Association of Blood Lipoprotein Levels With Incident Alzheimer Disease in Community-Dwelling Individuals

Abstract

Background and Objectives

Cardiovascular risk factors are important contributors to the risk of Alzheimer disease (AD). To further explore the physiologic links between cardiovascular health and AD risk, we studied the associations between various blood lipoprotein levels and AD risk in community-dwelling older adults.

Methods

This longitudinal analysis included participants aged 60 years or older without prevalent dementia and with available cognitive follow-up and lipoprotein marker data from the Framingham Heart Study. Levels of high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), small dense LDL-C (sdLDL-C), lipoprotein a (Lp(a)), apolipoprotein B (ApoB), and the ApoB isoform ApoB48 were measured in blood samples obtained from 1985 to 1988. Participants were under surveillance for incident AD until 2020. AD diagnosis was based on standard clinical criteria. The relationships between blood lipoprotein levels (expressed as both continuous variables and quartiles) and AD incidence were examined using Cox proportional hazard models adjusted for baseline age and sex.

Results

A total of 822 participants (mean [SD] age 72.5 [3.7] years, 538 [65.5%] women) were followed for a median (interquartile range) of 12.55 (7.34–15) years, during which 128 participants developed incident AD. An increase of 1 standard deviation unit (SDU) in ln(sdLDL-C) concentration was associated with a 21% increase in the risk of incident AD (hazard ratio [HR] 1.21, 95% CI 1.01–1.45), whereas a 1-SDU increase in ln(ApoB48) concentration was associated with a 22% decrease in the risk of incident AD (HR 0.78, 95% CI 0.66–0.93). Participants in the first HDL-C quartile were 44% less likely to develop AD compared with those in the second, third, and fourth HDL-C quartiles (HR 0.56, 95% CI 0.33–0.95). Participants with sdLDL-C concentrations below the median were 38% less likely to develop AD compared with those with sdLDL-C concentrations above the median (HR 0.62, 95% CI 0.44–0.86).

Discussion

Lower sdLDL-C and higher ApoB48 concentrations were associated with a lower AD risk. In addition, individuals with the lowest HDL-C concentrations were less likely to develop AD compared with the remaining sample. These findings underscore links between lipoprotein metabolism pathways and AD risk, emphasizing the potential role of blood lipoprotein markers in AD risk stratification and of lipid modification strategies in dementia prevention.

Introduction

Dementia is a leading source of morbidity and mortality in the aging population.¹ Worldwide, there were 57.4 million people living with dementia in 2019, a number that is expected to reach 152.8 million by 2050.² At the same time, there is a temporal trend toward decreasing incidence of Alzheimer disease (AD) and other dementias in the United States and other high-income countries, which has at least partially been attributed to better management of cardiovascular risk factors.^1,3

Cardiovascular risk factors are important contributors to the risk of dementia in general⁴ and AD in particular.⁵ The convergence of epidemiologic data with clinicopathologic evidence pointing toward a frequent presence of vascular pathology in the brain of individuals with AD⁶ has led to a growing interest in understanding the mechanistic interplay between cardiovascular risk factors and AD pathology and how the former can trigger or promote the latter.⁷

Pathways involved in lipoprotein metabolism may constitute a biophysiologic link between cardiovascular health and AD risk. For instance, genetic variations in the gene encoding apolipoprotein E (APOE gene)—a protein involved in lipid transport and lipoprotein clearance—have been associated with cardiovascular disease risk⁸ but are also the strongest identified genetic contributors to the risk of late-onset AD.¹ Among other lipoproteins, decreased levels of high-density lipoprotein cholesterol (HDL-C) and elevated levels of low-density lipoprotein cholesterol (LDL-C), small dense LDL-C (sdLDL-C), lipoprotein a (Lp(a)), apolipoprotein B (ApoB), and the ApoB isoform ApoB48—a specific marker of chylomicron metabolism⁹—have all been linked to an increased cardiovascular disease risk.^10-16 However, evidence regarding their associations with AD risk is either conflicting or lacking entirely. Specifically, although in some case-control and cross-sectional studies, HDL-C, LDL-C, and Lp(a) concentrations have been associated with AD^17,18 or cognitive performance,¹⁹ prospective studies have not corroborated these findings.^20-22 In addition, the associations of sdLDL-C and ApoB48 with AD incidence have not been explored.

In this work, we aimed to address the abovementioned literature gaps and shed more light on the relationships between blood lipoprotein levels and AD incidence. To this end, we examined the longitudinal associations between a variety of blood lipoprotein markers, namely HDL-C, LDL-C, sdLDL-C, Lp(a), and ApoB as well as its isoform ApoB48, and the risk of incident AD in a community-based sample of older adults.

Methods

Participants

The Framingham Heart Study (FHS) is an ongoing, community-based cohort study that was launched in 1948 in Framingham, Massachusetts.²³ Participation eligibility was determined by randomly selecting Framingham residents between 30 and 59 years from local census data, with a two-thirds sampling ratio. Family members residing in the same household with an individual selected for participation were also included (provided they were within the eligible age range). Individuals with definite signs of cardiovascular disease at baseline were excluded from participation. Participants of the Original Cohort have undergone up to 32 examinations, performed every 2 years, which have included detailed history taking by a physician, a physical examination, and laboratory testing.

This longitudinal analysis included participants from the Original Cohort of the FHS who were 60 years or older and free of dementia at examination cycle 19 (1985–1988) and had available cognitive follow-up and lipoprotein marker data.

Standard Protocol Approvals, Registrations, and Patient Consents

All participants have provided written informed consent. Study protocols and consent forms have been approved by the institutional review board of the Boston University Medical Center.

Dementia Surveillance and Diagnosis

Participants were administered a Mini-Mental State Examination (MMSE) at each biennial examination. Individuals with possible cognitive impairment or dementia—identified by combining information from education-specific MMSE performance and multiple other sources (eMethods 1)—were flagged for further evaluation, including a complete neurologic and neuropsychological assessment (eTable 1).

Identified cases of suspected cognitive impairment were followed by a neurologist and a neuropsychologist at periodic intervals while cases of suspected dementia were sent to the dementia review panel consisting of at least 1 neurologist and 1 neuropsychologist. Dementia diagnosis, type, and date of onset were established based on data from previously performed serial neurologic and neuropsychological evaluations, telephone interviews with caregivers, medical records, neuroimaging studies, and, when applicable and available, autopsies. After participant death, medical and nursing records were reviewed postmortem to assess for potential development of cognitive decline since the last examination. For dementia cases detected before 2001, a repeat review was completed to ensure application of up-to-date diagnostic criteria.

Dementia diagnosis was based on criteria from the Diagnostic and Statistical Manual of Mental Disorders, fourth edition. AD diagnosis was based on criteria from the National Institute of Neurological and Communicative Disorders and Stroke and the Alzheimer's Disease and Related Disorders Association.²⁴ Diagnostic criteria for other dementia subtypes are described in the Supplement (eMethods 2). The diagnostic algorithm allows participants to be diagnosed with more than 1 dementia subtype.

Blood Lipoprotein Levels

Lipoprotein marker measurements were performed between 2006 and 2008 in fasting venous blood samples collected during examination cycle 19 (1985–1988). The samples were collected in 0.1% EDTA tubes. HDL-C and LDL-C concentrations were measured using previously described methods.²⁵ The concentration of sdLDL-C was measured by automated standardized enzymatic analysis on a Hitachi 911 analyzer (Hitachi Corporation, Tokyo, Japan), using kits provided by Denka-Seiken Corp. (Tokyo, Japan).²⁶ ApoB and Lp(a) concentrations were measured by immunoturbidimetric assays by Wako Chemicals (Richmond, VA).²⁷ ApoB48 concentrations were measured using ELISA kits from Shibayagi (Gunma, Japan). The characteristics of the ApoB48 assay have been described elsewhere.²⁸ In brief, the assay uses a monoclonal antibody against the C-terminal decapeptide of ApoB48 and has been calibrated using recombinant ApoB48 antigen with more than 90% recovery of ApoB48; it exhibits no cross-reactivity with the other major ApoB isoform, ApoB100.

Demographic and Vascular Risk Factors

Demographic, anthropometric, and vascular risk factor data were updated at examination cycle 19 (1985–1988). Systolic blood pressure (SBP) was derived from the average of 2 blood pressure readings obtained 10 minutes apart with the participant resting in a sitting position. Diabetes mellitus was defined as random blood glucose levels ≥11 mmol/L or use of antidiabetic medications (insulin or oral glucose-lowering agents). Smoking was defined as current smoking within the year preceding the evaluation. Waist-to-hip ratio (WHR) was calculated by dividing the participant's waist circumference by hip circumference (cm/cm).

APOE Genotyping

APOE genotyping was performed by leukocyte DNA amplification and restriction isotyping, as previously described.²⁹ Participants were categorized as APOE ε4 carriers or noncarriers, according to the presence of ≥1 APOE ε4 allele.

Statistical Analysis

Lipoprotein marker variables (i.e., HDL-C, LDL-C, sdLDL-C, Lp(a), ApoB, and ApoB48) were natural log-transformed and subsequently converted to z-scores. Participants were also ranked into quartiles according to each lipoprotein marker distribution.

Primary Analyses

We used Cox proportional hazard models with AD as the dichotomous outcome. Follow-up was from the examination 19 evaluation through 2020. Time-to-event was time to AD diagnosis; participants who did not develop AD were right-censored at the time of their last evaluation or at the age at death (if AD-free at death). Lipoprotein marker variables were the main predictors (separate model for each marker) and were initially entered in the models in their continuous form. Subsequently, lipoprotein marker variables were entered in the models as quartiles to inspect their relationships with AD incidence for nonlinearity or presence of threshold effects; the fourth quartile (containing participants with the highest lipoprotein marker concentrations) was used as the reference and was compared with each of the other (first, second, and third) quartiles. Based on visual inspection of these results, we further constructed binary indicators for associations between lipoprotein marker quartiles and AD incidence exhibiting a nonlinear pattern, to better demonstrate potential threshold effects and AD risk discontinuity. Models were adjusted for baseline age and sex.

Stratified Analyses by Sex

To explore potential differences in the associations of lipoprotein markers with AD incidence between men and women, we stratified study participants based on sex and repeated the primary analyses in each stratum.

Sensitivity Analyses

We conducted the following sensitivity analyses: (1) we examined the associations between lipoprotein levels and all-cause dementia incidence; (2) we further considered educational attainment, vascular risk factors (i.e., SBP, use of antihypertensive medications, diabetes, current smoking, and WHR), and APOE ε4 carriership as potential confounders; and (3) in instances where results between primary analyses (adjusted for age and sex) and fully adjusted models differed, we recomputed the former after excluding participants with missing data on education, vascular risk factors, or APOE ε4 carriership, to explore whether result differences were due to confounding or changes in the analytical sample. Finally, to further assess the stability of our findings, we repeated the primary analyses after excluding participants who developed interim stroke during the study follow-up.

A 2-sided p value of ≤0.05 was considered statistically significant. All statistical analyses were performed using SAS software (SAS Institute, Cary, NC).

Data Availability

Anonymized data not published within this article may be shared on request from any qualified investigator for purposes of replicating procedures and results. FHS data can be accessed through the NIH database of genotypes and phenotypes (ncbi.nlm.nih.gov/gap/).

Results

Participant Characteristics

The Original Cohort of the FHS enrolled 5,209 Framingham residents. Of these, 822 participants were 60 years or older and free of dementia at examination cycle 19 and had available cognitive follow-up and lipoprotein marker data. During a median (interquartile range) follow-up of 12.55 (7.34–15) years, 158 participants developed incident dementia; among these, 128 (81%) developed AD (108 developed pure AD, 7 AD with stroke, and 13 mixed AD and vascular dementia) and 30 other types of dementia (eTable 2). Baseline characteristics of study participants overall and by sex are presented in Table 1. The mean (SD) age was 72.5 (3.7) years, and 538 (65.5%) participants were women; 56 (6.8%) participants had previous stroke.

Table 1.

Baseline Participant Characteristics

	Total (N = 822)	Men (n = 284)	Women (n = 538)
Demographics
Age, y, mean (SD)	72.48 (3.70)	72.28 (3.56)	72.58 (3.77)
Women, n (%)	538 (65.45%)
Educational attainment, n (%)
No high school degree	231 (28.41%)	85 (30.25%)	146 (27.44%)
High school degree	312 (38.38%)	106 (37.72%)	206 (38.72%)
Some college	158 (19.43%)	42 (14.95%)	116 (21.80%)
College degree	112 (13.78%)	48 (17.08%)	62 (12.03%)
Duration of follow-up, y, median (IQR)	12.55 (7.34–15)	10.45 (5.31–15)	13.50 (8.15–15)
Vascular risk factors
Systolic blood pressure, mm Hg, mean (SD)	142.73 (20.19)	142.65 (20.52)	142.76 (20.03)
Use of antihypertensive medications, n (%)	351 (42.80%)	107 (37.81%)	244 (45.44%)
Diabetes mellitus, n (%)	53 (6.48%)	26 (9.15%)	27 (5.06%)
Current smoking, n (%)	101 (12.32%)	29 (10.21%)	72 (13.43%)
Waist-to-hip ratio, mean (SD)	0.89 (0.09)	0.96 (0.06)	0.85 (0.07)
Previous stroke, n (%)	56 (6.81%)	29 (10.21%)	27 (5.02%)
Genetic risk factors
APOE ε4 carrier, n (%)	137 (20.03%)	42 (18.18%)	95 (20.97%)
Blood lipoprotein levels
HDL-C, mg/dL, mean (SD)	50.16 (17.38)	42.46 (14.11)	54.22 (17.57)
LDL-C, mg/dL, mean (SD)	140.63 (35.08)	133.30 (32.31)	144.51 (35.89)
Small dense LDL-C, mg/dL, mean (SD)	42.63 (20.25)	45.14 (18.90)	42.82 (20.89)
Lp(a), mg/dL, mean (SD)	24.29 (25.77)	24.52 (24.99)	24.16 (26.19)
ApoB, mg/dL, mean (SD)	109.43 (25.72)	106.77 (23.77)	110.83 (26.61)
ApoB48, mg/dL, mean (SD)	1.18 (0.71)	1.24 (0.70)	1.15 (0.72)

Abbreviations: ApoB = apolipoprotein B; HDL-C = high-density lipoprotein cholesterol; IQR = interquartile range; LDL-C = low-density lipoprotein cholesterol; Lp(a) = lipoprotein a.

Blood Lipoprotein Levels and AD Incidence

In Cox proportional hazard models with lipoprotein markers as continuous variables, an increase of 1 standard deviation unit (SDU) in ln(sdLDL-C) concentration was associated with a 21% increase in the risk of incident AD, whereas a 1-SDU increase in ln(ApoB48) concentration was associated with a 22% decrease in the risk of incident AD (Table 2).

Table 2.

Associations of Blood Lipoprotein Markers as Continuous Variables and as Quartiles With AD Dementia Incidence

Lipoprotein markera	Events, n/at risk, n	HR (95% CI)b	p Value	Quartile	HR (95% CI)b	p Value
HDL-C	128/821	1.14 (0.94–1.39)	0.18	Q1 Q2 Q3 Q4	0.55 (0.31–1.00) 1.00 (0.63–1.59) 0.94 (0.59–1.48) 1 (reference)
				Q1 vs Q2-Q3-Q4c	0.56 (0.33–0.95)	0.03
LDL-C	128/821	1.11 (0.93–1.33)	0.25	Q1 Q2 Q3 Q4	0.71 (0.43–1.18) 0.67 (0.41–1.10) 0.85 (0.54–1.36) 1 (reference)
Small dense LDL-C	128/822	1.21 (1.01–1.45)	0.04	Q1 Q2 Q3 Q4	0.57 (0.35–0.93) 0.49 (0.30–0.81) 0.73 (0.46–1.16) 1 (reference)
				Q1-Q2 vs Q3-Q4c	0.62 (0.44–0.86)	0.008
Lp(a)	128/821	1.05 (0.88–1.25)	0.58	Q1 Q2 Q3 Q4	0.94 (0.55–1.60) 1.08 (0.66–1.78) 1.44 (0.89–2.32) 1 (reference)
ApoB	128/821	1.15 (0.96–1.37)	0.13	Q1 Q2 Q3 Q4	0.73 (0.43–1.22) 0.80 (0.48–1.34) 1.15 (0.72–1.85) 1 (reference)
ApoB48	128/817	0.78 (0.66–0.93)	0.006	Q1 Q2 Q3 Q4	1.81 (1.07–3.07) 1.51 (0.87–2.63) 1.41 (0.82–2.45) 1 (reference)

Abbreviations: AD = Alzheimer disease; ApoB = apolipoprotein B; HDL-C = high-density lipoprotein cholesterol; HR = hazard ratio; LDL-C = low-density lipoprotein cholesterol; Lp(a) = lipoprotein a.

^aVariables expressing lipoprotein marker concentrations have been natural log-transformed and subsequently converted to z-scores.

^bResults from Cox proportional hazards models with AD dementia as the outcome and the respective lipoprotein marker (in a continuous or quartile form) as the main predictor. Models are adjusted for baseline age and sex.

^cBinary indicators constructed for associations between lipoprotein marker quartiles and AD incidence exhibiting a nonlinear pattern.

Models with lipoprotein markers as quartiles revealed nonlinear relationships of HDL-C and sdLDL-C concentrations with incident AD (Table 2). Specifically, participants with the lowest HDL-C concentrations (in the first HDL-C quartile) were 44% less likely to develop AD compared with those in the second, third, and fourth HDL-C quartiles. Moreover, participants with sdLDL-C concentrations below the median were 38% less likely to develop AD compared with those with sdLDL-C concentrations above the median. Finally, for ApoB48, the gradual increase in AD risk for lower quartiles suggested a linear relationship; thus, threshold effects using binary indicators were not further explored.

Stratified Analyses by Sex

In stratified analyses with lipoprotein markers as continuous variables, a 1-SDU increase in ln(sdLDL-C) and ln(ApoB) concentrations was associated with a 24% increase in the risk of incident AD, whereas a 1-SDU increase in ln(ApoB48) concentration was associated with a 20% decrease in the risk of incident AD in women (Table 3). No associations between lipoprotein markers as continuous variables and AD incidence were observed in men (Table 4).

Table 3.

Associations of Blood Lipoprotein Markers as Continuous Variables and as Quartiles With AD Dementia Incidence in Women

Lipoprotein markera	Events, n/at risk, n	HR (95% CI)b	p Value	Quartile	HR (95% CI)b	p Value
HDL-C	99/537	1.13 (0.92–1.39)	0.26	Q1 Q2 Q3 Q4	0.43 (0.19–0.97) 1.14 (0.69–1.86) 0.93 (0.56–1.53) 1 (reference)
				Q1 vs Q2-Q3-Q4c	0.43 (0.20–0.92)	0.03
LDL-C	99/537	1.19 (0.98–1.46)	0.09	Q1 Q2 Q3 Q4	0.57 (0.32–1.01) 0.55 (0.32–0.97) 0.75 (0.45–1.23) 1 (reference)
				Q1-Q2 vs Q3-Q4c	0.64 (0.43–0.97)	0.04
Small dense LDL-C	99/538	1.24 (1.01–1.53)	0.04	Q1 Q2 Q3 Q4	0.52 (0.30–0.89) 0.46 (0.26–0.81) 0.69 (0.41–1.17) 1 (reference)
				Q1-Q2 vs Q3-Q4c	0.59 (0.40–0.88)	<0.001
Lp(a)	99/537	1.07 (0.88–1.31)	0.49	Q1 Q2 Q3 Q4	0.91 (0.51–1.62) 0.97 (0.56–1.69) 1.24 (0.72–2.13) 1 (reference)
ApoB	99/537	1.24 (1.01–1.52)	0.04	Q1 Q2 Q3 Q4	0.55 (0.32–1.01) 0.77 (0.44–1.34) 1.16 (0.69–1.95) 1 (reference)
ApoB48	99/534	0.80 (0.66–0.98)	0.03	Q1 Q2 Q3 Q4	1.91 (1.00–3.64) 1.89 (0.98–3.67) 1.6 (0.82–3.11) 1 (reference)

Abbreviations: AD = Alzheimer disease; ApoB = apolipoprotein B; HDL-C = high-density lipoprotein cholesterol; HR = hazard ratio; LDL-C = low-density lipoprotein cholesterol; Lp(a) = lipoprotein a.

^aVariables expressing lipoprotein marker concentrations have been natural log-transformed and subsequently converted to z scores.

^bResults from Cox proportional hazard models with AD dementia as the outcome and the respective lipoprotein marker (in a continuous or quartile form) as the main predictor. Models are adjusted for baseline age.

^cBinary indicators constructed for associations between lipoprotein marker quartiles and AD incidence exhibiting a nonlinear pattern.

Table 4.

Associations of Blood Lipoprotein Markers as Continuous Variables and as Quartiles With AD Dementia Incidence in Men

Lipoprotein markera	Events, n/at risk, n	HR (95% CI)b	p Value	Quartile	HR (95% CI)b
HDL-C	29/284	1.23 (0.82–1.85)	0.32	Q1 Q2 Q3 Q4	0.54 (0.17–1.73) 0.60 (0.18–1.98) 0.86 (0.25–2.93) 1 (reference)
LDL-C	29/284	0.85 (0.58–1.25)	0.40	Q1 Q2 Q3 Q4	2.76 (0.58–13.03) 2.29 (0.51–10.34) 2.82 (0.60–13.30) 1 (reference)
Small dense LDL-C	29/284	1.12 (0.78–1.62)	0.55	Q1 Q2 Q3 Q4	0.81 (0.29–2.30) 0.62 (0.22–1.76) 0.86 (0.33–2.23) 1 (reference)
Lp(a)	29/284	1.01 (0.69–1.46)	0.99	Q1 Q2 Q3 Q4	1.10 (0.30–4.13) 1.63 (0.49–5.47) 2.55 (0.82–7.91) 1 (reference)
ApoB	29/284	0.85 (0.58–1.24)	0.40	Q1 Q2 Q3 Q4	2.01 (0.62–6.55) 1.09 (0.31–3.87) 1.30 (0.41–4.16) 1 (reference)
ApoB48	29/283	0.72 (0.52–1.00)	0.05	Q1 Q2 Q3 Q4	1.96 (0.77–4.96) 0.76 (0.23–2.53) 1.14 (0.41–3.14) 1 (reference)

Abbreviations: AD = Alzheimer disease; ApoB = apolipoprotein B; HDL-C = high-density lipoprotein cholesterol; HR = hazard ratio; LDL-C = low-density lipoprotein cholesterol; Lp(a) = lipoprotein a.

^aVariables expressing lipoprotein marker concentrations have been natural log-transformed and subsequently converted to z-scores.

^bResults from Cox proportional hazard models with AD dementia as the outcome and the respective lipoprotein marker (in a continuous or quartile form) as the main predictor. Models are adjusted for baseline age.

In models with lipoprotein markers as quartiles, relationships of HDL-C, LDL-C, and sdLDL-C concentrations with incident AD exhibited a nonlinear pattern in women (Table 4). Specifically, women with the lowest HDL-C concentrations (belonging in the first HDL-C quartile) were 57% less likely to develop AD compared with women in the second, third, and fourth HDL-C quartiles. Furthermore, women with LDL-C concentrations below the median were 36% less likely to develop AD compared with those with LDL-C concentrations above the median. Finally, women with sdLDL-C concentrations below the median were 41% less likely to develop AD compared with women with sdLDL-C concentrations above the median. No associations between lipoprotein markers as quartiles and AD incidence were observed in men.

Sensitivity Analyses

Results from analyses with all-cause dementia as the outcome are presented in Table 5. In models with lipoprotein markers as continuous variables, a 1-SDU increase in ln(sdLDL-C) concentration was associated with a 19% increase in incident dementia risk, whereas a 1-SDU increase in ln(ApoB48) concentration was associated with a 18% decrease in incident dementia risk. In models with lipoprotein markers as quartiles, the relationship of sdLDL-C concentrations with all-cause dementia incidence exhibited a nonlinear pattern. Specifically, participants with sdLDL-C concentrations below the median were 30% less likely to develop dementia compared with those with sdLDL-C concentrations above the median.

Table 5.

Associations of Lipoprotein Markers as Continuous Variables and as Quartiles With All-Cause Dementia Incidence

Lipoprotein markera	Events, n/at risk, n	HR (95% CI)b	p Value	Quartile	HR (95% CI)b	p Value
HDL-C	158/821	1.06 (0.88–1.21)	0.52	Q1 Q2 Q3 Q4	0.72 (0.64–1.38) 1.00 (0.65–1.53) 0.94 (0.61–1.43) 1 (reference)
LDL-C	158/821	1.03 (0.88–1.21)	0.72	Q1 Q2 Q3 Q4	0.97 (0.63–1.50) 0.65 (0.41–1.04) 0.88 (0.57–1.35) 1 (reference)
Small dense LDL-C	158/822	1.19 (1.01–1.40)	0.04	Q1 Q2 Q3 Q4	0.60 (0.39–0.94) 0.59 (0.38–0.92) 0.74 (0.48–1.13) 1 (reference)
				Q1-Q2 vs Q3-Q4c	0.70 (0.51–0.96)	0.03
Lp(a)	157/813	1.05 (0.89–1.23)	0.57	Q1 Q2 Q3 Q4	0.93 (0.59–1.48) 0.97 (0.62–1.51) 1.28 (0.83–1.97) 1 (reference)
ApoB	158/821	1.11 (0.94–1.30)	0.21	Q1 Q2 Q3 Q4	0.81 (0.51–1.28) 0.77 (0.48–1.22) 1.11 (0.72–1.70) 1 (reference)
ApoB48	158/817	0.82 (0.71–0.96)	0.01	Q1 Q2 Q3 Q4	1.50 (0.95–2.35) 1.16 (0.71–1.88) 1.10 (0.73–1.88) 1 (reference)

Abbreviations: ApoB = apolipoprotein B; HDL-C = high-density lipoprotein cholesterol; HR = hazard ratio; Lp(a) = lipoprotein a; LDL-C = low-density lipoprotein cholesterol.

^aVariables expressing lipoprotein marker concentrations have been natural log-transformed and subsequently converted to z-scores.

^bResults from Cox proportional hazard models with all-cause dementia as the outcome and the respective lipoprotein marker (in a continuous or quartile form) as the main predictor. Models are adjusted for baseline age and sex.

^cBinary indicators constructed for associations between lipoprotein marker quartiles and all-cause dementia incidence exhibiting a nonlinear pattern.

Results from analyses further adjusted for educational attainment, vascular risk factors, and APOE genotype are presented in eTable 3. In models with lipoprotein markers as continuous variables, a 1-SDU increase in ln(ApoB48) concentration was associated with a 19% decrease in the risk of incident AD. In models with lipoprotein markers as quartiles, the relationship of sdLDL-C concentrations with AD incidence exhibited a nonlinear pattern. Specifically, participants with sdLDL-C concentrations below the median were 35% less likely to develop AD, compared with those with sdLDL-C concentrations above the median. Repeating the primary analyses in participants with no missing data on educational attainment, vascular risk factors, or APOE ε4 status yielded similar findings (eTable 4). Sources of missingness are provided in eTable 5.

After excluding participants who developed interim stroke (n = 93), associations between lipoprotein markers as continuous variables and incident AD were similar to those observed in primary analyses (eTable 6). Findings were also similar in models with lipoprotein markers as quartiles, except for the higher AD risk among participants with the lowest HDL-C concentrations compared with the remaining sample, which, although in the same direction, did not remain statistically significant.

Discussion

In this investigation of older, community-dwelling individuals without dementia, we studied the associations of several blood lipoprotein makers previously linked to cardiovascular risk with AD incidence. Lower sdLDL-C and higher ApoB48 concentrations were associated with a reduced risk of incident AD. In addition, individuals with HDL-C concentrations below the 25th percentile were less likely to develop AD compared with the remaining sample. These findings further highlight the physiologic link between cardiovascular health and AD risk and could contribute to the development of blood lipoprotein profiles for AD risk stratification.

We observed that participants in the lowest 25% of HDL-C distribution were 44% less likely to develop AD compared with the remaining sample. This association was not significant in analyses with all-cause dementia as the outcome, or in models further adjusted for educational attainment, vascular risk factors, and APOE genotype. However, the latter was likely due to the lower power of the fully adjusted models, as indicated by the similar results obtained from primary models after excluding participants with missing data on covariates. Studies examining the relationship between HDL-C and AD incidence in other populations have yielded inconsistent findings. For instance, in a previous investigation including more than 111,000 Danish individuals, participants with the highest HDL-C levels were more likely to develop AD.³⁰ Similarly, in a post hoc analysis of clinical trial data from 18,688 community-dwelling older adults in Australia and the United States, high HDL-C concentrations were associated with an increased risk of all-cause dementia.³¹ In another study analyzing electronic health record and survey data of more than 184,000 participants, both individuals with the highest and those with the lowest HDL-C levels were more likely to develop dementia (defined as AD, vascular dementia, or nonspecific dementia), suggesting a U-shaped relationship.³² Conversely, in approximately 470,000 participants from the UK Biobank (UKB), neither a linear nor a nonlinear association between HDL-C levels and AD incidence was identified,²² while in 4,932 participants from the Offspring Cohort of FHS, higher HDL-C concentrations were associated with a decreased risk of incident AD in early and middle adulthood, but not in individuals older than 60 years.³³

The mechanisms through which higher HDL-C levels might increase AD risk in older adults are unclear. Aging is accompanied by an increase in cellular production of proinflammatory cytokines and mediators, leading to the institution of a chronic low-grade systemic inflammatory state, commonly referred to as inflammaging.³⁴ Although normal HDL is an effective antioxidant and anti-inflammatory molecule, in inflammatory microenvironments HDL proteins and lipids undergo oxidative and enzymatic modifications. These modifications may not only lead to ablation of its anti-inflammatory properties but also transform HDL into a proinflammatory particle,^35,36 potentially perpetuating aging-related proinflammatory phenotypes that have been implicated in the pathogenesis of AD.³⁷ On the contrary, higher HDL-C concentrations during midlife might offset part of the risk of developing AD later in life, potentially through protection from vascularly driven neurodegenerative pathways.⁷

Although a meta-analysis of 26 case-control studies revealed higher levels of LDL-C in individuals with AD than in dementia-free controls,¹⁸ findings from prospective investigations have been conflicting. In participants aged 65 years or older from the WHICAP study, there was no association between LDL-C levels and AD incidence over a mean follow-up of 4.8 years.²¹ Similarly, in approximately 470,000 UKB participants, LDL-C levels were not associated with AD incidence.²² On the contrary, in a nationwide population-based study of Korean individuals aged 40 years or older with a median follow-up of 8.33 years, compared with participants in the third LDL-C quintile, those in the first quintile exhibited the highest risk of developing incident AD, followed by individuals in the fifth quintile, suggesting the presence of a U-shaped relationship.³⁸ In our analysis, similar to the WHICAP and UKB studies, we did not observe an association between LDL-C concentrations and AD incidence in the total sample. The detection of an association in Korean individuals but not in other samples might be related to the large sample size of that study (over 6,883,000 participants), or to the different population characteristics. Another possibility is that sex could act as an effect modifier in the relationship between LDL-C levels and AD incidence. In stratified analysis, we observed that women in the lower half of LDL-C distribution were 36% less likely to develop incident AD than those in the higher half, but no association was present in men or the total sample.

Although higher sdLDL-C has been associated with accelerated cognitive decline in participants with a mean age of 63 years from the Atherosclerosis Risk in Communities (ARIC) study,³⁹ to our knowledge, its association with AD incidence has not been previously explored. In our sample, higher sdLDL-C concentrations were associated with an increased risk of incident AD. Quartile analyses also revealed that compared with individuals with sdLDL-C concentrations above the median, those with concentrations below the median had a 38% lower risk of incident AD.

Unequivocal evidence from clinical and genetic studies has causally linked LDL-C to cardiovascular disease.⁴⁰ Similarly, high blood levels of sdLDL-C have been consistently associated with an atherogenic lipid profile and an increased risk of incident coronary heart disease.¹² At the same time, clinicopathologic data indicate that vascular pathology frequently coexists with AD pathology and might have additive or synergistic effects on cognitive decline.⁶ Therefore, the relationships between LDL particles and AD incidence might be mediated by vascular pathways. In line with this, the association of sdLDL-C with cognitive decline in ARIC participants was driven by an accelerated decline in executive function, a domain commonly affected in vascular contributions to cognitive impairment and dementia.⁴¹ Finally, epidemiologic data also support this vascular mediation hypothesis, as they indicate a downward trend of AD incidence in the United States and other high-income countries that has at least partially been attributed to better management of cardiovascular risk factors.^1,3

In ARIC participants, compared with individuals in the lowest ApoB quintile, those in the highest one exhibited greater cognitive decline, which, similar to the sdLDL-C associations, was primarily driven by an accelerated decline in executive function test scores.³⁹ On the contrary, in participants from the Sydney Memory and Ageing Study, ApoB levels were not associated with baseline cognitive measures nor with cognitive decline.⁴² Associations between ApoB levels and AD incidence have only been examined in the UKB, where higher ApoB concentrations were associated with an increased risk of incident AD.²² In our analysis, ApoB concentrations were not associated with AD incidence in the total sample. Differences in population characteristics, sample size, or the age distribution of study participants might be responsible for these incongruent findings. Specifically, UKB participants had a younger mean age than those included in our analysis. It has been demonstrated that late-life dementia may be preceded by weight loss several years before the diagnosis.⁴³ Therefore, potential associations between lipoprotein markers and dementia incidence in younger individuals might not be reproducible later in life because weight changes that could in turn influence lipid profiles might have already been instated. Sex could also be an effect modifier in the relationship between ApoB and AD incidence. In stratified analyses, higher ApoB concentrations were associated with an increased AD risk in our women subsample. Similarly, sex-stratified analyses in UKB participants revealed that associations between ApoB and all-cause dementia incidence were only present in women.²²

We identified a relationship between higher blood ApoB48 concentrations and decreased AD risk that remained significant in all sensitivity analyses. In stratified analyses, this relationship was only present in women; however, the effect size was actually larger in men, suggesting that the nonsignificant result was likely secondary to lower power in the men subsample. To our knowledge, the relationship between ApoB48 and AD incidence has not been previously studied.

Trapping of ApoB particles within the arterial wall is the fundamental step that initiates and drives atherosclerosis, with ApoB concentration within the arterial lumen being the main determinant of the number of ApoB particles that will be trapped.¹³ It is now recognized that LDL-C predicts the atherogenic risk of lipoproteins because it strongly correlates with the ApoB particle number,⁴⁴ and ApoB, as the direct determinant, is a better predictor of atherogenic risk than LDL-C or non-HDL-C. Therefore, the mechanisms through which ApoB might be linked to AD risk could be similar to those of LDL particles, potentially involving atherogenic and vascular injury–related pathways. However, our finding that ApoB48 might confer protection against AD poses new biophysiologic questions and queries the unidimensionality of ApoB influences on AD risk. To contextualize, ApoB exists in 2 isoforms: the full-length ApoB100, consisting of 4,536 amino acids, and ApoB48, a truncated form consisting of the 2,152 N-terminal amino acids. Although both are encoded by the same gene (APOB gene), these isoforms play distinct physiologic roles. While ApoB100 is mainly synthesized and expressed in the liver, constituting an integral component of very low–density lipoprotein, intermediate-density lipoprotein, and LDL particles, ApoB48 is primarily synthesized and expressed within the intestine and is present in chylomicrons and their remnants.⁴⁵ Therefore, ApoB100 might be more clinically relevant in determining the level of circulating atherogenic lipoproteins and is also the isoform measured by ApoB assays; although these assays recognize both ApoB48 and ApoB100, there are very few circulating ApoB48 particles at any time, even in postprandial blood samples.¹³ In mice, it has been shown that overexpression of recombinant human ApoB48 leads to decreased secretion of the ApoB100 isoform due to decreased synthesis at steady state.⁴⁶ Therefore, relative overexpression of ApoB48 might lead to decreased ApoB100 concentrations. This hypothesis is also supported by the fact that in familial hypobetalipoproteinemia heterozygotes, where the average concentrations of plasma ApoB100 would be expected to be approximately 50% of normal, the levels are actually closer to approximately 25% of normal.⁴⁷ Consequently, increased expression of ApoB48 might confer protection against AD through decreasing ApoB100 secretion and, in turn, mitigating its vascular contributions to AD risk. Nevertheless, mechanistic studies are necessary to precisely delineate the underlying biology of these findings.

Previous studies exploring associations of Lp(a) levels with cognitive performance have yielded variable findings, with reports of positive,³⁹ negative,⁴⁸ and null associations.²⁰ In line with previous investigations examining the association with AD incidence, or differences in Lp(a) levels between individuals with AD and age-matched controls,^22,49 we did not find evidence of an association between Lp(a) levels and AD risk in the total sample nor in the sex subgroups.

Stratified analyses revealed that, with the exception of ApoB48, effect sizes for the observed relationships between lipoprotein levels and AD incidence were larger in women compared with men or the total sample. Furthermore, associations of LDL-C and ApoB concentrations with AD incidence were only observed in women, despite the smaller sample. Although lower power in the men subsample might have contributed to these findings, sex could be an effect modifier in the relationships between lipoprotein levels and AD incidence. This hypothesis would also be plausible from a physiologic standpoint because the women included in our analysis were postmenopausal; thus, their lipoprotein profiles could have been affected by menopausal-related changes. For example, after menopause, LDL-C levels rise, frequently exceeding those of age-matched men, with a shift to smaller, denser, and potentially more atherogenic particle sizes, such as sdLDL-C.⁵⁰ These unfavorable lipoprotein profile changes have been implicated in the increased cardiovascular disease risk observed in postmenopausal women⁵⁰ and could theoretically accelerate AD pathology through vascular injury–related pathways.⁷

Although the prospective, community-based design and the rigorous dementia surveillance and diagnostic algorithms of FHS instill confidence in our findings, certain limitations need to be considered. First, despite a median follow-up of 12.6 years, reverse causality cannot be entirely excluded because weight changes that could potentially affect lipid profiles might start up to 20 years before the development of dementia.⁴³ Second, there was an interval of approximately 20 years between blood collection and lipoprotein concentration measurements. While rigorous blood sample preprocessing and storage protocols to ensure preservation of biospecimen integrity over extended storage periods were followed, increased measurement error that could reduce statistical power and potentially bias certain results toward the null is a possibility. Third, lipoprotein marker measurements were performed at a single time point, limiting our ability to assess their temporal stability within the context of this study. Fourth, this analysis included only individuals of White European ancestry, and further studies are necessary to confirm the generalizability and external validity of the findings in more racially and ethnically diverse samples. Fifth, false positives are possible because of multiple testing; hence, results with CI bounds close to one should be interpreted with caution. Finally, similar to all observational studies, findings are subject to residual confounding; future studies with designs that allow for causal inferences (i.e., Mendelian randomization and randomized clinical trials) are needed to confirm the presence and directionality of the observed relationships.

To summarize, in a community-based sample of older adults without dementia, lower sdLDL-C and higher ApoB48 concentrations were associated with a reduced risk of incident AD. In addition, individuals with the lowest HDL-C concentrations (below the 25th percentile) were less likely to develop AD compared with the remaining sample. With the exception of ApoB48, effect sizes were greater in women. Our findings could inform public health policies and guide future targeted lipid modification strategies—through pharmacotherapy and/or lifestyle interventions—for populations at risk. They can also aid in the construction and improvement of multimodal AD risk stratification algorithms.

Glossary

AD

Alzheimer disease

ApoB

apolipoprotein B

ARIC

Atherosclerosis Risk in Communities

FHS

Framingham Heart Study

HDL-C

high-density lipoprotein cholesterol

HR

hazard ratio

LDL-C

low-density lipoprotein cholesterol

Lp(a)

lipoprotein a

MMSE

Mini-Mental State Examination

SBP

systolic blood pressure

sdLDL-C

small dense LDL-C

SDU

standard deviation unit

UKB

UK Biobank

WHR

waist-to-hip ratio

Author Contributions

S. Charisis: drafting/revision of the manuscript for content, including medical writing for content; study concept or design; analysis or interpretation of data. S. Lu: drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data. J.D. Melgarejo: drafting/revision of the manuscript for content, including medical writing for content. C.L. Satizabal: drafting/revision of the manuscript for content, including medical writing for content. R.S. Vasan: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design. A.S. Beiser: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; analysis or interpretation of data. S. Seshadri: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design.

Study Funding

This work was funded by the National Heart Lung and Blood Institute (Framingham Heart Study Contracts N01-HC-25195, HHSN268201500001I, and 75N92019D00031), the Boston University School of Medicine, and grants from the National Institute on Aging (R01 AG054076, R01 AG049607, U01 AG052409, R01 AG059421, RF1 AG063507, RF1 AG066524, U01 AG058589), the National Institute of Neurological Disorders and Stroke (R01 NS017950, UH2 NS100605, UF1 NS125513), and the National Institute of Diabetes and Digestive and Kidney Diseases (R01 DK080739).

Disclosure

C.L. Satizabal is supported by a New Investigator Research Grant to promote Diversity from the Alzheimer's Association (AARGD-16-443384) and also receives support from R01 AG059727. C.L. Satizabal and S. Seshadri are partially supported by the South Texas Alzheimer's Disease Research Center (P30 AG066546). S. Seshadri receives support from The Bill and Rebecca Reed Endowment for Precision Therapies and Palliative Care, and from the Barker Foundation as the Robert R Barker Distinguished University Professor of Neurology, Psychiatry and Cellular and Integrative Physiology. S. Charisis, S. Lu, J.D. Melgarejo, R.S. Vasan, and A.S. Beiser report no disclosures relevant to the present manuscript. The content of this article is solely the responsibility of the authors and does not necessarily represent the official views of the NIH. Go to Neurology.org/N for full disclosures.