Metabolic Syndrome and Cognitive Decline Among the Oldest Old in Okinawa: In Search of a Mechanism. The KOCOA Project

Abstract

The study aim was to test whether the metabolic syndrome or its components predicted cognitive decline among persons aged 80 years and older (mean 85.0 years). Participants were members of the “Keys to Optimal Cognitive Aging Project,” a prospective cohort study in Okinawa, Japan. Metabolic syndrome was assessed at baseline. Cognitive functions were assessed annually for up to 3 years. One hundred and forty-eight participants completed at least one follow-up with 101 participating in all three assessments and 47 participating in two of the three assessments. The mean and median duration of follow-up were 1.8 and 2 years, respectively. Metabolic syndrome and four components were not associated with decline in global and executive cognitive functions. However, high glycosylated hemoglobin was associated with decline in memory function at the second follow-up. Our study supports accumulating evidence that the positive association between metabolic syndrome and cognitive function might not hold for the oldest old.

The number of older adults with cognitive impairment or dementia is expected to rise dramatically due to an aging population (1). It is critical, therefore, to identify modifiable risk factors for cognitive decline. It is widely acknowledged that risk factors for cardiovascular disease play a role in the development of dementia, both Alzheimer’s disease (AD) and vascular dementia (VaD) (2–12). Although several studies reported that metabolic syndrome, defined as a cluster of cardiovascular disease risk factors, including abdominal obesity (7,11), hypertriglyceridemia (6), low high-density lipoprotein cholesterol (HDL-C) level (6), hypertension (2–5,8), and hyperglycemia (4,8,11), has been associated with increased risk of cognitive impairment, the association appears to depend on age. Metabolic syndrome usually increases risk of cognitive decline among younger old adults (ages 65–74 years), but as the age of study participants increases further, the association begins to diminish (6,10,11). Among those in their mid-80s and beyond, often referred to as the oldest old, the association can disappear or reverse (13,14).

Similarly, several components used to define metabolic syndrome have paradoxical effects in the oldest old. For example, at middle age, obesity is generally associated with increased risk of later life dementia (7), yet in the oldest old, both abdominal obesity and higher body mass index (BMI) have been associated with a reduced risk of later life dementia (15,16). Hyperlipidemia is associated with cognitive impairment among younger to mid-older persons (9). In the oldest old, however, at least one study did not support this association (17). Hypertension at middle age is also a risk factor for cognitive decline and dementia in later life (3–5,8). On the other hand, among the oldest old, low blood pressure rather than high blood pressure is found to be a risk factor for the development of dementia (18,19). In sum, the protective effects of lower triglycerides, blood pressure, and BMI in midlife to early old age against later life cognitive decline appear to be less robust or even reverse in the oldest old populations. The reasons are unclear. The paradoxical change in metabolic syndrome’s association with several age-related diseases and, in some studies, with overall mortality has been referred to as “reverse metabolic syndrome” and has been hypothesized to represent an evolutionary adaptation to prolong human life after reproduction (20). Interestingly, little is known about these relations in long-lived Japanese populations who have traditionally had more VaD and less Alzheimer’s dementia than Western populations (21) and thus might be more susceptible to metabolic syndrome-related cognitive effects.

Therefore, the aim of the current study was to prospectively explore the association between the metabolic syndrome (and its components) and later life cognitive decline among functionally independent community-dwelling older adults aged 80 years and older and free from cognitive impairment at baseline in Japan’s Okinawa prefecture—a population known for their longevity (22). The Okinawans are of particular interest because their cardiometabolic risk factors have risen at the fastest pace in Japan over the past few decades (23) yet little is known about the effects of this on cognitive function.

METHODS

Participants

Participants were members of the Keys to Optimal Cognitive Aging Project (KOCOA), a prospective pilot cohort study of community-dwelling older people aged 80 years or older living in Ginowan City in Okinawa, Japan. The study was conducted annually between November 2007 and March 2010 and consisted of three time points (baseline and two follow-ups). A detailed description of the recruitment process has been presented elsewhere (24). Briefly, researchers visited 22 senior centers, explained the study protocol, and asked them to participate in the study. Senior centers, funded by the municipal government, offer various activities to the local seniors, which include handicrafts, games, traditional dance, and occasional lectures on topics of interest to the seniors. A request to join the study was made at the conclusion of each presentation. There were 196 volunteers, aged 80 years and older (mean age: 85.2 years, range: 80–98 years), who participated in the KOCOA study. The participants underwent a clinical examination at baseline and a face-to-face interview either in their local senior center or in their home at baseline and follow-up.

In the current study, we excluded those with frank dementia defined as Clinical Dementia Rating Scale (CDR) (25) greater than or equal to 1.0 at baseline. Attrition rates were 29.0% from baseline to the first follow-up and 19.7% from the first to the second follow-up (Figure 1). Of 193 nondemented participants, 148 completed at least one follow-up and were featured in the current analysis. The study was approved by the Ethics Committee of University of the Ryukyus, the Institutional Review Board at the Oregon Health & Science University, and informed consent was obtained from all participants prior to the enrollment in the study.

Figure 1.

Flow of participants.

Metabolic Syndrome

Presence of the metabolic syndrome at baseline was determined according to a minimally modified version of the National Cholesterol Education Program Third Adult Treatment Panel guidelines (26) as having three or more of the following conditions: BMI 25 kg/m² or more, triglyceride level 150 mg/dL or more, HDL-C level less than 40 mg/dL in men and less than 50 mg/dL in women, systolic blood pressure 130 mmHg or more and diastolic blood pressure 85 mmHg or more or currently taking antihypertensive medication, and glycosylated hemoglobin (HbA1c) level 5.4% or more or currently taking diabetes medication.

For the current study, BMI was used instead of waist circumference, which was not measured, with a cutoff value of 25 kg/m² or more according to guidelines published by the Japan Society for the Study of Obesity (27). Because fasting blood glucose level was not available, HbA1c level of 5.4% or more was used instead of fasting plasma glucose level 110 mg/dL or more (28,29).

Cognitive Tests

Cognitive function was assessed by a trained interviewer at baseline and at the follow-up examinations using the Japanese version of the Mini-Mental State Examination (JMMSE) (30,31), the Verbal Fluency Initial Letter (VFL), and the Scenery Picture Memory Test (SPMT) (32). The JMMSE is a measure of global cognitive function and ranges from 0 to 30 points, with higher scores representing better cognitive function. Because the distribution of JMMSE was not normal, we transformed using the square root of the number of errors. The VFL (33) was used to measure executive function. This test requires the participant to generate words beginning with letter “Ka” (in Japanese) in 60 seconds and has been used widely in Japan as an equivalent test to Initial Letter Fluency by Lezak (33). The SPMT is a measure of memory function and has been described in detail elsewhere (32).

Statistical Analysis

Comparisons of baseline characteristics of those with and without the metabolic syndrome were performed using t test for continuous variables and Pearson chi-square test for categorical variables. We used mixed-effects models to examine the association between the metabolic syndrome, its components, and cognitive test scores (outcome variables) over time. Mixed-effects models account for within-subject correlations among repeated cognitive measurements and accommodate missing data without removing the individual from the analysis (34). The mixed-effects model consists of two parts: fixed effects and random effects. Fixed effects describe the population average of baseline or trajectories, and random effects describe the participant-specific heterogeneity in those measures. The trajectory of the cognitive test scores over time may be nonlinear because of the stress of the test situation at baseline and/or the learning effect over time. Therefore, the models included two dummy variables representing follow-up time periods (first follow-up, second follow-up, with baseline as a reference) instead of continuous time as random effects. The interaction of the time dummies (treated as random effects) with the metabolic syndrome or its components status was included to represent the additional impact of the metabolic syndrome or its components status on the cognitive test scores at each time point. The detailed statistical model appears in the Appendix 1. We ran separate models for each of the three cognitive test scores (as an outcome) and six metabolic syndrome states, that is, metabolic syndrome itself and five components (as predictors) and their interactions with time. Statistical significance was set as p < .0083, the Bonferroni multiple comparisons–adjusted p value. For the JMMSE, we used the square root of the number of errors, as expressed by (30 − JMMSE)^1/2, to normalize the distribution (35,36). Thus, an increase in this transformed variable with time indicated cognitive decline (ie, increased numbers of errors). The normality of the transformed JMMSE score and other two test scores was confirmed by visual inspection of the distributions and Q-Q plots. All models were adjusted for age, gender, and years of education. All analyses were conducted with SAS 9.1 (SAS Institute Inc., Cary, NC).

RESULTS

We compared the participants followed (n = 148) and not followed (n = 45). The two groups did not differ in demographic characteristics (age: p = .155, years of education: p = .332 based on Student’s t test, and gender: p = .669 based on Pearson chi-square test), the metabolic syndrome and its components (metabolic syndrome: p = .596, BMI: p = .845, triglycerides: p = .863, HDL-C: p = .849, blood pressure: p = .296, HbA1c: p = .253 based on Pearson chi-square test), and the score of CDR (0 vs 0.5) and SPMT (CDR: p = .2318 based on Pearson chi-square test, SPMT: p = .153 based on Student’s t test). Those not followed had lower baseline scores of JMMSE and VFL (JMMSE: p < .001, VFL: p = .018).

The mean (SD) age of the participants at baseline was 85.0 (3.2), and 74.3% were women. A total of 49 participants (33.1%) met the metabolic syndrome criteria of whom 33 participants (67.3%) had 3, 11 participants (22.5%) had 4, and 5 participants (10.2%) met 5 of the criteria used to define metabolic syndrome. High blood pressure and high HbA1c were very common. Baseline JMMSE error score was similar between the two groups (those with and without metabolic syndrome), but baseline VFL score differed between the groups (Table 1).

Table 1.

Baseline Characteristics of the Analytic Cohort (n = 148)

		Metabolic Syndrome
Variables	Total, n = 148	Without, n = 99	With, n = 49	p Value*
Demographics
Age, M ± SD	85.0 ± 3.2	85.0 ± 3.1	84.9 ± 3.4	.927
Women, n (%)	110 (74.3)	73 (73.7)	37 (75.5)	.816
Years of education, M ± SD	7.4 ± 2.3	7.3 ± 2.2	7.4 ± 2.4	.834
Individual criterion used for metabolic syndrome, n (%)†
High BMI	55 (37.2)	18 (18.2)	37 (75.5)	<.001
High triglycerides	28 (18.9)	4 (4.0)	24 (49.0)	<.001
Low HDL-C	18 (12.2)	2 (2.0)	16 (32.7)	<.001
High blood pressure	95 (64.2)	50 (50.5)	45 (91.8)	<.001
High HbA1c	93 (62.8)	47 (47.5)	46 (93.9)	<.001
Cognitive test
JMMSE, M ± SD	23.9 ± 3.8	23.8 ± 3.2	24.1 ± 3.6	.683
VFL, M ± SD	6.4 ± 2.9	6.7 ± 2.6	5.7 ± 2.1	.035
SPMT, M ± SD	7.2 ± 2.8	7.4 ± 2.9	6.8 ± 2.5	.257

Notes: BMI = body mass index; HbA1c = glycosylated hemoglobin; HDL-C = high-density lipoprotein cholesterol; JMMSE = Japanese version of Mini-Mental State Examination; SPMT = Scenery Picture Memory Test; VFL = Verbal Fluency Initial Letter.

^*Pearson chi-square test for categorical variables and Student’s t test for continuous variables.

^†High BMI is defined as BMI 25 kg/m² or more, high triglycerides as triglycerides level 150 mg/dL or more, low HDL-C as HDL-C level less than 40 mg/dL in men and less than 50 mg/dL in women, high blood pressure as systolic blood pressure 130 mmHg or more and diastolic blood pressure 85 mmHg or more or taking an antihypertensive medication, and high HbA1c as glycosylated hemoglobin level 5.4% or more or taking an antidiabetic medication.

Global Cognitive Function (JMMSE)

The participants generally showed fewer errors in JMMSE at the second follow-up, but the interaction of the time (first and second follow-up) and metabolic syndrome or each component of metabolic syndrome was all insignificant, suggesting that there was no difference in changes in number of errors of JMMSE between the two groups (Table 2).

Table 2.

Association Between the Metabolic Syndrome and the Square Root of the Number of Errors in the JMMSE^*

Model Terms†	Estimate	SE	p Value‡
Metabolic syndrome
Metabolic syndrome	−0.061	0.128	.632
Time1 (first follow-up)	−0.175	0.082	.035
Time2 (second follow-up)	−0.208	0.081	.011
Metabolic syndrome × Time1	0.111	0.144	.443
Metabolic syndrome × Time2	0.142	0.139	.310
BMI
High BMI	−0.018	0.126	.889
Time1 (first follow-up)	−0.090	0.085	.294
Time2 (second follow-up)	−0.228	0.082	.006
High BMI × Time1	−0.131	0.140	.350
High BMI × Time2	0.190	0.136	.165
Triglycerides level
High triglycerides	0.150	0.154	.330
Time1 (first follow-up)	−0.148	0.075	.050
Time2 (second follow-up)	−0.196	0.074	.009
High triglycerides × Time1	0.053	0.175	.762
High triglycerides × Time2	0.163	0.161	.315
HDL-C level
Low HDL-C	0.026	0.188	.889
Time1 (first follow-up)	−0.136	0.072	.060
Time2 (second follow-up)	−0.146	0.071	.043
Low HDL-C × Time1	−0.015	0.214	.943
Low HDL-C × Time2	−0.100	0.191	.600
Blood pressure
High blood pressure	0.047	0.127	.709
Time1 (first follow-up)	−0.118	0.112	.294
Time2 (second follow-up)	−0.086	0.112	.446
High blood pressure × Time1	−0.032	0.140	.819
High blood pressure × Time2	−0.114	0.139	.413
HbA1c level
High HbA1c	−0.016	0.125	.900
Time1 (first follow-up)	−0.240	0.111	.033
Time2 (second follow-up)	−0.190	0.105	.071
High HbA1c × Time1	0.159	0.140	.257
High HbA1c × Time2	0.047	0.135	.726

Notes: BMI = body mass index; HbA1c = glycosylated hemoglobin; HDL-C = high-density lipoprotein cholesterol; JMMSE = Japanese version of Mini-Mental State Examination.

^*An increase in the square root of the number of errors in the JMMSE with time indicates cognitive decline.

^†Separate mixed-effects models for the metabolic syndrome or each component controlling for age, gender, and years of education.

^‡Bonferroni correction for multiple comparisons of six metabolic syndrome states (metabolic syndrome itself and five components) sets significant level at p < .0083.

Executive Cognitive Function (VFL)

The metabolic syndrome and its components were not associated with baseline and changes in VFL score, the latter indicated by nonsignificant interaction terms with first and second follow-up (Table 3).

Table 3.

Association Between the Metabolic Syndrome and the Verbal Fluency Initial Letter

Model Terms*	Estimate	SE	p Value†
Metabolic syndrome
Metabolic syndrome	−1.098	0.443	.014
Time1 (first follow-up)	−0.073	0.258	.777
Time2 (second follow-up)	0.132	0.320	.680
Metabolic syndrome × Time1	−0.129	0.454	.777
Metabolic syndrome × Time2	−0.399	0.552	.471
BMI
High BMI	−0.805	0.439	.069
Time1 (first follow-up)	−0.095	0.268	.724
Time2 (second follow-up)	−0.004	0.326	.991
High BMI × Time1	−0.042	0.440	.924
High BMI × Time2	−0.018	0.547	.973
Triglycerides level
High triglycerides	−0.527	0.544	.335
Time1 (first follow-up)	−0.040	0.235	.866
Time2 (second follow-up)	0.091	0.294	.758
High Triglycerides × Time1	−0.389	0.548	.479
High triglycerides × Time2	−0.441	0.645	.495
HDL-C level
Low HDL-C	−0.813	0.658	.219
Time1 (first follow-up)	−0.183	0.225	.417
Time2 (second follow-up)	0.053	0.283	.851
Low HDL-C × Time1	0.637	0.673	.345
Low HDL-C × Time2	−0.341	0.747	.649
Blood pressure
High blood pressure	−0.839	0.443	.060
Time1 (first follow-up)	−0.146	0.350	.677
Time2 (second follow-up)	0.399	0.441	.367
High blood pressure × Time1	0.050	0.441	.909
High blood pressure × Time2	−0.614	0.545	.263
HbA1c level
High HbA1c	−0.561	0.439	.203
Time1 (first follow-up)	0.004	0.351	.991
Time2 (second follow-up)	−0.074	0.411	.857
High HbA1c × Time1	−0.177	0.441	.689
High HbA1c × Time2	0.106	0.534	.843

Notes: BMI = body mass index; HbA1c = glycosylated hemoglobin; HDL-C = high-density lipoprotein cholesterol.

^*Separate mixed-effects models for the metabolic syndrome or each component controlling for age, gender, and years of education.

^†Bonferroni correction for multiple comparisons of six metabolic syndrome states (metabolic syndrome itself and five components) sets significant level at p < .0083.

Memory Function (SPMT)

Only high HbA1c was associated with change in SPMT score indicated by significant interaction term with second follow-up (Table 4).

Table 4.

Association Between the Metabolic Syndrome and the Scenery Picture Memory Test

Model Terms*	Estimate	SE	p Value†
Metabolic syndrome
Metabolic syndrome	−0.649	0.444	0.146
Time1 (first follow-up)	0.047	0.227	0.838
Time2 (second follow-up)	−0.029	0.280	0.918
Metabolic syndrome × Time1	−0.035	0.401	0.931
Metabolic syndrome × Time2	−0.089	0.484	0.855
BMI
High BMI	−0.536	0.438	0.222
Time1 (first follow-up)	0.066	0.236	0.780
Time2 (second follow-up)	−0.204	0.284	0.474
High BMI × Time1	−0.080	0.387	0.837
High BMI × Time2	0.399	0.475	0.403
Triglycerides level
High triglycerides	−1.089	0.534	0.043
Time1 (first follow-up)	0.035	0.206	0.866
Time2 (second follow-up)	−0.036	0.256	0.888
High triglycerides × Time1	0.004	0.493	0.994
High triglycerides × Time2	−0.079	0.568	0.890
HDL-C level
Low HDL-C	−0.039	0.653	0.952
Time1 (first follow-up)	0.005	0.199	0.980
Time2 (second follow-up)	−0.047	0.246	0.851
Low HDL-C × Time1	0.268	0.590	0.651
Low HDL-C × Time2	−0.081	0.660	0.902
Blood pressure
High blood pressure	−0.264	0.441	0.551
Time1 (first follow-up)	0.081	0.309	0.794
Time2 (second follow-up)	−0.027	0.387	0.945
High blood pressure × Time1	−0.073	0.388	0.851
High blood pressure × Time2	−0.052	0.479	0.914
HbA1c level
High HbA1c	−0.095	0.435	0.828
Time1 (first follow-up)	0.023	0.306	0.939
Time2 (second follow-up)	0.654	0.349	0.064
High HbA1c × Time1	0.015	0.386	0.968
High HbA1c × Time2	−1.201	0.451	0.008‡

Notes: BMI = body mass index; HbA1c = glycosylated hemoglobin; HDL-C = high-density lipoprotein cholesterol.

^*Separate mixed-effects models for the metabolic syndrome or each component controlling for age, gender, and years of education

^†Bonferroni correction for multiple comparisons of six metabolic syndrome states (metabolic syndrome itself and five components) sets significant level at p < .0083.

^‡p Value significant after Bonferroni adjustment.

DISCUSSION

Metabolic syndrome in midlife increases risk for later life age-related diseases (12), physical decline (37), and cognitive decline (38)—essential factors for healthy aging (39). On the contrary, the results of the current study of older adults aged 80 years and older (mean age of 85.0 years at baseline) suggest that the metabolic syndrome is not associated with cognitive decline, although we cannot conclude definitely from this study due to the relatively small sample size. During the 3 years of follow-up, we did not find differences in declines between those with and without metabolic syndrome in all three cognitive outcomes (i.e., global cognitive function as measured by JMMSE, executive function measured by the VFL, and memory function measured by the SPMT). This lack of association was observed for four of the five components of the metabolic syndrome with the exception of HbA1c. High HbA1c (defined as ≥ 5.4%) was associated with decline in memory function after 2 years.

Several studies have examined the association between the metabolic syndrome (and/or the components used to define the metabolic syndrome) and cognitive impairment, cognitive decline, or dementia in old age. The study results tend to differ depending on the age of the study population: Longitudinal studies have shown consistently that midlife obesity (7), hypertension (3,8,40), elevated total cholesterol (3,8,41), and diabetes mellitus (4,8,42) are associated with reduced cognitive function in later life. It is now well established that prevalence of these components of the metabolic syndrome in middle age increases the risk of late-life cognitive decline and dementia—including both AD and VaD.

On the other hand, there are no consistent findings to date to indicate that prevalence of the metabolic syndrome (or most of its key components) in later old age (ie, in those aged 80 plus years) is associated with subsequent cognitive decline or dementia. Some studies showed an increased risk of dementia or cognitive impairment for adults in early old age with the metabolic syndrome compared with those without (6,11). However, consistent with our findings, two previous prospective studies with large numbers of oldest old failed to show a significant association between the metabolic syndrome and cognitive decline or dementia. One study was of individuals aged 85 years or older (14), and another was a multiethnic cohort with an average age of more than 75 years (13). Two additional studies of the old–old, one with individuals aged 80 plus years (43) and another with a mean age of 76.2 years (44), respectively, reported that cognitive decline was not independently linked to hypertension, a key component of the metabolic syndrome, but only with hypertension coexisting with diabetes.

Some studies have shown an “inverse” association between several components of the metabolic syndrome and dementia or cognitive decline. For example, studies have reported (a) a higher risk of dementia with low blood pressure (rather than high blood pressure) among adults aged 75 plus years (19), (b) a reduced risk of dementia with higher total cholesterol among adults aged 70 plus years (17,45), (c) a reduced risk for dementia with higher BMI among adults with mean age of 71.8 years (16), and a reduced risk of dementia for abdominal obesity among adults aged 75 plus years (15). The latter study also reported that adults aged 75 plus with the metabolic syndrome may be at lower risk for AD in addition to overall dementia (15).

An important question is why the association of metabolic syndrome and cognitive decline is attenuated at older ages. Regarding one component of the metabolic syndrome that changes direction with age, hypertension, one of the possible mechanisms for the null (or reverse) association between hypertension and cognitive performance among the oldest old may include changes in cerebral blood flow tied with cognitive performance (46). Cerebral blood flow decreases with advancing age (47–49), and demented older adults have lower cerebral blood flow than those with optimal cognitive function (49). Although reduced cerebral blood flow is a normal aging process and brain atrophy begins in relatively early adulthood (50,51), exaggerated reduction in cerebral blood flow may accelerate the atrophy. Hypertension effects on brain function, possibly accounting for reduced regional cerebral blood flow, have been shown to be cumulative, and the association between hypertension and regional cerebral blood flow has been further strengthened by duration of hypertension (52). On the other hand, in the oldest old, higher blood pressure may be required to maintain adequate cerebral blood pressure and to preserve cognitive functions (49,53).

Regarding another component of the metabolic syndrome that changes direction with age, cholesterol, experimental studies suggest that high cholesterol accelerates production of beta-amyloid (54,55), which is the putative causative agent for AD. However, cholesterol is required for the formation of synapses (56), and synaptic loss correlates with cognitive impairment in AD (57). Cholesterol is also essential to maintain the myelin sheath, which surrounds and insulates the nerve cells, is rich in cholesterol, and is used to build synapses. The myelin function declines with age after middle age, and AD is associated with more severe myelin breakdown (58). Therefore, low cholesterol levels may make the myelin more vulnerable to malfunction as age progresses to the upper limits.

On the other hand, one consistent component of the metabolic syndrome across ages is the link of hyperinsulinemia and diabetes to an increased risk of dementia in late life. This relation appears to be consistent regardless of the ages at which the measurements were taken in middle or old age. One explanation is that insulin can act directly on the brain. Peripheral insulin crosses the blood–brain barrier via a receptor-mediated active transport process (59), and insulin receptors have been found in the hippocampus (60), a brain area that is crucial to memory function and is damaged by AD. Moreover, high insulin levels in the brain may interfere with the metabolism of amyloid beta by insulin-degrading enzyme, which regulates the metabolism of insulin and the deposition of beta amyloid (61). Consistent with the deleterious effect of hyperinsulinemia and diabetes on brain function, we found that elevated HbA1c is associated with cognitive decline over a 3-year period, even in the oldest old. HbA1c is one common measure of glycosylated hemoglobin, which measures average blood glucose levels over the preceding 1–3 months and, as such, is a better marker of long-term glucose levels than a one-time fasting glucose measurement. Our finding lends support to previous work linking glucose to cognitive decline (62) and suggests a more focused effort on biological mechanisms linked to blood sugar and insulin, such as the insulin signaling pathway. This pathway influences multiple phenotypes associated with healthy aging (63), including cognitive health (64).

There are other potential explanations for our findings. One such explanation might be selection bias. The participants might represent a group of healthy survivors who are less susceptible to the adverse effects of the metabolic syndrome. Another possible explanation is reverse causation. The reduced risk of having metabolic syndrome may be a consequence rather than a cause of dementia or cognitive impairment. For example, BMI, total cholesterol, and blood pressure are shown to decrease several years preceding the onset of dementia (65,66). Weight loss may result from eating difficulty (67) or impaired olfaction (68) due to cognitive impairment, and/or sarcopenia and low blood pressure preceding onset of dementia may be secondary to the underlying dementia pathology (13,45,65). Finally, recent studies have led to the hypothesis that metabolic syndrome in the oldest old might even be an adaptive mechanism to protect against deleterious effects of aging (20).

This study has several limitations. First, our cohort consisted of older people who were relatively healthy at baseline and was limited to Japanese participants from Okinawa. Therefore, caution should be used when generalizing to other populations. Second, we may have an insufficient sample size and follow-up period to detect a meaningful association between the metabolic syndrome and cognitive decline, although the participants are aged 80 years or older (averaging 85 years and ranging up to 98 years) and are at higher risk of developing cognitive impairment or dementia. Using the estimated coefficients, standard errors, and the current sample size, we performed power analyses for the interaction terms between the metabolic syndrome or its components and time, which indicate the additional change in cognitive test scores from the baseline examination among the participants with the metabolic syndrome or its components. For all interactions, the power was below 30% except for the interaction between HbA1c and time 2 (power = 75.9%). However, despite the small sample size and short duration of follow-up, we still found a significant association between high HbA1c (defined as ≥ 5.4%) and decline in memory function after 2 years. Third, there is a lack of information on how long the participants had the metabolic syndrome before the study. Finally, we substituted BMI for waist circumference and glycosylated hemoglobin level for fasting glucose level, which may influence the overall results, even though these substitutions have been validated in other studies (27,69,70).

In conclusion, our results showed that during the 3 years of follow-up, those with or without metabolic syndrome or its components did not differ with regards to change in cognitive test scores, with the exception of HbA1c and memory decline. The bulk of evidence accumulated thus far in the oldest old suggests attenuated, null, or possibly even reversal risk of the metabolic syndrome and some of its components on cognitive function. Our study supports these past findings and also lends weight to the relevance of glucose, even in the oldest old, as a persistent risk factor for incident cognitive decline in the Okinawan population. A longer study with larger sample size is required to confirm our findings and to assess the utility of the metabolic syndrome (and its components) as a predictor for cognitive decline in the oldest old. Clinical trials will ultimately be needed to assess the efficacy of treating metabolic syndrome and its individual components as more and more individuals join the ranks of the oldest old (71).

FUNDING

This study was supported by the National Institute on Aging (K01AG023014, R01AG033581, P30AG008017; Dodge and R01 AG027060-01; Willcox), Linus Pauling Institute Research Grant (Dodge), and Japanese Ministry of Education, Culture, Sports, Science and Technology (MEXT) Grant-in-Aid for Young Scientists (B) (20790442; Katsumata).

CONFLICT OF INTEREST

All the authors have no conflicts of interest as regards this study.

APPENDIX I.

The separate mixed-effects model for the metabolic syndrome and each component controlling for age, gender, and educational levels is described below:

• β₀ is the population average intercept.

• ms is the dummy variable for the metabolic syndrome or its components status (coded ms = 1 for the presence and coded ms = 0 for the absence); β₁ indicates the cross-sectional population effect on cognitive test score at baseline.

• time1 and time2 are the dummy variables for time (coded time1 = 0 and time2 = 0 for the baseline and time1 = 1 and time2 = 0 for the first follow-up and time1 = 0 and time2 = 1 for the second follow-up); β₂ and β₃ are their associated fixed-effect coefficients and indicate 1-year and 2-year population change in cognitive test scores from the baseline examination among the participants WITHOUT the metabolic syndrome or its components, respectively.

• β₄ and β₅ are the fixed-effect coefficients for interactions between the metabolic syndrome or its components status and time and indicate the additional 1-year and 2-year population change in cognitive test scores from the baseline examination among the participants WITH the metabolic syndrome or its components, respectively.

• b₀ is the individual random intercept and b₁ and b₂ indicates 1-year and 2-year random change in cognitive test score from baseline examination, respectively.

• X is the vector of covariates to control age, gender, and years of education, and β is the vector representing their fixed-effect coefficients.