (1). Currently, there are no effective disease-modifying therapies; thus, there has been an increasing call for a focus on dementia prevention as the best strategy for attenuating disease incidence (2). Body mass index (BMI), as a modifiable risk factor, has been related to various poorer health outcomes (3), but its specific relationship to dementia risk remains a matter of debate.

Observational studies have reported positive, null, and inverse associations between BMI and dementia risk (4): Evidence from longitudinal prospective studies and meta-analyses suggests that high BMI or obesity in midlife is associated with an increased risk of dementia (5–12). However, the hypothesis of the causal link between obesity and dementia has been challenged by some recent findings (13). A study of 2 million adults showed lower rates of dementia in those who are obese (BMI, weight (kg)/height (m)², of >40) and progressively decreasing risk with increasing obesity (14). This new study challenges the presumed direction of the association found from previous meta-analyses due to the large number of participants (13). Results on the association between late life BMI and dementia is more mixed (15, 16). Some studies report that being underweight or having a normal BMI in late life is associated with an increased risk of dementia (17–20), while others indicate an opposite relationship (21, 22).

Factors influencing the heterogeneity of study results might include age at assessment of BMI and years of follow-up for incident diagnosis. In this study, we used data from the Framingham Offspring Cohort, which includes serial assessments of BMI and prospective follow-up for incident dementia over 38 years to evaluate dementia risk associated with BMI from mid- to late life.

METHODS

Participants

Initiated in 1948, the Framingham Heart Study (FHS) is an ongoing, multigenerational longitudinal cohort study. Beginning in 1971, a total of 5,214 children of the Original Cohort and their spouses were enrolled into the Offspring Cohort, and they have completed up to 9 cycles of health examinations, averaging about every 4 years (23).

At the second Offspring examination (1979–1983), all participants were screened for possible cognitive impairment, and their cognitive status was monitored regularly at subsequent health examinations and through ancillary studies of cognitive aging and dementia. Therefore, we chose the second Offspring examination as the baseline (n = 3,867) and included dementia status follow-up through 2017 for analysis. People with a record of dementia before the second examination (n = 1) or with clinical data only at baseline (n = 49) were excluded from the sample. We restricted the inclusionary age range because participants aged <20 years (n = 10) were low in numbers and, after 38 years, remain below age of typical dementia onset. Excluding participants aged >60 years (n = 175) better ensured a dementia-free sample at baseline. The final study sample (n = 3,632) was composed of Offspring participants aged 20–60 years at baseline.

All study participants provided written informed consent. The study protocols and consent forms were approved by the Boston University Medical Campus and Boston Medical Center Institutional Review Board.

Surveillance for dementia

As part of the second health examination, beginning in 1979, the FHS Offspring Cohort participants were administered subjective memory questions to establish a dementia-free cohort, and their cognitive status was monitored using objective assessment beginning in 1991, when the Mini-Mental State Examination (MMSE) was added to the health-examination protocol. Since 1999, they have been administered a comprehensive protocol of standard neuropsychological testing on average every 5 or 6 years. Several methods for identifying those with possible cognitive impairment or decline have been previously described and include, but are not limited to, performance on these assessments. A consensus adjudication panel comprising at least 1 neuropsychologist and 1 neurologist makes the determination of whether diagnostic criteria for dementia are met (23–25).

The information used to access and verify a participant’s cognitive status includes FHS clinical examinations, neuropsychological assessments and neurologic exams, medical records, and/or family interviews (26). The details of this consensus diagnostic process has been previously described and has been applied the entirety of FHS cohorts, ensuring that all diagnoses have been made using the same diagnostic standards (27, 28).

BMI and covariates

At each of the 8 health examinations, FHS participants underwent in-person interviews, thorough physical examinations, and laboratory investigations. Demographic information, lifestyle factors, and medical histories were also recorded through self-reporting. Weight and height measurements were obtained during the physical examination, and BMI was calculated using weight (in kilograms) divided by height (in meters) squared. For this study, the continuous measure of BMI was stratified into 3 categories using cutoffs of <25, 25–30, and >30. Age (in years) was recorded at each exam, while education (years of school), and sex (male/female) were obtained at the second Offspring examination.

Statistical analysis

We performed all statistical analyses using Stata, version 15.0 (StataCorp LP, College Station, Texas). Demographic variables at baseline are presented as means with standard deviations or counts with proportions, using the t test and χ² test for respective comparisons. To fully utilize the longitudinal BMI data collected from examinations 2–9, we performed Cox proportional hazards regression using different baseline examinations over time. BMI measurements at each examination were related to incident dementia, adjusting for age, sex, and education.

To further determine how BMI from ages that span midlife to later life might affect dementia risk, we varied what was considered baseline BMI so that the participant’s “baseline” was always aligned within that specific age group (see Web Figure 1, available at https://doi.org/10.1093/aje/kwab096, for illustration of BMI–age group matching methods). Visual inspection of plots for younger ages (20–29, 30–39 years) indicated that the proportional hazards assumption was violated, so we only used hazard ratios and 95% confidence interval for analyses for participants 40 years and older. Across the 8 health exams, we extracted a participant’s BMI at each of these age groups: 40–49, 50–59, 60–69, and ≥70 years. Univariate and multivariate Cox regression was performed separately for each of the age groups, with date of the health examination as date of entry. If more than 1 BMI value was identified for a participant, we used the first BMI record within each age stratum. In the analysis for dementia, those who died free of dementia were censored at death so that participants were followed until the record of dementia, death, or last evaluation (13). We first performed a univariate Cox regression analysis (model 1), followed by fitting a model adjusted for sex and education (model 2). We then considered mortality as a competing risk and developed a model for the subhazard function of a failure event of primary interest (dementia) adjusting for sex and education (model 3). Our rationale for this approach was based on the partial likelihood subdistribution hazards model proposed by Fine and Gray that considered death as a competing risk not fully accounted for by censoring (29). The risk set in the subdistribution hazards model included those who were at risk of the event of interest (e.g., dementia) at a given time and anyone who had a competing risk (e.g., death) before that given time (29).

A cubic spline basis for BMI was used to investigate possible nonlinearity in BMI-dementia associations. Separate models were fitted for each age group, with BMI of 22 as the referent. Three BMI values, 18.5, 25, and 30, were placed as knots in the spline analysis, adjusted for sex and education.

RESULTS

The study cohort included 3,632 participants (51.8% women), with mean age of 43.5 (standard deviation, 9.4) years and an average BMI of 25.5 (standard deviation, 4.4) at the second health exam. During 38 years of follow-up, the average BMI increased to 28.4 (standard deviation, 5.3), and 196 participants developed dementia. The mean age of dementia diagnosis was 77.5 (standard deviation, 7.8) years. Not surprisingly, age was strongly associated with risk of dementia (P < 0.0001). Individuals who developed dementia had on average 1 year less of self-reported education compared with those without dementia (P = 0.0002). In absolute numbers, a larger proportion of women developed dementia during the follow-up compared with men, but sex was not significantly associated with dementia risk in the univariate analysis (Table 1).

We found no association between BMI at examination 2 and subsequent dementia. At examination 9, higher BMI was found to be associated to lower dementia risk. Generally, there was a trend in the hazard ratio for dementia risk by BMI to decrease across the health examinations (Table 2 and Figure 1).

For the age-specific analyses, obesity at midlife (40–49 years) was associated with higher risk of dementia after adjusting sex and education (hazard ratio = 3.15, 95% confidence interval: 1.35, 7.33). After accounting for competing mortality risk, obesity at ages 40–49 years still showed a higher risk of dementia (subhazard ratios = 2.90, 95% confidence interval: 1.27, 6.60). In contrast, in later life (ages ≥70), it seems there was an inverse relationship between obesity and incident dementia but not at conventional levels of statistical significance (P = 0.061) (Table 3).

Figure 2 shows the estimated shape of each BMI-dementia association, allowing for nonlinearity. A higher BMI at ages 40–49 years was associated with increased dementia risk if BMI was above 30. After age of 50 years, low BMI (<22) was associated with increased risk of dementia, while higher BMI appeared to reduce the risk of dementia if BMI was >22. However, the nonlinear associations were not significant. Being overweight was significantly associated with lower risk of dementia in the age group of ≥70 years (see Web Table 1).

DISCUSSION

Leveraging 38 years of data from the Framingham Offspring study, this study assessed repeated BMI measures to evaluate the risk of dementia associated with BMI across both time epochs (e.g., 8 health exams) and age ranges (mid- to later life). Overall, we determined that the relationship of BMI to dementia is not static and in fact might modulate over the adult life-course. BMI measured distally to dementia onset showed a positive relationship, but that relationship flipped to a seemingly negative relationship when BMI was measured more proximally to the time of dementia onset. Obesity at midlife (40–49 years) was associated with higher risk of dementia, even after adjusting for sex and education (P = 0.008). There was a nonsignificant trend toward obesity at older age (≥70 years) associated with lower risk of dementia after accounting for competing mortality risk (P = 0.061). The analyses for the participants aged ≥70 might suffer from power considerations with additional years of follow-up warranted to confirm these findings.

The exact mechanisms explaining the risk of dementia associated with obesity are poorly understood (13). Obesity in midlife has been previously associated with greater brain atrophy (30), and poorer cognitive outcomes potentially reflecting neurodegenerative processes via vascular and dysmetabolic pathways and/or through cell-signaling proteins secreted by the adipose tissue (e.g., leptin and adiponectin) (9, 31, 32). The possible reverse association between obesity and risk of dementia at older ages might, in part, be because of preclinical changes in weight in the years preceding clinical onset of dementia (33). Weight loss, posited as a manifestation of dementia (such as functional inability to maintain eating habits), occurs with comorbidities at older ages and might predate dementia onset by 8–10 years (13, 27).

Consistent with our findings are several reviews and meta-analyses that also show obesity in midlife is associated with an increased risk of dementia (34, 35). Similarly, a number of studies report that people who are underweight in late life have an increased risk of dementia (18–20). In our study, we looked at the nonlinearity patterns and revealed a more nuanced picture. From ages 50–69 years, higher BMI appeared to not be associated with dementia risk. Again, this is consistent with findings reported by previous studies, In the Honolulu Asian Aging Study (HAAS), higher weight in late midlife (age 46–68 years) was not associated with an increased risk of dementia (36). In the Atherosclerosis Risk in Communities (ARIC) Study, where the participants had a mean age of 59 years at the age of assessment of BMI, there was no association between higher BMI and cognitive change (37). Our possible finding of higher BMI measured later in life and lower risk of dementia, which is also associated with a shorter period of follow-up, is similar to the findings mentioned above from the recent study of 2 million adults (14). Heterogeneity in the associations between BMI and dementia risk might help explain the mixed findings across studies. Narrower age ranges and short follow-up help reduce variability and can help account for the more consistent findings of higher BMI association with lower dementia risk in the older age ranges.

Unlike most studies that evaluated BMI at only a few time points, we were able to establish the associations between BMI and dementia risk both at different time periods of follow-up and at specific age ranges. The strength of our study is the use of longitudinal BMI data spanning over 38 years, in the same persons, allowing us to assess the dementia risks associated with BMI over the adult life course. However, because associations we reported at midlife and late life involved the same individuals, these findings might also have been affected by changes in BMI with age. There are other limitations in this study. First, BMI might be not an optimal measure of obesity in late life. We did not evaluate waist circumference—an anthropometric measurement of obesity used in other research (38)—due to missing data across multiple exams. Second, confounders such as physical activity and dietary factors were not included in our models. A previous FHS study using cross-sectional measures of self-reported physical activity as a categorical variable (e.g., slight, moderate, and heavy activity) found nonsignificant associations between physical activity and risk of dementia (16). Continuous measures of physical activity were not collected until later exams. Similarly, dietary questionnaires were also not administered until later exams. Third, data on education was obtained only from examination 2 and changes in education, which could affect socioeconomic risk factors, were not accounted for longitudinally. However, because the number of participants in each examination was not same, we still used education at baseline in this study as the best proxy measure available. The potential for genetic factors that could also account for heterogeneity of the BMI dementia risk was also a limitation. Last, because FHS Offspring participants are predominantly non-Hispanic White, these results might not be generalizable to other ethnicities.

In conclusion, our study explored and distinguished the associations between BMI and dementia risk at different time frames of follow-up and at age ranges from mid- to late life. Our findings highlight the importance of a life-course perspective when assessing the association between BMI and dementia and indicate that BMI history should be considered in future studies.

ACKNOWLEDGMENTS

Author affiliations: School of Population Medicine and Public Health, Peking Union Medical College, Beijing, China (Jinlei Li); Department of General Dentistry, Boston University Henry Goldman School of Dental Medicine, Boston, Massachusetts, United States (Prajakta Joshi); Department of Anatomy and Neurobiology, Boston University School of Medicine, Boston, Massachusetts, United States (Prajakta Joshi, Ting Fang Alvin Ang, Sherral Devine, Rhoda Au); Framingham Heart Study, Boston University School of Medicine, Boston, Massachusetts, United States (Prajakta Joshi, Ting Fang Alvin Ang, Chunyu Liu, Sanford Auerbach, Sherral Devine, and Rhoda Au); Department of Epidemiology, Boston University School of Public Health, Boston, Massachusetts, United States (Ting Fang Alvin Ang, Rhoda Au); Department of Biostatistics, Boston University School of Public Health, Boston, Massachusetts, United States (Chunyu Liu); Department of Neurology, Boston University School of Medicine, Boston, Massachusetts, United States (Sanford Auerbach, Rhoda Au); and Boston University Alzheimer’s Disease Center and Boston University Chronic Traumatic Encephalopathy Center, Boston University School of Medicine, Boston, Massachusetts, United States (Sanford Auerbach, Rhoda Au).

This work was funded by the National Natural Science Foundation of China (grant 71661167004), the Framingham Heart Study’s contract with the National Heart, Lung, and Blood Institute (contract N01-HC-25195), the National Institute on Aging (grants AG016495, AG008122, AG033040, AG049810, AG068753, and AG062109), the National Institute of Neurological Disorders and Stroke (R01-NS017950), and the Alzheimer’s Association (grant VMF-14-318524).

Ethical approval: All study participants provided written informed consent. The study protocols, and consent forms were approved by the Boston University Medical Campus and Boston Medical Center Institutional Review Board.

R.A. serves as a scientific consultant to Signant Health and Biogen that is unrelated to this research work. The remaining authors report no conflicts.

REFERENCES