Association of anthropometry and weight change with risk of dementia and its major subtypes: A meta‐analysis consisting 2.8 million adults with 57 294 cases of dementia

Summary Uncertainty exists regarding the relation of body size and weight change with dementia risk. As populations continue to age and the global obesity epidemic shows no sign of waning, reliable quantification of such associations is important. We examined the relationship of body mass index, waist circumference, and annual percent weight change with risk of dementia and its subtypes by pooling data from 19 prospective cohort studies and four clinical trials using meta‐analysis. Compared with body mass index–defined lower‐normal weight (18.5‐22.4 kg/m2), the risk of all‐cause dementia was higher among underweight individuals but lower among those with upper‐normal (22.5‐24.9 kg/m2) levels. Obesity was associated with higher risk in vascular dementia. Similarly, relative to the lowest fifth of waist circumference, those in the highest fifth had nonsignificant higher vascular dementia risk. Weight loss was associated with higher all‐cause dementia risk relative to weight maintenance. Weight gain was weakly associated with higher vascular dementia risk. The relationship between body size, weight change, and dementia is complex and exhibits non‐linear associations depending on dementia subtype under scrutiny. Weight loss was associated with an elevated risk most likely due to reverse causality and/or pathophysiological changes in the brain, although the latter remains speculative.

kg/m 2 ) levels. Obesity was associated with higher risk in vascular dementia. Similarly, relative to the lowest fifth of waist circumference, those in the highest fifth had nonsignificant higher vascular dementia risk. Weight loss was associated with higher all-cause dementia risk relative to weight maintenance. Weight gain was weakly associated with higher vascular dementia risk. The relationship between body size, weight change, and dementia is complex and exhibits non-linear associations depending on dementia subtype under scrutiny. Weight loss was associated with an elevated risk most likely due to reverse causality and/or pathophysiological changes in the brain, although the latter remains speculative.

| INTRODUCTION
Dementia, a disease primarily of aging, affects an estimated 47 million people globally. 1 It is a heterogeneous condition chiefly comprising Alzheimer disease (60-70% of cases) and vascular dementia accounting for about 15% of cases, although the two subtypes frequently co-occur. 1,2 Aging, family history, and sarcopenia are important risk factors for dementia, and there is growing evidence that vascular risk factors, such as diabetes, may also confer increased risk, particularly for vascular dementia, although findings are inconsistent. 3,4 Excess body weight, typically defined as having a high body mass index (BMI), has been causally linked to a large number of chronic conditions, particularly vascular disease. 5 While the relationship between BMI and dementia has been the subject of several largescale epidemiological studies, the findings have been inconsistent: some studies have reported that only individuals at the extreme ends of the body size spectrum (ie, underweight and obese) experience an increased dementia risk, 6,7 while others have documented positive 8 and even inverse 9 associations. Measures of central obesity, such as waist circumference (WC), have been argued to be more informative measures of obesity-related risk compared with BMI. Currently, little is known about the association between central obesity and dementia risk.
As nonvascular and vascular dementia have different pathophysiologies, any association with body size may similarly differ according to endpoint. Distinguishing between possible dementia subtypes in any analysis with measures of body size may, therefore, prove informative in explaining some of the observed heterogeneity. Further, whether sex differences exist, or if the association between body size and dementia risk differs in middle and later life remains unclear. 10,11 In this meta-analysis, we examined the relationships of body size with all-cause dementia and possible vascular and nonvascular dementia by sex and baseline age in participants free of dementia who had their body size assessed at baseline and were later followed up on dementia status. Using repeat measures, where available, we explored the association between standardized annual weight change during follow-up with subsequent dementia risk.

| METHODS
The investigators of studies that were either identified from previous systematic reviews and meta-analyses, 6,7,12-14 or who were known to Dementia Pooling Project 15 collaborators, were contacted and asked to contribute results for their studies ( Figure S1). Seven of the 14 studies from the Dementia Pooling Project contributed data that were included in the analyses. Additional relevant data from 28 studies were identified from five previous systematic reviews and meta-analyses of the association. Fourteen studies that had not contributed to previous overviews were identified through a PUBMED search by one investigator (CMYL). This was limited to human subjects and the

| Data analysis
Sex-specific hazard ratios (HRs) and 95% confidence intervals (CIs) were obtained for all-cause dementia in relation to (a) each of five BMI categories, with lower-normal weight as the referent group; (b) each fifth of WC, with the first fifth as the referent group; and (c) each of three annual percent weight change categories (greater than or equal to 0.5% annual weight loss, less than 0.5% annual weight change, and greater than or equal to 0.5% annual weight gain), with less than 0Á5% annual weight change as the referent group. Following a pre-specified common analytic protocol, effect estimates were adjusted for age (model 1); age, smoking, and education or socio-economic status (model 2; which was the primary model-used in the reporting of outcomes herein); and age, smoking, education or socioeconomic status, diabetes, systolic blood pressure, total cholesterol, blood pressure-lowering medication, cholesterol-lowering medication, and glucose-lowering medication where available (model 3). For studies with information on possible dementia subtype, study-specific estimates were requested for possible vascular dementia and possible nonvascular dementia.
A random effects meta-analysis was used to combine study-specific log HR to obtain an overall summary estimate and associated 95% CIs for BMI and WC in relation to all dementia endpoints investigated. Analyses were conducted for women and men combined and then separately. Heterogeneity between studies was quantified using the I 2 statistic. Sensitivity analyses were conducted by excluding the largest study and by excluding the studies that did not calculate BMI using objective measures of height and weight. To assess the potential effect of reverse causality, in the same studies, we compared data that were non-left censored with those that were. As the studies varied by the length of follow-up, maximum periods of left censoring requested (3, 5, Figure S2a; Table S3). Relative to the referent group, individuals with upper-normal BMI, overweight or obesity had a 10% to 15% lower dementia risk (Figure 1). Similar results were obtained after adjustment for age ( Figures S3-S6). Additional adjustment for cardiometabolic risk factors did not materially alter the relationship (model 3; Table S4). Neither were results significantly different in a range of sensitivity analyses (Table S5; Figure S7). Further, the associations were comparable between studies with baseline mean age younger than 60 and 60 years and older ( Figure S8) and before and after exclusion of the first 3, 5, or 10 years (median 10 years; Figure S2a; Table S3).
As with all-cause dementia, the association between BMI and nonvascular dementia risk was non-linear. Relative to lower-normal BMI, individuals categorized as underweight were at approximately one-quarter increased dementia risk (1.28 [1.06-1.54]; Figure 1; Table S4). Adjustment for cardiometabolic risk factors did not materially affect the association (Table S4). The findings were similarly robust when restricting the analysis to studies that used mea-  Table S3).
Both individuals with underweight or obesity were at increased vascular dementia risk. Compared with the referent group, those who were underweight had an approximate 80% greater dementia risk, and for those in the obese category, the risk was approximately 50% higher ( Figure 1) Table S4). Sensitivity analyses indicated findings compatible with the main result ( Figure S9). There was no evidence of a sex difference in the association between BMI and vascular dementia risk ( Figure 2). The association was broadly similar in studies with baseline mean age younger than 60 and 60 years and older ( Figure S8). Excluding the first few years of follow-up (median 7.5 years) did not materially influence the relationship ( Figure S2c;  Figure 3). Similar results were obtained when adjusted for age or after adjustment for cardiometabolic risk factors (Figures S10-S13; Table S6). The estimates tended to be larger for studies that used death records to ascertain dementia status rather than those that used medical examination ( Figure S14). Data from clinical trials produced similar results to those from nontrial populations ( Figure S15). The estimate of effect was more pronounced for studies with baseline mean age younger than 60 than 60 years and older especially at higher WC categories ( Figure S16).
Pooling data from ten studies (5319 nonvascular dementia cases) indicated that, compared with the lowest fifth, all WC categories were associated with lower nonvascular dementia risk ( Figure 3; Table S6).
F I G U R E 2 Associations between body mass index (BMI) and incident fatal and nonfatal dementia subtypes by sex. Hazard ratios (HRs) and 95% confidence intervals (CIs) adjusted for age, smoking, and education or socio-economic status Adjustment for cardiometabolic risk factors did not alter the association (Table S6) Table S6). No sex difference was evident (Figure 4). Data from clinical trial participants contained too few cases to draw meaningful conclusions ( Figure S15). Only the highest WC category in studies with baseline mean age 60 years and older was associated with increased vascular dementia risk ( Figure S16).  Table 2). In contrast, greater than or equal to 0.5% annual weight gain was not associated with dementia risk (1.00 [0.89-1.12]; Table 2). The results remained unchanged for models 1 and 3, and when studies were stratified by baseline mean age (Figures S17-S19; Table 2). For F I G U R E 3 Association between waist circumference and incident fatal and nonfatal dementia and its major subtypes. Hazard ratios (HRs) and 95% confidence intervals (CIs) adjusted for age, smoking, and education or socio-economic status greater than or equal to 0.5% annual weight loss, the estimate of effect was more pronounced for studies that ascertained dementia by medical examination than studies that used death records ( Figure S20).
As with all-cause dementia, greater than or equal to 0.5% annual weight loss was associated with higher nonvascular dementia risk ( loss was not associated with increased vascular dementia risk ( Alzheimer disease, suggesting that decreasing BMI could be a consequence, rather than a risk factor, of dementia. 43 An additional, albeit more speculative, explanation may be pathophysiological such as weight loss-induced cortical thinning, as cerebral atrophy is a characteristic of dementia. In a cohort of healthy elderly individuals, faster cognitive decline and accelerated atrophy rate were observed in those with relative weight loss greater than or equal to 5% (equivalent to greater than or equal to 1.2% annual loss) compared with those with relative weight loss less than 5% (equivalent to less than 1.2% annual loss). 44 Similarly, a Norwegian study that assessed percent change in BMI in midlife reported that while greater than or equal to 5% loss (equivalent to approximately greater than or equal to 0.6% annual loss) was associated with increased risk of dementia-related mortality, a gain of greater than or equal to 20% (equivalent to approximately greater than or equal to 2.2% annual gain) was associated with reduced risk. 45 Conversely, for vascular dementia, weight gain was associated with a modest 20% increased risk but only in those aged younger than 60 years at study baseline. This is consistent not only with what we know about weight gain being a risk factor for other vascular conditions such as coronary heart disease 46,47 but also with the diminution of the strength in the association between vascular risk factors such as diabetes and blood pressure with vascular risk at older ages. 48 Moreover, data from animal studies have indicated that weight gain is associated with increased vascular dementia risk. 49 For all-cause dementia, and its major subtype, that of nonvascular dementia, there was no evidence that carrying excess body weight (either in general or more centrally) conferred increased risk. Rather, individuals with a BMI of greater than or equal to 22.5 kg/m 2 (and higher WCs) had a slightly lower risk of dementia in later life. Conversely, and in agreement with some previous findings, 6