Study population
The Rotterdam Study is a Dutch population-based, prospective cohort study. This project was initiated in 1990–1993 in the Ommoord district of Rotterdam. The design and rationale of the Rotterdam Study have been described in detail elsewhere [18]. In summary, all inhabitants of this district aged 55 years and over were invited to participate, leading to a baseline cohort of 7983 participants (RSI). Over the years, two more allocation rounds were held. The first, in 2000–2001, included all inhabitants aged 55 years and over, leading to recruitment of an additional 3011 participants (RSII) [18]. In a second extension initiated in 2006, 3932 participants aged 45 years and over were included (RSIII) [18]. For follow-up, examinations were scheduled every 3–5 years [18]. The Rotterdam Study complies with the Declaration of Helsinki and has been approved by the Medical Ethics Committee of the Erasmus Medical Centre and also complies with the Dutch Ministry of Health, Welfare and Sport. All participants provided written informed consent to obtain and process data from their treating healthcare providers.
Population for analysis
The present study used data from the third visit of the first cohort (RSI-3) and the baseline examinations of the second (RSII-1) and third (RSIII-1) cohorts (electronic supplementary material [ESM] Fig. 1). A total of 6816 women were eligible for the analysis. Of these, 2053 women were excluded because (1) there was no information on their menopause status (n = 9); (2) they were not postmenopausal (n = 732); (3) age at menopause was not known (n = 145); (4) they did not give informed consent for type 2 diabetes follow-up (n = 56); (5) they had prevalent type 2 diabetes (n = 609); and (6) no information on incident type 2 diabetes was available (n = 502; Fig. 1). A further 1124 women were excluded because they had experienced non-natural menopause (n = 1109) or the type of menopause was not known (n = 15), leaving 3639 women included in the final analysis (Fig. 1).
Assessment of age at menopause
Menopausal status was evaluated using a subsection of the home interview questionnaire [18]. One set of questions was designed to obtain information on the timing of the last menstrual period, whether the respondent had experienced a natural menstrual period within the past 12 months and the age at last period for women who had not had a period for at least 12 months. Postmenopausal women were defined as women who reported an absence of menstrual periods for 12 months. For women who had experienced a natural menopause, age at menopause was defined as the self-reported age at the time of the last menstruation. For all women reporting menopause after gynaecological surgery or radiation therapy, and for those reporting any other operations before the age of 50 years that might have led to menopause, information on the exact date and type of operation was verified using general practitioner records, including correspondence from medical specialists.
Ascertainment of type 2 diabetes
Participants were followed up from the date of the baseline centre visit onwards. At baseline and during follow-up, cases of prevalent and incident type 2 diabetes were ascertained through active follow-up using general practitioner records, hospital discharge letters and glucose measurements from Rotterdam Study visits, which take place approximately every 4 years [19]. Prevalent and incident type 2 diabetes were defined according to recent WHO guidelines as a fasting blood glucose concentration of ≥7.0 mmol/l, a non-fasting blood glucose concentration of ≥11.1 mmol/l (when fasting samples were unavailable) or the use of glucose-lowering medication [20]. Information on the use of glucose-lowering medication was obtained from both structured home interviews and pharmacy records [19]. At baseline, more than 95% of the Rotterdam Study population was covered by pharmacies within the study area. All potential type 2 diabetes events were independently adjudicated by two study physicians. In the case of disagreement, consensus was sought with an endocrinologist. Follow-up data were complete until 1 January 2012.
Potential confounding variables
Information on current health status, medical history, medication use, smoking behaviour, socioeconomic status and other factors was obtained at baseline (RSI-3, RSII-1 and RSIII-1; ESM Fig. 1). Education level was defined as low (primary education), intermediate (secondary general or vocational education) or high (higher vocational education or university). Data on age at menarche were collected by asking women, ‘How old were you when you had your first menstrual period?’ The retrospective data on self-reported number of pregnancies of at least 6 months’ duration and use of hormone replacement therapy were collected by a questionnaire during the home interview. Participants were asked whether they were currently smoking cigarettes, cigars or pipes. Alcohol intake was measured in grams of ethanol per day. History of CVD was defined as a history of CHD (myocardial infarction, revascularisation, coronary artery bypass graft surgery or percutaneous coronary intervention), heart failure and stroke, and was verified from the medical records of the general practitioner. BP was measured in the sitting position at the right upper arm with a random-zero-sphygmomanometer. The mean of two consecutive measurements was taken. Medication use information was based on home interview. Antihypertensive medication use was defined as use of diuretics, β blockers, angiotensin-converting-enzyme inhibitors and calcium channel blockers. All biochemical variables were assessed in serum samples taken after fasting. Thyroid-stimulating hormone (TSH) was measured on the Vitros Eci (Ortho Diagnostics). Total cholesterol (TC), HDL-cholesterol (HDL-C), triacylglycerol (TG) and C-reactive protein (CRP) were measured on the COBAS 8000 Modular Analyzer (Roche Diagnostics). LDL-cholesterol (LDL-C) levels were estimated indirectly from measurements of TC, HDL-C and TG by means of the Friedewald equation [21]. The corresponding interassay CVs are: TSH, <13.2%; lipids, <2.1%; and CRP, <16.9%. Physical activity was assessed using the Longitudinal Aging Study Amsterdam Physical Activity Questionnaire and is expressed in MET-h/week [22].
Potential intermediate variables
All intermediate variables were assessed at baseline (RSI-3, RSII-1 and RSIII-1; ESM Fig. 1), and the height (m) and body weight (kg) of participants were measured without shoes and heavy outer garments. BMI was calculated as weight divided by height squared (kg/m2). Fasting insulin and glucose levels were measured using a COBAS 8000 Modular Analyzer (Roche Diagnostics). The interassay CVs are <8% and <1.4% for insulin and glucose, respectively. Total oestradiol levels were measured using RIA and sex hormone-binding globulin (SHBG) levels were measured using the Immulite platform (Diagnostics Products, Breda, the Netherlands). The minimum detection limit for oestradiol was 18.35 pmol/l. Serum levels of total testosterone were measured using liquid chromatography–tandem MS. The corresponding interassay CVs for total oestradiol, SHBG and total testosterone are <7%, <5% and <5%, respectively. Serum levels of dehydroepiandrosterone and dehydroepiandrosterone sulphate were estimated in 12 batches by coated-tube or double-antibody RIAs (Diagnostic Systems Laboratories, Webster, TX, USA). In self-reported white participants, genotyping was conducted using the Illumina 550K array. We selected 54 SNPs previously reported to have an association with age at natural menopause from a genome-wide association study of 70,000 women [23]. We calculated a weighted genetic risk score by multiplying the number of risk alleles at each locus by the corresponding reported β coefficient from the previous genome-wide association study and then summing the products. The total score was then divided by the average effect size multiplied by 100 to rescale the scores to a range of 0–100.
Statistical analysis
Main analyses
Person-years of follow-up were calculated from the study entry date (RSI-3, March 1997 – December 1999; RSII-1, February 2000 – December 2001; and RSIII-1, February 2006 – December 2008) to the date of type 2 diabetes diagnosis, death or the censor date (date of last contact), whichever occurred first. Participants were followed up until 1 January 2012. Cox proportional hazards models were used to evaluate whether age at natural menopause as a continuous or categorical variable (categories: premature menopause, <40 years; early menopause, 40–44 years, normal menopause, 45–55 years; and late menopause, >55 years [reference]) was associated with risk of type 2 diabetes. HR and 95% CIs were calculated. The proportional hazard assumption of the Cox model was checked by visual inspection of log minus log plots and by performing a test for heterogeneity of exposure over time. There was no evidence for violation of the proportionality assumption in any of the models (p for time-dependent interaction terms >0.05). To study relationships across increasing categories of age at natural menopause, trend tests were computed by entering the categorical variables as continuous variables in multivariable Cox proportional hazard models. To achieve normal distributions, skewed variables (CRP, dehydroepiandrosterone sulphate, insulin, testosterone, TG, TSH and SHBG) were natural log-transformed. In the base model (model 1), we adjusted for age, cohort (I, II and III), use of hormone replacement therapy and reproductive factors (age at menarche and number of pregnancies of at least 6 months’ duration). To examine whether the relationship of age at natural menopause with risk of type 2 diabetes was independent of potential intermediate factors, model 2 included the terms of model 1, as well as BMI (continuous), glucose concentration (continuous) and insulin concentration (continuous). Model 3 included all covariates included in model 2, along with the following additional potential confounding factors or intermediate factors: metabolic risk factors (total cholesterol, systolic BP [continuous], indication for hypertension [yes vs no] and use of lipid-lowering medication [yes vs no]), lifestyle factors (alcohol intake [continuous], smoking status [current vs former/never] and physical activity [continuous]), education level (low, intermediate and high), prevalent CHD (yes vs no) and CRP level (continuous). Moreover, to explore whether a nonlinear association was present, a quadratic term for age at natural menopause (continuous) was tested.
Sensitivity analyses
To explore whether sex hormone levels and common genetic factors could explain the association between age at natural menopause and type 2 diabetes, the models were further adjusted for these factors. Studies suggest that age at menopause and age-related disease risk are linked through common genetic factors [11]. We also performed a series of alternative sensitivity analyses. Since waist circumference is a better measure of visceral adiposity (an important risk factor for diabetes) and menopause is associated with accumulation of abdominal fat, we performed a sensitivity analysis substituting BMI with waist circumference [9]. To account for the specific effects of lipid particles on diabetes, we substituted TC with HDL-C, LDL-C and TG. We also restricted the analysis to participants who did not report using lipid-lowering medication. Information on parental history of diabetes was collected by trained research assistants during home visits at RSI and RSII, but not at RSIII. Therefore, we further adjusted the multivariable model for parental history of diabetes, but restricting the analysis to RSI-3 and RSII-1. Since smoking and hormone replacement therapy are important determinants of age at natural menopause and are associated with a risk of type 2 diabetes [24, 25], we restricted the analysis to women who were not current smokers and did not report use of hormone replacement therapy. To explore potential survival bias, we stratified the analysis by baseline age (<65 years and ≥65 years). We also reanalysed the data excluding the first 3 years of follow-up and excluding the participants with prevalent CVD. Moreover, we included women with non-natural menopause or unknown menopause type in the analysis to investigate the role of both age at natural and non-natural menopause in the risk of type 2 diabetes (selected characteristics are shown in ESM Table 1). Values were missing for one or more covariates (Table 1). As these values were likely to be missing at random, to prevent loss of efficiency, missing values were imputed using a multiple imputation technique (60 imputation sets; ESM Table 2). There were no significant differences in age at natural menopause or incident type 2 diabetes between participants with complete information for all covariates (n = 1884) and those who had missing values for at least one covariate in model 3 (n = 1755, 48%). Rubin’s method was used to calculate pooled coefficients (HR) and 95% CIs [26]. A p value of <0.05 was considered statistically significant. All analyses were done using IBM SPSS Statistics software (version 21.0, Chicago, IL, USA).
Table 1 Selected characteristic of study participants, the Rotterdam Study
Table 2 Associations of age at natural menopause with the risk of type 2 diabetes in postmenopausal women with natural menopause, the Rotterdam Study (n = 3639)