Study population
The present study was conducted using data collected from the TLGS, an ongoing community-based prospective study conducted on a sample of residents from district 13, Tehran, Iran [24]. The first phase of the TLGS was initiated in 1998 with participation of 15,005 individuals. Collecting data from participants had been repeated every 3 years [25]. We recruited 3687 men and women who had participated in the third TLGS phase (2006–2008), with complete dietary data (completed FFQ). The characteristics of participants who completed the FFQ were similar to those of the total population in the third phase of TLGS [26]. For the current analysis, 3052 adult men and women (age ≥ 19 years) with complete baseline data (demographics, anthropometrics, biochemical and dietary data), were included. After exclusion of participants with CVD, HTN or CKD outcomes, as well as exclusion of participants with under- or over-reports of energy intakes, or with specific diets, and participants who lost to follow-up or have missing data, final study population for CVD, HTN and CKD were 2369, 1878 and 1780 adults, respectively (Fig. 1); the remaining eligible participants were followed up to the fifth phase of TLGS (2012–2014). Mean period of follow-up for CVD outcomes, CKD and HTN was 6.7, 6.4 and 5.8 years from the baseline examination, respectively.
Anthropometric and demographic measures
Anthropometric data were collected by the trained interviewers. Body weight was measured, to the nearest 100 g, using digital scales (Seca, Hamburg, Germany), while the subjects were minimally clothed and without shoes. Height was recorded to the nearest of 0.5 cm, in a standing position, without shoes, using a tape meter. Body mass index (BMI) calculated as weight (kg) divided by height in square (m2). Waist circumference was measured using a soft measuring tape, midway over light clothing, between the lower border of the ribs and the iliac crest at the widest portion, without any pressure to the body.
For measurements of systolic (SBP) and diastolic (DBP) blood pressures, after a 15-min rest in a sitting position, two measurements of blood pressure were taken on the right arm, using a standard mercury sphygmomanometer calibrated by the Iranian Institute of Standards and Industrial Researches [27]. Depending on the participants arm circumference, a regular adult or large cuff was used. The cuff was placed at heart level on the participant’s right arm and inflated at an increment rate, until the cuff pressure was 30 mmHg above the level at which the radial pulse disappeared. There was at least a 30-s interval between the two blood pressure measurements and mean of the two measurements was considered as the participant’s blood pressure; participants were asked to avoid tea or coffee consumption, physical activity, and smoking and were also asked to empty their bladder 30 min prior to the measurements. Physical activity was assessed using the Modifiable Activity Questionnaire (MAQ); the frequency and time spent on light, moderate, hard and very hard intensity activities according to the list of common activities of daily life over the past year were documented. Reliability and convergent validity of the Persian version of the MAQ has previously been investigated. Physical activity levels were expressed as metabolic equivalent hours per week (MET-h/wk) [28].
Biochemical measures
Blood samples were taken after 12 to 14 h of overnight fasting, between 7:00 and 9:00 AM. Fasting plasma glucose (FPG) and 2-h post-challenge plasma glucose (2 h-PCPG) were assayed using glucose oxidase. Measurement of serum triglycerides was done by enzymatic colorimetric method using glycerol phosphate oxidase. High-density lipoprotein-cholesterol (HDL-C) was measured after precipitation of the Apo-lipoprotein B containing lipoproteins with phosphotungstic acid. All blood analysis was done at the research laboratory of the TLGS, using Pars Azmoon kits (Pars Azmoon Inc., Tehran, Iran) and a Selectra 2 auto-analyzer (Vital Scientific, Spankeren, The Netherlands). Serum creatinine levels were measured by kinetic colorimetric Jaffe methods. Both inter- and intra-assay coefficients of variations (CVs) were less than 5%.
Dietary assessment
A validated 168-item food frequency questionnaire (FFQ) was used to assess typical food intakes over the previous year. Trained dietitians, with at least 5 years of experience in the TLGS survey, asked participants to designate their intake frequency for each food item consumed during the past year on a daily, weekly, or monthly basis. Portion sizes of consumed foods reported in household measures were then converted to grams [26]. Participants were questioned about frequency of drinking coffee or tea in the preceding year, considering a given portion size (cups per day or week or month). Caffeine intake was calculated as mg/day, from the sum of caffeine content in tea, coffee, soft drinks and chocolates. We did not collect any information on the type of coffee or tea and their preparation methods in the current study.
The validity of the food frequency questionnaire has been previously evaluated by comparing food groups and nutrient values determined from the questionnaire with values estimated from the average of twelve 24-h dietary recall surveys and the reliability has been assessed by comparing energy and nutrient intakes from two FFQs; Pearson correlation coefficients and intra-class correlation for energy and nutrients showed that the correlation coefficients between the FFQ and multiple 24 h recalls were 0.53 and 0.39, and those between the two FFQs were 0.59 and 0.60 in males and females, respectively. So it was shown acceptable agreements between the FFQs and twelve 24-h dietary recall surveys, and FFQ1 and FFQ2. The correlation coefficients between the FFQ and multiple 24 h recalls were reported 0.39 and 0.47 for carbohydrate, 0.65 and 0.50 for protein, 0.59 and 0.38 for total fat in males and females, respectively. Those between the two FFQs were 0.45 and 0.47 for carbohydrate, 0.79 and 0.69 for protein, 0.43 and 0.42 for total fat in males and females, respectively [29].
Definition of terms and outcomes
Hypertension was defined as SBP ≥ 140 mmHg or DBP ≥ 90 mmHg, or self-reported usage of blood pressure lowering medications [30]. Details of the data collection for CVD outcomes have been described elsewhere [31]. Cardiovascular disease outcomes were defined as any CHD event, stroke (a new neurological deficit that lasted ≥24 h), or CVD deaths (definite fatal myocardial infraction (MI), definite fatal CHD, and definite fatal stroke) [32]. CHD events included cases of definite MI (diagnostic ECG and biomarkers), probable MI (positive ECG findings plus cardiac symptoms or signs plus missing biomarkers or positive ECG findings plus equivocal biomarkers), and angiographic proven CHD. History of CVD was defined as previous ischemic heart disease and/or cerebro-vascular accidents.
Chronic kidney disease was defined as estimated GFR (eGFR) < 60 mL/min per 1.73 m2 [33]. To calculate eGFR, the CKD-EPI creatinine equation, developed by the Chronic Kidney Disease Epidemiology Collaboration, was used. As a single equation CKD-EPI has been expressed as follows:
eGFR = 141 × min (Scr/κ,1)α × max (Scr/κ, 1)-1.209 × 0.993age × 1.018 [if female] × 1.159 [if black].
In this equation, Scr is serum Cr in mg/dL; κ is 0.7 and 0.9 for men and women, respectively, α is − 0.329 and − 0.411 for men and women, respectively; min indicates the minimum of Scr/κ or 1, and max indicates maximum of Scr/κ or 1 [34].
Statistical analyses
Mean and standard deviation (SD) values, and frequency (%) of baseline characteristics of participants were compared according to incidence of CVD, CKD and HTN, using analysis of variance or chi square test, respectively.
The incidence of CVD and HTN over the follow-up period was considered as a dichotomous variable (yes/no) in the models. Dietary intakes of caffeine, coffee and tea intake were entered in the models as both continuous and categorical variables. In the categorical model, intakes of caffeine and tea were categorized into tertiles, and the first tertile was considered as reference. Participants were categorized to two groups according to coffee drinking status (as “drinkers” and “non-drinkers”). In the continuous model, hazard ratio was calculated for each 100 mg/day caffeine, 1 cup/day tea and 1 cup/week coffee increases for each related variables.
Cox proportional hazards regression models with person-years as the underlying time metric were used to estimate HRs and 95% confidence intervals (CIs) for associations between intakes of caffeine, coffee and tea and the incidence of CVD. Time to event for CVD was defined as time to end of follow-up (censored cases) or time to having an event, whichever occurred first. We censored participants at the time of death due to non-CVD causes, at time of leaving the district, or end of study follow-up (March 2014). For the censored and lost to follow-up subjects, the survival time was the interval between the first and the last observation dates. The proportional hazard assumption of the multivariable Cox model was assessed using Schoenfelds global test of residuals. To estimate incidence of CKD and HTN multivariable logistic regression models were used.
To obtain the final multivariable models and determine confounding variables, we performed a univariate analysis. Variables with PE less than 0.2 in the univariate analyses were selected as confounders. Potential confounders, adjusted in the Cox models, include CVD risk score (continuous) [35], total energy (kcal/d), fat and fiber intakes (g/d) for CVD outcomes. The CVD risk score calculated based on age, total cholesterol, HDL-C, SBP, treatment for HTN, smoking, and type 2 diabetes status, which has been validated among Iranian population [36]. Adjustment of CVD risk score, as a continuous potential risk factor of CVD events, improved the stability of our models due to the limited number of events during the study follow-up. For HTN and CKD [37], the models were adjusted for age (years), sex (male/female), BMI (kg/m2), triglyceride to HDL-C ratio, smoking (yes/no), total energy (kcal/d) and fat (g/d), and total fiber intake (g/d). Analyses of coffee were adjusted for tea, and vice versa. To assess the overall trends across increasing tertiles of caffeine or tea intake and each of three outcomes and to determine P values for trend, the median of each tertile of caffeine or tea was used as a continues variable in the regression models. CVD, HTN or CKD incidence was used as a dependent variable in the linear regression models. P values obtained from regression models were considered as P values for trend.
All statistical analyses were performed using the Statistical Package for Social Science (version 20; IBM Corp., Armonk, NY, USA) and STATA version 12 SE (Stata Corp LP, College Station, TX, USA), P-values < 0.05 being considered significant.