Study sample
Data were derived from the on-going Longitudinal Aging Study Amsterdam (LASA) [15]. This is a longitudinal, population-based study in the Netherlands focusing on trajectories of physical, psychological, social and cognitive functioning in subjects aged 55 years and older. In 1992–1993 a random sample of men and women aged 55–85 years, stratified for age and sex, from three geographic areas of the Netherlands (Amsterdam, Zwolle and Oss) was included. Follow-up measurements were conducted about every 3 years. Data collection included a main and medical interview conducted in the homes of the subjects. The main interview was done by trained and supervised interviewers and the medical interview was performed by trained nurses. All subjects gave informed consent and the study was approved by the Ethical Review Board of the VU University Medical Center (VUmc), Amsterdam, the Netherlands and conducted according to the principles of the Helsinki declaration.
At the start of the study in 1992–1993, 3107 subjects were enrolled. To select cognitively healthy subjects at baseline, subjects with an age and education corrected MMSE lower than 27 points were excluded (this cut-off is based on the lowest 10th percentile of the MMSE in the Maastricht Aging Study (MAAS) [16], leaving 2527 subjects at baseline. In 1995–1996, 2545 subjects were re-examined. See for further details about the following cycles of this LASA cohort and for the sample size per risk factor Additional file 1: Table S1 and S2.
Measurements
Biomaterial
The ApoE phenotyping was done either in 1992–1993 or 1995–1996 at the Immunochemisch Laboratorium of the VUmc. The blood samples were frozen at -80 °C until determination in 1997–1998. The method used is described by Havekes et al. (1987) and consisted of isoelectric focusing of delipidated serum samples, followed by immunoblotting [17]. In the analyses, we used the presence of an ApoE ε4 isoform (phenotypes ε2/4, ε3/4, ε4/4) as a dichotomous variable [17]. The ApoE ε4 isoform was used as proxy for the presence of an APOE ε4 allele.
Cholesterol levels (total cholesterol, High-Density Lipoprotein (HDL) cholesterol and Low-Density Lipoprotein (LDL) cholesterol) and homocysteine (in combination with vitamin B12) were determined in morning blood samples collected in 1995–1996 (second LASA cycle). Subjects were allowed to eat toast and drink tea, but no dairy products. The EDTA plasma samples were stored at -80 °C and analyzed by the Department of Clinical Chemistry of the VUmc in 2001/2002 (homocysteine) and 2005 (cholesterol). For determination of total cholesterol and HDL cholesterol enzymatic colorimetric tests were used. LDL cholesterol was calculated as total cholesterol minus HDL-cholesterol minus VLDL-cholesterol; VLDL-cholesterol was calculated as total triglyceride concentration expressed in mmol/L multiplied by 0.456 [18]. This method is less reliable when the triglyceride level is ≥5.0 mmol/L. Therefore, this analysis was only done for triglyceride levels of < 5.0 mmol/L. Total homocysteine was determined with the Abbott IMx analyser which uses fluorescence polarization immunoassay (FPIA) technology. Serum levels of vitamin B12 were determined at the Endocrine Laboratory of the VUmc with a competitive immunoassay luminescence on the automated ACS 180 System (Bayer Diagnostics, Mijdrecht, The Netherlands).
For determination of the inflammation markers (interleukin-6 (IL-6), C-reactive protein (CRP) and a1-antichymotrypsin (ACT)) serum collected in 1992–1993 (only in Amsterdam and Zwolle) was stored at -80 °C until determination in 2002–2004. Sensitive regular immunoassays (ELISA) were used at Sanquin Research (Amsterdam) to determine IL-6, CRP and ACT. CRP was expressed in ug/ml, IL-6 in pg/ml and ACT in % of normal plasma. The normal human plasma pool (% NHP) used as a standard for ACT contained ~ 300 mg ACT per L. For part of the subjects, CRP levels were determined directly after blood sampling.
Both the cholesterol as the inflammation markers, were added to the analyses as continuous variables.
Comorbidity
Hypertension was defined as a blood pressure > 140/90 mmHg measured at the upper arm or the use of antihypertensive medication collected in the first follow-up measurement in 1995–1996 (at baseline blood pressure was measured only at the finger). Post-hoc we also analyzed the association of a measured high blood pressure and the use of antihypertensive medication with cognitive decline separately.
The presence of a history of myocardial infarction (MI), DM or stroke was assessed by self-report. The assessment of comorbidity by self-report was found to be comparable with the medical information reported by the general practitioner [19].
Depressive symptoms were assessed using the Center for Epidemiologic Studies Depression scale (CES-D) [20]. The CES-D is a self-report scale containing 20 items describing depressive symptoms. The maximum score is 60 with higher scores indicating more depressive symptoms. In the analyses, the CES-D was used as a continuous variable.
Lifestyle
The number of alcohol consumptions was categorized into three categories: 0 alcoholic drinks per day (‘none’ group), 1–2 alcoholic drinks for men and 1 alcoholic drink for women per day (‘minimal’ group) or > 2 alcohol drinks for men and > 1 alcohol drink for women per day (‘moderate’ group) [8]. Smoking was dichotomized in ‘yes (or stopped within one year)’ or ‘no’.
For the assessment of physical activity, the LASA Physical Activity Questionnaire (LAPAQ) was used addressing walking outdoors, bicycling, light household, heavy household, and two sports activities [21]. The subjects are asked how often and how long they carried out these activities in the past 2 weeks. In the analyses, total physical activity in minutes per day was used as a continuous variable.
Cognitive outcome measures
Two different neuropsychological tests were used as outcome measures: The Mini-Mental State Examination (MMSE) and 15 Words Test (15WT). The MMSE is the most used screening instrument for global cognitive dysfunction [22]. The score ranges from 0 to 30 points, with higher scores indicating better cognitive functioning. The 15WT is the Dutch version of the Auditory Verbal Learning Test [23]. Fifteen words have to be learned over five trials. In LASA the 15WT is restricted to three trials due to a limitation in time. In this study we used the maximum immediate recall score and delayed recall score, both ranging from 0 to 15 words. The delayed recall was assessed after 20 minutes of distraction.
Statistical analyses
Spline regression analyses
Former studies, have shown that the association between age and cognition is nonlinear [24]. Linear regression techniques are therefore not sufficient enough to estimate this association and spline regression analyses are indicated to fit the nonlinear longitudinal associations between age and the cognitive outcome measures more precisely (Additional file 2: Spline regression analyses) [25]. To achieve the best fit of the data with a spline regression model, either linear or cubic splines can be used. Based on the likelihood-ratio (LR) test we determined which of these two types of splines showed a better fit with our data. We examined the positions where the splines join smoothly together, referred to as knots in spline regression analyses. We identified the optimal position of the knots by testing both a model with one and two knots and moving those 5 years up and down. The ages that corresponded to the position of the knots were used to separate our sample into different age groups, to facilitate interpretation of results. Lastly, for all the different risk factors and outcome measures we compared our final model with a linear regression model without splines to test whether the model with splines showed a better fit (based on the LR test).
Differences between age groups at baseline
Statistically significant differences in baseline characteristics between the age groups were determined by using ANOVA for continuous variables, chi-square for categorical variables and Kruskal-Wallis test for skewed variables. Mixed model analysis was used to determine the difference in cognitive test score change per year in the age groups.
Association of risk factors with cognitive outcome measures
We performed three different analyses to determine the association of the risk factors (measured at baseline or, for hypertension and some biomaterial measurements, in the second cycle) with the cognitive outcome measures. In all these analyses, splines (determined as described in 2.3.1) were added to the model to estimate the association between age and the cognitive outcome measure. First, we determined the association of the risk factors with the three cognitive outcome measures in the total sample by using a linear mixed model (including a random intercept and fixed slopes). Secondly, we added the interaction of the risk factors with the splines to the analyses to assess the age dependency of the risk factors. Because the splines represent different age groups, a significant interaction means that the association of that risk factor with the cognitive outcome measure is different between age groups. If this interaction was statistically significant for a categorized risk factor, we visualized the association in a figure. Lastly, we determined the association coefficient per age group of the risk factors with the cognitive outcome measures. This last step helps us to interpret the results we found with the interaction analyses (we also performed these analyses for the risk factors that did not show a significant interaction). All the analyses were adjusted for sex and education (in years).
Selection during follow-up
To determine whether there was a selection towards healthier subjects during follow-up, we determined the baseline values of the risk factors and cognitive outcome measures of the subjects that were present in the sample during each LASA cycle. Decreasing baseline values during follow-up would be indicative of selection towards healthier subjects.
Statistical software
The spline regression analyses were performed with the statistical software R version 3.2.5 (http://www.r-project.org). The statistical significance of the association of the risk factor with cognitive decline per age group was determined with Stata version 15. The differences in baseline characteristics between the three age groups were analyzed with SPSS Statistics version 22. The level of significance was set to p = 0.05.