Figure1 provides an overview of the SLBDS relative validation process, from face validation and pre-testing to survey administration and statistical analyses. These steps are outlined in further detail below.
Relative validation process
Face validity, a process whereby the SLBDS was subjectively viewed as covering what it purported to measure, was assessed by six researchers at the University of Colombo, Sri Jayewardenepura General Hospital and Post Graduate Training Centre, and University of Oxford. This process was conducted in-person and via email and telephone interviews. The feedback and discussion generated through this process focussed on prescribed SLBDS intake units (for example, coconut spoons vs cups vs tablespoons) and how best to measure notoriously difficult to capture oil, salt, and coconut intake, with researchers divided on whether to measure intake frequency (i.e.number of times itemconsumed in a 24-h period) or intake amount. In the latter case, we decided to record both. Clarity of survey wording and format was further refined through pre-testing with 25 Sri Lankan adults aged 20 to 65 known to our research network.
The face validated and pre-tested version of the SLBDS was subsequently tested against the 24DR to determine criterion-related validity i.e. how well the new SLBDS estimated 24-h food intake and adherence to the SLFBDGs compared to the reference method (24DR). We selected this reference method because of its use in prior brief instrument validation studies and its statistically significant comparability to the gold standard self-report 7-day weighed-intake dietary record (7DWR) in the target population [19]. The 24DR has a similar objective to the SLBDS and measures intake over the same time frame (24h) whilst differing in its reliance on memory and portion size reporting format. These similarities and differences make the 24DR an appropriate reference method for this study [7, 20].
Between December 2018 and February 2019, we administered both the SLBDS and the 24DR to a sample of 94 Sri Lankan adults living in urban Colombo, and urban and rural sectors in Kalutara and Trincomalee. The Trincomalee district is located in the majority Moor and Muslim Eastern Province, whereas both Colombo and Kalutara are majority Sinhalese and Buddhist Western provinces. A sample size calculation was conducted to determine the sample size required to detect a low correlation between results from the test and reference surveys (r=0.3) with alpha and beta set to 95 and 80% respectively, and accounting for a 10% participant dropout rate. Divisional Secretaries of Colombo, Kalutara, and Trincomalee were contacted to obtain electoral lists for each Grama Niladhari Division (GND) within the district and gain permission to visit individual households for data collection. Adults aged 18years or older who were willing and able to provide informed consent were eligible for inclusion. Data collection started at a randomly selected location within the district. From that starting point, the nearest house appearing in the relevant electoral list was approached. If a consenting adult was present, the third house to the left was approached for the next interview, and so on. If more than one eligible adult was present in the household, the interviewee was selected by drawing lots. If an eligible adult was not present or did not consent, the house next door was approached. This recruitment method was followed until 56, 29, and nine participants in Colombo, Kalutara, and Trincomalee, respectively, were recruited.
Each participant completed two consecutive face-to-face interviewer-administered dietary surveys, the SLBDS (new tool) and a 24DR (reference method), in the participants preferred language: Sinhala, Tamil or English. As the surveys were administered one after the other during the same visit, administration order was randomised to avoid response-order effects. Data collection was undertaken in participants homes by two experienced female Sri Lankan researchers, with 50% of the study sample randomly allocated to each interviewer (i.e. the same interviewer applied both dietary surveys). Data collectors were consulted during the survey development phase and trained through role play and practice interviews with members of the research team to deliver the surveys uniformly, closely following the provided templates and corresponding instructions.
Both surveys are structured dietary assessment tools that ask participants to recall their food and beverage consumption during the previous 24h ashort recall period thathas proven useful for minimising recall bias [21]. The two surveys differ in length, degree of survey structure, memory requirements, recall process, detail captured, and analysis burden (Table1). The same standard household utensils were referenced to obtain information on portion sizes in both surveys and each queried whether or not the day being recalled was typical of participants' usual food intake, with interviewers prompting atypical responses for further explanation. Vegetarian status and special diets were also recorded on the paper-based templates provided.
Using the SLBDS, interviewers asked participants if they had consumed each of the following food groups: (1) Rice, bread, other cereals and yams; (2) Fruit; (3) Vegetables; (4) Fish, pulses, meat and eggs; (5) Milk or milk products; (6) Nuts and oil seeds; as well as (7) Sweetened drinks, sweets and desserts; (8) Fast food; (9) Salt; and(10) Tea and coffee in the past 24h, and if yes, what portions (in prescribed units) of specific foods (also prescribed) within each group they consumed. After a first pass of the survey, the interviewer revisited unanswered questions.
Detailed food and beverage consumption information was collected from participants using the 24DR method. Uninterrupted and in their own words, participants were asked to list everything they had consumed, including food and drink items and their corresponding quantities, the previous day (from waking to evening). The interviewer then probed this list for additional information: consumption time and location; item brand; further item description; and leftovers. To conclude the interview, respondents were given a further opportunity to provide additional information/detail on their 24-h intake.
All 188 surveys were verbatim transcribed, translated into English, and entered into Excel with 24DR data hand coded and summed to reflect SLBDS food groups/categories (1 to 10 listed above). Coding was blinded to the results of the SLBDS to avoid bias. This was achieved by coding the 24DR survey results before the researcher responsible for analysis gained access to SLBDS data. We used a chi-square test to determine whether participant characteristics differed by participant sex. As most of the dietary data were not normally distributed, we calculated the median and interquartile range (IQR) for intake of each food group and food/beverage item (based on serving size where specified in the SLFBDGs and portion size where unspecified) reported in the SLBDS and 24DR. Wilcoxon rank-sum tests were used to assess the statistical differences between medians. Correlation (r) between individual intakes collected by each measure was determined using Spearmans Rho tests. The use of these non-parametric tests ensures that spikes at zero consumption do not invalidate statistical assumptions. To detect differences and bias between the two methods, differences were plotted against means in Bland-Altman plots. We estimated Cohens kappa (k) with 95% confidence intervals to measure the inter-rater reliability for comparing achievement of recommended food group intake (where 0=not achieved and 1=achieved) based on the SLFBDGs between the new and reference method. For yes/no SLBDS questions: Did you consume Western or local fast food yesterday? and Did you add salt, sauce/ketchup or chutney/chilli paste to your breakfast/lunch/dinner/snack?, yes responses were assigned a score of 1 and no a zero. 24DR data were coded similarly: we assigned reporting of fast food (local and Western) and salt intake at specified meal times (breakfast, lunch, dinner, and snack) a score of 1 (if intake was reported) and 0 (if no intake was reported). The unweighted kappa statistic describes the level of agreement over and above chance agreement between the two measures as slight (00.20), fair (0.210.40), moderate (0.410.60), substantial (0.610.80), and almost perfect (>0.81) [22]. To assess whether the SLBDS was an equally valid measure of dietary intake for both female and male participants, we calculated the results for each of these agreement analyses separately in females and males as a secondary analysis. We considered a p value <0.05 as evidence against the null hypothesis. All statistical analyses were conducted in R version 4.0.1.
Ethics approval for this study was received from the University of Colombo (Faculty of Medicine) and the University of Oxford (Oxford Tropical Research Ethics Committee). Written informed consent was obtained from each participant prior to data collection. Compensation for participation was not provided.
