Study Design and population
The Tehran Lipid and Glucose Study (TLGS) is a population-based prospective study consisting of 15005 participants, aimed to estimate the prevalence and incidence of NCDs (22); the target population is a representative sample of an urban Iranian population, aged 3 to 69 years, living in Tehran, district No.13. The first examination cycle of the study started in 1999–2001 and after that, follow-up examinations have been repeated approximately every 3 years. Detailed description of rationale, design, and methodology of this study have been published (23). Only the participants who attended both the 5th and 6th consecutive examination cycles were included. We excluded participants younger than 20 or older than 69 years old (n= 1,181), had prevalent hypertension (n= 2,004), had prevalent cardiovascular disease (n= 278) or serum creatinine values> 2 mg/dL (n= 1), had prevalent diabetes mellitus (n=439), or had missing covariates (n= 1495) at examination cycle 5. After applying the exclusion criteria, 5,423 individuals remained eligible for the current analysis.
Assessment of Hypertension
Systolic blood pressure (SBP) and diastolic blood pressure (DBP) were measured following standardized protocols at each TLGS examination cycle (24). After resting for 15 minutes in sitting position, SBP and DBP were measured twice at the one-minute interval with a standard mercury sphygmomanometer calibrated by the Iranian Institute of Standards and Industrial Researches. The average of the two measures was taken as the systolic and diastolic blood pressures (22, 23).
Hypertension was defined as the SBP ≥ 140 mmHg or DBP ≥ 90 mmHg or the use of antihypertensive medications. We determined the incidence of hypertension by the presence of hypertension at examination cycle 6, among participants free of this condition at examination cycle 5 (Table 1).
Assessment of covariates
Weight was measured in minimal clothes and without shoes on an electronic scale, which was placed on a flat surface and calibrated to zero before measurement. Height was measured in a standing position and without shoes using a tape meter. Body mass index was calculated as “weight (kilograms)/height (meters) squared”. Current smoking, prevalent cardiovascular diseases and parental hypertension were self-reported. Current smokers were defined as a person who smokes cigarettes daily or occasionally. Prevalent cardiovascular diseases were defined as any coronary heart disease (CHD) (myocardial infarction or angiographic proven CHD) and cerebrovascular events (ischemic or hemorrhagic stroke) (22, 23). Diabetes mellitus was defined as a fasting plasma glucose ≥126 mg/dl or use of anti-hyperglycemic agents.
Framingham Hypertension Risk Score prediction model
Framingham hypertension risk score was derived from 1717 individuals (54% women), aged 20 to 69 years old, who were free of hypertension, cardiovascular diseases, and diabetes at the time of the baseline examination of the Framingham Offspring Study in 1979 followed to 2001. A Weibull regression model was computed along with covariates of age, BMI, SBP, and DBP as continuous variables, as well as sex (women vs. men), smoking (current vs. former or never smoker), and parental history of hypertension (both, one, or no parental history) as categorical variables, and an interaction term between age and DBP. The predicted risk of hypertension was calculated for each participant using the below equation:
Where t = time in years between examinations, βi = the regression coefficients of interested covariates and σ = scale parameter. The values of the coefficients and definitions of covariates are in Appendix A.
The 5,423 Participants were followed a single period from examination cycle five to six, contributing to a total of 12,855 person-years at risk. We examined the validity of the Framingham risk score in four stages (21). First, we calculated the Framingham risk score using the -coefficients derived in the Framingham study. Second, we recalibrated the Framingham risk score by updating the intercept; we replaced the intercept and scale parameter of the Framingham risk score with those of the TLGS, considering the linear predictor based on the original model as the offset in the model. Third, we recalibrated the Framingham risk score by another simple updating approach; we updated the intercept and calibration slope, considering the linear predictor as the only covariate in the model. Fourth, we revised the Framingham risk score by a more extensive updating approach (model revision), recalibration, and re-estimation of the coefficient of the sex covariate by fitting a Weibull model, in which the linear predictor and sex are the only covariates. This modeling choice was motivated by a difference between TLGS and Framingham regarding the hazard ratio of sex.
We assessed the performance of the Framingham risk prediction model among the TLGS population according to three evaluations: equality of regression coefficients (hazard ratio, HR); discrimination; and calibration. To compare the coefficients between the TLGS study and the Framingham study, we used a Weibull model using the same covariates in the Framingham model. A Z test statistic was calculated as:
Where βF and βT are the regression coefficients of the Framingham study and the TLGS, respectively, and SE2F and SE2T are the squares of the SEs for the two coefficients (25). Next, we assessed discrimination based on Harrell’s concordance statistic (c-index). Calibration included the comparison of the predicted hypertension incidence with the observed incidence for each decile of the risk score in a graphical assessment (calibration plot). The ratio of the predicted to observed risks was also calculated for the whole validation cohort. Furthermore, an additional analysis was conducted by including individuals with diabetes. All of the analyses were done with Stata version 14.2 (StataCorp. 2015. College Station, TX: StataCorp LP.).