Does a midlife dementia risk score relate to late-life structural brain changes on MRI?

(1)

Does a midlife dementia risk score relate to late-life structural brain changes on MRI?

Joona Koskinen LT6 jookos@student.uef.fi Lääketieteen koulutusohjelma Itä-Suomen yliopisto Terveystieteiden tiedekunta Lääketieteen laitos/Neurologia February 2019 Supervisors: Dosentti Alina Solomon MD, PhD; Anette Hall, PhD

(2)

Due to the aging of population worldwide, the number of people living with dementia is increasing. Since the drug trials to stop the progression of dementia have not been successful, the treatment for dementia should focus on prevention. LIBRA dementia risk score is a new tool to prevent dementia and to estimate risk for dementia already in midlife. This study was based on CAIDE (Cardiovascular diseases, aging and dementia) study and LIBRA (Lifestyle for brain health index) dementia risk score. CAIDE study was based on assessments of

sociodemographic and health-related factors of 2000 middle-aged individuals in 1972-1987 and reassessments and brain MRI scans of those who those who participated on the study in 1998 and 2005-2008. LIBRA dementia risk score consists of modifiable risk and protective factors for dementia with different weights. LIBRA score was calculated from the CAIDE data.

We found that higher LIBRA score in midlife was significantly associated with white matter lesions after a 20-year follow-up time period when adjusted with confounders in the original CAIDE population. Also, we found that LIBRA score had some association for gray matter volume but no significance was found after adjustments.

Tiivistelmä

Väestön ikääntyessä maailmanlaajuisesti dementian esiintyvyys kasvaa. Dementiaa aiheuttavien sairauksien lääkekokeilut eivät ole olleet onnistuneita, joten dementian hoidossa tulisi keskittyä ennaltaehkäisyyn. LIBRA dementian riskipisteytys on uusi työkalu ennaltaehkäisemään dementiaa ja arvioimaan sen riskiä jo keski-iässä. Tämä tutkimus pohjautui CAIDE-tutkimukseen (Cardiovascular diseases, aging and dementia) ja LIBRA (Lifestyle for brain health index) dementian riskipisteytykseen. CAIDE-tutkimus pohjautui vuosien 1972-1987 aikana 2000 keski-ikäisen osallistujan sosiodemografiseen ja terveyteen liittyvien tekijöiden arviointeihin sekä myöhemmin tutkimukseen osallistuneiden uudelleenarviointeihin ja aivojen MRI-kuvantamisiin vuosina 1998 ja 2005-2008. LIBRA riskipisteytys koostuu dementian muunneltavissa olevista riski- ja suojaavista tekijöistä eri painotuksella. LIBRA-pisteytys kerättiin CAIDE-tutkimuksen datasta. Havaitsimme, että korkeampi LIBRA-pisteytys keski-iässä oli tilastollisesti merkitsevässä yhteydessä valkean aineen leesioihin 20 vuoden seurannassa korjattuna sekoittavilla tekijöillä CAIDE tutkimusaineistossa. Lisäksi LIBRA-pisteytyksellä havaittiin olevan yhteyttä harmaan aineen tilavuuteen, mutta korjattuna sekoittavilla tekijöillä tilastollisesti merkitseviä tuloksia ei saatu.

(4)

1 Background

1.1 Dementia, cognitive impairment, and the role of brain MRI

Dementia is a syndrome usually occurring in elderly people and characterised by cognitive impairment and lack of capability in independently managing daily activities. Causes for the development of dementia vary, starting with Alzheimer’s disease (AD), a neurodegenerative disorder found in approx. 70% of patients with dementia (1). Other types of dementia are e.g.

vascular dementia (due to cerebrovascular disease), dementia with Lewy bodies, frontotemporal dementia, Parkinson disease dementia. Brain injury, inflammation or temporal lobe epilepsy can also cause cognitive impairment (2). Due to the aging of the population in Finland and worldwide, the number of people living with dementia or other conditions with memory impairment is increasing. Symptoms of memory loss or mild cognitive impairment (MCI) (3) may start long before dementia onset, with a prevalence of MCI estimated to be up to 40%, depending on the affected cognitive domains (4).

In order to diagnose dementia patients with symptoms of memory loss usually start with basic clarifications at a physician’s reception. These are mainly anamnesis for medical history, education and symptoms, questionnaires for the patient and for the patient’s relatives and medical examination including neurological examination done by the physician. It’s also important to rule out treatable diseases or conditions for memory disorders e.g. depression, hypo- or hypertyreosis or B12-vitamin deficiency. Patient’s cognitive skills are assessed via anamneses and questionnaires such as Consortium to Establish a Registry for Alzheimer’s Disease (CERAD) or Mini-Mental State Examination (MMSE). Also, the physician assesses for behavioural orders and depressive symptoms by using Geriatric Depression Scale (GDS). After basic clarifications patients are sent to a specialist (geriatrician or neurologist for patients of working age). Then the cause of dementia is specified using primarily magnetic resonance imaging (MRI) or if MRI is contraindicated specification is done using computed tomography (CT). MRI is more accurate and important to illustrate the characteristic structural changes in the brain of certain diseases causing dementia. Usually a radiologist assesses the MRI scannings and uses different scales for the characteristic changes such as gray matter (GM) atrophy in cortical, temporomesial and in particular hippocampal regions with Schelten’s scale, and white matter lesions (WML) or white matter hyperintensities (WMH) with Fazekas scale. WMLs and lacunar infarcts are considered to be linked to small vessel disease and vascular cognitive

(5)

impairment. WMLs have also been linked to increased risk of stroke, dementia and death.

Posterior atrophy is assessed with Koedam scale (1).

Stopping the progression of AD and dementia has been difficult as many drug trials have not been successful. Current AD medications (cholinesterase inhibitors donepezil, galantamine and rivastigmine, and N-methyl-D-aspartate receptor blocker memantine) have mainly symptomatic effects (5). There is thus a lot of interest in strategies for preventing dementia- related diseases. Although age is a main risk factor in development of dementia and AD, these are multifactorial and complex conditions, where several risk factors have been associated with disease development and progression. In addition to non-modifiable risk factors for late-onset AD and dementia such as age, sex, family history, and genetic factors (primarily apolipoprotein E (APOE) ε4 allele), several modifiable risk factors have been described (6).

1.2 Modifiable risk factors for dementia and cognitive impairment

Many modifiable risk factors for developing dementia have been indicated by a large number of observational studies in different populations. Examples of such modifiable risk factors are vascular and metabolic factors (e.g. hypertension, overweight/obesity, hypercholesterolemia, diabetes mellitus, at-risk pre-diabetes stages, cardiovascular diseases, and cerebrovascular lesions), lifestyle-related factors (e.g. physical inactivity, smoking, alcohol overconsumption, and unhealthy diet), socioeconomic factors (e.g. education, socioeconomic status), and psychological factors (e.g. depression and chronic stress). Traumatic brain injury, occupational exposure (e.g. heavy metals) and infective agents (herpes simplex I, Chlamydia pneumoniae, and spirochetes) have also been linked to dementia risk (6, 7). In addition, genetic risk factors for dementia and AD (e.g. family history, APOE ԑ4 allele) seem to interact with environmental factors. For example, the detrimental effects of physical inactivity, high dietary saturated fat intake, and lower intake of polyunsaturated fats, smoking and alcohol overconsumption in midlife seem to be more pronounced among APOE ԑ4 allele carriers. While genetically susceptible individuals such as APOE epsilon4 carriers may be more vulnerable to negative effects of environmental risk factors, they may also benefit more from preventive lifestyle interventions (8).

(6)

Prevention is particularly important, as cognitive impairment and dementia/AD have become a public health problem. Even if a risk factor may have a small influence on disease risk, if the factor is very common it has a large effect at general population level (6). A more recent study of associations between seven risk factors (diabetes, midlife hypertension midlife obesity, physical inactivity, depression, smoking and low educational attainment) and dementia took into account the tendency of such factors to occur together (i.e. not independently). The study estimated that a 10% decline per decade on each of the seven risk factors would lower the prevalence of AD with 8.3% by 2050. It also estimated that about a third of AD cases worldwide may be related to these seven modifiable risk factors which could be targeted by dementia preventive interventions (9).

Recently the World Health Organization (WHO) has published the first guidelines for reducing the risk of cognitive decline and dementia following systematic reviews of available evidence for interventions targeting physical activity, smoking cessation, nutrition, alcohol consumption disorders, cognitive and social activities, weight management, management of hypertension, diabetes, dyslipidaemia, depression and hearing loss (10). The guidelines also mention the importance of identifying early the people who have increased risk of dementia, and who may have benefits from preventive interventions.

1.3 Developing risk scores from individual risk factors

While many studies have provided important knowledge by focusing on separate risk factors, it is essential to investigate also the impact of combinations of risk and protective factors on the overall risk for developing dementia/AD (e.g. risk scores). This could facilitate early identification of people who have higher risk and who may benefit most from prevention strategies. There are different dementia risk scores which have been developed using different risk factors or biomarkers and methods (11). One approach used in most available risk scores is data-based, by using existing studies with collected data from different populations.

Cardiovascular risk factors, Aging and Dementia (CAIDE) is a population-based follow-up study that was conducted in Finland and was used to develop a midlife dementia risk score including factors such as age, sex, education, systolic blood pressure, BMI, total cholesterol and physical activity with and without the presence of APOE (7) ԑ4 allele. The CAIDE Dementia Risk Score based on this midlife risk profile was shown to predict well the development of dementia 20

(7)

years later (7). Several other dementia risk scores have been developed using this approach, but they have focused primarily on older people (11).

Another way to develop a risk score is the evidence-based approach, i.e. reviewing published literature on dementia risk factors in order to identify which risk factors to include in the risk score, and how much weight to give to each factor. Evidence-based approach thus focuses on literature rather than on analysis of data from a cohort study. The Australian National University Alzheimer’s Disease Risk Index (ANU-ADRI) includes both risk and protective factors, and focuses on factors that can be assessed by self-report (12). ANU-ADRI authors created the risk score from reported AD associations in extant literature. Odds ratios for included factors were modified into an algorithm for calculating the risk score. In creating the ANU-ADRI score, decisions had to be made to exclude some risk factors because ANU-ADRI focused on risk assessment among older adults (>60 years), e.g. hypertension was not included because of varying reported effects at older ages compared to mid-life.

There are several important methodological issues to be considered with each approach. In data-based risk scores an important issue is overfitting the models, i.e. models that perform very well in the initial dataset/population used to develop them but perform poorly outside the initial dataset/population. Too complex models in small datasets are more likely to be overfitted. To make sure that a risk score is trustworthy and accurate, external validation is essential. For example, a risk score developed in a population from one country can be tested in a population from another country to see if it still performs well, or if modifications need to be made. Many studies aiming to develop dementia risk scores use at least internal validation, where the data for developing the model and the data for testing the model are both random samples from the same initial dataset/population. However, most available dementia risk scores lack external validation. The CAIDE Dementia Risk Score has been validated in a different population from USA (13).

An advantage of the evidence-based approach is that it develops risk scores based on factors or biomarkers consistently reported in the literature (and thus more reliably associated with dementia risk). One main limitation is that not all relevant risk factors/biomarkers may be

(8)

reported in the literature at a given time point, or the number of studies reporting them may be too small for a literature review or meta-analysis to draw any conclusions.

The Lifestyle for brain health index (LIBRA) is a relatively new dementia risk score developed with an evidence-based approach (14). After the CAIDE Dementia Risk Score, it is the second risk estimation tool to focus on a midlife profile of risk and protective factors. The goal is thus longer-term prediction of dementia development, and identification of at-risk individuals already in midlife. Another unique characteristic of the LIBRA score is the fact that it only includes factors which are modifiable, or at least manageable with existing treatments (alcohol intake, coronary heart disease, physical inactivity, chronic kidney disease, diabetes, hypercholesterolemia, smoking, obesity, hypertension, diet, depression, and cognitive activity).

It is thus the only risk score with a primary focus on the potential for prevention.

CAIDE and ANU-ADRI are so far the only risk scores including modifiable risk factors that have been validated in relation to not only dementia, but also brain MRI measures (7, 15). Because of the uniqueness of the LIBRA score, and the fact that it has not been studied so far in relation to dementia and AD-related biomarkers, the present study focuses on the LIBRA score and especially its associations with structural brain MRI measures.

1.4 LIBRA score, dementia and cognitive impairment

The LIBRA score was developed within the EU project (EU FP7) Innovative, Midlife INtervention for Dementia Deterrence (IN-MINDD, http://www.inmindd.eu) as a tool to evaluate the risk for dementia for individuals in midlife (14, 16). The included risk factors are coronary heart disease, physical inactivity, chronic kidney disease, diabetes, cholesterol, smoking, midlife obesity, midlife hypertension and depression. In addition, it takes in notice protective factors such as low/moderate alcohol intake, healthy diet and high cognitive activity.

Each factor has been assigned a weight (coefficient), which is positive for risk factors and negative for protective factors (Table 1). The sum of the coefficients for an individual provides the individuals total LIBRA score.

In a previous study, LIBRA score predicted the risk for dementia and cognitive impairment after 16 years follow-up (17). An increase of one point in the LIBRA score was related to higher dementia risk by 19% and for cognitive impairment by 9%. A prospective cohort study

(9)

investigating associations between socioeconomic status (education and wealth) and dementia with LIBRA score showed that higher wealth resulted in lower LIBRA score and lower wealth resulted in higher LIBRA score; results were similar for education and LIBRA score. Lower wealth (and higher LIBRA score) was related to higher risk of developing dementia, and higher wealth to lower dementia risk. A one-point increase in LIBRA score was associated with a 13 % increase in dementia risk. The association of LIBRA score with dementia risk was stronger in participants aged 50-70 years than in participants aged 70 years or older (18). Another prospective cohort study reported that higher LIBRA score in midlife predicted a faster cognitive decline in 10 years in the domains of verbal memory, cognitive flexibility and mental speed, and the results were mostly unrelated to gender or educational level. LIBRA score also predicted steeper decline in mental speed in participants with baseline cognitive impairment, suggesting its usefulness in secondary prevention of cognitive disorders (19).

LIBRA score was reported to be mainly useful in predicting dementia in midlife (age 45-59 years) rather than in late-life (age 60-75 years). It was associated with cognitive decline in a health-seeking clinical sample including individuals with mild cognitive impairment (20).

Another study emphasized the importance of LIBRA score in midlife (55-69 years) and late-life (70-79 years) for estimating the risk for subsequent dementia development, but not in oldest- old age groups (21).

Recently, LIBRA was also investigated for predicting the risk for dementia and MCI in the Finnish CAIDE study. Higher LIBRA score in midlife was related to higher risk for dementia and MCI 30 years later. The same study showed that high LIBRA score in late life was related to higher risk for MCI and higher risk for dementia among APOE ε4 non-carriers (16).

Table 1. LIBRA components and corresponding number of points.

Risk/protective factor Points Low/moderate alcohol intake -1.0 Coronary heart disease +1.0

Physical inactivity +1.1

Chronic kidney disease +1.1

Diabetes +1.3

High Cholesterol +1.4

Smoking +1.5

Midlife obesity +1.6

(10)

Midlife hypertension +1.6

Healthy diet -1.7

Depression +2.1

High cognitive activity -3.2

1.5 Impact of individual LIBRA components on dementia-related changes on brain MRI

1.5.1 Healthy diet

A recent study found that midlife changes in diet, such as fat quality, increased vegetable consumption and decreasing salt and sugar consumption, were related to lower risk for dementia in later life (22). Another observational study reported that a long-term healthy diet was associated to larger volume of hippocampus. This association was independent from other factors such as physical inactivity, smoking or cardiometabolic health factors; low alcohol intake was the key component for larger hippocampal volume independently (23). A longitudinal study of 4 years found a significant inverse association of vegetable intake with brain gray matter volume loss, in the temporal region in particular (24).

1.5.2 Coronary heart disease

In a cross-sectional study WML volume was higher in patients with newly diagnosed atherosclerotic disease (including coronary artery disease) compared with the general population. Larger WML volumes were observed in patients with cerebrovascular disease and abdominal aortic aneurysm than in those with peripheral artery disease or coronary artery disease (25). The same study also showed that WML volumes increased with age and were higher in men than women. The decrease in brain volumes at older ages were however reported to be comparable to the general population. Other cross-sectional studies have found links between coronary heart disease and both WML and gray matter changes on brain MRI (26, 27, 28). In the longitudinal CAIDE study in Finland, coronary heart disease was associated with lower cortical thickness, and lower total gray matter volume, especially with longer duration of the disease, and in people who also had hypertension (29). However, no association was found with WML.

1.5.3 Physical inactivity

(11)

Hippocampal atrophy has been associated with leisure time inactivity along with age and lower education level. The same study also showed that physical inactivity tended to associate with total gray matter atrophy (30). A study of self-reported inactivity showed that individuals with less physical activity had smaller gray matter volumes, although physical activity had no associations to the volume of white matter hyperintensities or global white matter (31). In the CAIDE study, leisure-time activity in midlife has been linked to larger total brain volume and gray matter volume, especially in the frontal lobes, although the association with WMH was not significant after adjusting for several confounders (32).

An increase in hippocampus volume has been reported in several trials where the participants received physical activity interventions (33, 34, 35, 36).

1.5.4 Renal dysfunction

Renal dysfunction measured by decreased estimated glomerular filtration rate (eGFR) has been shown to have an association with deep and subcortical white matter hyperintensities (37). A systematic review supported an association between decreased GFR and WML but most significant results were that in the general population, people without additional cardiovascular risk factors or history of stroke or TIA (transient ischemic attack) were not more likely to have WMLs (38). The same study also suggested that age, hypertension and diabetes mellitus might be confounders, since they are related to chronic kidney disease (CKD) and brain lesions. Because of such confounding factors, there might be an overestimation of the association between CKD and brain lesions. A 5-year follow-up study with repeated MRI scans indicated that CKD was an independent risk factor for deep WMLs, and that hypertension might be relevant in promoting deep WMLs and decreasing cognitive functioning (39). Another study reported that impaired kidney function was related to smaller GM and hippocampal volume and cortical thickness independently (40).

1.5.5 Diabetes

A study of an elderly population showed that in individuals with diabetes, in particular 2-hour post load glucose levels were associated to hippocampal atrophy even after adjusting for confounding factors, and that midlife diabetes and diabetes of longer duration were risk factors for hippocampal atrophy (41). Another study showed that individuals with longer diabetes duration and HbA1c ≥ 7.0% had smaller total and regional brain volumes (deep gray matter and hippocampus, and also frontal and occipital lobes). In addition, individuals with

(12)

prediabetes and HbA1c value <7.0% had no difference in brain volumes compared to individuals without diabetes (42). Also, the same study showed that severe diabetes (based on HbA1c level) was linked with increased white matter hyperintensities compared to participants without diabetes or with less severe diabetes. Here, evidence of mediation through cerebrovascular disease between smaller brain volumes and diabetes was not seen. Thus, diabetes might lead to smaller brain volumes independently. Another population-based study of individuals without dementia showed that midlife diabetes had associations to ischemic and atrophic changes of the whole brain, regional hippocampus atrophy, and affected cognition later in life (43). Late-life diabetes had no effect on cognition. Middle-aged individuals with type 1 diabetes mellitus (T1DM) of childhood-onset have been shown to have greater severity of clinically relevant white matter hyperintensities compared to similarly aged individuals without T1DM (44), independently of other risk factors, such as hyperglycemia or hypertension.

1.5.6 Smoking

A study investigating dose-response relations of smoking with WMH progression reported that age when smoking started and time since quitting had no associations, but increasing pack- years of smoking was associated to greater risk (45). A review of long-term smoking and brain changes on MRI reported that smokers had smaller cortical gray matter volume (46). Another systematic review including 4 studies indicated that compared to non-smokers, smokers had lower GMV in multiple cortical and subcortical regions, and atrophy in multiple prefrontal cortex regions (47). The decrease in GMV was thought to be responsible for cognitive impairment in smokers. Two of the studies found lower GMV in the thalamus. Another study found that smoking resulted in thinner cortex in several brain regions, including medial and lateral orbitofrontal cortex, middle temporal region and atrophy in early AD (48).

1.5.7 Obesity

In a cross-sectional study, it was found that abdominal obesity may partially mediate the association between T2DM and total GMV. Waist-hip-ratio (WHR) was independently associated to lower total GMV, but not BMI or physical activity measured in mean steps/day (49). Also, mean steps/day seemed to independently associate with smaller total hippocampal volume. Another study using cross-sectional data reported that obesity (as BMI and WHR) was associated to lower gray matter volumes (50). Associations between obesity and white matter

(13)

were not observed. BMI in midlife was not associated with cortical thickness in the CAIDE study (51), although midlife overweight and obesity, as well as elevated BMI from midlife until late life were related to more severe WML in late-life (52).

1.5.8 Hypertension

A population-based longitudinal study showed that high systolic and diastolic blood pressure were associated to WML progression, but after baseline WML was taken into account only high systolic blood pressure was associated to WML progression (53). Also, treated hypertensive individuals had less WML progression even with higher WML load at baseline. Another population-based study with a 4-year follow-up showed similar results, i.e. baseline blood pressure level and changes in blood pressure were strong predictors of WML progression independently, mainly periventricular and total WML (54). Antihypertensive treatment initiation during the follow-up was associated to less WML progression, especially with higher baseline systolic blood pressure, with a trend of association with smaller WML load in treated controlled individuals compared to treated but uncontrolled. In the CAIDE study, midlife hypertension was related to lower cortical thickness in late life in several brain regions, e.g.

insula, frontal, and temporal cortex (51). Both midlife hypertension and sustained hypertension from midlife until late life were associated with more pronounced WML in late life (52).

A systematic review and meta-analysis showed that in most studies (26 studies out of 28, 92.9%) higher BP levels were associated to lower global and regional brain volume, primarily in hippocampus and prefrontal cortex (55). Associations to lower brain volume were found with high SBP and/or DBP, or with low BP, although a few studies observed no associations between BP and brain volume. All longitudinal cohort studies showed that high BP predicted greater volume reduction in brain.

Another study in patients with manifest arterial disease (coronary artery disease, cerebrovascular disease, peripheral artery disease, or abdominal aortic aneurysm) showed that at baseline low DBP (irrespective of the course of BP during follow-up) was associated to progressive subcortical atrophy (56). Also, patients with higher SBP, DBP and mean arterial pressure at baseline and declining BP levels during follow-up had less progressive subcortical atrophy. Another finding was that, compared to increasing BP levels over time, declining BP in patients with low baseline BP was linked to similar or more progression of subcortical atrophy.

These findings suggested that, although BP lowering seemed beneficial in the patients with

(14)

elevated BP levels, in those with lower BP more caution should be taken regarding BP control (56).

1.5.9 High cognitive activity

Very few studies have investigated the effects of cognitive activity on brain MRI measures. One study showed that greater cognitive activity and physical activity were related to lower WML volumes independently (57). Higher cognitive activity in midlife and currently predicted lower WML volumes. Cognitive and physical activity was hypothesized to reduce Aβ-dependent and - independent pathways for AD development, which may help in maintaining brain integrity.

1.5.10 Depression

A study of depressive and non-depressive AD patients with GM volume loss showed that AD patients with comorbid depression had greater GM volume loss mainly in the sensory and motor areas and right thalamus (58). Another study of subclinical depression and structural brain alterations showed that middle-aged men (age ≈51 years) with depressive symptoms had association with volume loss in hippocampus at the age of 59 years (59). A systematic review reported that early-onset depression was associated to reversible hippocampus atrophy, whereas late-onset depression appeared to indicate early neurodegeneration and cerebral abnormalities including WMH (60). A longitudinal study of 9 years showed that depressed mood predicted more WMLs in men and women, and somatic symptoms also predicted increased WML volume in men (61).

1.5.11 High cholesterol

A study found that higher HDL-cholesterol was associated with higher hippocampal volume in elderly patients aged 75-85 years. Thus, high HDL levels may be protective against dementia and hippocampus atrophy. This was thought to be due to brain lipoproteins and cholesterols facilitation for synaptic plasticity, although the association may also have been explained by reverse causality. No associations were found between total cholesterol or LDL-cholesterol and hippocampus volume (62). A similar association was not found in a larger sample (63). Midlife total serum cholesterol levels were not associated with cortical thickness in late life in the CAIDE study (51), although lipid-lowering medication seemed to be protective against WML (52).

(15)

A pilot study of cholesterol-lowering treatment in AD patients found a significant shrinkage in right hippocampus with atorvastatin therapy compared to placebo group, but no statistically significant reduction in total hippocampus volume was found. Atorvastatin therapy improved some cognitive measures compared to placebo (64). Another study found that low LDL- cholesterol levels were associated to smaller GM volumes in frontal and posterior cingulate in participants with hypertension, and in addition hypertensive participants had highest periventricular WMH (65).

1.5.12 Alcohol

One study reported that, while some regions of the brain were unharmed by the effects of alcoholism, regions at risk were the prefrontal cortex and subjacent white matter, cerebellar sites and white matter structures and tracts (66). MRI scans from people with excessive alcohol use showed smaller gray matter volumes in the cerebral cortex, and older alcoholics had greater shrinkage of gray and white matter volumes in frontal lobes compared to younger alcoholics. Also, abstinence from alcohol even in the shorter-term seemed to be linked to some improvement in volumes of cortical gray matter, overall brain tissue and hippocampal structures, although the mechanisms for brain tissue volume loss through excessive alcohol use, or the restoration by abstinence are not exactly known.

One review reported that MRI studies in general found a negative linear effect of alcohol consumption on brain volumes, although there appeared to be a U-shaped relationship with more white matter integrity in elderly low to moderate drinkers (67). The negative effects of alcohol seemed more pronounced in men. Another review (68) on postmortem and in vivo neuroimaging studies described that longitudinal studies suggested benefits of alcohol abstinence on cortical gray matter, overall brain, and hippocampal volumes. Longer-term abstinence was related to benefits on brain volumes especially in frontal and temporal regions.

Two other studies reported a U-shaped association between WML burden and alcohol intake (69), and an association between alcohol dependence and lower gray matter volumes in mesocorticolimbic system, possibly linked to neural changes which could lead to impulsive behavior (70).

2. Aims and relevance of this study

Due to the current lack of effective disease-modifying drugs, it is important to focus on prevention of dementia through lifestyle-related modifications. This study aimed to investigate

(16)

associations between LIBRA dementia risk score in midlife and late-life, and brain MRI findings from the CAIDE 1998 and 2005-2008 follow-ups. This study is relevant to dementia prevention programs because the LIBRA score contains modifiable risk and protective factors, and thus emphasizes the potential for dementia risk reduction.

3 Materials and methods 3.1 CAIDE study

Cardiovascular Risk Factors, Aging and Dementia (CAIDE) is a follow-up study that was based on random population-based samples of middle-aged individuals in 1972, 1977, 1982 and 1987 from Eastern Finland (part of the FINMONICA study and North Karelia Project for cardiovascular disease prevention). The study design has been previously described in detail (7, 8, 22, 32, 51, 52, 71), and is summarized at www.uef.fi/caide. Participants in the study were re-examined first in 1998 (2000 individuals were invited, and 1449 participated) and then in 2005-2008 (1426 of the initial 2000 individuals were still alive and living in the Kuopio and Joensuu areas, and 909 participated). All visits included assessments of sociodemographic, vascular, lifestyle and other health-related factors. For the LIBRA score, CAIDE data on all factors except diet and cognitive activity were available.

In 1998 and 2005-2008, CAIDE re-examinations also included detailed cognitive assessments and dementia and MCI diagnoses. These were done with a three-step protocol: screening, clinical phase and differential diagnostic phase. Screening in the first re-examination led to further evaluation in participants with ≤24 points on Mini-Mental State Examination (MMSE) (72). In the second re-examination, for screening; participants with ≤24 points or ≥3 point decline in MMSE or if participants had <70% delayed recall in the CERAD word list (73) or concerns for individual’s cognition lead for further evaluations. Clinical phase, in both re- examinations, consisted of medical and neuropsychological assessments. Differential diagnostic phase included MRI/CT for brain imaging, blood tests and cerebrospinal fluid (CSF) analysis if needed. Primary diagnosis was verified by a review board including the study physician, neuropsychologist, senior neuropsychologist and a senior neurologist. Based on available information diagnoses for dementia and MCI were ascertained by established criteria (74-76).

3.2 Assessment of LIBRA factors in the CAIDE study

The descriptions for assessment of the LIBRA factors in the CAIDE study are shown in Table 2.

Each variable was categorized, and each category was assigned points according to the LIBRA

(17)

index described in Table 1. The total LIBRA score was obtained by calculating the sum of all points for a specific CAIDE participant.

Low/moderate alcohol intake was based on answers to the question “Do you use alcohol?” with the following alternative answers: once or twice per year or less frequently, 3-4 times a year or approximately once in two months were given –1 LIBRA point; other alternatives such as not using alcohol, approximately once per month, once or twice per month, once per week, once or twice per week were given 0 LIBRA points. For alcohol the LIBRA weight was negative for low or moderate consumption because it was considered a protective factor.

Coronary heart disease was based on self-report of myocardial infarction or angina pectoris diagnosed by a physician, or a record of diagnosed ischemic heart disease or atrial fibrillation or flutter in the Finnish hospital discharge register based on ICD codes. In ICD-10 codes I20-I25 (ischemic heart diseases) and I48 (atrial fibrillation or flutter), in ICD 9 codes 410-414 (ischemic heart diseases) and 427.3A (atrial fibrillation or flutter) and in ICD 8 codes 410-414 (ischemic heart diseases) and 427.92 (atrial fibrillation) were given 1 LIBRA point. Absence of these ICD codes or self-report of no coronary heart disease were coded as 0.

Physical inactivity was based on the question “How often do you participate in leisure time physical activity that lasts at least 20-30 minutes and causes breathlessness and sweating”. Answer alternatives once a week, 2-3 times a month, a few times a year or less frequently, or unable to exercise due to injury or sickness were given 1.1 LIBRA points. The other answer alternatives (daily or 2-3 times a week) were given the value of 0.

Chronic kidney disease was based on ICD codes for chronic renal failure, proteinuria or chronic nephritis from the Finnish hospital discharge register: in ICD 10 codes N18 (chronic renal failure) and R80 (proteinuria), in ICD 9 codes 585 (chronic renal failure) and 791.0 (proteinuria), in ICD 8 code 582 (chronic nephritis). Diabetes was based on self-report for history of diabetes diagnosed by a physician or a record of diabetes in the Finnish hospital discharge register based in ICD codes: in ICD 10 codes E10-14, in ICD 9 code 250 and in ICD 8 code 250. These alternatives were given the value of 1.1.

Cholesterol was based on total serum cholesterol. Levels equal or higher than 6.50 mmol/l were coded as 1.4 LIBRA points, and levels lower than 6.50 mmol/l were coded as 0. Smoking was based on the question “Have you ever smoked?”; answering yes was coded as 1.5 points, and answering no was coded as 0. Midlife obesity was based on measuring body-mass-index; BMI same or higher than 30 kg/m² was given the value of 1.6 points, otherwise it was coded as 0 (if BMI data were available).

Midlife hypertension was defined based on measuring systolic and diastolic blood pressure from participants’ right arm after being seated for 5 minutes. Systolic blood pressure over 140 mmHg or

(18)

diastolic blood pressure over 90 mmHg were given the value of 1.6 points, and all other alternatives were given the value of 0 (if blood pressure data were available).

Healthy diet was based on CAIDE healthy diet index which was previously published (71), based on a questionnaire covering healthy and unhealthy dietary components. Data on dietary habits was collected only for a smaller subsample of CAIDE participants, and thus was not available for the present study.

Depressive symptoms were assessed based on two statements about hopelessness; “I feel it is impossible to reach goals that I would like to strive for” and “The future seems hopeless to me, and I can’t believe that things are changing for the better”, which participants were able to answer with 5 alternative options (Likert scale): 0-absolutely agree, 1-somewhat agree, 2-cannot say, 3- somewhat disagree or 4-absolutely disagree. Answers of “somewhat agree” or “absolutely agree” to both statements were coded as 2.1 LIBRA points. All other alternatives were coded as 0 and missing values as missing. Data on cognitive activity were not available, and this variable was not included in the LIBRA score calculations.

Table 2 Assessment of LIBRA factors from the CAIDE study.

Risk/protective

factor CAIDE assessment Low/moderate

alcohol intake Frequency of alcohol consumption question with alternative answers: once or twice per year or less frequently, 3-4 times a year, or approximately once in two months were given -1 LIBRA points.

Answer options not using alcohol, approximately once per month, once or twice per month, once per week, once or twice per week, daily were coded as 0.

Missing values were coded as missing.

Coronary

heart disease Self-report of myocardial infarction or angina pectoris diagnosed by physician, or records of diagnoses of ischemic heart disease or atrial fibrillation based on the Finnish Hospital Discharge Register ICD codes (ICD 10, ICD 9 or ICD 8, listed in the text above) were given 1 LIBRA point. Absence of self-reported or register-based coronary heart disease was coded as 0.

Physical

inactivity Answering to the question “How often do you participate in leisure time physical activity that lasts at least 20-30 minutes and causes breathlessness and sweating” with alternatives once a week, 2-3 times a month, a few times a year or less frequently or unable to exercise due to injury or sickness were given the value 1.1. Alternatives daily or 2-3 times a week were given value 0.

Missing values were coded as missing.

Chronic

kidney disease This variable was based on record of certain ICD codes from the Finnish

hospital discharge register (listed in text above); diagnoses in ICD-10, ICD-9 or ICD-8 of chronic renal failure, proteinuria or chronic nephritis were given the value of 1.1. Absence of these diagnostic codes was given the value of 0.

(19)

Diabetes Self-report of diabetes diagnosed by a physician or record of diagnoses of diabetes based on the Finnish Hospital Discharge Register of certain ICD codes (listed in text above) were given the value of 1.1. Absence of diabetes was coded as 0.

Cholesterol Serum cholesterol values < 6.5 mmol/l were coded as 0, values ≥ 6.50 mmol/l were coded as 1.4, and missing values as missing.

Smoking If participants had ever been smoking they were given value 1.5 if not value 0 and missing values were coded as missing

Midlife obesity If participants body-mass index was ≥ 30 kg/m² they were given value 1.6, all other values were given value 0 and missing values coded as missing

Midlife

hypertension If participants had either SBP ≥ 140 mmHg or DBP ≥ 90 mmHg they were given value 1, if less than these value 0 and missing values as missing.

Healthy diet Not available Depressive

symptoms Measuring hopelessness was based on two questions: “I feel it is impossible to reach goals that I would like to strive for” and “The future seems hopeless to me, and I can’t believe that things are changing for the better”. Participants rated their agreement or disagreement to the questions using Likert scale: 0- absolutely agree, 1-somewhat agree, 2-cannot say, 3- somewhat disagree or 4- absolutely disagree. Replies of “somewhat agree” or “absolutely agree” were coded as 2.1. All other alternatives were coded as 0 and missing values as missing.

High cognitive

activity Not available

3.3 MRI assessments in the CAIDE study

In 1998, MRI scans were taken only from individuals who needed further evaluation in the Kuopio cohort (39 with dementia, 31 with MCI) and 42 cognitively normal controls (total 112 participants). In 2005-2008 MRI scans were conducted in 113 participants from Kuopio (during the differential diagnostic phase), of which 69 did not have dementia and were considered for this study. Figure 1 below shows the number of participants included in the present study based on availability of both MRI and LIBRA data.

T1-weighted images were acquired using a three-dimensional magnetization prepared rapid acquisition gradient echo (3DMPRAGE) sequence on 1.5T MR scanners. One MRI scanner was used in the first re-examination (Siemens Magnetom Vision), and two scanners were used in the second re-examination (Siemens Magnetom Vision or Siemens Avanto). Axial fluid- attenuated inversion recovery images (FLAIR) were available from both re-examinations. The imaging parameters for both sequences and image check procedures have been previously described (32, 51, 52).

In the first re-examination, GM volume was measured using FSL FAST (FMRIB’s Automated Segmentation Tool) (77). WML were assessed from FLAIR images using the Wahlund semi-

(20)

quantitative visual rating scale (78). The lesions were rated separately from zero to three for five brain regions (frontal, parieto-occipital, temporal, infratentorial and basal ganglia) in both hemispheres. MTA was rated from zero (no atrophy) to four (end-stage atrophy) bilaterally from T1-weighted images according to the Scheltens visual rating scale (79).

In the second re-examination, brain volumes and cortical thickness were assessed with the FreeSurfer software package, version 5.0 (http://surfer.nmr.mgh.harvard.edu/) (80). WML volumes were calculated using CASCADE, an automatic pipeline developed at Karolinska Institute, Stockholm, Sweden (81, 82). Visual MTA assessment was done using the Scheltens scale. Alzheimer’s disease signature cortical thickness was calculated as the average cortical thickness of left and right entorhinal, inferior temporal, middle temporal and fusiform regions (83).

Figure 1 demonstrating design of this study with baseline CAIDE population and MRI populations in 1998 and 2005-2008 with available midlife LIBRA score.

3.4 Statistical analyses

Analyses were conducted using SPSS software version 23. Level of significance was p<0.05 in all analyses.

1) Midife LIBRA and MRI measures in 1998

(21)

MRI measures included WML visual rating total score (continuous variable, zero-skewness log- transformed), total gray matter volume (continuous variable, TIV-adjusted and zero-skewness log-transformed), and MTA visual rating (dichotomous, average of left and right MTA 0 or 1 versus 1.5 and above). Linear regression models or binary logistic regression models with MRI measures as dependent variables were used for continuous or dichotomous variables, respectively. Model 1 was adjusted for follow-up time, and model 2 was additionally adjusted for age in midlife, diagnosis, gender and education.

2) Midlife LIBRA and MRI measures in 2005-2008

MRI measures included WML visual rating total score (continuous variable, zero-skewness log- transformed), total gray matter volume (continuous variable, TIV-adjusted and zero-skewness log-transformed), medial temporal atrophy visual rating (dichotomous, average of left and right MTA 0 or 1 versus 1.5 and above), hippocampus volume (continuous variable, TIV-adjusted and zero-skewness log-transformed) and AD signature thickness (continuous variable, zero- skewness log-transformed). Linear regression models or binary logistic regression models with MRI measures as dependent variables were used for continuous or dichotomous variables, respectively. Model 1 was adjusted for follow-up time from midlife to 2005-2008 and scanner type, except for visually rated MTA. Model 2 was additionally adjusted for age in midlife, gender and education.

3) 1998 LIBRA and MRI measures in 1998

MRI measures included WML visual rating total score (continuous variable, zero-skewness log- transformed), total gray matter volume (continuous variable, TIV-adjusted and zero-skewness log-transformed), and MTA visual rating (dichotomous, average of left and right MTA 0 or 1 versus 1.5 and above). Linear regression models or binary logistic regression models with MRI measures as dependent variables were used for continuous or dichotomous variables, respectively. Model 1 was unadjusted, and model 2 was adjusted for age in 1998, gender and education.

4)1998 LIBRA and MRI measures in 2005-2008

MRI measures included WML visual rating total score (continuous variable, zero-skewness log- transformed), total gray matter volume (continuous variable, TIV-adjusted zero-skewness log-

(22)

transformed), medial temporal atrophy visual rating (dichotomous, average of left and right MTA 0 or 1 versus 1.5 and above), hippocampus volume (continuous variable, TIV-adjusted zero-skewness log-transformed) and AD signature thickness (continuous variable, zero- skewness log-transformed). Linear regression models or binary logistic regression models with MRI measures as dependent variables were used for continuous or dichotomous variables, respectively. Model 1 was adjusted for follow-up time, and model 2 was additionally adjusted for age in 1998, gender and education.

4 Results

4.1. Population characteristics

Characteristics of participants with available LIBRA score and MRI are shown in Table 3. In the CAIDE 1998 MRI population participants were older than in 2005-2008 MRI population, with mean (SD) age in midlife 51.09 (4.61) years vs 47.33 (4.69) years, and had less years of education 6.96 (2.53) vs 7.53 (2.46). In the 1998 MRI population participants had higher midlife LIBRA score mean (SD) 3.98 (2.04) vs. 3.57 (1.68), and more participants had depression N (%) 12 (14.6) vs. 4 (8.9), midlife hypertension 63 (76.8) vs. 30 (66.7), midlife obesity 21 (25.6) vs.

10 (22.2), high cholesterol 58 (70.7) vs. 29 (64.4) and coronary heart disease 8 (9.8) vs. 3 (6.7) when compared to the MRI population in 2005-2008.

Table 3. Characteristics of the 1998 and 2005-2008 CAIDE MRI populations with available data on LIBRA score

N CAIDE 1998 MRI

population N CAIDE 2005-2008 MRI population

Midlife examination

Age 82 51.1 (4.6) 45 47.3 (4.7)

Women (N. %) 82 50 (61) 45 28 (62)

Education. years 82 7.0 (2.5) 45 7.5 (2.5) APOE4 carriers (N, %) 81 32 (40) 44 16 (36) Follow-up time, years 82 22,9 (2,4) 45 30,8 (2,5) Low/moderate alcohol intake

(N, %) 82 26 (32) 45 13 (29)

Coronary heart disease (N, %) 82 8 (9.8) 45 3 (6.7) Physical inactivity (N, %) 82 51 (62) 45 28 (62)

(23)

Chronic kidney disease (N, %) 82 0 0

Diabetes (N, %) 82 1 (1) 45 1 (2)

High cholesterol (N, %) 82 58 (71) 45 29 (64)

Smoking (N, %) 82 31 (38) 45 17 (38)

Midlife obesity (N, %) 82 21 (26) 45 10 (22) Midlife hypertension (N, %) 82 63 (77) 45 30 (67)

Depression (N, %) 82 12 (14) 45 4 (8)

LIBRA score 82 4.0 (2.0) 45 3.6 (1.7)

MRI measures 1998 examination 2005-2008 examination

Hippocampus vol. N.A. 44 6.70 (1.22)

Total gray matter vol. 64 342 (56) 44 508 (61)

TIV 64 1386 (126) 44 154 (173)

AD signature thickness N.A. 44 2.56 (0.32)

MTA* 65 1,0 (0,0-2,5) 45 1,0 (0,0-2,5)

Deep WML N.A. N.A.

Periventricular WML N.A. N.A.

WML (Wahlund)* 82 5,0 (0,0-22,0) N.A.

WML volume N.A. 45 31.4 (28.6)

Values are means (SD) unless otherwise specified. * Median (range)

In the CAIDE 1998 re-examination, participants with available LIBRA score were younger compared to participants without complete data on LIBRA score (mean (SD) 51.09 (4.61) vs.

56.17 (5.00), p<0.001) (Table 4). Participants with available LIBRA score had longer follow-up time compared to those without available LIBRA score (22.93 (2.38) vs. 16.72 (4.51), p=<0.001). Lower percentage of women had available LIBRA score compared to participants without LIBRA (number (percentage) 50 (60.98) vs. 23 (76.67), p=0.123)

Table 4 - Comparison between participants with and without available data on LIBRA score in the CAIDE 1998 MRI population

(24)

N LIBRA score

available N LIBRA score

not available P-value

Age 82 51,1 (4,6) 30 56,2 (5.0) <0.001

Women (N. %) 82 50 (61) 30 23 (77) 0.123

Education. years 82 7.0 (2.5) 29 7.0 (3.0) 0.949 Follow-up time, years 82 22.9 (2.4) 30 16.7 (4.5) <0.001 Total gray matter

volume ⁶⁴ 0.36 (0.03) 26 0.36 (0.02) 0.640

Visually rated white

matter lesions ⁸² ^{6.0 (4,4)} ³⁰ ^{6.4 (4.0)} ^0.642 More severe medial

temporal atrophy (N, %) 65 14 (22) 26 8 (31) 0.353

Values are means (SDs) unless otherwise specified. P-values are shown from t-test for continuous variables, and chi2 test for categorical variables.

In the CAIDE 2005-2008 re-examination, participants with available LIBRA score were younger compared to participants without available data for LIBRA score (mean (SD) 47.33 (4.69) vs.

55.71 (4.37), p<0.001). Participants with available LIBRA score had longer follow-up time compared to those without available LIBRA (30.80 (2.53) vs. 22.76 (3.23), p<0.001).

Participants with available LIBRA score tended to have more severe MTA compared to those without available LIBRA (N (%) 11 (24.44) vs. 11 (45.83), p=0.069) (see Table 5).

Table 5 - Comparison between participants with and without available data on LIBRA score in the CAIDE 2005-2008 MRI population

N LIBRA score

available N LIBRA score not

available P-

value

Age 45 47.3 (4.7) 24 55.7 (4.4) <0.001

Women (N. %) 45 28 (62) 24 14 (58) 0.753

Education, years 45 7.5 (2.5) 22 8.3 (2.8) 0.246

Follow-up time, years 45 30.8 (2.5) 24 22.8 (3.2) <0.001 White matter lesion

volume ⁴⁵ 31357 (28589) 24 35868 (32372) 0.553

Hippocampus 44 0.0044 (0.00075) 22 0.0045 (0.00055) 0.523

(25)

Gray matter volume 44 0.33 (0.030) 22 0.34 (0.021) 0.268 AD signature cortical

thickness ⁴⁴ 2.56 (0.32) 23 2.66 (0.13) 0.161

More severe MTA (N,

%) ⁴⁵ ^{11 (24)} ²⁴ ^{11 (45)} ^0.069

Values are means (SDs) unless otherwise specified. P-values are shown from t-test for continuous variables, and chi2 test for categorical variables.

4.2. Midlife LIBRA score and MRI measures in the 1998 and 2005-2008 CAIDE re- examinations

Associations between midlife LIBRA score and MRI measures in the CAIDE 1998 re- examination are shown in Table 6. Model 1 was only adjusted for follow-up time, and model 2 was additionally adjusted for follow-up time, diagnosis (control/MCI/dementia), age in midlife, gender and education. Higher LIBRA score in midlife was significantly associated with more severe white matter lesions; standardized β-coefficient (p-value) was 0.240 (0.032) after all adjustments (model 2). For total gray matter volume the relation to LIBRA score was also significant in model 1, with standardized β-coefficient (p-value) -0.256 (0.047), indicating that higher LIBRA score was related to lower total gray matter volume. No significant association was found for the fully adjusted model with gray matter volume. There was also no significant association of midlife LIBRA score with visually rated MTA in 1998.

Table 6. Midlife LIBRA and MRI measures in 1998 CAIDE visit

Model 1 Model 2

Standardized β-coefficient (p-value) Visually rated white

matter lesions 0.211 (0.053) 0.240 (0.032) Total gray matter

volume -0.256 (0.047) -0.048 (0.703)

OR [95% CI], p-value More severe medial

temporal atrophy 0.892 [0.655-1.214], 0.466 0.764 [0.521-1.12], 0.170 Model 1: Adjusted for follow-up time

Model 2: Model 1 + diagnosis + age in midlife, gender and education

Linear regression models were used for white matter lesions and total gray matter volume; and binary logistic regression models for medial temporal atrophy.

(26)

Associations between midlife LIBRA score and MRI measures in the CAIDE 2005-2008 re- examination are shown in Table 7. MRI measures were WML volume, hippocampus volume, gray matter volume, cortical thickness in Alzheimer’s disease signature areas and visually rated MTA.

Associations between LIBRA score and MRI measures in 2005-2008 were examined by two models. Model 1 was adjusted for follow-up time from midlife to 2005-2008 and scanner type, except for visually rated MTA. Model 2 was additionally adjusted for age in midlife, gender and education.

Table 7 shows that no statistically significant associations were found between LIBRA score in midlife and WML volume, hippocampus volume, gray matter volume, or AD signature cortical thickness in 2005-2008. Higher LIBRA score was significantly related to more severe MTA when adjusted for follow-up time; OR (95% CI) was 1.797 (1.033-3.127), p-value 0.038.

However, this association was not significant after full adjustments in model 2; OR (95% CI) was 1.365 (0.699-2.663), p=0.362. There was a trend of association for higher LIBRA score in midlife and lower gray matter volume; standardized β-coefficient (p-value) was -0.232 (0.109), but this was not significant in model 2.

Table 7. Midlife LIBRA and MRI measures in 2005-2008 CAIDE visit

Standardized β-coefficient (p-value) White matter lesion

volume 0.077 (0.615) -0.086 (0.595)

Hippocampus volume -0.195 (0.220) -0.164 (0.343) Gray matter volume -0.232 (0.109) -0.105 (0.494) AD signature cortical

thickness -0.058 (0.670) 0.028 (0.847)

temporal atrophy 1.797 [1.033-3.127], 0.038 1.365 [0.699-2.663], 0.362 Model 1: Adjusted for follow-up time from midlife until 2005-2008, scanner type (EXCEPT MTA) Model 2: Model 1+ age in midlife, gender and education

Also potential influence of APOE genotype on associations between midlife LIBRA score and 1998 or 2005-2008 MRI measures was examined, but no significant effect was found (data not shown).

(27)

4.3. Associations between LIBRA score in 1998 and MRI measures in the 1998 and 2005- 2008 CAIDE re-examinations

Cross-sectional associations between LIBRA and MRI measures in the 1998 CAIDE re- examination are shown in Table 8. LIBRA score in 1998 was not significantly associated with MRI measures in 1998. Higher LIBRA score in 1998 tended to be related to lower total GM volume; standardized β-coefficient (p-value) was -0.20 (0.094), but this association was not significant with full adjustments.

Table 8. LIBRA in 1998 and MRI measures in 1998 CAIDE visit

Standardized β-coefficient (p-value) Visually rated white

matter lesions 0.12 (0.272) 0.12 (0.285)

Total gray matter

volume -0.20 (0.094) -0.083 (0.497)

temporal atrophy 1.02 [0.73-1.41], 0.928 0.93 [0.63-1.38], 0.722 Model 1: Unadjusted

Model 2: Model 1 + diagnosis + age in 1998, gender and education

Table 9 represents associations between LIBRA score in 1998 and MRI measures in 2005-2008.

There were no statistically significant associations in any of the models.

Table 9. LIBRA in 1998 and MRI measures in 2005-2008 CAIDE visit

Standardized β-coefficient (p-value) White matter

lesion volume -0.01 (0.953) -0.06 (0.694)

Hippocampus -0.15 (0.330) -0.22 (0.171)

(28)

Gray matter

volume -0.21 (0.149) -0.17 (0.280)

AD signature

cortical thickness -0.15 (0.287) -0.17 (0.257) OR [95% CI], p-value

More severe medial temporal atrophy

1.01 [0.716-1.419], 0.962 0.94 [0.602-1.470], 0.788

Model 1: Adjusted for follow-up time from 1998 until 2005-2008, scanner type (except MTA) Model 2: Model 1+ age in 1998, gender and education

5. Discussion

This was the first time LIBRA score was studied in relation to brain MRI measures. Statistical analyses in this study showed that higher LIBRA score in midlife was significantly associated with more severe white matter lesions approx. 21 years later (1998 CAIDE re-examination), but not approx. 31 years later (2005-2008 CAIDE re-examination) in the fully adjusted models.

This may be explained by several reasons: smaller MRI population in 2005-2008 reducing statistical power; different white matter lesions measures (in 1998 visual rating was used, and in 2005-2008 the volume of the lesions was measured); and different MRI populations. The 1998 population included people with dementia and, although diagnosis was adjusted for in the models, it is possible that this population may have had more severe white matter lesions.

The 2005-2008 population did not include dementia but only at-risk participants who may have had less severe white matter lesions.

Higher LIBRA score in midlife was associated with lower total gray matter volume approx. 21 years later, and there was a trend for association also approx. 31 years later in models adjusted to follow up time. However, the fully adjusted models were not statistically significant. Midlife LIBRA score showed no association with more severe medial temporal atrophy 21 years later, but the association was significant 31 years later. This suggests that a higher LIBRA score may impact MTA over a longer period of time. However, the fully adjusted models with medial temporal atrophy as outcome were not statistically significant, suggesting that e.g. age, gender, or education level may have an impact on the relations between midlife LIBRA score and total gray matter volume and medial temporal atrophy.

(29)

No statistically significant associations were observed between midlife LIBRA score and other measures of gray matter atrophy, or between late-life LIBRA score and any of the investigated MRI measures in the present study.

Results of this study are in line with earlier findings regarding relations between CAIDE dementia risk score and brain MRI measures. An observational study using CAIDE dementia risk score and MRI scans from the FINGER study showed that higher CAIDE dementia risk score was associated with more severe deep WML, lower gray matter and hippocampal volume and lower cortical thickness (84). Another study investigating associations between CAIDE dementia risk score and MRI scans found that higher CAIDE score was associated to more severe MTA and white matter changes (85). Another study found that a higher CAIDE dementia risk score was mostly associated with more severe white matter hyperintensities and higher MTA score (7). The CAIDE score is different from the LIBRA score in that it also includes age, gender and education.

The effects of LIBRA dementia risk and protective factors independently in previous literature seem to be in line with the results of this study for the most part. Healthy diet has been shown to associate with larger volume in hippocampus (23) and higher vegetable intake to less GMV loss (24). Atherosclerotic cardiovascular diseases have been shown to result in increased WML volume and decrease in GMV (25). Physical inactivity has associations to hippocampal atrophy (30) and association to smaller gray matter volumes (31). Renal dysfunction or decreased eGFR has been associated to subcortical (37) and deep (37, 39) white matter hyperintensities but DM and CHD (38) and hypertension are thought to promote the changes of CKD, although renal dysfunction may lead to smaller GM and hippocampal volume and thinner cortices independently (40). Midlife diabetes and longer time of diabetes showed risk for hippocampal atrophy (41, 43) and longer duration also showed smaller total and regional deep GM and hippocampal volume and severe diabetes was linked to WM hyperintensities (43). Long-term smoking has been shown to result in smaller cortical (46) and subcortical GMV (46, 47) and thinner cortices (48). Obesity as BMI and waist-hip-ratio (50) and waist-hip-ratio independently (49) have been shown to associate with lower gray matter volume. High blood pressure has been associated with WMLs (53, 54) and lower global and regional brain volume, in particular hippocampus and prefrontal cortex (55). High cognitive activity along with physical activity were independently associated with lower WML volumes (57). Depression

(30)

has been shown to associate with hippocampal volume loss (59) and GM volume loss in AD patients (58). Early onset depression was associated with reversible hippocampus atrophy and late onset depression with WMH (60). Depressed mood predicted WMLs (61). Alcohol use was associated with smaller GMVs in cerebral cortex and older alcoholics seem to have smaller GMVs and WMVs in frontal lobes (66). Alcohol dependence was related to lower GMVs in mesocorticolimbic system (70).

Strengths of this current study are the long follow up time and data on several modifiable and lifestyle related risk factors included in the LIBRA score. By focusing on these factors in dementia prevention programmes it may be possible to lower the risk for developing dementia.

Also, LIBRA score includes chronic conditions, such as cardiovascular diseases, chronic kidney disease, diabetes, high cholesterol, obesity, hypertension and depressive symptoms. This study shows their combined influence on MRI scan findings from the CAIDE population. Another strength is that in addition to risk factors LIBRA takes in to account protective factors. This would allow future dementia prevention programmes to favour e.g. low or moderate alcohol consumption and healthy diet.

Limitations of this study should be noted. The main one is that statistical power was limited due to small sample of participants with available data from MRI scans and LIBRA score (N=82 in 1998 and 45 in 2005-2008). The 1998 MRI population was selected using a case-control design, i.e. it included a larger proportion of dementia and MCI compared with the entire CAIDE population, although diagnosis was accounted for in the analyses. MRI participants without dementia in the 2005-2008 follow-up were not significantly different from the original CAIDE population (7).

High cognitive activity was not included in statistical analyses or LIBRA scoring because for high cognitive activity there was no data available in the CAIDE population. LIBRA score does not include e.g. age, which is a strong risk factor for dementia. The reason why unmodifiable risk factors were not included is because LIBRA was designed to focus only on modifiable and lifestyle-related risk and protective factors. However, in the present study adjustments were made in the analyses for follow-up from the time from where LIBRA was calculated to the time to MRI scanning in question, e.g. follow-up time from midlife to 1998 MRI scanning, or from LIBRA in 1998 to MRI scanning in 2005-2008. Adjustments were also made for age, gender, and education.

Does a midlife dementia risk score relate to late-life structural brain changes on MRI?

Does a midlife dementia risk score relate to late-life structural brain changes on MRI?

Table of contents