Objectives: The prevalence of diabetes is increasing rapidly in developing countries. We aimed to estimate the prevalence of diabetes and to describe its main correlates and associated dietary intake in urban adults from Colombia.

Setting: The Colombian Study of Nutritional Profiles (COPEN) was a population-based, cross-sectional, multi-stage probabilistic sampling survey designed to represent the five main Colombian cities.

Participants: Between June and November 2018, we studied 736 non-pregnant participants aged 18 or older. Diabetes was defined as a random plasma glucose ˃= 200 mg/dL, self-reported prior diagnosis of diabetes or use of any oral or injectable antidiabetic agent(s). Participants also fulfilled a detailed 157-item food frequency questionnaire (FFQ).

Primary and secondary outcome measures: Prevalence of diabetes, dietary intake of key nutrients, achievement of dietary goals among individuals with diabetes.

Results: The overall estimated prevalence of diabetes was 10.1%, with no difference by sex (9.6% in women, 10.8% in men, p=0.43). Socioeconomic level (SEL) correlated positively with diabetes prevalence, the absolute difference in prevalence for the highest vs lowest SEL was 5.6%. The association between diabetes and education level depended on sex, diabetes was more prevalent among more educated men and less educated women. Abdominal obesity was associated with a 65% increase in diabetes prevalence among men, and a 163% increase in women. The proportion of non-achievement of dietary intake goals among participants with diabetes was 94.4% for saturated fats, 86.7% for sodium, 84.4% for fiber and 80% for trans fats. In multivariate logistical regression models, age was the strongest independent correlate of diabetes.

Conclusions: Self-reported diabetes was highly prevalent among Colombian adults, much more than described in most official reports. There were large differences by abdominal obesity status, region of residence, SEL and educational level. The proportion of individuals with diabetes meeting dietary recommendations was alarmingly low.

COPEN (Estudio Colombiano de Perfiles Nutricionales – Colombian Study of Nutritional Profiles) was a population-based, cross-sectional, multi-stage sampling survey designed to represent five cities, one from each of Colombia´s major regions: Bogotá (Central plateau), Barranquilla (Caribbean region), Cali (Pacific region), Medellín (Northwest or "paisa" region) and Bucaramanga (Northeast/Andean region). The sampling frame was obtained from the last census of the Colombian population, cartography was obtained from the national geostatistical frame developed by the Colombian National Department of Statistics (Departamento Administrativo Nacional de Estadística - DANE) and data on socioeconomic level (SEL) came from the National Superintendence of Public Services. In the first stage of sampling we selected cartographic sectors, within sectors we selected blocks (on average 8 per cartographic sector), within blocks we selected households, and within households we selected individual participants. Within each household, individuals were randomly selected employing a Kish grid. The sample was stratified by city, sex, age group and SEL. With this design and including the design effect, the complete study sample yielded an overall sampling error of 2.2%. The sampling errors for each city were respectively: Bogotá 4.0%, Medellín 5.0%, Cali 5.0%, Barranquilla 5.6% and Bucaramanga 6.8%. We excluded foreigners living in Colombia, individuals in hemodialysis or peritoneal dialysis therapy and persons with disabilities that precluded a reliable fulfillment of the study questionnaire. The complete study for COPEN was 1942 individuals, from which a random subsample of 736 non-pregnant participants aged 18 or older (representing 47.8% of all non-pregnant adults in COPEN) participated in the analyses reported in this paper.

Information was captured using a tablet device containing digital forms with proper validation rules, developed for the study. All staff in charge of data collection was extensively trained by the study Principal Investigator. A random 10% of participants were re-contacted by phone in order to double-check the accuracy of the information provided on date of birth, sex, city of residence, marital status, job status, educational level and date of initial contact. All data were collected between June and November 2018.

Patient and Public Involvement

Patients were not involved in the design of the study, but aggregated results will be presented to local and national authorities to inform public health policies concerning nutrition and primary prevention of diabetes.

Measurements

We collected information on sex, date of birth, SEL, marital status, educational level and employment status using a standardized questionnaire. The SEL that we employed for analyses was the one registered in DANE for that particular block. After a brief introduction about the importance of the accuracy of the measurements to be performed, we measured height and weight in all participants, and waist circumference in patients aged 18 and older. Height was measured using a portable stadiometer supported on a firm surface, taking care that the patient was barefoot, standing right and with heels and calves touching the stadiometer. Weight was measured in a solar digital scale with 100g sensitivity and 200 Kg capacity, all study scales were calibrated simultaneously the day before the study start, and every week afterwards. Waist circumference was measured by a sitting observer, directly over the participant´s skin, at the midpoint between the las rib and the anterosuperior iliac crest, using a flexible metallic measuring tape. All measurements were performed in duplicate, and if there was a between-measures discrepancy greater than 1 cm for height, 100g for weight or 1 cm for waist circumference, a third measurement was collected. For analyses we used the average of each anthropometric measure.

Socioeconomic level is classified in Colombia by the Statistics Department DANE in 6 strata according to characteristics of the residence (with stratum 1 being the lowest and stratum 6 being the highest) (19). Residential dwellings are classified according to their physical characteristics and environment. The methodology for this classification creates homogeneous strata taking as input information about land use, public utilities, access routes, topography, land valuation and property characteristics. The stratification unit is the sub-zone, corresponding generally to a block. Residential dwellings are classified in the predominant stratum of the sub-zone, as long as their characteristics do not differ ostensibly from the predominant conditions in the group. Otherwise, they are considered outliers and their stratum is assessed based on their particular characteristics. This information is very well established, updated and freely accessible for all the country. Given that sociodemographic, income and human development indicators are more similar for individuals living in strata 4 to 6 than among the other strata (19), we analyzed SEL in three groups, corresponding to strata 1-2 (low SEL), 3 (medium SEL) and 4-6 (high SEL). We interpreted BMI according to the cut points proposed by the World Health Organization (WHO): Underweight (BMI˂18.5 Kg/m2), normal weight (BMI >=18.5 and ˂25 Kg/m2), overweight (BMI>=25 and ˂30 Kg/m2) and obesity (BMI>=30 Kg/m2). We defined abdominal obesity as a waist circumference >= 90cm for women, and >= 94cm for men, according to the proposed cutoffs for Latin American adults (20).

Capillary blood specimens were collected by trained staff following standardized procedures, blood glucose levels were promptly measured and registered using an Accu-Check meter. Since fasting could not be guaranteed, we considered that an individual had diabetes if he/she met one of these three conditions: 1. A capillary blood glucose level ˃= 200 mg/dL, 2. A self-reported prior diagnosis of diabetes or 3. Self-reported use of any oral or injectable antidiabetic agent (s).

Usual dietary intake was assessed employing a 157-item semi-quantitative food-frequency questionnaire (FFQ). The FFQ was an enhanced and adapted version of an earlier FFQ specifically designed for the Colombian population (21). In a prior validation against four independent 24-hour dietary recalls, a shorter version of the FFQ showed a percent of classification in the same quartile of nutrient intake between 61 and 83%, and Pearson correlation coefficients between 0.51 for protein and 0.77 for carbohydrate (I). Portion sizes were established according to the reference unit most frequently consumed for each food. There were 9 possible ingestion frequencies: i. Never, ii. One to three times/month, iii. At least once/week, iv. Two to four times/week, v. Five to six times/week; vi. Once a day, vii. Two to three times a day, viii. Four to five times a day and ix. Six or more times a day. Participants were asked to make their selections based on their usual intake over the last year. FFQs were individually administered by study staff. The nutrient contribution of each food was calculated according to composition tables by the Colombian Institute for Family Welfare (Instituto Colombiano de Bienestar Familiar - ICBF), the United States Department of Agriculture and manufacturer´s information.

Data analysis

All prevalence estimations were projected to the target study population using city, sex, age group and SEL-specific expansion factors according to the study multi-stage sampling design. We did not have any missing data points for the analyses reported in this paper (sociodemographic factors, diabetes status and dietary intake variables). The univariate associations between nominal predictors and diabetes status were examined using chi-square independence tests. To test for a linear trend in the association between ordinal predictors and diabetes status, we report the p-value associated with a rank-correlation (Spearman) test between predictor and outcome. We also ran multivariable logistical models in which sex, age group, SEL and educational level were the independent variables and diabetes status was the outcome. We initially compared mean consumption of macronutrients and micronutrients of interest between individuals with or without diabetes using a one-way ANOVA, with diabetes as fixed factor. Since a higher BMI is associated with diabetes risk and also with a higher dietary nutrient intake, we performed multivariable linear regression analyses in which age, sex, BMI, SEL and diabetes status were the predictors and the daily consumption of each nutrient was the dependent variable (one model per nutrient). We explored the achievement of dietary recommendations among individuals with diabetes, expressed as the percent of individuals with diabetes who met the protein (˃=15% of total caloric intake [TCI]), saturated fat (SFA) (˂7% of TCI), monounsaturated fat (MUFA) (˃=12% of TCI) and trans fat (˂1g/day) recommendations set by the by the Latin-American Diabetes Association (22) and the fiber (14 g per each 1,000 Calories) and sodium (˂2300 mg/day) goals set by the American Diabetes Association (23). In order to explore factors associated with achievement of dietary goals, we also built a series of nested multivariable logistic models, in  which achievement of each dietary goal was the outcome. Model 1 had as predictors only sex and age, model 2 had all variables in model 1 plus SEL, model 3 had all variables in model 2 plus city, model 4 had all variables in model 3 plus BMI, and model 5 had all variables in model 4 plus diabetes status. All analyses were performed in SPSS for Windows, v.21 (Cary, NC, USA).

Ethical aspects

All participants provided written informed consent. All study procedures were performed according to the principles of the Helsinki Declaration, and to local rules and regulations as provided by Resolution 8430 of 1993 of the Colombian Ministry of Health. The study was approved by the IRB of Universidad de los Andes (Comité de Ética de la Vicerrectoría de Investigaciones), according to minute 1016 of April 27, 2018. 

Please see the attached EXCEL file "Variable list COPEN Diabetes" to see the type and labels of variables in the dataset.