Population and sample
This cross-sectional population-based study originated from the health survey EpiFloripa Adults 2009 conducted in Florianópolis, Santa Catarina, southern Brazil, between September 2009 and January 2010. Florianópolis is the capital of the Brazilian state of Santa Catarina and has close to 420,000 inhabitants; it is also the Brazilian capital with the best social and health indicators .
The sample size was calculated to estimate the prevalence of each health outcome investigated in the survey, given a reference population of 249,530 adults from 20 to 59 years of age, a confidence level of 95%, 50% prevalence for unknown health outcomes, an error sample of 3.5 percentage points, a design effect (deff) estimated at 2 due to cluster sampling, and percentage of losses estimated at 10%. Based on these parameters, a sample size of 1,720 individuals was obtained. However, considering the multiple objectives of the EpiFloripa study, this study interviewed 32 adults in each of the 63 census sectors, thus increasing the sample size to 2,016 individuals.
Sampling was conducted in two stages. In the first, the 420 urban census sectors of the city were stratified according to deciles of the family head income (R$ 192.80 to R$ 13.209.50 [Brazilian currency], US$ 1 = R$ 1.7 during the data collection period); 60 sectors were randomly selected (sampling fraction equal to seven), consisting of six sectors in each decile. In the second stage, the sampling units were the households. In this step, the number of permanent private households in each sector was updated by the study supervisors. The number of dwellings ranged from 61 to 810, giving a variation coefficient of occupied households between sectors of 55%. To decrease this, and considering the geographical proximity and income decile, it was decided to group some census sectors and divide others, resulting in 63 census sectors with a variation coefficient of 32%. Eighteen households from each of these geographical units were randomly selected.
Eligibility, exclusion, and loss criteria
All adults aged 20 to 59 years living in selected households were eligible for the study. The exclusion criteria for carrying out anthropometric measurements and blood pressure measures were bedridden subjects, amputees, plastered and pregnant women, and those who had given birth in the six months preceding the interview. Individuals with neurological disorders that might interfere with their understanding the questions regarding the survey interview were also excluded. One resident who was not found in at least four visits, one in the weekend and another at night, was considered lost.
Data collection was carried among all adults living in selected households. To this end, 35 interviewers were selected. A personal digital assistant (PDA) was used to record and store data. All interviewers were intensively trained for the fieldwork; the survey pretest was conducted among 30 adults not sampled, and the pilot study was conducted among 100 individuals.
The dependent variable was hypertension (yes/no), which was defined as systolic pressure ≥ 140 mmHg and/or diastolic pressure ≥ 90 mmHg, and/or self-report of taking some antihypertensive medication, and/or diagnosis of hypertension by a doctor .
Blood pressure levels were measured twice, and the average measurements were used in the study. The resting time was approximately 30 minutes before the first measurement and approximately 15 minutes between measures. Blood pressure was measured in accordance with the recommendations of the Brazilian Guidelines on Hypertension . For this purpose, blood pressure was measured on the right arm, which rested on a table at heart level with the palm facing upward. The individual remained sitting with his feet planted on the ground. At the beginning of the interview, the individuals were advised to refrain from smoking and drinking coffee, chimarrão, or black tea, and to empty their bladder. Electronic sphygmomanometers with a digital readout system (Techiline®, São Paulo, Brazil), previously and properly calibrated by the National Institute of Metrology, Standardization and Industrial Quality (Inmetro), were used to measure blood pressure levels.
Anthropometric measurements of body mass, height, and WC were evaluated according to standardized procedures , and the mean of two measurements was used in the study.
Body mass was measured with the digital scale model HCM 5110M (GA.MA Italy Professional®, Bologna, Italy), with a resolution of 100 grams and 150 kg capacity, which was calibrated before the survey. Height was measured using a stadiometer with a measuring tape at a resolution of 1 mm. The WC, taken without any clothing, was measured with an inextensible anthropometric tape (Sanny ®, São Bernardo do Campo, Brazil) with a resolution of 1 mm at the narrowest waist.
%BF was calculated based on equations with BMI  and WC . The equation using BMI developed with a sample composed of white Americans and Africans showed a determination coefficient of R = 0.86 and a standard error of 4.98% when compared to the body composition analysis of the four-compartment model evaluated by means of DEXA (bone density) and hydrostatic weighing (water and body volume) . The equation of Gallagher et al.  is as follows: %BF = 64.5 - 848 x (1/BMI) + 0.079 x age - 16.4 x sex + 0.05 x sex x age + 39.0 x sex x (1/BMI). In the indication of sex, the values are 1 for men and 0 for women; age is given in complete years.
The equation with WC developed with healthy adults from Glasgow, Scotland, showed a determination coefficient of R = 77.8%, with %BF estimated by means of hydrostatic weighing, and a standard error of 4.10% for men, and a determination coefficient of R = 70.4% and a standard error of 4.70% for women . The equation of Lean et al.  differs between men and women. For men: %BF = (0.567 x WC) + (0.101 x age) - 31.8. For women: %BF = (0.439 x WC) + (0.221 x age) - 9.4. WC must be in cm, and age in complete years.
BMI was calculated (weight/height 2) and ranked according to the literature  into obesity (BMI ≥ 30 kg/m2), overweight (BMI of 25.0 to 29.9 kg/m2) normal weight (BMI ≥ 18.5 and <25 kg/m2), and underweight (BMI <18.5). WC was assessed based on cutoff points in relation to risk of metabolic complications, and ranked into very high risk (men ≥ 102 cm, women ≥ 88 cm), high risk (men ≥ 94 cm, women ≥ 80 cm), and no risk (men <94 cm, women <80 cm) . The inter-and intra-examiner technical error of measurement (TEM) was calculated according to the recommendations of Gore et al. . The maximum inter-examiner TEM (1.86%) and intra-examiner TEM (1.18%) was found in the WC measure, which indicated an adequate level of interviewers for anthropometric measurements.
The control variables included demographic data, such as age, which was expressed in complete years and categorized as 20–39 and 40–59 years. The self-reported skin color was collected based on the categories proposed by the IBGE  and classified as white, light-skinned black, dark-skinned black, yellow, and indigenous. The socioeconomic variables were educational level assessed by the complete years of schooling and per capita family income in reais (R$, the Brazilian currency; US$ 1 = R$ 1.7 during the data collection period).
Smoking was assessed by the categories nonsmoker, former smoker, light smoker (less than 10 cigarettes daily), and moderate/heavy smoker (more than 20 cigarettes daily). The Alcohol Use Disorders Identification Test (AUDIT) was used to identify people with problematic alcohol use, using the cutoff point to classify subjects into no (score 0–7) and yes (score ≥ 8) . Physical activity was assessed according to the leisure domain of the questionnaire used in the surveillance system of risk and protective factors for chronic diseases through a telephone survey (VIGITEL), considering as inactive those who practiced no physical exercise or practiced less than once a week in the three months preceding the interview . Food habits were assessed based on the regular consumption of fruits and vegetables (≥ 5 days per week) according to the VIGITEL questionnaire .
Descriptive statistics using the mean, standard deviation, and absolute and relative frequency was applied. To compare continuous variables between sexes, Student’s t-test for independent samples was used. To determine differences between sexes in the distribution of categorical variables, the chi-square test was used. The ROC curve was calculated to analyze the discriminatory power of %BF in the identification of hypertension and to find the best cutoffs to identify this association . In this study, this cutoff point was the one with the best accuracy, i.e., with fewer false positives and false negatives. The larger the area under the ROC curve (AUC), the greater the discriminatory power of the %BF equation to identify hypertension. The confidence interval was calculated at 95% (95% CI), which determines whether or not the predictive capacity is due to chance, and the lower limit should not be less than 0.50 . The differences between the AUC values of the different %BF equations were compared using the nonparametric test . The sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), positive likelihood ratio (LR +), and negative likelihood ratio (LR-) of all the best cutoffs of %BF were calculated to identify hypertension.
The association between hypertension and high %BF estimated by the cutoff points calculated in this study was analyzed by Poisson regression, estimating the crude and adjusted prevalence ratios and 95% CI. Three models of adjusted analysis were developed to determine the magnitude of association between excess fat and hypertension. Model 1 was adjusted by sociodemographic variables (age, skin color, educational level, and per capita family income). Model 2 was adjusted by sociodemographic variables and health behaviors (smoking, problematic alcohol use, practice of physical activities during leisure time, and regular consumption of fruits and vegetables). Model 3 was adjusted by sociodemographic variables, health behaviors, height, and BMI (for %BF equation with WC) or WC (for %BF equation with BMI). All analyses were stratified by sex and carried out considering the design effect and sampling weight. To calculate the screening properties of %BF equations, the MedCalc software version 12.1.4 was used. For association analyses, the Stata 9.0 software was used.
The study was approved by the Ethics Committee on Human Research of the Federal University of Santa Catarina (No. 351/08). All the subjects who participated in the study signed an informed consent form.