Subjective cognitive decline, APOE e4 allele, and the risk of neurocognitive disorders: age- and sex-stratified cohort study

Abstract

OBJECTIVE:

Subjective cognitive decline (SCD) and APOE e4 allele (APOE4) are known predictors of mild cognitive impairment and dementia (MCI/dementia), with recent evidence showing interaction between SCD and APOE4 in amplifying the risk of MCI/dementia. However, the literature is unclear whether the interaction effect is seen across various age and sex strata. This study examined the interaction between SCD and APOE4 – across different age and sex strata – on the risk of MCI/dementia.

METHODS:

This cohort study included 16,221 participants aged ≥50 years and had normal cognition at baseline. Participants were evaluated for SCD and APOE4 at baseline, and followed-up almost annually for MCI/dementia (median follow-up=4.5 years). Interaction effects were examined in Cox regression using Relative Excess Risk due to Interaction (RERI), stratified by age (≤70 versus >70 years) and sex.

RESULTS:

SCD and APOE4 were independently associated with MCI/dementia (HR 1.4–1.8), with the highest risk when SCD and APOE4 co-occurred (HR 2.6). APOE4 amplified the association between SCD and MCI/dementia in older women (RERI 1.0; 95% CI 0.3–1.6), but not in other age or sex strata. Among older women, half of them developed MCI/dementia by 12.1 years in the absence of SCD or APOE4. This duration shortened to 8.1–10.3 years in the presence of either SCD or APOE4, and to 4.4 years in the presence of both SCD and APOE4. Interaction effect among older women remained consistent when alternate outcomes were used (i.e. MCI and dementia due to Alzheimer’s disease; dementia; and Alzheimer’s dementia) (RERI 1.2–2.5).

CONCLUSIONS:

APOE4 amplifies the association between SCD and neurocognitive disorders in older women, with the findings suggesting the need for further research to delineate underlying neurobiology. APOE4 may potentially have a role in facilitating further risk stratification of older women with SCD in clinical practice.

INTRODUCTION

Subjective cognitive decline (SCD) refers to the subjective perception of a decline in cognition (typically in the memory domain) among individuals with normal cognition (i.e. in the absence of objective cognitive deficits) (Jessen et al., 2014a; Jessen et al., 2020). It is increasingly common at older ages, with a reported prevalence of 50–80% among community-dwelling older persons (Jessen et al., 2020). In clinical practice, SCD can often present concurrently with depressive or anxiety symptoms (Reid and Maclullich, 2006; Jessen et al., 2014a; Hill et al., 2016), and is not uncommonly the chief complaint of older persons with depression (Reid and Maclullich, 2006); this has led to prior uncertainty on whether SCD is merely an alternate manifestation of depressive or anxiety disorders in older persons (Reid and Maclullich, 2006). However, in recent literature, SCD has been shown to represent a disease process that is distinct from those of depressive or anxiety disorders, with studies demonstrating the independent role of SCD in predicting neurocognitive disorders (Liew, 2019b; Liew, 2020c; Liew, 2020d). Given these recent evidence, SCD has been suggested to be useful in the diagnosis of prodromal neurocognitive disorders (Jessen et al., 2014a; Jessen et al., 2020), and has most recently been incorporated into the 2018 NIA-AA (National Institute on Aging–Alzheimer’s Association) research criteria for Alzheimer’s disease (Jack et al., 2018) as a transition phase between normal cognition and early neurocognitive disorders.

Among individuals with SCD, APOE e4 allele (APOE4) has been shown to be among the strongest predictors for conversion to neurocognitive disorders (Jessen et al., 2014b; Hong et al., 2015), given the detrimental effects of APOE4 on blood-brain barrier function, myelination, and neural repair (Chew et al., 2020). At the same time, APOE4 carriers have also been shown to be more likely to report SCD (Krell-Roesch et al., 2015). Considering such close link between SCD and APOE4, it is not surprising that recent evidence has demonstrated the interaction effect between SCD and APOE4, whereby individuals with both SCD and APOE4 showed greater cognitive decline than those with either SCD or APOE4 alone (Dik et al., 2001). However, extant literature is not yet clear on whether the interaction effect between SCD and APOE4 is present across all demographic strata. A recent study (Muller-Gerards et al., 2019) provided suggestive evidence that the interaction effect was mainly seen in women and not in men, but findings from this study remained inconclusive due to a relatively smaller study sample, with non-significant p-values being reported despite large magnitude of interaction effects in women. In addition, there has not been any study that examined the interaction effect of SCD and APOE4 across different age strata.

A clear understanding of the interaction effect across the demographic strata of age and sex can have pertinent implications to clinical and research practice. It may allow clinicians to have a clearer understanding on the usefulness of SCD and APOE4 – especially when both are used in combination and in the relevant demographic strata – in predicting neurocognitive disorders. It may also compel researchers to further clarify the biological relationship between SCD and APOE4 in specific demographic strata, which can enrich our understanding on the biological progression of neurocognitive disorders especially in the context of the different demographic groups. Using a large sample and a cohort study design, this study sought to provide more definitive evidence on the interaction effect of SCD and APOE4 – across different age and sex strata – on the risk of developing mild cognitive impairment and dementia.

METHODS

Study population

This study involves secondary analysis of the National Alzheimer’s Coordinating Center (NACC) database. The data were obtained with permission from NACC, using the data request form as available at https://naccdata.org/. Data in the NACC database were based on participants recruited from approximately 41 Alzheimer’s Disease Centers (ADCs) across the United States between 2005 and August 2021. Majority of the participants (86.6%) visited the ADCs primarily to volunteer in research, while the rest visited ADCs to seek clinical evaluation. On an approximately annual basis, the participants took part in standardized assessments (which included clinical history, physical examination and detailed neuropsychological testing) to evaluate for incident mild cognitive impairment (MCI) and dementia. The study included participants who fulfilled the following criteria at baseline: (1) aged ≥50 years; (2) diagnosed as having normal cognition at baseline (i.e. participants had completed diagnostic evaluations and found not to have MCI or dementia); and (3) provided information on SCD and APOE4. All contributing ADCs obtained informed consent from their participants, as well as received approval by their local institutional review boards.

Measures

The exposure-of-interest in this study was defined using four mutually exclusive levels, based on the presence of SCD and APOE4: (1) SCD absent and APOE4 noncarrier; (2) SCD absent and APOE4 carrier; (3) SCD present and APOE4 noncarrier; and (4) SCD present and APOE4 carrier. SCD was evaluated with a single yes/no question based on whether the participant perceived “a decline in memory relative to previously attained abilities”. The focus on memory domain is not inconsistent with current evidence in the literature, particularly in the recently proposed SCD framework, where memory concerns have been suggested to demonstrate better likelihood (than other non-memory concerns) in detecting prodromal neurocognitive disorders (Jessen et al., 2014a). APOE4 carrier was defined by individuals who have at least 1 copy of APOE4 allele. APOE genotyping was conducted independently by respective ADCs and reported to NACC using a standardized form. Where available, information on APOE genotype was also obtained from 2 additional sources, namely the Alzheimer’s Disease Genetics Consortium and the National Cell Repository for Alzheimer’s Disease.

Several baseline covariates were also captured in this study, namely, age, sex, ethnicity, years of education, current smoking, untreated hearing loss [i.e. hearing loss without the use of hearing aid], obesity, hypertension, hyperlipidemia, diabetes mellitus, history of traumatic brain injury, history of stroke, Geriatric Depression Scale (GDS) (Yesavage and Sheikh, 2008), and anxiety symptoms. These covariates are briefly described here. Information on age, sex, ethnicity, years of education, hearing loss and current smoking were obtained from participants and informants. Obesity was defined by Body Mass Index (i.e. weight in kilograms / squared of height in meters) of ≥30, using participants’ weight and height that were measured at baseline. Presence of hypertension, hyperlipidemia, diabetes mellitus, traumatic brain injury and stroke were determined based on subjective reports by participants or informants, or as assessed by clinicians after taking in account all available information (i.e. participant/informant report, medication use, physical examination and medical records). GDS (Yesavage and Sheikh, 2008) assesses the level of depressive symptoms over the past week using 15 yes/no questions. Responses are summed to produce a total score, with higher scores indicating higher levels of depressive symptoms. Anxiety symptoms were evaluated with a single yes/no question based on whether the participants have experienced “any signs of nervousness such as shortness of breath, sighing, being unable to relax, or feeling excessively tense” in the past month.

The diagnoses of MCI and dementia were made based on all available information from standardized assessments, with 72.5% made via consensus conference and the remainder made by single clinicians. MCI was diagnosed using the modified Petersen criteria (Petersen and Morris, 2005); which in essence, involves the presence of cognitive concerns, objective cognitive deficits, and normal functional activities. Dementia was diagnosed using the 2011 NIA-AA criteria (McKhann et al., 2011) from March 2015 onwards (when version 3.0 of the Uniform Data Set was introduced in NACC); while before March 2015, the diagnostic criteria for dementia were not specified in the NACC database, with some ADCs using the NINCDS-ADRDA (National Institute of Neurological and Communicative Disorders and Stroke–Alzheimer’s Disease and Related Disorders Association) criteria (McKhann et al., 1984), DSM-IV (Diagnostic and Statistical Manual of Mental Disorders–Fourth Edition) criteria (American Psychiatric Association, 2000) or other standard criteria for dementia. In essence, the diagnosis of dementia requires the presence of cognitive concerns, objective cognitive deficits, and impaired functional activities. The primary aetiology of Alzheimer’s disease was further identified from participants with MCI and dementia when:

1. Participants fulfilled NINCDS-ADRDA criteria for possible or probable Alzheimer’s disease (McKhann et al., 1984), or 2011 NIA-AA criteria for probable Alzheimer’s disease (McKhann et al., 2011; Albert et al., 2011); and

2. Alzheimer’s disease was deemed by clinicians to be the primary cause of the observed cognitive impairment.

Statistical analyses

In the primary analyses, Cox proportional hazard regression (Liew, 2021) was conducted to evaluate the risk of MCI and dementia. Time-to-event was defined as the duration from baseline to the diagnosis of either MCI or dementia. Cox regression was conducted in the overall sample, as well as in four different age- and sex-stratified samples: (1) Men ≤70 years; (2) Women ≤70 years; (3) Men >70 years; and (4) Women >70 years. Of note, the median age of the study sample was used to split the participants into 2 equal-sized age-groups (i.e. ≤70 years and >70 years). Cox regression adjusted for demographic information (age, sex and ethnicity), established risk factors of neurocognitive disorders (years of education, current smoking, untreated hearing loss [i.e. hearing loss without the use of hearing aid], obesity, diabetes mellitus, hypertension, hyperlipidemia, history of traumatic brain injury, and history of stroke) (Livingston et al., 2020), as well as potential confounders (Rothman, 2012) that may predict both the exposure-of-interest (SCD) and the outcome-of-interest (neurocognitive disorders), namely, GDS score and presence of anxiety symptoms. GDS score was included as a potential confounder because depressive symptoms are strongly associated with SCD (exposure-of-interest) (Reid and Maclullich, 2006; Jessen et al., 2014a), while at the same time, the presence of depressive symptoms has been shown to be an independent predictor of neurocognitive disorders (outcome-of-interest) (Diniz et al., 2013; Liew, 2019b). Similarly, anxiety symptoms are known to correlate with SCD (exposure-of-interest) (Hill et al., 2016), and the presence of anxiety symptoms has also been reported to predict neurocognitive disorders (outcome-of-interest) (Gulpers et al., 2016; Liew, 2020c). Further details on the conduct of Cox regression are available in Supplementary Material 1. The risk estimates from Cox regression were reported as hazard ratios (HR), as well as presented in Kaplan-Meier curves.

Interaction analyses between SCD and APOE4 were conducted, by computing the Relative Excess Risk due to Interaction (RERI) (Rothman, 2012; VanderWeele and Knol, 2014). RERI is a measure of biological (additive) interaction between two exposures (i.e. SCD and APOE4)(Rothman, 2012; VanderWeele and Knol, 2014). It reflects the excess risk for individuals with both exposures which is not explained by the independent effects of either exposure(Rothman, 2012; VanderWeele and Knol, 2014). In general, RERI can be interpreted as follows: RERI=0 implies no interaction between SCD and APOE4; RERI>0 indicates possible biological interaction; while RERI>1 implies stronger evidence of biological interaction (VanderWeele and Knol, 2014). Computation of RERI was done using the following formula (Rothman, 2012; VanderWeele and Knol, 2014), based on the HR from Cox regression:

$$\mathit{\text{RERI}}=H{R}_{SCD\&\mathit{\text{APOE}}4}–H{R}_{SCD\mathit{\text{only}}}–H{R}_{\mathit{\text{APOE}}4\mathit{\text{only}}}+\mathit{1}$$

Five sensitivity analyses were conducted to evaluate the robustness of the results when some parts of the primary analyses were modified. They included:

1. using the alternate endpoint of MCI and dementia due to the primary aetiology of Alzheimer’s disease, given that APOE4 (the subject of current study) is a known risk factor for Alzheimer’s disease;

2. using the alternate endpoint of dementia, instead of the composite endpoint of mild cognitive impairment or dementia;

3. using the alternate endpoint of dementia due to the primary aetiology of Alzheimer’s disease, given that APOE4 (the subject of current study) is a known risk factor for Alzheimer’s disease;

4. adjusting additionally for the global Z-score of neuropsychological tests at baseline. Further descriptions on the neuropsychological tests are available in Supplementary Material 2;

5. adjusting for neuropsychiatric symptoms at baseline (using the total score of Neuropsychiatric Inventory–Questionnaire), instead of depressive and anxiety symptoms in the main analyses. Further description on Neuropsychiatric Inventory–Questionnaire is available in Supplementary Material 2.

All analyses were conducted in Stata (version 16).

RESULTS

The total sample size was 16,221, with a median age of 71 (interquartile range, IQR 65–77) and a median education of 16 years (IQR 14–18). Figure 1 presents the flow diagram related to participant selection, while Table 1 shows the participant characteristics as well as the comparison between participants who did and did not develop MCI or dementia. One-fifth of the participants (21.4%) only had baseline data and did not contribute to the follow-up data, while the rest of the participants had a median duration of follow-up of 4.5 years (IQR 2.2–8.0 years). At baseline, 4,443 (27.4%) participants reported the presence of SCD and 3,937 (24.3%) were APOE4 carriers. During the period of follow-up, 1,894 (11.7%) participants developed MCI (of which 963 or 50.8% had presumptive aetiology of Alzheimer’s disease); while 956 (5.9%) developed dementia (of which 747 or 78.1% had Alzheimer’s dementia). Among participants who had presumptive aetiology of Alzheimer’s disease (n=1,710), 997 (58.0%) were based on NINCDS-ADRDA criteria, while 719 (42.1%) were based on 2011 NIA-AA criteria.

Figure 1.

Participant enrolment and exclusion details.

NACC, National Alzheimer’s Coordinating Center; MCI, mild cognitive impairment; NC, normal cognition; SCD, subjective cognitive decline.

Table 1.

Demographic information of the study participants at baseline (n=16,221), and comparison between those did and did not develop mild cognitive impairment and dementia during the follow-up period.

Variable	Overall sample (n=16,221)	Participants who did not develop MCI or dementia (n=13,371)	Participants who developed MCI or dementia (n=2,850)	p valuea
Age, median (IQR)	71 (65–77)	70 (65–76)	76 (70–82)	<0.001
Years of education, median (IQR)	16 (14–18)	16 (14–18)	16 (13–18)	<0.001
Male sex, n (%)	5,542 (34.2)	4,456 (33.3)	1,086 (38.1)	<0.001
Ethnicity, n (%)				<0.001
White	12,609 (77.7)	10,277 (76.9)	2,332 (81.8)
African American	2,387 (14.7)	2,032 (15.2)	355 (12.5)
Others/Unknown	1,225 (7.6)	1,062 (7.9)	163 (5.7)
Current smoker, n (%)	660 (4.1)	560 (4.2)	100 (3.5)	0.096
Untreated hearing loss, n (%)	1610 (9.9)	1288 (9.6)	322 (11.3)	0.007
Obesity, n (%)	4,135 (25.5)	3,518 (26.3)	617 (21.6)	<0.001
Diabetes mellitus, n (%)	2,029 (12.5)	1,673 (12.5)	356 (12.5)	0.980
Hypertension, n (%)	7,965 (49.1)	6,448 (48.2)	1,517 (53.2)	<0.001
Hyperlipidemia, n (%)	8,110 (50.0)	6,669 (49.9)	1,441 (50.6)	0.510
History of traumatic brain injury, n (%)	1,833 (11.3)	1,548 (11.6)	285 (10.0)	0.016
History of stroke, n (%)	1,042 (6.4)	787 (5.9)	255 (8.9)	<0.001
GDS score, median (IQR)	1 (0–2)	1 (0–2)	1 (0–2)	<0.001
Presence of anxiety symptoms, n (%)	1,448 (8.9)	1,144 (8.6)	304 (10.7)	<0.001
Presence of SCD, n (%)	4,443 (27.4)	3,428 (25.6)	1,015 (35.6)	<0.001
APOE4 carrier, n (%)	3,937 (24.3)	3,049 (22.8)	888 (31.2)	<0.001

MCI, mild cognitive impairment; IQR, interquartile range; GDS, Geriatric Depression Scale; APOE4, APOE e4; SCD, subjective cognitive decline.

^aTest of difference between participants who did and did not develop mild cognitive impairment and dementia: chi-square test for categorical variables, and Mann-Whitney U test for continuous variables. Bold-faced p-values are ≤0.05.

Results from the primary analyses are presented in Table 2. In the overall sample, risk of MCI and dementia was associated with the presence of either SCD or APOE4 alone (HR 1.4–1.8), and became much higher in the presence of both SCD and APOE4 (HR 2.6). There was suggestive evidence of interaction effect between SCD and APOE4, with a RERI of 0.4 (95% CI 0.0–0.8). The differential risks are also reflected in the survival time. Among individuals without SCD or APOE4, half of them would have developed MCI or dementia by 14.5 years. This duration shortened to 10.4–13.0 years in the presence of either SCD or APOE4 alone, and 8.1 years in the presence of both SCD and APOE4.

Table 2.

Risk of mild cognitive impairment and dementia, based on the overall sample as well as stratified by age and sex (n=16,221).

Different combinations of presentation	No. of MCI and dementia / Total (%)	Risk of MCI and dementia		Effect modification		Survival (50th centile) in years (95% CI)c
		Hazard ratio (95% CI)a	P value	RERI (95% CI)b	P value
(a) Overall sample
SCD absent and APOE4 noncarrier	1288 / 8972 (14.4)	1.0 (Ref)	Ref			14.5 (13.8, 15.2)
SCD absent and APOE4 carrier	547 / 2806 (19.5)	1.4 (1.3, 1.6)	<0.001			13.0 (11.6, 14.4)
SCD present and APOE4 noncarrier	674 / 3312 (20.4)	1.8 (1.6, 2.0)	<0.001			10.4 (9.5, 11.2)
SCD present and APOE4 carrier	341 / 1131 (30.2)	2.6 (2.3, 3.0)	<0.001	0.4 (0.0, 0.8)	0.030	8.1 (6.8, 9.4)
(b) Men ≤ 70 years
SCD absent and APOE4 noncarrier	91 / 1300 (7.0)	1.0 (Ref)	Ref			Not availabled
SCD absent and APOE4 carrier	72 / 524 (13.7)	1.7 (1.3, 2.4)	0.001			14.3 (13.8, 14.8)
SCD present and APOE4 noncarrier	68 / 455 (15.0)	2.1 (1.5, 3.0)	<0.001			14.2 (12.6, 15.7)
SCD present and APOE4 carrier	46 / 196 (23.5)	3.2 (2.1, 4.7)	<0.001	0.3 (−0.9, 1.5)	0.613	10.5 (7.8, 13.2)
(c) Women ≤ 70 years
SCD absent and APOE4 noncarrier	181 / 2840 (6.4)	1.0 (Ref)	Ref			Not availabled
SCD absent and APOE4 carrier	120 / 1051 (11.4)	1.4 (1.1, 1.8)	0.003			15.1 (14.6, 15.5)
SCD present and APOE4 noncarrier	118 / 1095 (10.8)	2.0 (1.6, 2.6)	<0.001			Not availabled
SCD present and APOE4 carrier	71 / 421 (16.9)	2.5 (1.8, 3.4)	<0.001	0.0 (−0.8, 0.8)	0.949	14.5 (11.8, 17.2)
(d) Men > 70 years
SCD absent and APOE4 noncarrier	393 / 1724 (22.8)	1.0 (Ref)	Ref			10.9 (8.8, 13.1)
SCD absent and APOE4 carrier	123 / 451 (27.3)	1.1 (0.9, 1.3)	0.329			10.0 (8.1, 11.9)
SCD present and APOE4 noncarrier	199 / 681 (29.2)	1.7 (1.4, 2.0)	<0.001			6.5 (5.3, 7.6)
SCD present and APOE4 carrier	94 / 211 (44.6)	2.0 (1.6, 2.6)	<0.001	0.3 (−0.3, 0.8)	0.334	5.7 (4.1, 7.3)
(e) Women > 70 years
SCD absent and APOE4 noncarrier	623 / 3108 (20.1)	1.0 (Ref)	Ref			12.1 (11.1, 13.0)
SCD absent and APOE4 carrier	232 / 780 (29.7)	1.6 (1.4, 1.9)	<0.001			10.3 (9.0, 11.6)
SCD present and APOE4 noncarrier	289 / 1081 (26.7)	1.7 (1.5, 2.0)	<0.001			8.1 (6.5, 9.7)
SCD present and APOE4 carrier	130 / 303 (42.9)	3.3 (2.7, 4.0)	<0.001	1.0 (0.3, 1.6)	0.003	4.4 (3.8, 5.0)

MCI, mild cognitive impairment; CI, confidence interval; RERI, Relative excess risk due to interaction; SCD, subjective cognitive decline; APOE4, APOE e4; Ref, reference group.

^aModel adjusted for baseline variables of age, sex and ethnicity, years of education, current smoking, untreated hearing loss, obesity, diabetes mellitus, hypertension, hyperlipidemia, history of traumatic brain injury, history of stroke, total score on Geriatric Depression Scale, and presence of anxiety symptoms. Significant estimates (with p≤0.05) are highlighted in bold.

^bIn general, RERI=0 implies no interaction between SCD and APOE4; RERI>0 indicates possible biological interaction; while RERI>1 implies stronger evidence of biological interaction. Significant RERI (with p≤0.05) are highlighted in bold.

^cThe estimated time that is needed for half of the participants to develop MCI or dementia. The 95% CI was computed with 1000 bootstrap sampling.

^dNot available, because less than half of participants in this group developed the outcome by the end of the follow-up period.

The results remained largely similar in the 4 subgroups stratified by age and sex (Table 2). However, it became apparent that the interaction effect between SCD and APOE4 was primarily seen in older women >70 years and not in the other age or sex strata. Notably, the magnitude of RERI now became larger in women >70 years (RERI 1.0; 95% CI 0.3–1.6). This interaction effect in older women can be visible in the Kaplan-Meier curve in Figure 2. Among women >70 years, half of them would have developed MCI or dementia by 12.1 years in the absence of SCD or APOE4. This duration shortened to 8.1–10.3 years in the presence of either SCD or APOE4 alone, and became as short as 4.4 years in the presence of both SCD and APOE4. The interaction effect can also be seen in the interaction plot in Figure 3, whereby the risk of SCD and APOE4 became amplified when both exposures were present concurrently in older women. In particular, the risk was in excess of the independent risks of either SCD or APOE4 alone, with the excess risk demarcated as RERI on the interaction plot.

Figure 2.

Kaplan-Meier curves reflecting the risk of mild cognitive impairment and dementia, based on the overall sample as well as stratified by age and sex.

SCD, subjective cognitive decline; APOE4, APOE e4; MCI, mild cognitive impairment.

Figure 3.

Interaction plot to demonstrate the excess risk due to interaction among older women >70 years, for the primary outcome of mild cognitive impairment and dementia.

SCD, subjective cognitive decline; APOE4, APOE e4; RERI, relative excess risk due to interaction; MCI, mild cognitive impairment.

In the first three sensitivity analyses, alternate outcomes were used focusing on: (1) MCI and dementia due to Alzheimer’s disease; (2) dementia; and (3) dementia due to Alzheimer’s disease. The results remained consistent, with RERI magnitude of >1 among older women (RERI 1.2–2.5) (Supplementary Material 3 to 5). In the fourth and fifth sensitivity analyses, the Cox regression included additional covariate adjustment of neuropsychological tests and neuropsychological symptoms. The results also remained similar, with the interaction effect primarily seen in older women >70 years and not in the other age or sex strata (Supplementary Material 6 to 7). The interaction effects among older women, based on results from the five sensitivity analyses, can also be visible in the interaction plots in Figure 4.

Figure 4.

Interaction plot to demonstrate the excess risk due to interaction among older women >70 years, based on results from the five sensitivity analyses.

SCD, subjective cognitive decline; APOE4, APOE e4; RERI, relative excess risk due to interaction; MCI, mild cognitive impairment.

DISCUSSION

Summary of findings

This study examined the interaction effect between SCD and APOE4 across different age and sex strata, using a large sample, a longitudinal study-design and after accounting for a wide range of potential confounders. SCD and APOE4 were independently associated with the risk of MCI and dementia, with the risk being highest among those with both SCD and APOE4. Interaction effect between SCD and APOE4 was seen in older women (but not in the other age or sex strata) – that is, the risk became amplified in older women when SCD and APOE4 were present concurrently, with the risk being in excess of the independent risks of either SCD or APOE4 alone. This interaction effect among older women remained consistent in the five sensitivity analyses, when alternate outcomes were used (i.e. MCI and dementia due to Alzheimer’s disease; dementia; and dementia due to Alzheimer’s disease) and when additional covariates were adjusted for (i.e. neuropsychological tests; and neuropsychiatric symptoms).

Interpretation of findings

The findings on interaction effect in women are consistent with a related literature which showed that SCD may be a more relevant predictor of neurocognitive disorders in women and not in men (Peres et al., 2011; Heser et al., 2019). Women are often more involved than men in health and health care all along their life, being responsible for family members’ health, watching for signs of illness, helping family members when ill and making appointments or escorting family members to physician, and with routine visits for reproductive health care (Verbrugge, 1982). Given their social roles, women may be more body-sensitive, being more likely to finely perceive symptoms, more willing to talk about health problems, and more informed about health and health services (Peres et al., 2011). As such, it is not inconceivable that women may possibly have a better ability to perceive early and subtle cognitive changes, with higher utility of SCD in women than in men (Peres et al., 2011; Heser et al., 2019).

The findings on interaction effect in women are also consistent with the literature which showed that APOE4 is a stronger predictor of neurocognitive disorders in women than in men (Ferretti et al., 2018; Zhu et al., 2021; Fisher et al., 2018), possibly related to the mediating effects of vascular diseases and depression. Although APOE4 is best-known as a genetic risk factor for Alzheimer’s disease, it is also widely recognized as a risk factor for vascular diseases (Eichner et al., 2002), and can increase the risk of cerebrovascular diseases and vascular dementia. The differential effect of APOE4 on women may possibly be related to the different manifestations of vascular diseases across sexes (Leening et al., 2014), with prior evidence showing that men were more likely to develop coronary heart disease as their first vascular event; while women are more likely to develop cerebrovascular diseases as their first manifestation, and hence may possibly be more prone to vascular dementia. Similarly, APOE4 has also been associated with a higher risk of depression in the older population (Tsang et al., 2017); with late-life depression as another possible explanation to the differential effect of APOE4 on women, given that depression is more likely to affect women (Ferretti et al., 2018; Zhu et al., 2021) and depression has been associated with incident neurocognitive disorders (Diniz et al., 2013; Liew, 2019b).

The interplay among APOE4, age and sex also mirrors some of the findings from recent literature. In a study by Mortensen et al (Mortensen and Hogh, 2001), APOE4 allele was only associated with cognitive decline among older women (>70 years), and not among younger women (between 50–70 years) or among men. In another age- and sex- stratified cohort study (Rasmussen et al., 2018), APOE4 visibly resulted in the highest risk of dementia among the older women (≥80 years) when compared to men or to the younger age-groups. Similarly in a recent meta-analysis (Neu et al., 2017), APOE4 also demonstrated a higher risk of Alzheimer’s dementia in older women (between 65–75 years) when compared to men and those <65 years, although the sex differences diminished among the oldest-old group (>75 years). It is understandable to see such interplay among APOE4, age and sex – especially from the epidemiological perspective – given that APOE4, age and sex are among the key risk factors of dementia. Yet from the biological perspective, it remains unclear how these 3 risk factors interplay with each other within the complex biological system, although a plausible hypothesis has been proposed in recent literature, related to the role of sex hormones on brain function (Gamache et al., 2020; Riedel et al., 2016; Zhu et al., 2021; Fisher et al., 2018). Estrogen is known to have a neuroprotective effect on women, due to the critical roles of estrogen in the brain in regulating gene expression, energy production and neuronal plasticity (Gamache et al., 2020; Fisher et al., 2018). During post-menopausal period, glucose hypometabolism occurs in the brain as the female brain switches its energy source from glucose to ketone bodies in the face of declining estrogen (Gamache et al., 2020; Riedel et al., 2016). Among those with APOE4, the glucose hypometabolism in the brain may be further worsened given the association between APOE4 and defective estrogen signalling (Gamache et al., 2020). On top of that, as ketone bodies are primarily derived from white matter, the switch of energy source to ketone bodies can lead to reduction in white matter volume and disruption of myeline sheath (Gamache et al., 2020; Riedel et al., 2016). Over time, these changes in the brain may possibly manifest with cognitive decline and neurocognitive disorders, with the risk probably being most prominent among older women who report SCD (which can be a very early indicator of the onset of neurocognitive disorders). Notwithstanding all the above explanations, the complex interplay among age, sex, APOE4 and SCD remains an area that will benefit from further research exploration, which may potentially enrich our understanding on the neurobiological mechanisms linking these separate factors.

Clinical implications

The findings may potentially have implications to older women who present with SCD in clinical practice, highlighting a plausible utility of APOE genotyping specific to this subgroup of individuals. In recent years, there have been growing interests in identifying individuals at risk of neurocognitive disorders for further interventions (Ngandu et al., 2015), given current understanding that interventions to preserve brain functions are more likely to be beneficial when provided early and before any irreversible neuronal cell death has occurred (Cummings and Fox, 2017). Although APOE4 allele has been a well-established risk factor for neurocognitive disorders, routine APOE genotyping has not been recommended in the literature due to the non-deterministic nature of APOE4 allele (i.e. not all APOE4 carriers will develop neurocognitive disorders) and the concerns that routine testing of APOE genotype can lead to undue worries and potential discriminations (Farrer et al., 1995). Similar to the issue with APOE4, SCD has been shown to increase the risk of neurocognitive disorders (Jessen et al., 2020; Jessen et al., 2014a), but it is also non-deterministic in nature given the other non-neurodegenerative causes of SCD such as personality traits, psychiatric conditions, and excessive self-attention (Jessen et al., 2020).

However, the predictive utilities of APOE4 and SCD rise dramatically when they are used in combination and among older women. This is evident in the findings on survival time in this study (Table 1), whereby the time to incident neurocognitive disorders markedly shortened in the presence of both APOE4 and SCD, and particularly among older women. In other words, among older women who report SCD, APOE genotyping may allow us to further stratify the individuals into two contrasting risk groups – the higher risk group (i.e. APOE4 carrier) had half of the individuals developing neurocognitive disorders within 4.4 years; which was substantially shorter than the responding 8.1 years among the lower risk group (i.e. APOE4 noncarrier). Subject to further validations in other study populations, APOE genotyping may potentially have a role in providing more granular risk information among older women who present with SCD in clinical practice. Such risk stratification can allow clinicians to employ more personalized approaches in patient counselling and management (Liew, 2020c; Liew, 2020d), whereby those at imminent risk of neurocognitive disorders may be identified for more intensive interventions (e.g. risk-factor modification, physical exercise and cognitive training) (Ngandu et al., 2015; Jessen et al., 2020; Livingston et al., 2020), enrolment into preventive trials (Liew, 2019e; Liew, 2020b), as well as closer monitoring of cognitive function over time to allow timely diagnosis of cognitive impairment (Liew, 2019c; Liew, 2020a; Liew, 2019a; Liew, 2019d).

Limitations

Several limitations should be considered. First, participants in this study largely involved those who volunteered at ADCs, which may not necessarily be similar to those who seeks clinical evaluation in routine healthcare settings. Supplementary Material 8 provides a comparison of the demographic information between those who volunteered for research and those who sought clinical evaluation. In particular, those who volunteered for research were more likely to be White, more likely to have hypertension and APOE4, as well as less likely to have diabetes mellitus, hyperlipidemia, anxiety symptoms or SCD. Second, the SCD measure in this study was based on a single-question and focused on the memory domain. Such SCD measure may not have captured the full range of memory concerns or other non-memory domains (Liew et al., 2019). Third, it may be arguable that the association between SCD and MCI can be inflated, given the fact that the “subjective reporting of cognitive decline by individuals or informants” is also one of required criteria in diagnosing MCI. However, this may less likely be the case in the current study, as the participants with SCD at baseline had been assessed and found not to have MCI or dementia (i.e. fulfilling the current operational criteria of SCD, which is the subjective perception of cognitive decline in the absence of objective cognitive deficits) (Jessen et al., 2014a; Jessen et al., 2020). In addition, the findings remained consistent in the sensitivity analysis when alternate endpoint of dementia was used, which provided some assurance on the validity of the findings. Fourth, the presumptive diagnosis of Alzheimer’s disease was made using available clinical criteria (i.e. NINCDS-ADRDA criteria and 2011 NIA-AA criteria) (McKhann et al., 1984; McKhann et al., 2011; Albert et al., 2011), and may not be as accurate as those made using the biomarkers of Alzheimer’s disease. Fifth, although the study analyses adjusted for many of the established risk factors of neurocognitive disorders (Livingston et al., 2020), there could still be the effect of residual confounding due to other known risk factors that were not captured in NACC database (e.g. social isolation, physical inactivity, air pollution) (Livingston et al., 2020) and hence could not be adjusted for in the study.

Conclusion

SCD and APOE4 are independently associated with the risk of MCI and dementia, with the risk being highest among those with both SCD and APOE4. Interaction effect between SCD and APOE4 was seen in older women, with the risk being in excess of the independent risks of either SCD or APOE4 alone. While it is not yet clear why the interaction effect only occurs in older women, there may be several plausible explanations related to better perception of subtle cognitive changes among women, mediating effects of cerebrovascular disease and depression in women, as well as glucose hypometabolism in the brain during the post-menopausal decline in estrogen. The findings highlight the need for further research to enrich our understanding on the neurobiological mechanisms linking age, sex, APOE4 and SCD. The findings may also indicate a potential role of APOE4 in further risk stratifying older women who present with SCD in clinical practice, to facilitate risk-based, personalized interventions.

DATA AVAILABILITY

The data were obtained from the National Alzheimer’s Coordinating Center (NACC). For further information on access to the database, please contact NACC (contact details can be found at https://naccdata.org/).