Twenty-six–Year Change in Total Cholesterol Levels and Incident Dementia: The Honolulu-Asia Aging Study

Background  The relationship between total cholesterol levels and dementia is unclear.

Objective  To compare the natural history of change in total cholesterol across 26 years between men who did and did not develop dementia 3 years after the last measurement.

Design, Setting, and Participants  In the Honolulu-Asia Aging Study, 1027 Japanese American men had total cholesterol levels assayed on 5 occasions between 1965 and 1993 and were screened for dementia on 2 occasions between 1991 and 1996.

Main Outcome Measure  The slope of 26-year change in serum total cholesterol levels was estimated by a repeated-measures analysis and was compared between men with incident dementia (n = 56) and those without dementia (n = 971) at the end of the follow-up period.

Results  Cholesterol levels in men with dementia and, in particular, those with Alzheimer disease had declined at least 15 years before the diagnosis and remained lower than cholesterol levels in men without dementia throughout that period. The difference in slopes was robust to adjustment for potential confounding factors, including vascular risk factors, weight change, alcohol intake, and use of lipid-lowering agents.

Conclusion  A decline in serum total cholesterol levels may be associated with early stages in the development of dementia.

Although the importance of several vascular factors in the evolution of dementia and Alzheimer disease (AD) has become established over the past decade, associations with cholesterol levels remain controversial. Two studies^1,2 have found associations between higher midlife total cholesterol levels and increased risk of AD 20 to 30 years later. However, 2 studies with shorter follow-up periods have found associations between lower total cholesterol levels and incident AD,^3,4 and cross-sectional studies have most often found associations with lower rather than higher total cholesterol.⁵

To understand the cholesterol-AD link, physical health status is an important potential confounding or modifying factor because low cholesterol levels are associated with increased frailty and mortality in late life.⁶ Alzheimer disease is a catabolic condition associated with weight loss and declining blood pressure, even before the onset of clinical dementia.^7,8 A decline in total cholesterol level may accompany these processes, but the natural history of such a change has not been established.

In the Honolulu-Asia Aging Study, total cholesterol levels were measured in a subsample of men on 5 occasions during a 26-year period, and the men were examined for incident dementia 3 years after the last measurement. In this report, we compare the natural history of cholesterol level change between men with and without incident dementia.

The Honolulu-Asia Aging Study has been described previously.⁷ The baseline sample consisted of Japanese American men who were born between 1900 and 1919 and residents on the island of Oahu, Hawaii, in 1965. Participants were examined between 1965 and 1968 and between 1971 and 1974; 3734 men (80% of surviving cohort members) participated in a later examination between 1991 and 1993 that included an assessment for dementia. The analysis described herein was restricted to a subsample of 1430 men who had previously participated in a lipoprotein substudy in 1970-1972 and 1980-1982; of these, 70% were selected at random from the baseline sample and 30% were selected on the basis of raised baseline cholesterol or triglyceride levels.⁹ Thus, cholesterol levels were measured on 5 previous examinations: in 1965-1968, 1970-1972, 1971-1974, 1980-1982, and 1991-1993.

Of the 1350 participants who were dementia free in 1991-1993, 1027 (76.1%) were reassessed between 1994 and 1996. In this analysis, we compared those with incident dementia (n = 56) with those without dementia (n = 971) at that assessment. The assessment of dementia in the Honolulu-Asia Aging Study late-life examinations followed a 3-stage standard protocol that has been described in detail previously.⁷ Dementia evaluations included neuropsychological assessment, a proxy interview, a neurological examination, neuroimaging, and consensus diagnoses. Dementia was diagnosed according to Diagnostic and Statistical Manual of Mental Disorders, Revised Third Edition criteria¹⁰; probable and possible AD, according to National Institute of Neurological and Communicative Disorders and Stroke–Alzheimer's Disease and Related Disorders Association criteria¹¹; and vascular dementia, according to California Alzheimer Disease and Treatment Centers criteria.¹²

Blood samples for total cholesterol levels were drawn with participants in the nonfasting state for the first 4 examinations and in the fasting state for the fifth. Intervals between examinations were calculated for individual participants from the recorded examination dates. Data on a number of potential confounding factors were available. Previous stroke and ischemic heart disease were ascertained through continual surveillance of the cohort that was established in 1965. Physical function and depressive symptoms were measured at the 1991-1993 late-life examination.^7,13 Systolic and diastolic blood pressure and weight were measured at each examination. Alcohol intake, use of any prescribed lipid-lowering medication, and smoking status were assessed at each examination by answers to a standard questionnaire. Apolipoprotein E (APOE) genotype (presence or absence of the ε4 allele) and self-reported health status (excellent or good vs fair or poor) in 1991-1993 were considered potential modifying factors.

Individual trajectories of change in total cholesterol levels over the 5 examinations were estimated from linear random-effects models, using SAS statistical software, version 8 (SAS Institute Inc, Cary, NC). Cholesterol level was entered as the dependent variable. Dementia, time (years since study entry to the nearest month), and a dementia × time interaction term were entered as independent variables. The coefficient for the interaction term represents the additional change in cholesterol level over time associated with dementia. Because the trajectory of cholesterol change in the sample was nonlinear, time² and time³ variables were also included in the model to estimate the interaction between time and dementia status.

All models were adjusted for baseline age entered as a linear continuous variable. Other independent variables were entered to investigate confounding effects. The analyses were repeated for dementia subtypes and stratified by APOE status (for AD vs no dementia) and self-reported health (for any dementia vs no dementia). To formally test the interactions between self-reported health and dementia, models were fitted with all 2-way and 3-way interactions among self-reported health, dementia, and rate of change (ie, time, time², and time³). Likelihood ratio tests were used to assess the combined effect of the interaction terms. The same procedures were carried out to investigate whether APOE status (presence or absence of the ε4 allele) modified the dementia × time interactions of interest.

To account for between-person heterogeneity in the changes of cholesterol levels, intercept, time, and time² were modeled as random effects; this was not done for the stratified analyses, since the random effect for time² was dropped because the model did not converge. Goodness-of-fit criteria, including log likelihood ratio, the Akaike Information Criteria (AIC), the Akaike Information Criteria (corrected) (AICC), and Schwartz Bayesian Information Criteria (BIC), were used for model selection. Three outlying individual trajectories with large deviations from the population mean trajectory were identified using the method of Lesaffre and Verbeke.¹⁴ However, none of the outlying trajectories qualitatively changed the fixed effects estimates, nor did they change the significance levels. Therefore, the sample size was left unchanged in the final models.

Compared with those without dementia, men who developed dementia were older and had lower levels of education and higher levels of vascular risk factors at the previous late-life examination (Table 1). In those with dementia, total cholesterol levels were lower at all previous examination points, with the greatest difference at the 1980-1982 examination. Proportions of the sample receiving any lipid-lowering agent were 2.2% in 1965-1968, 3.4% in 1980-1982, and 16.7% in 1991-1993. Of the 56 men with incident dementia, 31 had AD, 15 had vascular dementia, and 10 had other subtypes. In our analytic sample, 2.1% (n = 22), 18.2% (n = 187), and 79.6% (n = 818) underwent 3, 4, and 5 cholesterol level measurements, respectively. The proportion of incident dementia cases was 18% (n = 4) for men who underwent 3 measurements, compared with 5.1% (n = 51) for those who underwent 4 or 5 measurements.

The most parsimonious model of cholesterol level change over time (in years) prior to 1994 was expressed by the following equation: cholesterol level (in milligrams per deciliter) = 229.9 − (0.97 × time) + (0.039 × time²) − (0.0025 × time³). On average, cholesterol levels declined with age.

Adjusted slope coefficients for total cholesterol levels are displayed in Table 2. The first 3 rows display slope coefficients for any dementia, AD, and vascular dementia, respectively, relative to slopes in participants without dementia. The following 4 rows display stratified analyses for self-reported health status (for the slope difference between any dementia and no dementia) and for APOE genotype (for the slope difference between AD and no dementia). The final rows display slope coefficients for any dementia relative to no dementia, adjusted successively for other covariates.

At the 1965 baseline examination, cholesterol levels did not differ by later dementia status. However, there were significant interactions between dementia status and all components of the slope (linear, quadratic, and cubic), suggesting an increased decline in men who went on to develop dementia (Figure 1). This trend of change in cholesterol level was similar for dementia subtypes and was, if anything, stronger when adjusted for potential confounding factors (Table 2). The decline in cholesterol levels was strongest in those with dementia and the APOE ε4 allele (Table 2) and those with dementia and self-reported worse general health at the time of the final cholesterol measurement (Table 2, Figure 2). Although the stratified analyses were indicative of meaningful differences between subgroups, especially for those with incident dementia and previously self-reported fair or poor health, the likelihood ratio test for the interaction terms was not statistically significant (P = .49). The interaction between APOE ε4 and rate of change on incident AD was also not significant (data not shown). However, only 18 of the 293 men in the group with self-reported fair or poor health had incident dementia, and 14 of the 194 men with APOE ε4 had incident AD.

To assess the effect of missing data on the population mean trajectory, 2 analyses were carried out. We refitted the model for the subset of those with at least 4 measurements. The results were similar to those from the entire sample, suggesting that the missing data bias was likely to be minimal (data not shown). In addition, we compared the trajectories of those with missing data at the last examination with the trajectories of those with and without dementia who had a complete follow-up. No substantial differences were found (data not shown). Three-way interaction terms between baseline age, time, and dementia status were tested for all 3 time terms (linear, quadratic, and cubic), but the addition of these age interaction terms did not significantly alter the models. We also found no improvement in the goodness of fit when we incorporated a quadratic term for age (data not shown).

The role of total cholesterol levels in the etiology of dementia remains controversial. Previous studies in which cholesterol levels have been measured on a single occasion in midlife have suggested that high midlife cholesterol levels predict increased risk of dementia,^1,2 whereas individuals who reach the dementia threshold have normal or lower levels.^2,4,5 Ascertaining the natural history of change in cholesterol levels is important for clarifying underlying etiologic processes. In this cohort of men who were examined during a 26-year period, we found that those who went on to develop dementia had a greater decline in cholesterol levels during a 10- to 15-year period.

Risk factors such as blood pressure and weight decline with age. However, dementia-associated additional decline in these factors becomes detectable approximately 3 to 6 years before the onset of the clinical syndrome.^7,8 Our findings for cholesterol suggest a different pattern with a much earlier decline and little subsequent acceleration prior to the onset of dementia. Instead, levels tended to converge between the comparison groups in late life. Therefore, the decline is unlikely to be explained by behavioral or metabolic changes associated with mild cognitive impairment. The plotted trajectories (Figure 1) suggested a relative rise in cholesterol levels early in the follow-up period followed by a decline shortly afterward. From examination of the individual trajectories, this rise and fall did not appear to be purely an artifact of the models used. However, the exaggerated decline was not confined to those with an initial rise. One explanation is that these men were exposed initially to higher levels of cholesterol, which have been suggested to be a risk factor for later dementia.^1,2 Another is that temporal instability, rather than actual levels, may be a marker of risk. No evidence for differences between age groups was found.

Several factors might influence the findings. Participants were all men of Japanese origin living in Hawaii; it is possible that our results are specific to this ethnic group, although we believe this to be unlikely. Use of lipid-lowering medication was not common in this cohort until late in the follow-up period and is an unlikely explanation for differences in slopes. The final cholesterol level was measured in the fasting state. However, differences between fasting and nonfasting total cholesterol levels are small (about 3-4 mg/dL [0.08-0.10 mmol/L])¹⁵ and unlikely to account for our findings.

The observed associations may not represent direct causal pathways. Associations, for example, appeared fairly similar between AD and vascular dementia. The group differences in slopes, although significant, were small and there was considerable overlap between individual trajectories. We do not believe that our findings have implications for lipid-lowering interventions because decline in cholesterol levels may have been caused by other processes. Hypocholesterolemia is recognized to be associated with frailty and poor general health.⁶ It also has been found to be specifically associated with inflammatory markers and poor nutritional status.¹⁶ The difference in slope between those with and without dementia was weak and not significant in participants who previously reported good health; however, the low number of cases reporting poor health lowered our power to investigate this finding. If the decline in cholesterol level was purely secondary to neurodegenerative processes underlying dementia, this difference should be equally apparent in men reporting good and poor health, which did not appear to be the case. Similarly, the results do not suggest that the decline in cholesterol level was simply a marker of later health status because it was not associated with self-reported poor health in men who did not develop dementia (Figure 2). It is possible that the decline in cholesterol levels is a marker for early processes that reflect neurodegenerative changes and also lead to a decline in general health status.

Further research is required to clarify the processes underlying the decline in cholesterol levels prior to dementia because they may represent occurrences early in the course of the disease. Decline in cholesterol level may also be a marker for factors underlying both physical and cognitive decline in old age. It is likewise important that research into associations between cholesterol levels and risk of dementia take into account physical health status because it may have a modifying role.

Correspondence: Robert Stewart, MD, Section of Epidemiology, Campus Box 60, Institute of Psychiatry, De Crespigny Park, London SE5 8AF, England (r.stewart@iop.kcl.ac.uk).

Accepted for Publication: July 24, 2006.

Author Contributions: Dr Launer had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Launer. Acquisition of data: White. Analysis and interpretation of data: Stewart, Xue, and Launer. Drafting of the manuscript: Stewart, Xue, and Launer. Critical revision of the manuscript for important intellectual content: Stewart, White, Xue, and Launer. Statistical analysis: Stewart, Xue, and Launer. Obtained funding: White. Administrative, technical, and material support: White. Study supervision: White and Launer.

Financial Disclosure: None reported.

Funding/Support: The Honolulu-Asia Aging Study is supported by the National Institutes of Health (by contract N01-AG-4-2149 and grant 5U01-AG019349 from the National Institute on Aging [NIA] and by the Intramural Research Program of the NIA) and by contract N01-HC-05102 from the National Heart, Lung, and Blood Institute.

Role of the Sponsors: The funders approved the design of the study but have had no further input into the analysis, interpretation, or reporting of findings.