Feasibility and sustainability of dietary surveillance, Bosnia and Herzegovina

Abstract National dietary surveillance systems are necessary for monitoring people’s intake of foods and nutrients associated with health and disease, and for implementing national and global dietary goals. However, these systems do not exist in many low- and middle-income countries. The development of a model of dietary surveillance for Bosnia and Herzegovina, described here, provides insights into the feasibility and sustainability of dietary surveillance systems in resource-constrained settings and illustrates the challenges involved. In 2016, a year-long dietary survey was initiated in collaboration with the country’s Institute for Statistics using a subsample of households that participated in the 2015 national Household Budget Survey. Interviewers collected lifestyle, anthropometric and health data and participants answered two 24-hour dietary recall questionnaires. The survey included a representative sample of 853 participants and was performed efficiently by a small team of highly motivated, well-trained staff. Conducting a high-quality dietary survey was found to be feasible despite constrained resources. In addition, the ability to link dietary intake and regular household survey data provided an effective way of associating dietary variables with socioeconomic determinants of health. This dietary survey, the first conducted by an official institution in Bosnia and Herzegovina, represents an important starting point for building a sustainable nutritional surveillance system for the country. The cost–effective, low-burden approach to dietary surveillance described here could be applied in other low- and middle-income countries, many of which already carry out regular economic surveys.


Introduction
Diet and its associated health conditions are the most important modifiable risk factors for morbidity and mortality globally. The influence of diet on health is greater than that of tobacco, alcohol and drug use, and unsafe sex combined. 1,2 Moreover, there is evidence that even small dietary modifications can improve nutrition and health. [3][4][5][6] The availability of accurate, individual data on a population's diet is critical for developing dietary policies and programmes and for monitoring their implementation, both at a national (e.g. adherence to dietary guidelines) and a global level; for example, to achieve targets such as the sustainable development goal 2 to "end hunger and all forms of malnutrition." 7 Nevertheless, recent reviews of dietary surveillance worldwide reveal that data are scarce in many parts of Africa, Asia and eastern and south-eastern Europe. [8][9][10] The dietary surveillance systems developed in high-income countries, such as the National Health and Nutrition Examination Survey (NHANES) in the United States of America and the United Kingdom's National Diet and Nutrition Survey are expensive, complex and burdensome. 9,[11][12][13] Typically they require a large sample, a team of experts and substantial investment in equipment and software. However, many countries with resource constraints are left without these crucial decision-making aids for promoting population health. Without population data on food and nutrient intake, policy-makers risk making the wrong decisions or may ignore the problem of inadequate food systems and diet altogether. The United Nation's Secretary-General's Independent Expert Advisory Group on a Data Revolution for Sustainable Development has called for urgent action to enhance national capacity for efficient data collection to bridge the gap between developed and developing countries. 14 Although the so-called data revolution taking place globally as a result of rapid technological development provides both a challenge and an opportunity, some countries have been left out due to a lack of resources, knowledge, capacity or opportunity. 14 In 2017, the United Nation's World Data Forum highlighted the need to modernize national statistics systems and to expand them to include data domains currently beyond the scope of official statistics. 15 In particular, the accurate analysis of diet in disadvantaged population groups (e.g. lowincome groups, single-parent families and minorities) depends on linking dietary surveillance systems to existing demographic and socioeconomic data.
Seeking new ways to obtain good-quality dietary data in countries with financial and workforce constraints is, therefore, an important goal for researchers. Here, we describe our experience with developing dietary surveillance for Bosnia and Herzegovina and consider the challenges associated with establishing a feasible and sustainable dietary surveillance system in resource-constrained settings.

Policy & practice
Dietary surveillance in Bosnia and Herzegovina Selma Gicevic et al.
population slightly over 3.5 million. 18 In 2012, one quarter of children and one half of adults were overweight or obese, half of all deaths were due to stroke or heart disease, type-2 diabetes was on the rise, and a quarter of women of reproductive age were at risk of iron-deficiency anaemia. 19,20 However, data on diet were sparse. Apart from the food balance sheets produced by the United Nations' Food and Agriculture Organization, the only sources of data on food and diet in Bosnia and Herzegovina were the 2015 Household Budget Survey, which included food expenditure data, 21 and occasional population health surveys, which included a short-form, foodfrequency questionnaire. 20 Both these data sources were heavily dependent on donor funding and external expertise. We were unable to identify any food consumption surveys in the country that included data from individuals. Studies have identified several considerations that should be taken into account when setting up a national or global dietary and nutrition surveillance system: (i) demand for information; (ii) the institutional base for dietary surveys; (iii) resource needs (i.e. cost and staff time); (iv) personnel capacity; (v) technical capacity; (vi) the diet assessment method; (vii) a countryspecific food composition database; and (viii) data user participation. 13,22 Table 1 details how some of these considerations were tackled in establishing a dietary survey for Bosnia and Herzegovina and briefly describes future challenges.

Institutional base
The lack of a natural institutional base, or home, for dietary surveys that is independent of policy-makers was identified as a key challenge. 13,22 In Bosnia and Herzegovina, we asked the Institute for Statistics of the Federation of Bosnia and Herzegovina (a government body that is also an independent producer of official data), if it would take charge of implementing dietary surveys. 27 Subsequently, the government approved the inclusion of a pilot diet survey in the Institute's annual plan for 2017. The survey relied on the institute's existing infrastructure, including staff, regional offices, telephones, printing facilities and links with municipal authorities, the police and local media. The institute's involvement also provided access to the sampling frame used in previous surveys and enabled dietary data to be linked to other survey data at the individual level. One of the authors, who was a full-time nutrition scientist at the institute, underwent professional training at the Harvard T.H. Chan School of Public Health and, thereafter, developed the survey protocol, trained survey interviewers, supervised fieldwork, analysed data and prepared reports. The lead statistician at the Agency for Statistics of Bosnia

Resource utilization
The large sample size, the high cost of travel for interviewers and the large team of experts traditionally required to implement dietary surveys were additional challenges. To minimize sampling costs, we drew a cluster sample of households from the sampling frame used for the 2015 Household Budget Survey. The sample was nationally representative and comprised a two-stage, stratified random sample of households: (i) first, a sample of enumeration units (i.e. primary sampling units) was selected, with each unit having an equal probability of selection; and (ii) second, a sample of households (i.e. secondary sampling units) was drawn by applying Bethel's optimal allocation algorithm using data for 24 variables from the previous household budget survey across strata. 28 Restricting our selection of households to respondents in the 2015 Household Budget Survey enabled us to link dietary and socioeconomic data. We aimed for the minimum sample that was large enough to achieve a precision of 5 to 10% for the mean intake of each nutrient of interest. For the majority of nutrients, a sample of 100 to 250 individuals was sufficient, assuming a single day of data per person (vitamin A required a larger sample because of high within-person variability in intake). 29 The European Food Safety Agency recommends including at least 130 individuals in each subgroup of interest (e.g. 130 males and 130 females in each age group). 30 Consequently, a sample of around 1000 individuals was sufficient to provide high precision for the survey overall and reasonable precision for subgroups defined by age, gender, type of residence and other characteristics.
First, we selected primary sampling units that were within 90 minutes' travel from the interviewer's residence; then, we randomly selected households within these clusters. For methodological and financial reasons, we excluded children

Box 1. Considerations for a model of dietary surveillance for countries with constrained resources
• Cover a representative sample of the population, ideally a nationally representative sample would provide results generalizable to the entire population. • Collect data from individuals (in contrast to household spending surveys or food balance sheets) and link to data from existing surveys (e.g. economic surveys) so that diet can be analysed in terms of demographic and socioeconomic variables. • Collect data continuously or, at least, periodically (e.g. four-yearly) to enable trend analysis. • Use dietary assessment methods that are standardized and harmonized with those in other countries to permit cross-country comparisons. 13 • Analyse both food and nutrient intake to identify nutrients that are consumed either inadequately or in excess; • Focus on the "usual intake" of the population, not on intake on any one day.
• Provide data on the intake distribution (i.e. on the prevalence of inadequate and excessive intake) in addition to the mean intake (the primary goal). was selected as the institutional base; (ii) the government of Bosnia and Herzegovina approved the pilot survey as part of the Institute's 2017 annual plan (i) The Institute for Statistics provided access to detailed data in the master sampling frame used in previous surveys and to infrastructure and local connections (e.g. regional offices, telephones, printing and links to police, municipal authorities and local media); (ii) one local staff member provided by an academic partner trained to become a local contact for future dietary surveys; (iii) an academic partner provided financial resources to cover interviewer fees; (iv) clear protocols and operational procedures were developed for the survey The government should include a dietary survey in routine data collection (e.g. four-yearly)

Resource utilization Sample
Large sample size needed for traditional dietary surveys (i) Adopt a multiple-stage, cluster sampling approach; (ii) use the minimum sample size required for monitoring dietary intake in the population; (iii) use the minimum sample to achieve a precision of 5 to 10% for the sample means of each nutrient; (iv) employ existing local expertise, when possible (i) Sampling frame for latest Household Budget Survey in 2015 used; (ii) sample design performed by the Household Budget Survey statistician (ii) lack of a local food composition database (i) Procure software for nutrition assessment linked to the local Balkan food composition database; (ii) train local staff to use the software (i) Diet-assessment software and a food composition database -our dietary study used Diet Assess and the Balkan food composition database; 25 (ii) one trained analyst to code food items, assign portion sizes and run basic descriptive statistics using SAS codes; (iii) food atlas for estimating portion sizes; (iv) SAS codes for analysis and adolescents, pregnant and breastfeeding women and elderly people with cognitive decline (each of these groups requires a different data collection protocol and a separate data analysis). 30 Eligibility was determined using presurvey questions. Finally, we chose a sample size of 980 households on the assumptions that two people could be interviewed in around 50% of households and that the nonresponse rate would be around 40%. 31 To minimize fieldwork costs, the first eligible male and female in each household were interviewed whenever possible. However, we reached the quota for female participants aged 60 years and older within 9 months, whereas young male participants were underrepresented. In the last 3 months of the survey (i.e. September to November 2017), we focused on young male participants to obtain sufficient numbers in each sex and age subgroup. We obtained written informed consent from each participant at recruitment. After three unsuccessful contact attempts, a household was regarded as unavailable. The refusal rate was the percentage of households that were successfully contacted but refused to participate. Subsequently, appropriate sampling weights were applied to adjust for design and participation effects.

Personnel capacity
Local capacity was built among staff already employed by the Institute for Statistics. In addition to the nutritionist, a statistician and an information technology technician were available part time for sample design, calculating weightings, online questionnaire development and testing, and technical troubleshooting. The dietary and ancillary data collection protocol was developed by synthesizing the best practice of NHANES and the European Food Safety Agency. 11,30 The protocol provided detailed instructions for interviewers on contacting participants, keeping records, conducting both dietary interviews (with particular attention to the multiple-pass method) 32 and use of the food atlas. 33 The protocol developed could be used in the next survey and passed on to any new members of staff. Interviewers were carefully selected from the Institute's existing roster using predefined criteria and underwent training. The key requirements were: (i) a high level of motivation; (ii) computer literacy; (iii) ownership of a vehicle with the ability to travel within a 90-minute radius of their place of residence; (iv) an assertive personality; (v) an understanding of the importance of high data quality; (vi) cultural sensitivity; and (vii) professional demeanour. The funds required for implementing the dietary study were covered by the Institute for Statistics' budget and there was no need to rely on donor funding (Box 2). Our approach was markedly different from previous practice in post-war Bosnia and Herzegovina, when donor-funded activities might have artificially inflated the cost of surveillance because costs were often determined by the amount of external funding, rather than by local needs and economic conditions.

Diet assessment method
Selecting the most appropriate diet assessment method was another challenge because no single method is optimum for all purposes: for example, the best method for nutrition counselling may not be best for dietary surveillance or for evaluating the relationship between diet and disease. Inevitably, the choice of method involves a trade-off between statistical precision, the period covered, the level of detail and cost. 34,35 For our study, interviewers used the five-step, multiple-pass method based on 24-hour dietary recall to obtain dietary data on two occasions: 32 a household interview was followed by a phone call 4 to 30 days later. This approach is regarded as the preferred survey method for obtaining a good level of detail on diet. 13,36 To determine the quantities of food consumed, we used the Diet Assess food atlas, 33 which features colour images of eating utensils, dishes, individual foods and meals.

Food composition database
Another common obstacle for dietary surveys is the lack of a local food composition database. 13 Our data were analysed using the Balkans Food Composition Database, 23,33 which has been developed over the past decade using data from the European Food Information Resource Network, 37 from the local food industry and from chemical  25 As the local database is a work in progress, before performing our analyses, we supplemented the Balkan database with data from the United States' Department of Agriculture's database and incorporated some new foods (e.g. chia seeds), some Bosnian composite dishes and some important nutrients that were missing from existing foods (e.g. docosahexaenoic acid and eicosapentaenoic acid). 38 In addition, as the Balkans Food Composition Database also incorporates the European Food Safety Agency's FoodEx2 standardized food classification and description system, 39 it was possible to compare our survey results with those of other surveys conducted in Europe and to share data with the Global Dietary Database. 36,40 Our survey covered all four seasons of the year and data were collected on week days and weekends in the ratio of 70:30. We used Diet Assess software to convert dietary data into data on nutrients and food groups. 25,41 Composite dishes were disaggregated into separate food groups and each food item was analysed for nutrient intake, taking into account the food preparation method. Mean nutrient and food intakes were calculated using data from the first survey interview only to stay consistent with other nutrition surveys, 42 whereas data from both interviews were used to model the usual intake distribution.

Data collection
As paper-based data collection is slow and relatively expensive, we developed our questionnaires using a Qualtrics computer-assisted personal and telephone interviewing platform with an offline data collection option (Qualtrics, Provo, United States of America). 24 Interviewers collected dietary, demographic, lifestyle, health and anthropometric data using computer tablets and uploaded the data when an internet connection became available. Anthropometric measures were self-reported: we provided measuring tapes, scales and instructions to enable interviewees to take measurements in privacy. We validated this approach by verifying waist and hip circumference measurements in around 10% of participants (they were not informed about validation in advance to avoid bias). The intra-class correlation coefficient between the interviewers' and participants' measurements was 0.97 (P < 0.0001) for waist circumference and 0.91 (P < 0.0001) for hip circumference. Blood pressure was measured by the interviewer using an Omron M6 Comfort automated monitor (Omron, Kyoto, Japan) validated for field use in epidemiological studies. 43,44 Initial survey findings For our survey, four interviewers carried out fieldwork between November 2016 and November 2017 that covered 980 households in 200 enumeration units and collected data on 872 adults (380 men and 492 women). Residents in 24.0% (235/980) of households could not be contacted because they had died or moved or for some other reason, whereas 10.0% (98/980) refused to participate and 66.0% (647/980) participated. Complete data were collected on 96.0% (837/872) of participants and data were lost during transfer from the offline survey application on 3.0% (26/872). In addition, 0.2% (2/872) refused blood pressure measurement, 0.9% (8/872) failed to provide anthropometric measurements and 0.5% (4/872) did not respond to the phone call for the second interview. A systematic error was found in data collected on 19 participants (2.2%) by one interviewer, nutrient values were significantly higher than for other interviewers. Consequently, these observations were excluded from the analysis and the final data sample comprised 853 participants (371 men and 482 women). The final respondents' profile was representative of the general adult population, excluding pregnant and breastfeeding women ( Table 2). The nonresponse rate for households was 34.0% (333/980), which is within the range observed in other dietary surveys. 31 The initial results show a high estimated prevalence of overweight or obesity (69.0%), abdominal adiposity (63.0%) and hypertension (53.0%). 45 There were also indications that inadequate intake of some key nutrients and excessive intake of nutrients associated with noncommunicable diseases were common, which underlines the importance of dietary surveillance in this population.

Future challenges
During our study, we identified several outstanding issues (Table 1). First, a strategy for finding and retaining reliable and conscientious interviewers is needed to ensure good-quality data are collected. The Institute for Statistics could hire full-time interviewers for multiple surveys rather than recruiting different individuals for each. Second, our use of an outdated sampling frame resulted in a relatively high nonresponse rate and the overburdening of respondents because the same households were repeatedly sampled. The Institute could build longterm relationships with households in the master sampling frame by providing incentives (e.g. gift vouchers) in partnership with local supermarkets. Participants' details could then be kept up to date and they would be motivated to participate in multiple surveys. Third, because of our household sampling strategy and our use of an outdated sampling frame, we reached our quota of middle-aged women half way through the study but lacked sufficient data on young men. Given financial and time constraints, we changed our strategy and focused on missing age and sex groups. Ideally, future sampling should be based on individuals to avoid this problem. Fourth, our high interviewer costs could be reduced by integrating the dietary survey into the Household Budget Survey as a separate module in a way that does not overburden participants. We were unable to evaluate this approach as our dietary study took place 2 years after the last Household Budget Survey. Fifth, our approach to waist and hip measure-ment, which showed good correlations between interviewers' and respondents' measurements, should be further tested in different population groups as it may be pragmatic in contexts where body contact is inappropriate or where a male interviewer must collect data from female participants. Finally, we did not explore ways of generating a demand for data among users or of encouraging data users to participate in data production, both of which may make dietary surveillance more sustainable. 22 This was beyond the scope of our study.
In conclusion, we were able to conduct a survey of population dietary habits in a resource-constrained setting with a relatively small investment by relying on an efficient, highly motivated team and by using surveillance structures that commonly exist in many low-and middle-income countries. In particular, collaboration with the national statistics office enabled us to integrate data from multiple surveys and thus identify potential health disparities. The incorporation of dietary surveys into the regular activities of a country's statistical institutions makes it possible to monitor trends, ideally on a continuous basis. In the future, we plan to extend data collection to children, adolescents and pregnant and lactating women and to foster the participation of data users. We hope our approach can be applied in other low-and middle-income countries to obtain high-quality dietary data for both national and international use. ■

Viabilidad y sostenibilidad de la vigilancia alimentaria, Bosnia y Herzegovina
Los sistemas nacionales de vigilancia de la alimentación son necesarios para controlar la ingesta de alimentos y nutrientes asociados con la salud y las enfermedades, así como para alcanzar los objetivos alimentarios nacionales y mundiales. Sin embargo, estos sistemas no existen en muchos países de ingresos bajos y medios. La elaboración de un modelo de vigilancia alimentaria para Bosnia y Herzegovina, que se describe aquí, permite comprender mejor la viabilidad y la sostenibilidad de los sistemas de vigilancia alimentaria en entornos con recursos limitados e ilustra los problemas que se plantean. En 2016, se inició una encuesta alimentaria de un año de duración en colaboración con el Instituto de Estadística del país a partir de una submuestra de hogares que participaron en la Encuesta nacional de presupuestos familiares de 2015. Los entrevistadores recopilaron datos de estilo de vida, antropométricos y de salud y los participantes respondieron a dos cuestionarios de recordatorio alimentario de 24 horas. La encuesta incluyó una muestra representativa de 853 participantes y fue realizada eficazmente por un pequeño equipo de personal altamente motivado y bien capacitado. La realización de una encuesta alimentaria de alta calidad resultó factible a pesar de la escasez de recursos. Además, la capacidad de vincular la ingesta alimentaria con los datos de las encuestas periódicas en los hogares era una manera eficaz de asociar las variables alimentarias con los factores socioeconómicos determinantes de la salud. Esta encuesta sobre la alimentación, la primera realizada por una institución oficial de Bosnia y Herzegovina, representa un punto de partida importante para la creación de un sistema sostenible de vigilancia alimentaria para el país. El enfoque rentable y de baja carga de la vigilancia alimentaria que aquí se describe podría aplicarse en otros países de ingresos bajos y medios, muchos de los cuales ya llevan a cabo encuestas económicas periódicas.