Setting and study population
This study was embedded within the Rotterdam Study, a prospective population-based cohort study. In 1990, all residents aged 55 and older residing in Ommoord, a district of Rotterdam, the Netherlands, were invited. Of the 10,215 invited inhabitants, 7,983 agreed to participate in the baseline examinations. In 2000, the cohort was extended: again all residents aged 55 and older of the same district were invited, except for the participants that were already participating in the original cohort. Of the 4,472 invitees, 3,011 agreed to participate. Follow-up examinations take place every 3 to 4 years [16].
From the original cohort, we excluded 455 participants because they were not properly screened for dementia, 482 participants because they had prevalent dementia, and 43 participants for lack of follow-up information on the dementia diagnosis. Finally, 7,003 participants were included in the analyses. In the extended cohort, we excluded 29 participants because they had prevalent dementia and 29 for lack of follow-up information on the dementia diagnosis. Finally, 2,953 participants were included in the analysis. Even though a part of the original cohort still participated in 2000, those participants were only included in the original, and not in the extended cohort. We aimed to avoid overlapping individuals, since such a correlation would limit the assessment of cohort effects.
The Rotterdam Study has been approved by the medical ethics committee according to the Population Study Act Rotterdam Study and written informed consent was obtained.
Selection and measurement of risk factors
Potentially causal and modifiable risk factors for dementia were chosen on the basis of previous literature [1–3]. The following risk factors were selected: overweight and obesity, hypertension, diabetes mellitus, unfavorable cholesterol levels, smoking, and low educational level.
All risk factors were measured at baseline. Weight and height were measured at the research center visit and body mass index (BMI) was calculated as weight in kilograms divided by length in meters squared. BMI was categorized into four categories: underweight (<18.5), normal weight (18.5–25), overweight (25–30), and obesity (>30). Blood pressure was measured in sitting position on the right arm and calculated as the average of two measurements using a random-zero sphygmomanometer. Hypertension was defined as a blood pressure ≥140/90 mmHg or use of blood pressure lowering medication, prescribed for the indication of hypertension. Diabetes mellitus was defined as a fasting serum glucose level ≥7.0 mmol/L, non-fasting serum glucose level ≥11.1 mmol/L, or use of anti-diabetic medication. Serum glucose, total cholesterol, and high-density lipoprotein (HDL)-cholesterol levels were acquired by an automated enzymatic procedure (Boehringer Mannheim System). Medication use, educational level, and smoking habits were assessed by interview. For people not taking lipid-lowering medication, total cholesterol/HDL-cholesterol ratio was divided in quartiles using the lowest category as the reference category. People using lipid-lowering medication were added as the fifth category and also compared to the reference category. Educational level was categorized as low (primary education or lower vocational education), intermediate (secondary education or intermediate vocational education), and high educational level (higher vocational education or university). Smoking habits were categorized as current, former, and never smoking.
Although stroke, coronary heart disease, heart failure, and atrial fibrillation are cardiovascular diseases that may result from cardiovascular risk factors and can be modified only via secondary prevention, several studies have shown that these diseases are related to dementia or cognitive impairment, independently of cardiovascular risk factors [17–20]. We therefore assessed the PAR of these factors in an additional analysis. A history of stroke, coronary heart disease (myocardial infarction or revascularization procedure), heart failure, and atrial fibrillation was evaluated using home interviews and confirmed by reviewing medical records [21, 22].
Assessment of dementia
Participants were screened for dementia at baseline and at follow-up examinations using a three-step protocol [11]. Screening was done using the Mini-Mental State Examination (MMSE) and the Geriatric Mental Schedule (GMS) organic level. Screen-positives (MMSE <26 or GMS organic level >0) subsequently underwent an examination and informant interview with the Cambridge Examination for Mental Disorders in the Elderly [11]. Participants who were suspected of having dementia underwent extra neuropsychological testing if necessary. Additionally, for persons not visiting the research center, the total cohort was continuously monitored for dementia through computerized linkage of the study database and digitized medical records from general practitioners and the Regional Institute for Outpatient Mental Health Care. When information on neuroimaging was required and available, it was used for decision making on the diagnosis. Ultimately, a consensus panel, led by a neurologist, decided on the final diagnosis in accordance with standard criteria for dementia (DSM-III-R) [11]. Both the original and the extended cohort were followed for dementia until 10 years after baseline examinations: until 2000–2003 for the original cohort and until 2010–2012 for the extended cohort. Follow-up for dementia was complete for 99.5 % of potential person-years in the original cohort and for 98.5 % of potential person-years in the extended cohort.
Statistical analyses
We imputed missing data on the investigated risk factors (3.3 % in the original cohort and 6.7 % in the extended cohort) using the mean of five imputations. Missing data were imputed on age, sex, and all other investigated risk factors. Differences between the original and the extended cohort were calculated using logistic regression models, adjusting for age and sex where appropriate. Analyses were performed using statistical software package SPSS 20.0.
We calculated PARs and corresponding 95 % confidence intervals (CIs) using the Interactive Risk Attributable Program (US National Cancer Institute) [23]. This metric is also referred to as population attributable fraction, or PAR%. We provide logit transformed 95 % CIs accompanying the PARs as these are better interpretable and more stringent with regard to our combined PAR estimates [24]. The PAR was estimated and adjusted for confounding factors according to the following formulae:
$$ \mathrm{P}\mathrm{A}\mathrm{R}=1-{\displaystyle {\sum}_{i=1}^{\mathrm{I}\ }}{\displaystyle {\sum}_{j=1}^J}{\rho}_{ij}{R}_{i\Big|j}^{-1} $$
(1)
where
$$ \left.Ri\right|j=\frac{ \Pr \left(D=1\Big|X= xi,C=cj\right)}{ \Pr \left(D=1\Big|X=x1,C=cj\right)} $$
(2)
and
$$ {\uprho}_{\mathrm{i}\mathrm{j}} = \Pr \left(\mathrm{X}={\mathrm{x}}_{\mathrm{i}},\mathrm{C}={\mathrm{c}}_{\mathrm{j}}\left|\mathrm{D}=1\right.\right) $$
(3)
with D = 1 denoting presence of disease, X denoting exposure with i levels, and C denoting a confounder with j levels. Participants were censored at date of dementia, date of death, or last date of follow-up, whichever came first. The relative risk was estimated from a multivariable Poisson model [23]. First, we calculated the PAR for overweight and obesity, hypertension, diabetes mellitus, unfavorable cholesterol levels, smoking, and low educational level. Second, we calculated the PAR of stroke, coronary heart disease, heart failure, and atrial fibrillation. We calculated the PAR for each risk factor separately as well as the combined PAR per cohort. Many risk factors have a coinciding effect on the etiology of dementia. Therefore, the combined PAR cannot be estimated by the sum of the separate PARs as this would lead to an overestimation. The advantage of the statistical program used in this study is that it takes into account that a disease case can simultaneously be attributed to more than one risk factor. Hence, it takes into account the overlap between the PARs of each risk factor when estimating the combined PAR. The combined PAR included the PARs of modifiable risk factors which were associated to dementia in the expected direction in our study, meaning the relative risk was above one. PARs cannot be calculated using a relative risk below one as this will result in a PAR which cannot be interpreted [25].
Every PAR was adjusted for age, sex, and all other risk factors included in the model. Age was added per 5-year categories into the models. Because of a small number of people in the oldest age categories, the oldest age category of both cohorts comprised people 85 years of age and older.
