2002 Taiwan Survey of Hypertensive, Hyperglycemia, Hyperlipidemia Survey (2002 TwSHHH) was a national representative prospective cohort was established in 2002. Briefly, participants of the TwSHHH cohort were included on the basis of multistage, stratified, random sampling from the National Health Interview Survey, which recruited 7578 individuals (age ≥ 15 years) and obtained detailed information from a face-to face questionnaire, anthropometric measurements, and blood sample analyses, which has been described in previous articles[36, 37]. Each participant’s baseline data that were collected from March 11, 2002 to August 10, 2002 were linked to the Taiwan Statistics of Causes of Death and National Health Insurance Research Database (NHIRD) until December 31, 2015, and were obtained from a universal, single-payer, and compulsory health insurance system that covers 99% of the 23 million residents of Taiwan, with diseases identified according to the International Classification of Disease-9 and 10 (ICD-9 and ICD-10) codes. All eligible participants in this study were excluded if recruited prior to the enrollment date of the 2002 Taiwan’s Triple High Survey and if they: (1) were younger than 20 years; (2) had a pregnancy within the previous 1 year; (3) had a recorded history of coronary artery disease and ischemic stroke in the National Health Insurance system; and (4) had missing data for identical numbers linked to the Taiwan National Health Interview Survey or National Health Insurance Research Database. A total of 6048 participants were included in the final analysis dataset used for the current analyses. Informed consent was obtained from each participant. The protocol was reviewed and approved by the Research Ethics Committee of National Taiwan University Hospital.
Assessment of healthy lifestyle factors
The body mass index (BMI) was calculated as the weight in kilograms divided by the square of height in meters from self-reported data in 2002 and participants were categorized as underweight (BMI < 18.5), normal weight (18.5 ≤ BMI < 25), obesity I (25 ≤ BMI < 30), obesity II (30 ≤ BMI < 35), obesity III (35 ≤ BMI < 40), and obesity IV (40 ≤ BMI) according to the recommendations of the World Health Organization.
Data used to generate the healthy diet patterns were derived from a simplified food frequency questionnaire with 20 items of food. We used the alternative Mediterranean diet score as our healthy dietary score (supplement), which was calculated by the frequency of intake and added scores across all 11 components of the 17 primary criteria contained in the Mediterranean dietary score for the following items39: fresh vegetables, legumes, fresh fruits, dairy products (milk, goat's milk, fermented milk, cheese, yogurt, Yakult), grains (rice or noodle), meat (beef, pork, goat, chicken), fish, eggs, sweets (cookies, candies, chocolate, cakes, bread, ice cream, milkshake), nonalcoholic beverages (cola, soda or sweet-beverage), and saturated lipid (burger, French frizzed, pizza). Participants were further classified according to the different levels of the alternative Mediterranean diet score (0–3, 4–5, 6–7, and 8–11 points). Participants with an alternative Mediterranean healthy diet score of 6 or more than 6 points were assigned to the adherence of healthy diet group, whereas those with scores less than 6 were assigned to the nonadherence of healthy diet group. Physical activity during the past 2 weeks were categorized as adequate active (1–50, 51–100, and 101–150 minutes/week) and nonoptimal physical activity, including inactive (0 minute/week) or overactive (> 150 minutes/week) grading (supplemental table 3). Smoking status was categorized as current smoking ≥ 20 years, current smoking < 20 years, quit smoking < 1 year, quit smoking ≥ 1 year, and never smoking (supplemental table 4). The participants were questioned about the usually drinking status and categorized as having frequency alcohol consumption (dinking every day with undrunk, half-drunk, or drunk status; drinking per 2 days with half-drunk or drunk status; and drinking once a week, with drunk status) or few (drinking less than once a week or drinking per 2 days, with an undrunk status) or no alcohol consumption (supplemental table 5). A detailed description of the questions and definition on ideal BMI, healthy diet, adequate physical activity, non-smoking status, and frequency alcohol consumption has been presented in supplemental tables 6 was based on the current literature, recommended guidelines but also based on levels realistically obtainable within the general population.
Healthy lifestyle scores
We created a simple pragmatic combined healthy lifestyle score. We created a healthy lifestyle score to sum each dichotomous lifestyle factor as "optimal" versus "nonoptimal" as follows: normal BMI (BMI < 25 kg/m2) versus obese (BMI ≥ 25 kg/m2), alternative Mediterranean diet 6 or higher points versus less than 6 points, adequate physical activity (1–150 min/week) versus non-optimal physical activity (0 or > 150 min/week), never smoking versus current or quit smoking, and frequent drinking versus few or no drinking. The participants received 1 point for each optimal criterion met, and points were summed to obtain a Taiwan healthy lifestyle score ranging from 0 (nonoptimal) to 5 (optimal). A weighted healthy Taiwan healthy lifestyle score also was created, where each dichotomous lifestyle factor was first weighted according to its independent magnitude of effect (e.g., beta coefficient adjusted for the other dichotomized lifestyle factors) on cardiovascular disease risk, and ranged from 0 to 17.
In accordance with the WCRF/AICR 2018 definition, the WCRF/AICR lifestyle score was created, which was a composite numerical measure of the adherence of health lifestyle and consisting of 7 main components, with each score based on a 0, 0.25, 0.5, and 1 scale representative from less healthy to most healthy. We defined the WCRF/ACIR healthy lifestyle score as the sum of scores across of all seven main components, including healthy weight, physically active, a diet rich in wholegrains, vegetables, fruit, and beans, and limiting consumption of “fast foods”, red and processed meat, sugar-sweetened drinks and alcohol. Based on the 2019 AHA update criteria of cardiovascular health, the Life's Simple 7 in our study included four core health behaviors (BMI, healthy diet (supplemental table 9), physical activity and non-smoking) and three health factors (cholesterol, blood pressure, and glycemic control). (supplemental tables 10). Each healthy heart behavior and factors providing 2 points for an ideal metric, 1 point for an intermediate metric, and 0 points for a poor metric were added to obtain the Life's Simple 7 score, ranging from 0 to 14.
At baseline, participants reported on sociodemographic factors and medical history, including educational level, monthly income, marital status, menopause status, history of estrogen exposure, and parental history of cardiovascular disease. Additionally, a history of the diagnosis of diabetes mellitus, hypertension and hyperlipidemia at baseline were obtained based on the measurement in 2002 or by ICD-9 or prescription of drugs from NHIRD prior to enrolled date. Diabetes at baseline was defined as a fasting serum glucose ≥ 126 mg/dL and hemoglobin A1c ≥ 6.5 mg/dL or records with two consistent diagnosis of diabetes by the ICD-9 codes or prescription of antidiabetic drugs for more than 28 days in data from the NHIRD prior to the enrollment date. Hypertension was defined as systolic blood pressure ≥ 140 mmHg or diastolic blood pressure ≥ 90 mmHg or records with two repeated diagnoses of hypertension or prescription of anti-hypertensive drugs for more than 28 days in data obtained from the NHIRD prior to enrollment date. Data on the use of lipid-lowering agent and aspirin were obtained from the drug register and defined as yes while prescriptions of more than 28 days prior to the enrolled date were included. Abdominal obesity was indicated as a waist circumference ≥ 80 cm in women and ≥ 90 cm in men. The adjusted factors included systolic blood pressure, diastolic blood pressure, and serum biomarkers obtained during the 2002 interview. The biomarkers comprised fasting glucose, glycated hemoglobin, triglyceride, and non-high-density lipoprotein cholesterol (HDL) as the continuous variables.
Outcome ascertainment and prospective follow-up
Follow-up information was obtained from the NHIRD and the Taiwan Cause of Death Register for fatal outcomes by record linkage using the personal identification numbers assigned to every citizen on Taiwan. ICD-9 codes were used to identify cardiovascular disease, such as coronary artery disease (ICD-9 codes 410–411, 414 and V45.81-82) or ischemic stroke (ICD-9 codes 433–436, 4371, 4379), with the first hospitalization with a diagnosis of the abovementioned interest events and the event date defined as the first date of hospitalization (Supplemental Table 1). We ascertained the occurrence of coronary artery disease- and stroke-related deaths from the death certificate. All participants were flagged for death at the department of Household Registration, with coded death certificates using the ICD-9 codes. The diagnoses of coronary artery disease and ischemic stroke were made by the treating physicians, based on a clinical assessment and examinations as considered relevant by the clinician in charge of treatment.
Participants were categorized into four groups among each healthy lifestyle scores, based on the simple and weighted Taiwan healthy lifestyle score, the WCRF/AICR healthy lifestyle score, and Life's Simple 7 score (Supplemental Table 12). The continuous variables are presented by the mean, standard deviation, or median levels; categorical data are presented in a contingency table with ANOVA to test for differences among quintiles. Multivariate Cox regression models were constructed for the combined health lifestyle scores, with the lowest score category used as the reference category (Supplemental Table 13). The linear trend test for lifestyle scores was performed by treating the number of low-risk factors as a continuous variable. The population attributable risk (PAR) was estimated using hazard ratio (HRs) obtained from the different Cox regression models in our cohort with that in the fully adjusted model. We tested potential effect modifiers based on the age category (< 60 and ≥ 60 years) by using the likelihood ratio test to compare models with and without a cross-product term.
To further investigate the role of a combination of healthy lifestyle factors to predict the CVD risk, we compared the four models with the healthy lifestyle score and tested the prediction performance using the calibration and discrimination abilities based on the Hosmer–Lemeshow statistic, comparisons of the Harrell C-index of survival data,[39–41] calibration curves, the net reclassification improvement (NRI), and integrated discrimination improvement (IDI) statistic. All statistical tests were two-tailed with a type I error. Statistical significance was considered at two-sided p-values < 0.05. The SAS version 9.4 (SAS Institute, Cary, NC, USA) and Stata version 12 (Stata Corporation, College Station, TX, USA) were used for statistical analysis.