Abbreviations: AD, Alzheimer's disease; APOE, Apolipoprotein E; BMI, body Mass Index; CI, confidence interval; MCI, mild cognitive impairment; MMSE, Mini Mental Status Examination; RR, relative risk; SD, standard deviation; VaD, vascular dementia; WC, waist circumference.
KJ Anstey, Building 63, Eggleston Rd, Centre for Mental Health Research, Australian National University, Canberra, ACT 0200, Australia. E-mail: firstname.lastname@example.org
The relationship between body mass index (BMI) (in midlife and late-life) and dementia was investigated in meta-analyses of 16 articles reporting on 15 prospective studies. Follow-ups ranged from 3.2 to 36.0 years. Meta-analyses were conducted on samples including 25 624 participants evaluated for Alzheimer's disease (AD), 15 435 participants evaluated for vascular dementia (VaD) and 30 470 followed for any type of dementia (Any Dementia). Low BMI in midlife was associated with 1.96 [95% confidence interval (CI): 1.32, 2.92] times the risk of developing AD. The pooled relative risks for AD, VaD and Any Dementia for overweight BMI in midlife compared with normal BMI were 1.35 (95% CI:1.19, 1.54), 1.33 (95% CI: 1.02, 1.75) and 1.26 (95% CI: 1.10, 1.44), respectively. The pooled relative risks of AD and Any Dementia for obese BMI in midlife compared to normal BMI were 2.04 (95% CI: 1.59, 2.62) and 1.64 (95% CI: 1.34, 2.00), respectively. Continuous BMI in late-life was not associated with dementia. Small numbers of studies included in pooled analyses reduce generalizability of findings, and emphasize the need for publication of additional findings. We conclude that underweight, overweight and obesity in midlife increase dementia risk. Further research evaluating late-life BMI and dementia is required.
Dementia and obesity are public health problems that are increasing and cause a significant burden of disease (1). Although there are some biological reasons for expecting body fat to be positively associated with cognitive function, there is also evidence that higher body mass index (BMI) is associated with chronic diseases that increase the risk of dementia. Excess weight is associated with insulin resistance, hypertension and changes in coronary arteries leading to an increase in risk for cardiovascular disease (1,2). Obesity is a feature of metabolic syndrome which has been identified as a risk factor for neurocognitive disorders (3,4). However, body fat also contains leptin and oestrogen which are potentially neuroprotective providing plausible mechanisms to explain observations from studies of better cognitive performance among those with higher BMI (5,6). Leptin has been shown to enhance hippocampal plasticity in rats (7) to inhibit cell death (8) but its dys-regulation has been speculated to impact negatively on cognitive function (8,9).
Despite reports of overweight BMI in midlife being a risk factor for dementia in late-life (10), some benefits of overweight in late-life have been identified (5). Therefore there is a need to clarify how these two conditions may be linked or related. Since the publication of previous reviews on body weight and dementia (11,12), several high quality, large studies have been published that have yet to be synthesized with existing data. The present study therefore aimed to synthesize these new studies into the existing data on BMI and risk of dementia, and differs from previous reviews in distinguishing results for BMI measured at midlife and late-life.
The methodology used follows three previous reviews in this series (13–15) which were included in the 25 systematic reviews selected for inclusion in the National Institutes of Health report on prevention of dementia (16). A literature search was conducted utilizing the relevant databases PubMed (1950 to November 2009), PsycINFO (1872 to November 2009) and the Cochrane Library (1800 to November 2009), with searches being limited to studies in English and focused on humans. The reference lists of the retrieved articles were also hand searched for other applicable publications. The following combination of selected body weight/physical activity terms and cognition terms were used for the search where an asterisk (*) indicates a word truncation. The body-weight terms included: Fat, Adipo*, Anthropometric, Anthropometry, Body composition, Body size, Body Mass, Body mass index, BMI, Metabolic syndrome, Metabolic disorder, Obes*, Skinfold thickness, Syndrome X, Thinness, Underweight, Waist circumference, Waist-hip ratio, waist hip ratio, Weight. The physical activity terms included: Activity, Exercise, Physical, Fit*, Activity level, Aerobic, Condition.
The dementia and cognition terms included: Cognit*, Memory, Attention, Reaction time, Speed of processing, Processing speed, Crystallized ability, Crystallized intelligence, Fluid ability, Fluid intelligence, General mental ability, GMA, Intelligence, Executive function, Neuropsychological testing, Mini mental stat* exam*, MMSE, Dementia, Alzheimer (auto explode), Mild cognitive impairment, MCI.
Study inclusion criteria
The study inclusion criteria ensured that all articles included in the review met the Oxford Centre for Evidence-Based Medicine Level of Evidence 1B (http://www.cebm.net/index.aspx?o=1025). Additional quality ratings were conducted for all studies meeting criteria using a checklist adapted from previous reviews (17,18). Studies were required to be prospective, longitudinal, population based studies with a minimum follow-up period of 1 year. Studies were required to measure body weight or waist circumference (WC) at baseline or during a follow-up period that preceded the final follow-up examination. The outcome measure had to include either dementia or cognitive decline with body weight, BMI, or WC as the main predictor or covariate in the analyses. Studies on dementia needed to have screened for dementia at baseline or adjusted for incident dementia and/or baseline cognition performance in analyses, unless two standard deviations from the mean age of participants at baseline was less than 60 years old. Cross-sectional, experimental and clinical studies were excluded.
Citations downloaded into an Endnote reference database were screened in two stages. In an initial screening based on the selection criteria, relevant abstracts were identified by one of the authors. If a decision was unable to be reached from information reported, the abstract was tentatively included. All remaining abstracts were screened a second time by another independent reviewer. The selected journal articles retrieved were examined by at least two of the authors and rated against the selection criteria. Two further articles were recommended by contacted authors.
Authors were contacted via email for any missing information. Data relating to BMI were extracted for potential meta-analysis. Studies used different types of comparisons within body-weight measures including BMI (continuous variable), change in BMI and BMI categories (e.g. 18.5–24.9 vs. <18.5; 18.5–24.9 vs. 25–29.9; 18.5–24.9 vs. ≥30; BMI < 30 vs. ≥30, etc.). Information was extracted relating to whether BMI and WC were measured at midlife or late-life. Midlife was defined at 40 to 59 years. Late-life was defined as 60 years or older. Where results for more than one follow-up period were reported for the same study, the estimate from the longest follow-up was selected. Adjusted results were used where available.
Meta-analysis was conducted where there was more than one study compatible in terms of the period at which the body-weight measure was taken (midlife or late-life), the type of comparison conducted, and the outcome measure. One study divided the overweight BMI category into BMI 25 to 27.49, and 27.50 to 29.99 and the latter was used as the comparison category in analyses reported (19) although results were similar with the former category. In some cases data were pooled between studies using slightly different cut-offs for low, normal, overweight and obese. For example, in the comparison of normal and overweight adults in midlife, the categories for overweight included BMI ranges 25–30, 25–29.9 and 27.50–29.99. All studies defined obese as BMI > 30 or BMI ≥ 30. Details of the specific cut-offs used in each study in each analysis are noted in the relevant tables. There were no compatible data on WC or cognitive decline that allowed meta-analysis.
Potential dementia outcomes included Alzheimer's disease (AD) (20,21), vascular dementia (VaD) (22), and Any Dementia, which potentially included AD, VaD, dementia with Lewy bodies, frontotemporal dementia, Picks disease, alcohol related dementia, mixed and other dementias. Information collated included: study design (sample source, number of participants, observation period), sample characteristics (country, percentage female, average age, age range, years of education), measurement of body weight (BMI, weight, WC), physical activity, covariates included in statistical models, measurement of dementia or cognition, unadjusted and adjusted estimates of association, hazard ratio (HR), relative risk (RR), beta and P-values with 95% confidence intervals (CIs). Data extraction was checked by a second reviewer.
The RR and HR were treated the same and are referred to as RR. This combining step is based on the assumption that dementia is a relatively rare event, and that the two different measures are therefore valid estimates of RR (23) and has been used previously (24). The data points for the meta-analysis of binary outcomes were the logarithms of the RRs and their standard errors (i.e. 2 data points per included study). The standard errors of the RRs are estimated from the 95% CI of the log RR by dividing the width of the interval by 3.92 (which is twice the 97.5th percentile of the standard normal distribution).
One study (25) reported data by subgroup. Prior to inclusion the meta-analyses, effect sizes reported for subgroups categorized by age and gender were pooled using fixed effects analysis. Effects were calculated for the midlife age range for men and women for underweight, overweight and obese categories (total 17 effects, i.e. 34 data points) compared to normal BMI.
Heterogeneity among studies was examined using standard I2 which estimates the proportion of variance because of heterogeneity (26). Fixed effects analysis is recommended for small numbers of studies where it is difficult to estimate variance (27) and was used for analyses with two or three studies. If there was evidence of heterogeneity and four or more studies were available, random effects meta-analysis was used (11,28). Both fixed effects and random effects analyses are weighted such that the weight is inversely proportional to the standard error of estimate of each study, accounting for sample sizes. Analyses were conducted using Revman 4.2 (29).
Seventy-nine articles meeting inclusion criteria were identified, of which 26 articles included sufficient data for meta-analysis. However, only 16 of these reports included data compatible with another study (10,11,19,30–42). Ten studies included data (or for which data were supplied by authors), could not be included in the meta-analysis (see Appendix S1). Figure 1 shows the stages in identifying studies for inclusion in the review. Characteristics of the studies from the 16 articles included in meta-analyses are shown in Table 1. Studies were drawn from the USA, Sweden, France, Finland and Japan. There were no combinations of compatible studies using WC so these data were not analysed.
Table 1. Articles included in meta-analyses
Study: first author, year (no. of subjects)
Study name (source)
Observation period, years (SD)
Body weight measures
Mean age, years (SD)
AD, Alzheimer's disease; BMI, body Mass Index; CASI, computer assisted self-interviewing; CCSMHA, Cache County Study on Memory and Health in Aging; COD, cause of death; DAS, dementia associated with stroke; DQ, Dementia Questionnaire;
DSM, Diagnostic and Statistical Manual of Mental disorders; IQCODE, informant questionnaire on cognitive decline in the elderly; MMSE, Mini Mental Status Examination; N/A, not applicable; NINCDS-ADRDA, National Institute of Neurological and Communicative Diseases and Stroke/Alzheimer's disease and Related Disorders Association; NINDS-AIREN, National Institute of Neurological Disorders and Stroke and the Association Internationale pour la Recherche et l'Enseignement en Neurosciences; SD, standard deviation; VaD, vascular dementia; WC, waist circumference.
Beydoun, 2008 (2 322)
Baltimore Longitudinal Study of Aging (Baltimore, USA)
BMI < 18.5, 18.5–24.9, 25.29.9, ≥30 Change in BMI WC Q1, Q2, Q3, Q4 Change in WC
AD = 68.6 (12.1) No AD = 56.9 (15.4)
AD = 38.0 No AD = 36.4
AD = 16.8 (2.7) No AD = 16.6 (2.9)
Borenstein Graves, 2001 (1 869)
Ni-Hon-Sea Project (Japanese-Americans aged 65 years+ living in King country, WA,USA)
Incident AD 3.8 (1.5) No AD 3.8 (1.7)
CASI DSM-IV NINCDS-ADRDA
Buchman, 2005 (918)
Religious orders study (Catholic clergy members from different religious communities across USA)
Measure of global cognitive function computed from scores on 19 tests
Developed AD = 80.2 (6.36) Did not develop AD 74.0 (6.79)
Developed AD = 68.6 Did not develop AD = 68.1
Developed AD = 18.1 (3.6) Did not develop AD = 18.2 (3.3)
No dementia = 42.45 (1.71) Dementia = 42.89 (1.66)
No dementia = 54.0 Dementia = 55.5
Level of education completed: Grade school (to age 12) No dementia = 13.8% Dementia = 19.5% High school (to age 18) No dementia = 34.4% Dementia = 32.3% Trade or technical school No dementia = 7.1% Dementia = 6.3% College or university No dementia = 44.8% Dementia = 41.9%
Underweight 81.75 Normal 64.09 Overweight 37.94 Obese 57.13
% with high school or less Underweight 45.99 Normal 45.24 Overweight 50.89 Obese 59.26
Yoshitake, 1995 (826)
N/A (Hisayanma residents aged 65 years+, Japan)
DSM-III-R NINCDS-ADRDA NINDS-AIREN
Male = 73 ± 5.6 Female = 74 ± 6.1
One study reported data on a male sample (19) and another included only female participants in analyses because of low numbers of males in the sample (34). Most studies did not stratify results by sex.
Quality ratings of included studies were all high. All met the following criteria: definition of exposure variable, definition of outcome, prospective design, sample size >100, multivariate statistics described, and data collection was standardized. Description of study period was reported in all but two (10,40) although this information was available from related publications. Description of sampling procedure and data collection was included in all but two studies (25,42). Few studies reported the psychometric properties of instruments and only about half of the studies were reproducible from the article alone. All prospective studies on which the articles were based had other sources of information on study design such as previous publications and websites.
Number of studies and participants included in pooled estimates according to outcome
Eleven studies reported compatible data on BMI and risk of AD (25,30,31,33–37,39,41,42). There were 15 256 participants included in the two studies that assessed BMI in midlife for later AD risk, and these studies had an average follow-up of 29.7 years. There were 13 166 participants included in the nine studies that assessed BMI in late-life for AD risk and these studies had an average follow-up of 6.82 years.
Four studies reported compatible data on BMI in late-life and risk of VaD (33,34,36,42), including a total of 5299 participants with an average follow-up of 9.55 years. Three studies reported on BMI and Any Dementia in midlife (10,19,33) (total n = 20 476; average follow-up 18.27 years) and six studies reported on BMI and Any Dementia in late-life (n = 12 792) with average follow-up of 8.09 years (32–34,36,37,43). There were no studies with compatible data on BMI and cognitive decline.
Low body mass index vs. normal body mass index and risk of Alzheimer's disease
The RR of AD derived from individual studies and the pooled estimates for low BMI vs. normal BMI are shown in Table 2. Compared to those with normal BMI, those with low BMI in midlife had 1.96 times the risk of developing AD in late-life. Note that this included data from one study where midlife BMIs were reported retrospectively (33) and that the small number of studies reduces the robustness of the pooled estimate.
Table 2. Risk of Alzheimer's disease in low vs. normal BMI in midlife
Study: first author
Study weight (%)
Adjusted for age, race, sex, years of education, c-reactive protein level. Interleukin 6 level, hypertension, diabetes, coronary heart disease cholesterol, ankle arm index, smoking activity, APOE; BMI categories for this study were underweight (<20), normal weight (20–25), overweight (25–30) and obese (>30); comparison for this particular relative risk was BMI 20–25 vs. <20; report of midlife BMI is retrospective.
Adjusted for age, age in midlife, education, race, sex, marital status, smoking, hyperlipidemia, hypertension, diabetes, ischemic heart disease and stroke; BMI categories were obese (≥30), overweight (25.0–29.9), normal (18.6–24.9), underweight (≤18.5); comparison for this particular relative risk was BMI 18.5–24.9 vs. <18.5.
Adjusted for years of education, ethnicity and smoking status; BMI categories for this study were <18.5 (underweight), 18.5–24.9 (normal), 25–29.9 (overweight) and >30 (obese); the relative risk estimate reported in the present study was derived by pooling results from gender subgroup analyses; comparison for this relative risk was BMI 18.5–24.9 vs. <18.5.
APOE, Apolipoprotein E; BMI, body mass index; CI, confidence interval.
Overweight vs. normal body mass index and risk of Alzheimer's disease and Any Dementia
The individual study and pooled RRs for AD and Any Dementia for overweight BMI vs. normal BMI are shown in Table 3. Compared with normal BMI, overweight BMI in midlife is associated with 35% increased risk of developing AD, 33% increased risk of developing VaD in late-life and 26% increased risk of Any Dementia.
Table 3. Risk of dementia in overweight vs. normal BMI groups in midlife
Study: first author
Study weight (%)
Adjusted for age, race, sex, years of education, c-reactive protein level. Interleukin 6 level, hypertension, diabetes, coronary heart disease cholesterol, ankle arm index, smoking activity, APOE; BMI categories for this study were underweight (<20), normal weight (20–25), overweight (25–30) and obese (>30); comparison for this particular relative risk was BMI 20–25 vs. 25–30; report of midlife BMI in this study is retrospective.
Adjusted for age, age in midlife, education, race, sex, marital status, smoking, hyperlipidemia, hypertension, diabetes, ischemic heart disease and stroke; BMI categories for this study were obese (≥30), overweight (25.0–29.9), normal (18.6–24.9) and underweight (≤18.5); comparison for this particular relative risk was BMI 18.5–24.9 vs. 25–29.9.
Adjusted for years of education, ethnicity and smoking status, age, diabetes, systolic BP, serum cholesterol; the relative risk estimate reported in the present study was derived by pooling results from gender subgroup analyses; BMI categories for this study were <18.5 (underweight), 18.5–24.9 (normal), 25–29.9 (overweight) and ≥30 (obese); comparison for this particular relative risk was BMI 18.5–24.9 vs. 25–29.9.
BMI categories for this study were <20, 20–22.49, 22.50–24.99, 25–27.49, 27.50–29.99 and >30; comparison used for this particular relative risk was 20–22.49 vs. 27.50–29.99.
APOE, Apolipoprotein E; BMI, body mass index; BP, blood pressure; CI, confidence interval.
Obese vs. normal body mass index and risk of Alzheimer's disease and Any Dementia
The individual study and pooled RRs for AD and Any Dementia for obese BMI vs. normal BMI measured in midlife are shown in Table 4. Compared with normal BMI, obese BMI in midlife is associated with a RR of 2.04 for AD and 1.64 for Any Dementia. In late-life, obese BMI was not associated with increased risk of AD in pooled analyses. Compared with non-obese adults in late-life, obese older adults did not have an increased risk of dementia.
Table 4. Obesity in midlife and late-life and risk of dementia
Study: first author
Study weight (%)
Adjusted for age, race, sex, years of education, c-reactive protein level, interleukin 6 level hypertension, diabetes, coronary heart disease cholesterol, ankle arm index, smoking activity, APOE; BMI categories were underweight (<20), normal weight (20–25), overweight (25–30) and obese (>30); comparison used for this particular relative risk was BMI 20–25 vs. >30; midlife BMI is retrospective.
Adjusted for age, age in midlife, education, race, sex marital status, smoking, hyperlipidemia, hypertension, diabetes, ischemic heart disease and stroke; BMI categories were obese (≥30), overweight (25.0–29.9), normal (18.6–24.9) and underweight (≤18.5); comparison for this particular relative risk was BMI 18.5–24.9 vs. ≥30.
Adjusted for years of education, ethnicity and smoking status; the relative risk estimate reported in this study was derived by pooling results from gender subgroup analyses; BMI categories were <18.5 (underweight), 18.5–24.9 (normal), 25–29.9 (overweight) and >30 (obese); comparison for this relative risk was 18.5–24.9 vs. ≥30.
Adjusted for age, sex, education, APOE e4 allele, hypertension, high cholesterol, diabetes, stroke, CABG, MI, systolic BP, serum cholesterol; obese BMI was classified as >30; comparison for this particular relative risk was BMI < 30 vs. ≥30.
Adjusted for age, sex, education, ethnic group, APOE status; BMI quartiles were used: <23.4, 23.4–26.2, 26.3–29.6 and >29.6; comparison used for this relative risk was BMI < 23.4 vs. >29.6.
BMI categories for this study were <20, 20–22.49, 22.50–24.99, 25–27.49, 27.50–29.99 and >30; 20.00–22.49 vs. ≥30.
Continuous body mass index and risk of Alzheimer's disease, vascular dementia and Any Dementia in late-life
Table 5 shows the RRs for continuous BMI measured in late-life in relation to dementia outcomes. Change in late-life BMI was not associated with AD, and actual BMIs were not significantly associated with risk of VaD or any Dementia. Heterogeneity statistics were significant for AD (χ2 (7) = 28.94, P < 0.01, I2 = 75.8) and Any Dementia (χ2 (5) = 17.79, P < 0.01, I2 = 71.9) but not VaD (χ2 (3) = 6.14, P = 0.11, I2 = 51.1). Funnel plots for analyses of late-life BMI and risk of AD and Any Dementia are shown in Fig. S1.
Table 5. Continuous BMI in late-life and risk of dementia
Adjusted for age, race, sex, years of education, c-reactive protein level, interleukin 6 level, hypertension, diabetes, coronary heart disease cholesterol, ankle arm index, smoking activity, APOE; report of midlife BMI in this study is retrospective; report of midlife BMI in this study is retrospective.
Adjusted for age.
Adjusted for sex, education, APOE.
Adjusted for diastolic blood pressure, cardiovascular disease, smoking, SES, treatment for hypertension.
Adjusted for age, sex, education, ethnic group, APOE status.
Adjusted for recruitment cohort, age, gender, education, ethnicity, caloric intake, APOE, physical activity, adherence to the Mediterranean diet, duration between dietary assessment and physical activity assessment.
Adjusted for age, sex, education, cardiovascular disease, diabetes, smoking, alcohol use
Adjusted for MeDi score (0–9), age, sex, education, marital status, total energy intake, practice of physical exercise, taking ≥5 drugs d−1, Center for Epidemiological Studies-Depression scale score, APOE genotype, hypertension, hypercholesterolemia, diabetes, stroke and tobacco and their interaction with time.
APOE, Apolipoprotein E; BMI, body mass index; CI, confidence interval; SES, socioeconomic status; TIA, transient ischemic attack.
Risk of Alzheimer's disease with change in continuous BMI
Selected analyses for males and females were possible for AD as two studies provided effect sizes for men and women (25,34). Underweight BMI in midlife was not associated with increased or decreased risk of AD in females (RR: 1.51; CI: 0.88, 2.60), but only one study reported relevant data for males (data not shown).
Overweight BMI in women in midlife appeared to be associated with an increased risk of AD when analyses were used to pool two studies (RR: 1.83; CI: 1.45, 2.29). In men, overweight BMI in midlife did not increase the risk of AD (RR: 0.98; CI: 0.96, 1.00). Obese BMI in midlife women (RR: 3.08; CI: 2.16, 4.37) and men (RR: 2.45; CI: 1.51, 3.95) was associated with an increased risk of AD in late-life (data not shown).
Studies meeting criteria that were not compatible for pooling and inclusion in meta-analysis
Ten studies (three with BMI measured in midlife and seven with BMI measured in late-life) reported data where the combination of BMI ranges and outcome measures were not compatible with any other study. The findings from these studies are summarized here.
A study of 169 adults aged 68 and older found that BMI ≥ 23 was associated with reduced risk of cognitive decline measured on the Mini Mental Status Examination (MMSE) (44). Continuous BMI was not associated with cognitive decline in two studies (n = 2509) (45,46). Another study examined BMI (in a sample aged 68 years and older), and cognitive decline according to BMI cut-points. It found that participants with BMI ≥ 23 had reduced risk of cognitive decline compared with participants with BMI < 23, in the subsequent 5 years, after adjusting for age and sex (OR: 0.28; CI: 0.09, 0.90) (44). Another study found that high BMI in late-life was not associated with increased risk of dementia (43). The only study reporting findings on continuous BMI and mild cognitive impairment (n = 1393) found no association (40). Another 8-year prospective study (n = 1351) found that WC in late-life (but not overweight or obesity) was associated with increased risk of dementia (47). A study of adults aged 33 to 62 found that over a 5-year follow-up period, higher BMI at baseline was associated with greater decline on word-list learning but not other cognitive measures (48).
Non-linear relationship of BMI and dementia risk, and change in BMI and dementia risk
One study showed that among participants aged under 76, the second and third quartiles of BMI were associated with the lowest risk of dementia (37). However, in participants aged 76 and older, there was a linear relationship between BMI and dementia risk such that increasing BMI was associated with decreasing risk (37).
Consideration of covariates
Factors that have been associated with dementia risk in previous reviews (13,15,49,50) were not adjusted for consistently among studies included in meta-analyses. In some cases this was because authors identified no effect of the covariate at baseline. Only one study controlled for diet or dietary components (40) despite the association of diet with weight and dementia (40). Only one study adjusted for alcohol consumption (39) and four adjusted for smoking (11,34,39,41). The relationship between body weight and cognition was adjusted for Apolipoprotein E genotype in six studies (30,32,33,35,37,40).
Our meta-analysis synthesized all available, compatible data from prospective studies reporting on the association of BMI with risk of dementia in late-life. In midlife, we found that underweight BMI, overweight BMI and obese BMI were all associated with increased risk of dementia compared with normal BMI. Risks appeared to be highest for underweight and obese BMI, suggesting a U-shaped relationship between midlife BMI and dementia risk, consistent with previous systematic reviews and a meta-analysis on body weight and dementia (11,12), that were based on fewer studies. Our findings were similar for AD and Any Dementia, probably because AD is the most prevalent form of dementia and so would have accounted for the majority of cases in the Any Dementia category. The lower prevalence of VaD means that it is more difficult to obtain enough information to conduct meta-analyses. Nevertheless, we still found that overweight BMI in midlife increased the risk of late-life VaD by 33%.
The small numbers of studies in our analyses, the use of fixed effects meta-analysis and the large values of I2, suggest that caution should be used in generalizing these results to broader populations (27). However, analyses were based on very large combined sample sizes and are consistent with two previous reviews (11,12). Nevertheless, it is critically important for more longitudinal studies to report results on BMI and late-life dementia so that more robust and generalizable pooled estimates can be obtained.
Most of the studies in late-life used BMI as a continuous measure, which precluded analysis by BMI category. If there is a U-shaped association between BMI and dementia risk in late-life that mirrors the pattern evident for BMI in midlife, then use of continuous BMI in linear models may not model this relationship accurately. It is therefore unclear whether the non-significant associations between continuous BMI in late-life and VaD and Any Dementia are true null findings or whether there is a weak protective effect of higher BMI in late-life. The one study meeting our criteria but not compatible for meta-analyses used categorical BMI and found that higher BMI in late-life was protective against dementia (43). Another difference between the studies of late-life and midlife is the follow-up period. Midlife studies had a much longer period of follow-up prior to dementia diagnosis. The average length of follow-up for the late-life studies was 7.13 years, which would include the prodromal phase of AD in particular (51). The only late-life study with a long follow-up period did find a significant association between BMI and AD risk (34).
Analyses conducted for men and women separately showed similar patterns to pooled analyses with a suggestion of stronger effects for women than men. For example, obese BMI in midlife was associated with a 3.08 times increased risk of AD for women and a 2.45 times increased risk for men. Fewer results were significant in these subgroup analyses, probably because of the smaller sample sizes.
A concern for interpretation of the significance of BMI as a dementia risk factor is the lack of consistent adjustment for health and demographic variables as well as factors such as diet, smoking, alcohol and physical activity. It is possible that these other factors interact with or are indirectly influencing the association between obesity and dementia risk. For example, a high-fat, high-calorie diet may increase BMI and dementia risk independently, or affect dementia risk via BMI. Luchsinger et al. (37) found in a prospective study of 939 individuals aged 65 and over with a follow-up of 6.3 years that those with greater caloric intake were at higher risk of AD. Other dietary factors have been reviewed and found potentially protective against dementia but results remain inconclusive (16). These include B-vitamins (52), omega-3 fatty acids (49) and antioxidants (53). There is relatively strong evidence that low levels of alcohol consumption are associated with decreased dementia risk (14).
The strengths of this meta-analysis include large sample size, long follow-up for the midlife studies, clinical diagnoses of dementia and exclusion of individuals with dementia at baseline. There was some variability in the cut-offs of BMI chosen by researchers with most using the World Health Organization guidelines (54) but others investigated smaller ranges of BMI (19). The overweight range in particular may include individuals who are slightly overweight up to those nearly obese. There may be a more specific threshold within this BMI range at which BMI becomes a risk factor. This meta-analysis is limited by the small numbers of studies available for some comparisons, and the lack of data on interactions with other risk factors for dementia. The comparatively small number of studies reporting data on underweight BMI and dementia risk in late-life may reflect publication bias and hence it is possible that our results overestimate the risk of underweight BMI. One article included self-report data for BMI in midlife (33) and the effect sizes for this study were in most cases smaller than studies where BMI had been measured in midlife. The inclusion of publications where BMI was not the main variable of interest may have counteracted publication bias. We lacked sufficient studies to conduct a robust evaluation of publication bias and note that current methods for evaluating publication bias have been criticized (55).
Cohorts differ in the pattern of fat distribution, and associated risk factors and this may contribute to differences in results (56). In this review, cohorts included in the midlife assessments were born between 1915 and 1933 and the cohorts included in the late-life assessments of BMI were born between 1901 and 1930. It is possible that improved chronic disease management (e.g. hypertension) means that overweight adults in later cohorts will have a relatively reduced risk of dementia. Body weight may also affect dementia risk via its effect on other medical conditions such as coronary events, and may interact with other cardiovascular risk factors. Body weight may also be increased in depression and anxiety, which have been linked with increased dementia risk (57).
Our meta-analysis suggests a U-shaped relationship between midlife BMI and late-life risk of AD and Any Dementia. Normal BMI in midlife confers the lowest risk of dementia, and it appears that obese BMI in midlife confers the greatest risk. Our study found no association between continuous BMI and late-life dementia but because of methodological factors described above we are hesitant to conclude that there is no association between late-life continuous BMI and dementia. It is possible that the length of follow-up of cohorts is critical to the evaluation of the risk of BMI for dementia in late-life. Further data are required from studies with longer follow-ups in late-life, reporting outcomes for BMI categories.
Without reduction in population-levels of obesity, our results suggest that we can expect to see an increase in the prevalence of dementia as the current cohorts age. Reducing the impact of obesity on dementia prevalence and incidence should be a priority for governments, health providers and the general public.
Conflict of Interest Statement
No conflict of interest was declared.
Funding: Australian Dementia Collaborative Research Centres; NHMRC fellowships 366756 and 471501. We thank Dr Richard Burns, Sayuri Prior and Lyndall Kopp. We thank the authors who kindly provided data for meta-analyses.