Data sources
This study utilised data from nationally representative surveys conducted in Bhutan, Eswatini, Georgia, Guyana, Kenya, Nepal and St Vincent and the Grenadines; all upper-middle, lower-middle, or low-income countries [29] at the time the surveys were conducted. The method of data acquisition and pooling have previously been described [30,31,32]. In brief World Health Organization (WHO) Stepwise Approach to Surveillance (STEPS) surveys [33] conducted in low, low- middle, or upper-middle income countries since 2005 were searched for. The search was limited to surveys conducted since 2005, as these studies were considered contemporary enough to be included in the same analysis. WHO STEPS surveys use a standardised questionnaire and protocol to monitor non-communicable disease risk at a population level, with the questionnaire comprising three steps: step one “behavioural measurements”, step two “physical measurements” and step three “biochemical measurements” [21, 33, 34]. Survey contacts were approached for the de-identified individual level data to be pooled for analyses. Data was pooled if signed agreement was made and they had a response rate ≥ 50%; participants were aged 15 years or older; included data on waist circumference, and/or a biomarker for diabetes (either a glucose measurement or HbA1c), and/or a measurement of blood pressure. For the current analyses surveys were included if questions on salt behaviour, fruit and vegetable intake, and the use of fats and oils for cooking were asked, seven out of 46 surveys. The surveys used a two-stage cluster random sampling design, with one person from each household (within the defined age range) randomly selected to complete the survey. All surveys were carried out by a trained data collection team member in the household setting, or at a conveniently-located health center and data on the three questionnaire steps were collected during the same visit.
Terminology – sex - gender
A person’s sex is recorded in the WHO STEPS surveys by the interviewer documenting the observed sex of the participant (binary, male or female) [21]. While acknowledging that the self-report of dietary behaviours is likely to be influenced by a person’s identity and social constructs, and therefore also related to a person’s gender, to be in line with the data collected, the term “sex”, and corresponding terms “male” and “female”, are used throughout this paper [35].
Classification of dietary behaviours
Diet behaviours [36] of salt use, fruit and vegetable consumption and type of oil and fat used in cooking are included within “Step 1 – Behavioural Measurements” of the questionnaire, and are the only dietary behaviour variables included in STEPS [21].
Salt use behaviours
There are seven salt use behaviour questions included in STEPS [21]: 1. How often do you add salt or salty sauce such as soy sauce to your food right before you eat it or as you are eating it? 2. How often is salt, salty seasoning or a salty sauce added in cooking or preparing foods in your household? Do you do any of the following on a regular basis to control your salt intake: 3. Limit consumption of processed foods? 4. Look at the salt or sodium content on food labels? 5. Buy low salt/sodium alternatives? 6. Use spices other than salt when cooking? 7. Avoid eating foods prepared outside of a home? The first two questions used a 5-point Likert response scale with options of: always, often, sometimes, rarely, or never. These answers were assigned a value of 0, 0.25, 0.5, 0.75 or 1, respectively. The other five questions used a “yes” or “no” response, which was assigned a value of 1 and 0, respectively. To investigate the prevalence of positive (good) compared to poor salt used behaviour, the response values for all the seven questions were summed, and individuals with a score of 0.5 (50%) or greater were labelled as having positive (good) salt use behaviour. Another method of scoring salt use behaviour and categorising into positive vs. poor behaviour was not identified in the literature, and therefore other options of quantification were tested. These included an ordinal 4-point score (categorising into of 25, 50, 75 and 100% of the salt behaviour questions answered positively) and a 7-point score (“1” being one question answered positively, through to “7”, being all questions answered positively). Given the low prevalence of positive salt use behaviour the 50% cut-off was used in the main analyses, with the 4-point score and 7-point score used in sensitivity analyses for the association of salt use behaviour with undiagnosed hypertension.
Fruit and vegetable intake
In the surveys, participants were asked to report the number of days per week they consume fruits and vegetables. If participants reported that they consumed fruits or vegetables on one or more days a week, they were then asked to state on any given day how many portions of fruits and vegetables they consume. To aid their response, they were shown pictures of local fruits and vegetables to refer to as a portion, corresponding to approximately 80 g. Fruit and vegetable intake (per day) was then calculated using the methods of Frank S et al. [31]. Briefly, individuals were categorised as meeting, or not meeting, the fruit and vegetable recommendations, based on the WHO- recommendation of five 80 g portions of fruit and vegetables, or more, on a given day, equivalent to 400 g or more a day [18].
Oil and fat use
Participants were asked to pick the main oil or fat used to prepare meals in their home. Options, specific to the types of oils and fats used in each country, were provided to the participant. Responses were categorized as: vegetable, animal, other, none in particular, or none used. For analysis, this was further collapsed into vegetable oil, all other oils and fats, and no fat or oil used, given the small number of individuals who reported using other types of fats and oils or no use of fats or oils. “Vegetable oil” was used as the reference (or “positive behaviour”) category, based on evidence that suggests plant-based oils are protective for heart health [13, 17].
Classification of cardiovascular risk factors
Waist circumference
Waist circumference in each survey was conducted following the STEPS data collection manual [37]. Data collectors used constant tension tape to measure waist circumference directly against the participant’s skin where possible, or over light clothing if direct contact was not possible. Measurement was taken with a participant in a standing position, with arms relaxed at their sides and at the end of a normal expiration. The point of measurement was the midpoint between the lower section of the last palpable rib and the top of the hip bone. Waist circumference was then recorded to the nearest 0.1 cm, and only one measurement per participant was recorded. Participants were classified as having a “high waist circumference” if their measured value was ≥102 cm for males and ≥ 88 cm for females [38].
Hypertension
Detailed country-specific methods of blood pressure measurement are described elsewhere [32]. Briefly, the included surveys followed the STEPS data collection manual [37], which specifies measures to be conducted using digital, automated upper arm monitors, following 15 min of rest. The majority of participants had three blood pressure readings taken, with 3 min rest between each measure. The average of the last two readings were then taken. For individuals with only two measures, the mean of both available measurements was taken; for individuals with only one measure that measure was taken. A person was classified as having hypertension if their average systolic blood pressure (SBP) measurement was greater than 140 mmHg, or their average diastolic blood pressure (DBP) measurement was greater than 90 mmHg, or they reported taking medication for hypertension. We defined a categorical variable of non-hypertensives (reference), undiagnosed hypertension, and diagnosed hypertension. Individuals with self-reported diagnosed hypertension were those who met the criteria for hypertension and also reported a diagnosis of hypertension. Undiagnosed individuals were those who had a high SBP (> 140 mmHg) or a high DBP (> 90 mmHg), did not report taking hypertension medication, and did not report a hypertension diagnosis.
Diabetes
Detailed country-specific methods of diabetes measurement are described elsewhere [30]. Briefly, point-of-care fasting capillary glucose measurement was the diabetes biomarker in all surveys apart from the survey conducted in Nepal, where laboratory-based assessment of fasting plasma glucose was used. For the six countries that measured capillary glucose, plasma equivalents were provided. Individuals were asked if they fasted or not prior to the measurement, for those who reported not fasting their blood glucose level was interpreted as a random blood glucose measure. Diabetes was defined as having an average fasting blood glucose (FBG) level of 7 mmol/L or greater, or having a random blood glucose (RBG) level of 11.1 mmol/L or greater, or on medication for diabetes. We evaluated a categorical variable of non-diabetics (reference), undiagnosed diabetes, and diagnosed diabetes. Individuals with self-reported diagnosed diabetes were those who met the criteria for diabetes and also reported a diagnosis of diabetes. Undiagnosed individuals were those who had a high FBG (> 7 mmol/L) or a high RBG (> 11.1 mmol/L), did not report taking diabetes medication, and did not report a diabetes diagnosis.
Sociodemographic and behavioural variables
Sociodemographic and behavioural factors of interest were sex, age, education, working status, physical activity levels, alcohol use and tobacco use [21].
Sociodemographic variables
Age was defined based on the dates of an individual’s birth and the survey, or self-reported age. Age was then categorised into 10-year categories: 15–24, 25–34, 35–44, 45–54, 55–64 and 65 or older. For education a range of options were given including: no formal schooling, less than primary school, primary school completed, secondary school completed, high school completed, college/university completed and post graduate degree. For analysis, education was categorised into “no formal schooling/education”, “primary school attendance only” and “secondary schooling or above”. For working status, a range of occupations were reported including: government employee, non-government employee, self-employed, non-paid, student, homemaker, retired, and unemployed. Of these we classified the self-report of any paid occupation as “working” and any unpaid occupation (for example homemaker) as “not working”.
Behavioural variables
STEPS surveys include physical activity questions, covering physical activity at work, for transport and for recreation. For physical activity at work or for recreation, participants were asked if they participate in vigorous or moderate intensity activity, on how many days during the week, and for how long. For transport participants were asked if they walk or cycle for at least 10 min at a time to get to/from places. If they answered “yes” to this question they were then asked on how many days, and during the day how long, they walked or cycled for transport. Answers to these questions were translated into metabolic equivalents (METs), and the WHO recommendation of achieving at least 600 METs [18] used as the cut-off for individuals to be categorised as physically active.
Alcohol consumption is also self-reported, participants were asked if they consumed alcohol in the past 12 months, and then if so the frequency of consumption in the past week. For analyses individuals were classified as “non-drinkers” (had not consumed alcohol in the past 12 months, or did not report consuming alcohol in the previous week) or “drinkers” (reported consuming at least one alcoholic beverage in the past week).
Tobacco use was based on reported frequency of smoking tobacco (cigarettes) and/or using smokeless tobacco (for example snuff or chewing tobacco), in a similar manner to questions on physical activity and alcohol use. Individuals were also asked if they previously used tobacco. Therefore, this variable was categorised as “no reported tobacco use”, “past tobacco use” and “current tobacco use”.
Analyses
Analyses for the population and dietary behaviour characteristics were performed on the sample of individuals with data on all three dietary behaviours from the seven countries. The complex survey design was accounted for, via the Stata svy command [39], and data were weighted so that data from each country contributed equally to the results. Percentages for categorical variables and means for continuous variables of demographic, behavioural and disease characteristics, by sex, were described and differences between sexes tested using Pearson’s chi-squared test for categorical variables and regression analysis for continuous variables.
Generalized linear models with country-level fixed effects were used to investigate cross-sectional associations between the dietary behaviours and waist circumference. Given that our outcome variables were discrete (i.e. dichotomous), we have fitted our generalized linear models using the binomial family distribution. For the hypertension and diabetes outcomes, separate multinomial logistic regression models with country-level fixed effects were used, comparing undiagnosed and self-reported diagnosed hypertension or diabetes with non-hypertensives or non-diabetics, respectively. For the waist circumference outcome models were adjusted for age, educational attainment, working status, physical activity, alcohol use and tobacco use. For the hypertension and diabetes outcomes, models were adjusted for age, educational attainment, working status, physical activity, alcohol use, tobacco use and waist circumference. Complete case analyses were conducted. Information on the number and proportion of participants with missing data on the outcome, independent or confounding variables is provided overall and by country in Additional file 1: Table S1.
To investigate the interaction of sex with the dietary behaviours on the outcomes, interaction terms were used and marginal estimates (proportion of males and females with the outcome for the dietary behaviour) were calculated. For these interactions a more lenient p- value of ≤0.10 was used to identify significance. Given the high proportion of respondents who reported using vegetable oil in cooking (93%) we have not presented the results by type of oil used, as findings were not informative. For the hypertension outcome two sensitivity analyses were conducted using the 4-point, and the 7-point salt behaviour score.
The results are presented with 95% confidence intervals. All analyses were conducted in Stata version 15.1 (StataCorp, College Station, Texas, US).