The Joint Effect of Body Mass Index and Serum Lipid Levels on Incident Dementia among Community-Dwelling Older Adults

Abstract

Objectives

This study aimed to explore the joint effect of body mass index (BMI) and serum lipids levels on incident dementia.

Methods

We prospectively followed up with 1,627 dementia-free community residents aged ≥60 for 5.7 years on average. At baseline, weight, and height were measured, and total cholesterol (TC), triglyceride (TG), low-density lipoprotein cholesterol (LDL-C), and high-density lipoprotein cholesterol (HDL-C) were detected in serum. Demographic characteristics were collected through questionnaires. Dementia was based on consensus diagnosis of neurologists and neuropsychologists using DSM-IV criteria. Additive Cox proportional model was used to assess the exposure-response relationship between BMI and serum lipid levels and dementia risk. Interactions and further classifications of BMI and serum lipid levels were further presented by bivariate surface models and decision-tree models.

Results

The joint effects of TC with BMI, TG with BMI, and LDL-C with BMI on the risk of incident dementia shared a similar pattern, different from their independent exposure-response curves. The joint effect of HDL-C with BMI showed an S-surface but without statistical significance. Participants with TC<5.4 mmol/L and BMI<21 kg/m² (Hazard Ratio(HR) 1.93, 95% Confidence Interval (CI) 1.05-3.53), TC<5.4 mmol/L and BMI≥21 kg/m² (HR 1.73, 95% CI 1.09-2.72), and TC≥5.4 mmol/L and BMI<21 kg/m² (HR 4.02, 95% CI 2.10-7.71) were identified to have the increased risk of incident dementia compared to those with TC≥5.4 mmol/L and BMI≥21 kg/ m². Participants with TG<1.7 mmol/L and BMI<21 kg/m² had an increased risk of incident dementia compared to those with TG≥1.7 mmol/L and BMI≥21 kg/m² (HR 1.98, 95%CI 1.17-3.3). Participants with LDL-C≥3.3 mmol/L and BMI<21 kg/m² were identified to have an increased risk of incident dementia compared to those with LDL-C≥3.3 mmol/L and BMI≥21 kg/m² (HR 3.33, 95%CI 1.64-6.78).

Conclusions

Our study showed that low BMI combined with low or high levels of serum lipids may increase the risk of dementia among older adults. This finding suggests the potential impacts of these two metabolic indexes on the risk of dementia.

Introduction

The prevalence of dementia is rapidly increasing along with an aging global population. The number of people with dementia is predicted to reach 82 million in 2030 and 152 million in 2050 globally (1), which will pose a great burden on society and economics. Although age is well recognized as the determinant for cognitive decline, dementia is not in a natural stage or an inevitable consequence of aging. Therefore, early screening of dementia risk is needed to promote timely preventive strategies.

The individual impacts of BMI and cholesterol with incident dementia have been investigated vastly in previous studies (2, 3). The pathogenesis of Alzheimer's Disease (AD) and coronary heart diseases mediated by the apolipoprotein E (ApoE) gene through BMI and lipids suggested a combination of BMI and serum lipids may better predict dementia incidence than BMI or lipids alone (4). In this study, we hypothesized that there might be a potential correlation between BMI and lipoproteins and the nonlinear relationship of these two metabolic indexes. We used additive Cox proportional model, a non-parametric modeling technique, to investigate the potential association between these variables with new-onset dementia in an urban community-based older adults cohort.

Research methodology

Study design and participants

The target population was recruited from the 11 neighborhoods in the Jing'An District of Shanghai between 2010 and 2012. Eligible participants were: (1) registered residents in this community; (2) ≥60 years; (3) without schizophrenia or mental retardation based on their medical records; or (4) able to communicate and conduct physical and cognitive examinations; (5) non-dementia based on the Diagnostic and Statistical Manual Disorder, 4th edition (DSM-IV) (5). Participants were not eligible if any of them (1) lived in nursing homes or other institutions; (2) were afflicted with a DSM-IV diagnosis of mental retardation or severe schizophrenia; or (3) had severe impairment of hearing, vision, or verbal and could not accomplish the neuropsychological evaluation. A detailed recruitment procedure has been published elsewhere (6, 7).

Measurements at baseline

At baseline, height and weight were measured. Body mass index (BMI) was calculated by dividing the weight in kilograms (kg) by height in meters (m) square. A 2ml blood sample was collected from each participant by a research nurse in the morning after 12 hours of overnight fasting. Serum lipids profiles (TC, TG, LDL-C, and HDL-C) were recorded using the Hitachi 7600 analyzer (Hitachi, Tokyo, Japan) at the central laboratory in Huashan Hospital (7).

Participants were interviewed via a questionnaire for their demographic characteristics and lifestyle, including age, sex, formal education year, cigarette smoking, and alcohol consumption. A current smoker was defined as a person who had smoked daily within the past month. Alcohol consumption was determined if the participants had at least one serving of alcohol weekly during the past year. Medical conditions such as type 2 diabetes mellitus (T2DM), hypertension, and stroke were asked for and verified by their medical records. The Center for Epidemiologic Studies Depression Scale (CES-D) was administered to assess depressive symptoms if CES-D ≥16 (6). Items from the Lawton and Brody Activity of Daily Living (ADL) scale-16 were used to evaluate physical self-maintenance and instrumental activities of daily living, such as eating, using the telephone, preparing meals, handling money, and completing chores. Participants were considered functionally intact if the ADL score was>16 (8). DNA was extracted from the blood or saliva samples, and ApoE genotyping was assayed through the TaqMan SNP method (6). ApoE ε4 positivity was defined if there was at least one ε4 allele.

Neurological and neuropsychological assessment and diagnosis at baseline

At baseline, cognitive functions including global cognition, executive function, spatial construction function, attention, memory, and language, were evaluated through a battery of tests: (1) the Mini-Mental State Examination (MMSE) for the global cognition, (2) the Conflicting Instructions Task (Go/ No Go Task), (3) the Stick Test, (4) the Modified Common Objects Sorting Test, (5) the Auditory Verbal Learning Test, (6) the Modified Fuld Object Memory Evaluation, (7) the Trail-making Test A&B, and (8) the Renminbi (the official currency of China) Test, translated from the EURO test. Based on education level, neuropsychologists provided different tests to the participants: tests 1 to 5 and 7 were given for those with at least six years of education, whereas for the others, tests 1 to 4, 6, and 8 were performed. More administration details have been reported elsewhere (6, 7).

The neurologists examined the participants' motor responses, reflexes, mood, and ability to perform daily activities (6). After each clinical assessment, neurologists and neuropsychologists reviewed each participant's functional, medical, neurological, psychiatric, and neuropsychological data. They reached a consensus regarding the presence or absence of dementia using the criteria from DSM-IV.

Follow-up procedure

One thousand six hundred twenty-seven participants without dementia at baseline were scheduled for follow-up from 2014 to 2016. The same neurological examinations and neuropsychological tests at baseline were used to evaluate participants' cognitive function during the follow-up. The same expert panel made a consensus diagnosis of the incidence of all-type dementia with the same criteria used at the baseline.

Statistical analysis

Participants who were followed for less than two years were excluded from the data analysis. Demographic characteristics were described using median and interquartile range (IQR) or mean and standard deviance (SD) for continuous variables and counts and proportion (%) for categorical variables. They were compared using a t-test or the Wilcoxon signed-rank test for continuous variables, and the Chi-squared test for categorical variables.

Independent exposure-response models

The independent exposure-response relationships between BMI or serum lipids were assessed by additive Cox proportional model. We used it to explore the separate nonlinear relationship between BMI/serum lipids levels and incident dementia. This model used the nonparametric smoothing functions to explore the linear and nonlinear association between explanatory and outcome variables. Because TC, TG, LDL-C, and HDL-C were correlated (Spearman's rank correlation tests were statistically significant), separate models were built by including either TC/TG/LDL-C/HDL-C to address the concern of collinearity. A typical form of a model with BMI as a covariate is as follows:

$$\begin{array}{l}\text{h(follow-up year,dementia)}\hfill \\ =h_0(follow-upyear)\times exp(s(TC)+s(BMI)+sex+age+educationyear\hfill \\ +ApoE\epsilon 4+depression+hypertension+diabetes\hfill \\ +stroke+baselineMMSE+smoke+alcoholconsumption)\hfill \end{array}$$ (1)

Where h(follow-up year, dementia) denotes the expected risk of dementia incidence. s(TC) and s(BMI) denote the thin plate smoothing functions of TC and BMI, respectively. The smoothness selection of additive Cox model was made by restricted maximum likelihood (REML) (9). Models were also developed for TG/LDL-C/HDL-C by replacing s(TC) with s(TG)/ s(LDL-C)/ s(HDL-C) in Model (1) in order to assess all the effects of different lipid levels.

Bivariate surface model

Previous studies have suggested the potential association between BMI and serum lipids with dementia (4, 10, 11). We assessed the interaction between BMI and TC/TG/LDL-C/ HDL-C in the subsequent analyses. A bivariate surface model(Model 2) under the same Model (1) assumption was built. Its typical form is as follows:

$$\begin{array}{l}=h_0(follow-upyear,dementia)\hfill \\ =h_0(follow-upyear)\times exp(s(TC,BMI)+sex+age+educationyear\hfill \\ +ApoE\epsilon 4+depression+hypertension+diabetes\hfill \\ +stroke+baselineMMSE+smoke+alcoholconsumption)\hfill \end{array}$$ (2)

Where s(TC, BMI) represents a bivariate spline smoothing function of TC and BMI. The degrees of freedom of smoothing functions for TC and BMI was the same in Model (2). We then drew the bivariate surface plots derived from Model (2) to examine the interaction between TC and BMI. Models were also developed for TG/LDL-C/HDL-C by replacing s(TC, BMI) with s(TG, BMI)/ s(LDL-C, BMI)/ s(HDL-C, BMI) in Model (2) in order to assess all the interactive effects of BMI and different lipids.

Cox proportional hazard model based on CART

The interactions between BMI and cholesterol profiles were statistically significant. Thus, we examined the relationship further. We adopted a decision-tree model (classification and regression tree, CART) to assess the interaction between BMI and blood cholesterols since the bivariate response model could not provide the effect estimates and their statistical significance. Using the recursive partitioning approach, CART was performed to generate predictive rules and optimal cut-off values for BMI and serum lipids for discriminating between individuals who do and do not develop dementia at follow-up. Multivariable analyses using Cox proportional hazard model to estimate the hazard ratio (HR) of different combinations of BMI and TC/TG/LDL-C/HDL-C based on CART. All the p values and 95% Confidence Intervals were calculated in two-tailed tests. The statistical analyses were performed using the mgcv package in R software (version 4.1.0).

Results

Baseline characteristics of the study population

The baseline characteristics of the 1627 participants, classified by cognition status (dementia-free /all-cause dementia) at follow-up, were shown in Table 1. They had a median age of 71.0, a mean BMI of 24.62±3.51 kg/m², and 45.9% were male. The means of TC, TG, LDL-C, and HDL-C were 5.41±1.01,1.74±1.03, 3.35±0.88, and 1.33±0.33 mmol/L respectively. Over half of the participants had hypertension, and the prevalence rates of other conditions (stroke, T2DM, and depression) were all less than 15%. Compared with dementia-free participants, those with new-onset dementia were older, had a lower BMI, were more likely to have hypertension and stroke, had lower education years, and had lower MMSE scores at baseline.

Table 1.

Baseline characteristics for the cohort and individuals with and without incident dementia

	Overall (n=1,627)	Non-dementia (n=1,491)	Incident dementia (n=136)	P-value
Age, year, median [IQR]	71.0 [65.0, 77.0]	70 [64.0, 76.0]	79 [75.8, 83.0]	<0.001*
Male, n (%)	747 (45.9%)	697 (46.7)	50 (36.8)	0.03184*
Body Mass Index, kg/m2, mean (SD)	24.62 (3.51)	24.68 (3.46)	23.96 (4.01)	0.0423*
Total cholesterol, mmol/L, mean (SD)	5.41 (1.01)	5.42 (1.03)	5.29 (0.85)	0.0812
TG, mmol/L, mean (SD)	1.74 (1.03)	1.76 (1.04)	1.62 (0.90)	0.1016
Low density lipoprotein, mmol/L, mean (SD)	3.35 (0.88)	3.36 (0.89)	3.24 (0.76)	0.09273
High density lipoprotein, mmol/L, mean (SD)	1.33 (0.33)	1.33 (0.33)	1.36 (0.35)	0.4452
History of stroke, n (%)	209 (12.8%)	174 (11.7)	35 (25.7)	<0.001*
History of hypertension, n (%)	872 (53.6%)	786 (52.7)	86 (62.3)	0.02352*
History of T2DM, n (%)	216 (13.3%)	196 (13.1)	20 (14.7)	0.7029
History of depression, n (%)	238 (14.6%)	210 (14.1)	28 (20.6)	0.05387
Current smoker, n (%)	165 (10.1%)	152 (10.2)	13 (9.6)	0.9309
Alcohol assumption, n (%)	147 (9%)	143 (9.6)	4 (2.9)	0.01496*
ApoE ε4 positive, n (%)	275 (16.9%)	246 (16.5)	29 (21.3)	0.1876
Education, years, median [IQR]	12.0 [9, 15]	12.0 [9, 15]	9.0 [6, 12]	<0.001*
Total MMSE score, median [IQR]	29 [28, 30]	29 [28, 30]	26 [14, 28]	<0.001*

IQR, Interquartile Range; SD, standard deviation; * p<0.05.

Independent exposure-response models

Figure 1 shows the predicted log(hazard ratio of incidence dementia) in the y-axis and continuous variable in the x-axis using spline function on BMI (Figure 1A, #1) and TC (Figure 1A, #2) BMI (Figure 1B, #1) and TG (Figure 1B, #2) BMI (Figure 1C, #1) and LDL-C (Figure 1C, #2) BMI (Figure 1D, #1) and HDL-C (Figure 1D, #2) separately. For instance, a log(hazard ratio) of 0 indicates no impact on the risk of incident dementia, while a log (hazard ratio) of 1 shows a 2.7 times higher risk of dementia incidence. The dose-response curves of the four BMI models revealed a stable L-shape. As BMI increased from 15–23 kg/m², the risk of dementia incidence decreased but remained relatively constant as BMI continued to increase.

Figure 1.

Exposure-response curves of independent effects of BMI (kg/m²) and TC/TG/LDL-C/HDL-C (mmol/L) using spline function, respectively

(A): the smoothing function of BMI and TC based on Model(1); (B): the smoothing function of BMI and TG based on Model (1); (C): the smoothing function of BMI and LDL-C based on Model (1); (D): the smoothing function of BMI and HDL-C based on Model (1). BMI, body mass index; TC, total cholesterol; TG, triglyceride; LDL-C, low-density lipoprotein cholesterol; HDL-C, high-density lipoprotein cholesterol.

However, the risk of incident dementia varied as serum lipid levels changed, as shown by the dose-response curves of four lipid levels. The confidence intervals of the linear predictor of TC (Figure 1A, #2), TG (Figure 1B, #2), and LDL-C (Figure 1C, #2) always included zero, suggesting no statistical significance for these three smoothed variables in their models. In contrast, the predictors of HDL-C (Figure 1D, #2) were larger at around 2.5 (mmol/L) and had statistical significance.

Bivariate surface model

Figure 2 displayed the exposure-response surface plots for the joint effects of baseline BMI and TC (Figure 2A)/ TG (Figure 2B)/LDL-C (Figure 2C)/HDL-C (Figure 2D) on dementia incidence. In general, we found a U-shape relationship between BMI and incident dementia risk across TC (Figure 2A), TG (Figure 2B), and LDL-C (Figure 2C). The bottoms of estimated dementia incident risk were observed at moderate BMI or relatively high BMI with high TC/TG/LDL-C. As for HDL-C (Figure 2D), the exposure-response surface plots demonstrated an S-shape relation between BMI and incident dementia risk. However, it is of note that HDL-C is not comparable with the other three lipids.

Figure 2.

The predicted dose-response surface plot of the joint effect of BMI (kg/m²) and TC/TG/LDL-C/HDL-C (mmol/L) on incident dementia

A): the bivariate spline smoothing function of TC and BMI based on Mode l(2); (B): the bivariate spline smoothing function of TG and BMI based on Model (2); (C): the bivariate spline smoothing function of LDL-C and BMI based on Model (2); (D): the bivariate spline smoothing function of HDL-C and BMI based on Model (2). BMI, body mass index; TC, total cholesterol; TG, triglyceride; LDL-C, low-density lipoprotein cholesterol; HDL-C, high-density lipoprotein cholesterol.

The joint effects of TC with BMI, TG with BMI, and LDL-C with BMI on the risk of incident dementia (Figure Figure 2, Figure 2) shared a similar pattern. The pattern was different from their independent exposure-response curves. The bottom of dementia risk was at moderate BMI (22.9-26.0 kg/m²) with the highest part of TC (>6.2 mmol/L)/ TG (>2.0 mmol/L) /LDL-C (>1.5 mmol/L).

The joint effect of HDL-C and BMI on the risk of incident dementia showed an S-surface with no statistical significance. The risk of dementia incidence remained almost unchanged with HDL-C while it decreased from the lowest BMI (BMI= 14.0 kg/m²) until the turning point of low and moderate BMI (BMI<25 kg/m²), increased until another turning (BMI=29 kg/ m²), and then decreased again in the high BMI range (BMI>29 kg/m²).

Cox proportional hazard model based on CART

Participants were categorized into four or six groups using CART models that included combinations of BMI, TC, TG, LDL-C, and HDL-C values. For the CART model with BMI and TC included, two nodes with a TC value of 5.4 and BMI of 21 divided participants into four groups. For the model with BMI and TG, two nodes with a TG value of 1.7 and BMI of 21 split participants into four groups. For the model with BMI and LDL-C, two nodes with an LDL-C value of 3.3 and BMI of 21 divided participants into four groups. Three nodes with an HDL-C value of 1.3, BMI of 21 and 29 split participants into six groups for the CART model, including BMI and HDL-C. The nodes of these indexes were approaching the recommended cut-off points of lipids used to help define atherosclerotic cardiovascular disease. According to 2016 Chinese guideline for the management of dyslipidemia in adults (12), TC<5.4 mmol/L is defined as a normal to borderline elevated cholesterol; TG <1.7 mmol/L or LDL-C <3.3 mmol/L is referred to an ideal to adequate lipid level; HDL-C<1.3 mmol/L partly indicate a low cholesterol concentration among adults. And BMI<21 kg/m² facilitates lipid control. According to Table 2, when adjusted for age, gender, years of education, ApoE ε4, medical history of stroke, hypertension, T2DM, depression, the behavior of cigarette smoking and alcohol drinking, and baseline MMSE score, group of TC<5.4 mmol/L with BMI<21 kg/m² (HR 1.93, 95%CI 1.05-3.53), TC<5.4 mmol/L with BMI≥21 kg/m² (HR 1.73, 95%CI 1.09-2.72), and TC≥5.4 mmol/L with BMI<21 kg/m² (HR 4.02, 95%CI 2.10-7.71) were identified to be associated with increased risk of dementia incidence compared to the group of TC≥5.4 mmol/L with BMI≥21 kg/m². The group of TG<1.7 mmol/L with BMI<21 kg/m² demonstrated an increased risk of dementia incidence compared with the group of TG≥1.7 mmol/L with BMI≥21 kg/m² (HR 1.98, 95%CI 1.17-3.3). Group of LDL-C≥3.3 mmol/L with BMI< 21 kg/m² indicated an increased risk of dementia incidence compared with the Group of LDL-C≥3.3 mmol/L with BMI≥21 kg/m² (HR 3.33, 95%CI 1.64-6.78).

Table 2.

Hazard ratios of incident dementia associated with BMI and TC/TG/LDL-C/HDL-C values categorizations

BMI and TC/TG/LDL-C/HDL-C Categorizations	HR	95%CI
TC< 5.4, BMI< 21	1.93	(1.05, 3.53)*
TC≥5.4, BMI< 21	4.02	(2.10, 7.71)*
TC< 5.4, BMI≥21	1.73	(1.09, 2.72)*
TC≥5.4, BMI≥21	1.00 (reference)	-
TG< 1.7, BMI< 21	1.98	(1.17,3.36)*
TG>=1.7, BMI< 21	1.36	(0.72, 2.56)
TG< 1.7, BMI≥21	0.92	(0.61, 1.38)
TG≥1.7, BMI≥21	1.00(reference)
LDL-C< 3.3, BMI<21	1.73	(1.00, 3.01)
LDL-CL≥3.3, BMI<21	3.33	(1.64, 6.78)*
LDL-C< 3.3, BMI≥21	1.29	(0.85, 1.97)
LDL-C≥3.3, BMI≥21	1.00(reference)	-
HDL<1.3, BMI<21	2.87	(0.86, 9.62)
HDL≥1.3, BMI<21	1.33	(0.54, 3.31)
HDL<1.3, 21≤BMI<29	0.93	(0.39, 2.21)
HDL≥1.3, 21≤BMI<29	0.74	(0.31, 1.77)
HDL<1.3, BMI>29	1.08	(0.34, 3.38)
HDL≥1.3, BMI>29	1.00(reference)	-

BMI, body mass index, kg/m²; TC, total cholesterol; TG, triglyceride; LDL-C, low-density lipoprotein cholesterol; HDL-C, high-density lipoprotein cholesterol, mmol/L; * with statistical significance.

Discussion

This study aimed to comprehensively investigate the linear, nonlinear, and joint effects of BMI and lipoproteins on dementia risk.

The bivariate surface models indicated that the relationship between BMI and dementia risk remained a U-shape when considering the joint effects of TC/TG/LDL-C. Cox models based on CART showed that low BMI with low or high TC, low BMI with low TG, and low BMI with high LDL-C were significantly associated with increased dementia incidence. However, the present study did not find any association between dementia incidence and the combination of BMI and HDL-C. These BMI-related associations contrasted with some typical expectations of the same directions of these effects. Many studies suggested that being overweight or obese was a risk factor for AD (13, 14, 15, 16, 17). Because being obese increases the risk for vascular dysfunctions, which may be an initial trigger state leading to AD and vascular forms of dementia (18).

However, it is commonly overlooked that the effects of being overweight on brain health might be independent of vascular effects due to adipocyte hormones and cytokines. Lower levels of BMI may not be advantageous during the late-life (19) because BMI represents not only adipose tissue but also skeletal muscles, which decreases with aging due to sarcopenia, cachexia, and starvation (20). One of the first studies to report a decline in body weight preceding a dementia diagnosis was the Ranchoo Bernardo study (21). Additionally, among some individuals, weight loss may be a potential preclinical marker for AD, particularly when measured 6–10 years before a clinical diagnosis. A study containing four longitudinal cohorts also indicated that a lower BMI might be a marker of increased risk of AD by a meditation analysis (4).

Our study's findings on serum lipids partly align with previous studies. A cohort study of nearly 4,000 participants conducted by Helzner et al. shows that higher pre-diagnosis TC and LDL-C concentration is associated with more rapid cognitive decline in AD patients (22). In a mendelian randomization study, Østergaard and colleagues observed that higher TC and LDL-C were associated with increased AD risk (23), which was also in line with our study. But they also reported that higher HDL-C was associated with reduced incident dementia risk, while ours was not significant. Interestingly, after variants in the ApoE locus were excluded from the analysis, the cholesterol levels were no longer significantly associated with AD risk (23). As for TG, our results share similarities with previous studies conducted in China (24, 25) and are supported by accumulating evidence (26, 27). Low TG has also been shown to be a serological marker of frailty (28). Our findings are similar to the results of a Netherland longitudinal study with community-dwelling older individuals aged 70–78 years at baseline published in 2022 (11). This study observed that interaction existed between late-life low BMI and low non-HDL cholesterols over a median follow-up of 10.3 years.

When considering these variations from various studies, it's important to account for differences in tastes, diets, and perception due to ethnicity. Additionally, lipids levels might be influenced by the timing of lipids measurement and ApoE genotype should be considered in different pathological stages of dementia (29). Potential mechanisms regarding adipose tissue and blood lipids may influence or interact with the brain and dementia risk during the dynamic period of life (30). Firstly, changes in BMI may be biomarkers for changes in energy metabolism that may influence the risk of AD, progression, and ultimately death. The hypothalamus regulates energy homeostasis by controlling hunger and satiety, regulating energy expenditure, and releasing hormones that increase the use of energy stores (31). Secondly, many studies have found that high cholesterol levels in the brain play an essential role in the process of beta-amyloid (Aβ)-induced AD (32). In addition to high cholesterol in the brain, animal and human experiments have reported a link between serum cholesterol and Aβ deposition. For example, amyloid precursor protein (APP) transgenic mice fed with a high-cholesterol diet and hypercholesterolemic rabbits had an accumulation of intracellular immunolabeled Aβ proteins in the brain (33, 34). Previous research also indicated that the senile plaque theories might provide a link between high LDL-C and AD (35). In this theory, elevated levels of LDL-C and TC cause the extracellular deposition of Aß proteins, hindering neuronal synaptic connections in the brain and increasing the risk of AD (36). Indeed, various studies have demonstrated that high concentrations of LDL-C cholesterol are associated with coronary heart disease and carotid artery atherosclerosis, which, in turn, may lead to cognitive decline through cerebral embolism or hypoperfusion (37, 38, 39, 40). Third, cholesterol and BMI are typical physical phenotypes linked with AD through common genome-wide association study (GWAS) signals (41).

A typical example is the role of the ε2/ ε3/ ε4 polymorphism of the ApoE gene in these traits (42, 43, 44, 45, 46). This polymorphism has been strongly associated with serum lipids and BMI (44, 45). The main function of the ApoE gene is also associated with lipolysis and clearance of lipids through the bloodstream and additional functions in adipocytes (47, 48). However, despite more than 40 years of research, mechanistic pathways, and biological mechanisms which can underline causal connections between the ε2 and ε4 alleles, lipids, BMI, and AD remain poorly understood. The progress is tempered partly by a crucial role of heterogeneity in genetic predisposition to non-Mendelian diseases and traits (49).

In this study, additive Cox proportional model was employed to explore whether there were interactive effects between BMI and serum lipids levels on dementia incidence. The most important advantage of additive Cox proportional model is that it allows nonparametric smoothing functions to account for potential nonlinear effects of some confounding factors on the dependent variable. Rather than focusing on individual factors of BMI or lipid profiles, this study provides novel evidence on the joint effects of these two cardiometabolic parameters on the susceptibility of all-cause dementia in late life. By adding a bivariate smoothing function, the models may avoid the artificial definition of exposure categories and the strong assumption of linearity (50). CART was used to identify the cut-offs of input variables that could be administered quickly and maintained high levels of sensitivity and specificity for the classification of dementia. And it is often able to uncover complex interactions between predictors that may be difficult or impossible to uncover using traditional multivariate techniques (51). This method also has a high sensitivity value among data mining methods in the predictions of dementia (52).

However, there are some limitations to our study. First, blood cholesterol levels and BMI are both dynamic phenotypes that change over time. BMI is not a perfect measure of adiposity. It is affected by factors such as ethnicity and age - the ratio of fat-free mass to height begins to decrease after 45 years old (53). Better measures of adiposity and its effects are needed. For example, the concentration of an adipocyte hormone in the blood. Second, it should be noted that the results of our study are based on our specific dataset and a single set of tuning parameters on CART. It is well known that for neural networks and machine learning, the performance of the classifier and the properties of the resulting predictions are heavily dependent on the chosen values for the tuning parameters (54, 55, 56). Third, due to the limited follow-up time, we cannot determine the causal relationship between metabolic factors and dementia. Some researchers also suggested that reverse-causation effect makes a higher BMI appear protective, and a lower BMI seems to be a risk, especially in late age (57, 58). We excluded participants with a follow-up time of less than two years to ensure enough lighten the reverse-causation impact. Fourth, the finding of our study has limited generalizability. The participants in our research population were a selected group of older adults. The average BMI was much lower than that of older adults in the US (59). And the prevalence of comorbidities (T2DM and depression) was also lower than that in the US (60, 61). Additionally, it was possible that participants were smoking significantly without meeting our criteria of being current smokers. This might impact the relevance of any clinical implications for the majority of older adults in many other countries. Lastly, there existed unmeasured confounders that could influence cognitive function. For example, there was insufficient data to examine other cholesterol sub-fractions or variations in ApoE gene interactions. Also, we found several known dementia risk factors, including more advanced age, shorter length of education, and stroke history, to be positive in the participants who developed dementia. However, the impact of these on the development of dementia in subjects, compared to the impacts of low BMI and lipids as the main contributing factors, cannot be determined.

Conclusions

Our study showed that low BMI combined with high or low TC/ low TG/ high LDL-C increased the risk of dementia incidence among older adults in urban Shanghai. The results of the present study indicate the joint effects of BMI and serum lipids may predict the risk of dementia incidence among the dementia-free older population. With these available and accessible measurements, prognostic factors may be established for clinical decisions, including continuous follow-up and additional adjunctive prevention, to improve the cognition status of community-dwelling older adults. This study suggested that low BMI and dyslipidemia might be a combined risk factor for cognitive function for the older population. Older adults might need to improve their diet and get involved in physical activity to protect their cognitive function. Moreover, this finding suggests that TC/TG/LDL-C may serve as a valuable predictor for the cognition status of the older population.

The interaction of different metabolic measurements should be taken into consideration in future hypotheses of dementia studies. Further population and experimental studies are warranted to reveal the mystery of metabolic effects on cognition decline. Continued identification of adiposity-associated biomarkers that are more sensitive than BMI or serum lipids in indicating the amount of adipose in the body or the blood is needed to determine the effect of this critical issue.

Funding:

The study was funded by the Shanghai Municipal Science and Technology Major Project (2018SHZDZX01) and ZJ LAB, National Natural Science Foundation of China (82173599), Shanghai Municipal Health Commission (2020YJZX0101), Key Project of the Ministry of Science and Technology, China (2020YFC2005003, 2021YFE0111800).

Data Availability Statement:

The data supporting the findings of this study are not publicly available due to privacy or ethical restrictions. The data are available on request from the corresponding authors. Additional institutional approvals, such as ethics approval, would be required to enable sharing of these data. Requests to access the datasets should be directed to dingding@huashan.org.cn.

Conflict of Interest:

The authors declare that the research was conducted without any commercial or financial relationships that could be construed as a potential conflict of interest.

Ethics Statement:

This study was reviewed and approved by the Medical Ethics Committee of Huashan Hospital, Fudan University, Shanghai, China (No.2009-195). Written informed consent was obtained from all participants or their legal guardians.

Author Contributions:

Qiqi Lei reviewed the literature, performed statistical analyses, and wrote the first draft of the manuscript. Zhenxu Xiao provided clinical advice. WW collected data and provided statistical advice. Xiaoniu Liang curated database. Qianhua Zhao conceived the study and performed clinical diagnoses. Ding Ding conceived the study and provided overall supervision. Wei Deng provided statistical advice and provided general supervision. All authors interpreted the results and contributed to the final version of the manuscript. All authors have approved the final manuscript.