Nienke Legdeur

In this thesis, we describe how cognition, brain pathology biomarkers, physical parameters and risk factors for cognitive impairment are related to age and which determinants are associated with cognitive functioning in individuals aged 90 years and older, the oldest-old.

Main findings of this thesis:
- We provide mean cognitive test scores (SD) and cut-off scores to define cognitive impairment in the oldest-old for twelve widely used cognitive tests based on existing literature.
- We show that markers measuring brain pathology, cognitive and physical processes are differently susceptible for the aging process. Most importantly, hippocampal atrophy is almost inevitable with aging, whereas approximately half of the cognitively normal oldest-old remains free of amyloid aggregation.
- The negative effect of vascular disorders and other risk factors on cognitive decline and incident dementia decreases with older age.
- In the oldest-old, higher handgrip strength, physical performance, nutritional status and hemoglobin A1c (HbA1c) are associated with better cognitive performance.
- More white matter hyperintensities (WMH), hippocampal atrophy and amyloid aggregation relate to worse cognitive functioning and faster cognitive decline in the oldest-old. Higher past cognitive activity, lower muscle mass, physical performance and body mass index (BMI) are related to a higher prevalence of these brain pathology biomarkers.

In this chapter, we summarize the findings of this thesis and place them into context of existing literature. Methodological considerations regarding the different studies of this thesis are discussed. Implications and future directions are described.

Age-related changes in cognition

In chapter 1 we review 20 articles that report on the cognitive functioning of the oldest-old without dementia. We pool data and provide mean cognitive test scores and cut-off scores to define cognitive impairment in the oldest-old of twelve cognitive tests: mini-mental state examination (MMSE), Trail Making Test A and B (TMT-A and -B), Digit Span forward and backward, modified MMSE (3MS), California Verbal Learning Test (CVLT), Category (Animal), Letter F and FAS Fluency, Constructional Praxis and Boston Naming Test (BNT)-15. Compared to literature on younger individuals, the mean cognitive test scores in this review are generally lower, although the performance on some cognitive tests seems to be more age-dependent than others [1–6]. For example, scores on the CVLT and TMT-A and -B are potentially more sensitive for an increasing age than scores on the Digit Span forward and backward and fluency tasks.

Age-related changes in brain pathology biomarkers and physical parameters

Our aim to study age-related changes in cognition is extended to brain pathology biomarkers and physical parameters in chapter 2. We show that markers measuring brain pathology, cognitive and physical processes differently associate with age in a cognitively normal population aged 60-102 years. Aging patterns of the markers reflecting the same process of interest, namely the brain pathology, cognitive or physical process, show no overlap. This implicates that there is no clear sequence in which these processes become abnormal with older age. An important difference in aging patterns is the almost inevitable presence of hippocampal atrophy at older age and the substantial number of oldest-old individuals that remain free of amyloid aggregation. These results endorse earlier findings implicating that neurodegenerative pathology can emerge through non-amyloid pathways [7] and that hippocampal atrophy in the oldest-old is also related to other pathologies than amyloid aggregation, such as hippocampal sclerosis, argyrophilic grain disease and vascular pathology [8,9]. Furthermore, abnormality of the brain pathology biomarkers does not clearly precede abnormality of the markers measuring cognitive and physical processes. It might be that with older age, a composite of different brain pathologies, including pathologies that were not part of the present study such as infarcts, tangles and arteriosclerosis, is necessary to compose a driving factor that explains the age-related increase of abnormal cognitive functioning [10]. Additionally, aging patterns are dependent on sex showing a higher prevalence of hippocampal atrophy, abnormal memory and abnormal handgrip strength in males across almost the entire age range and a higher prevalence of abnormal WMH volume in females. Possible underlying mechanisms for this difference include variations in risk factor exposure and cognitive activities throughout the life course between males and females and hormonal effects [11]. The association of amyloid abnormality with age is similar for males and females across the age range. This is somewhat unexpected as dementia prevalence is higher in females than in males [12,13]. A possible explanation might be a higher prevalence of other neuropathologies in females, leading to faster cognitive decline in females with the same amyloid burden as males [14,15]. However, the higher dementia prevalence in females compared to males might also be caused by the longer life expectancy in females [16]. Dementia incidence between males and females is almost comparable after correcting for the difference in life expectancy making the similar prevalence of amyloid abnormality less surprising.

Age-related changes in risk factors for cognitive impairment

In chapters 3 and 4 we study whether risk factors for cognitive impairment are dependent on age. In a cohort of 2527 cognitively normal individuals aged 55-85 years at baseline, we find that the association of LDL cholesterol, homocysteine, hypertension, history of stroke, depressive symptoms, interleukin-6, α1-antichymotrypsin, alcohol use and smoking with cognitive decline significantly differs between the three age groups (≤ 70 years, 70–80 years and > 80 years). In general, the presence of these risk factors is associated with less cognitive decline in the individuals aged > 80 years compared to the individuals aged ≤ 70 years and 70-80 years. Per age group, the risk and protective factors associated with cognitive decline differs and are presented in Figure 1a. Furthermore, in a primary care database of 442,428 individuals aged ≥ 65 years without dementia, the risk of hypertension, diabetes mellitus, dyslipidemia, stroke, myocardial infarction, heart failure and atrial fibrillation for dementia decreases with increasing age and is no longer significant in the oldest-old individuals (Figure 1a). To some extent these findings are not new as earlier literature already indicated age-dependent effects on cognition of various risk factors, mainly of hypertension and cholesterol [17]. However, we extent these earlier results by showing the same pattern for other risk factors, such as inflammatory markers and a history of stroke.

Figure 1a. The age-dependent associations of risk (orange box) and protective (green box) factors with cognitive decline (CD) and incident dementia (ID) (white box indicates no effect)
CD: Hypertension, Myocardial infarction, Interleukin-6, α1-antichymotrypsin, Alcohol ≤ 2 /day, Physical activity, Smoking, APOE ε4, LDL cholesterol, Homocysteine.
ID: Diabetes mellitus, Hypertension, Dyslipidemia, Stroke, Heart failure, Myocardial infarction, Atrial fibrillation.

The results from chapter 3 and 4 show an interesting paradox. In both studies, the prevalence of most risk factors for dementia, such as a history of stroke or heart failure, and the prevalence of dementia increase with older age. This suggests a possible correlation between the age-related increase of both the risk factors and dementia. Moreover, earlier literature shows that at older ages, dementia is mostly caused by a mixture of different pathologies including Alzheimer’s disease (AD) pathology and vascular brain pathologies [18,19]. However, our results indicate a diminished association between risk factors and dementia at older ages. Methodological aspects might have contributed to these findings, most importantly the selective loss of individuals during follow-up. Survival bias is an example of selection bias and a significant factor that needs to be considered, especially in studies with older individuals. We will therefore discuss this more extensively under ‘methodological considerations’. Apart from the methodological aspects, other explanations need to be considered as well. In case of hypertension, older individuals may be more sensitive to blood pressure drops and higher blood pressures are potentially necessary to ensure cerebral blood flow [20]. Furthermore, higher cholesterol levels might reflect a better nutritional status in older individuals which subsequently is an important determinant of overall well-being including good cognitive functioning [21,22].

Risk factors for cognitive impairment in the oldest-old

The age-related changes in risk factors for cognitive impairment as described in the previous paragraphs, highlight the importance to focus research on the largest growing segment of the population with the highest dementia prevalence: the oldest-old. In chapter 5 we summarize the literature regarding the risk and protective factors for cognitive impairment in the oldest-old and describe the design of the EMIF-AD 90+ Study. In chapter 7, the first results of the EMIF-AD 90+ Study are presented. We show that higher handgrip strength, physical performance, nutritional status and HbA1c levels are associated with better cognition in the oldest-old, although no implications about causality can be made, particularly nutritional status is more likely to be a consequence of than a risk factor for cognitive impairment (Figure 1b and Figure 2). Past cognitive activity, muscle mass, BMI, c-reactive protein (CRP), blood pressure and cholesterol level are not associated with cognition in this age group.

Our results with regard to cholesterol are consistent across studies as we find no association with cognition or incident dementia in the older individuals in both LASA, IPCI and the EMIF-AD 90+ Study. Our results regarding hypertension are somewhat mixed across studies with a protective effect of hypertension on cognitive decline in LASA in individuals over age 80 years, no effect of hypertension on incident dementia in the oldest-old in IPCI and no effect of blood pressure on cognition in the EMIF-AD 90+ Study. Hypertension prevalence in LASA and IPCI is comparable with 74% prevalent hypertension in individuals over age 80 years in LASA and 79% in individuals aged 80-85 years in IPCI, so this does not explain the different findings. Another possibility is the difference in outcome measure. There may be a higher chance of underreporting of dementia diagnosis in IPCI as in this observational study not all individuals are regularly evaluated by their general practitioner (GP). This might be especially the case for individuals without hypertension as they are probably less frequently seen by their GP and therefore a possible protective effect of hypertension on incident dementia may have been missed.

Figure 1b. The age-dependent associations of risk (orange box) and protective (green box) factors with cognition (C) cognitive decline (CD) and incident dementia (ID) (white box indicates no effect)
CD: Hypertension, Myocardial infarction, Interleukin-6, α1-antichymotrypsin, Alcohol ≤ 2 /day, Smoking, APOE ε4, LDL cholesterol, Homocysteine.
C: Handgrip strength, Physical performance, Nutritional status, High hemoglobin A1c level, Past cognitive activity, Muscle mass, Body mass index, C-reactive protein, Blood pressure, Total cholesterol.
ID: Diabetes mellitus, Dyslipidemia, Stroke, Heart failure, Hypertension.

Brain pathology biomarkers in relation to cognition and risk factors in the oldest-old

In chapters 6 and 7 we study in two different cohorts of oldest-old individuals the association between risk factors, brain pathology biomarkers and cognitive functioning. Results between these two studies are consistent as we find in both that more WMH and hippocampal atrophy are associated with worse cognition. In chapter 6 we additionally find that more WMH and hippocampal atrophy are independently, and not synergistic, associated with faster cognitive decline. In chapter 7 we extend the brain pathology biomarkers with amyloid BP ND and indicate that more amyloid aggregation also negatively effects cognition in the oldest-old (Figure 2). These findings are somewhat in contrast to earlier post-mortem neuropathological studies that indicate that the association between brain pathologies and cognitive impairment becomes weaker at older ages [23]. Although we cannot directly compare our results with studies performed in younger individuals, the implication from post-mortem research that amyloid aggregation is not distinctive between oldest-old individuals with and without cognitive impairment does not seem to hold in in-vivo studies. A possible explanation may be the time lag between the moment of cognitive testing and neuropathological evaluation which might underestimate the effect of brain pathologies on cognition in post-mortem research.

In the EMIF-AD 90+ Study we also study risk factors in relation to brain pathology biomarkers and show that high physical performance is associated with less WMH and hippocampal atrophy in the oldest-old (Figure 2). The direction of this association is unclear but WMH might interfere with specific motor pathways in the brain which may lead to lower physical performance, or WMH may affect the ability to process sensory information and thereby disturb physical performance [24]. Another possibility is that WMH, hippocampal atrophy and low physical performance are driven by the same risk factors, for example by physical inactivity or the presence of cardiovascular diseases [25]. It is also possible that physical performance is the driving factor as earlier literature indicated that physical activity can increase hippocampal volume, potentially by increasing the secretion of myokines and subsequently the level of brain-derived neurotrophic factor (BDNF) [26]. This hypothesis is underlined by our finding that the association between high physical performance and better cognition is partially mediated by hippocampal volume. However, brain pathologies do not seem to completely explain the association between physical performance and cognition, indicating that other underlying molecular mechanisms still need to be identified [27].

The counterintuitive association between higher past cognitive activity and more amyloid aggregation (Figure 2) might be explained by the concept of cognitive reserve [28]. We find that this association was driven by the cognitively normal individuals in whom higher past cognitive activity may protect against the harmful effects of amyloid aggregation on cognition. In younger cognitively normal elderly, higher cognitive activity across the lifespan was associated with less amyloid aggregation in one study [29] but other studies showed no association [30–32]. The lack of an association between cognitive activity and amyloid aggregation in these studies are somewhat in line with our results as they both indicate that the preventive effect of cognitive activity on cognitive deterioration is potentially explained by a mechanism that is independent of amyloid aggregation.

Furthermore, we find an association between lower BMI and more amyloid aggregation (Figure 2). This may be related to systemic disease manifestations, such as weight loss, in the early phases of AD. Possible mechanisms responsible for these changes in the early phases of AD include neuropathological alterations in brain areas that are important for energy metabolism regulation and lifestyle changes in individuals with AD [33].

Figure 2. Established correlates of cognition in the oldest-old based on the EMIF-AD 90+ Study
Amyloid aggregation, WMH volume, Hippocampal atrophy, Past cognitive activity, BMI, Muscle mass, Physical performance, Cognition, Handgrip strength, Nutritional status, HbA1c.
Green box: higher value -> lower level of brain pathology biomarker; Red box: higher value -> higher level of brain pathology biomarker. BMI: body mass index; HbA1c: hemoglobin A1c; WMH: white matter hyperintensities.

Methodological considerations

Overall strengths and limitations
An overall strength of this thesis is that we approached our main research aim, to unravel the determinants of cognitive functioning in the oldest-old, in different ways. We use five different cohort studies, namely the Longitudinal Aging Study Amsterdam (LASA), the Integrated Primary Care Information (IPCI) database, the PreclinAD Study, The 90+ Study in the USA and the EMIF-AD 90+ Study, and apply different statistical methods, namely spline regression analyses, Cox regression models, Fine and Gray regression models and generalized estimating equations (GEE). Last, in chapters 1 and 5 we review the literature on cognition in the oldest-old, facilitating the implementation of results from other chapters in existing literature. Another strength of this thesis is the focus on possible clinical implementations of our results. For example, in chapter 1 we do not only review the literature but also make the results useful for clinicians by providing mean cognitive test scores and cut-off scores.

The most important possible limitation that needs to be considered in all chapters of this thesis, is the role of survival bias [34]. Risk factors for cognitive impairment, such as stroke or hypertension, are related to an increased mortality risk. This leads to the selection of more healthy individuals, particularly in studies including at higher ages, who are less susceptible for the negative consequences of these risk factors. In cross-sectional studies, individuals who survive the risk factor are over-represented in the study sample compared to individuals who do not survive the risk factor. It is more likely that individuals who survive the risk factor do not have cognitive impairment as cognitive impairment also relates to mortality [35]. The consequence is that the association between the risk factor and cognitive impairment is underestimated [36]. Potentially this effect increases when research is focused on older individuals as mortality rates are higher. In longitudinal research, which is conducted in chapter 3, 4 and 6 of this thesis, survival bias, and other forms of selection bias, may attenuate the estimated association when a risk factor of interest is not only associated with cognitive decline but also with a higher study drop-out rate due to mortality or physical consequences. Various approaches are proposed to address potential selection bias, for example by applying a case control study design in which follow-up time is equalized between the cases and controls [37], but it is difficult to evaluate the effect of these approaches and no solution has been found ideal [34].

Another bias in longitudinal research which is specific for Kaplan-Meier and Cox regression analyses, is the effect of competing risk by mortality. In contrast to the selection bias described above, competing risk by mortality in Kaplan-Meier and Cox regression analyses lead to an overestimation of the effect of a risk factor on the outcome of interest [38,39]. In Kaplan-Meier and Cox regression analyses, individuals who die and individuals who are lost to follow-up are censored. Censored individuals are considered ‘at risk’ to develop dementia, which is of course not true for deceased individuals. Failing to account for mortality as competing risk will therefore overestimate dementia risk. A possible solution to account for mortality as competing risk is to use the Fine and Gray approach instead of Cox regression analyses [40]. In chapter 4 we apply both methods but our results do not differ between the Cox and Fine and Gray regression method. Possibly, the follow-up time in our study is too short to show a difference between the Cox and Fine and Gray regression method [38], but it might also be hypothesized that the Fine and Gray method is not sufficient enough to correct for competing risk by mortality.

Apart from methodological solutions, the exploration of mechanisms that may explain the diminished effect of a risk factor on cognition at high age, might also foster our understanding of this effect. If low blood pressure can indeed be linked to a reduced cerebral blood flow and thereby to worse cognition, this will strengthen the concept that the age dependency of risk factors is not (only) explained by methodological aspects. So far, previous literature does not establish a direct and firm connection between blood pressure, cerebral blood flow and cognition [41–43]. This might be related to the absence of a reliable measure for the response of cerebral perfusion to blood pressure changes, for example when standing up. Another possible limitation that needs to be considered when including individuals over a wide age range is the possibility of a cohort effect. Individuals included at age 60 years are from a later birth cohort than individuals included at age 90 years and it has been shown that later birth cohorts show better physical and cognitive functioning than earlier birth cohorts, designated as the Flynn effect [44]. This might overestimate the age-related changes in brain pathology, cognitive and physical markers we find in this thesis (chapter 2).

Cohort specific strengths and limitations
The age dependency of risk factors for cognitive impairment is established in two different cohorts (LASA and IPCI) using two different statistical methods (spline and cox regression analyses) with two different longitudinal outcome measures (cognitive decline and incident dementia). Despite these variations in methods, both studies show similar results, namely a decreasing effect of risk factors on cognition with increasing age. The strength of LASA is that it is a well characterized cohort, whereas the strength of IPCI is the large sample size (there are over two million individuals in the complete database). Both cohorts also have some limitations. First, they do not include information about midlife (age 40-55 years). Earlier