Investigating foods and beverages sold and advertised in deprived urban neighbourhoods in Ghana and Kenya: a cross-sectional study

Objectives The aim of this study was to characterise the local foods and beverages sold and advertised in three deprived urban African neighbourhoods. Design Cross-sectional observational study. We undertook an audit of all food outlets (outlet type and food sold) and food advertisements. Descriptive statistics were used to summarise exposures. Latent class analysis was used to explore the interactions between food advertisements, food outlet types and food type availability. Setting Three deprived neighbourhoods in African cities: Jamestown in Accra, Ho Dome in Ho (both Ghana) and Makadara in Nairobi (Kenya). Main outcome measure Types of foods and beverages sold and/or advertised. Results Jamestown (80.5%) and Makadara (70.9%) were dominated by informal vendors. There was a wide diversity of foods, with high availability of healthy (eg, staples, vegetables) and unhealthy foods (eg, processed/fried foods, sugar-sweetened beverages). Almost half of all advertisements were for sugar-sweetened beverages (48.3%), with higher exposure to alcohol adverts compared with other items as well (28.5%). We identified five latent classes which demonstrated the clustering of healthier foods in informal outlets, and unhealthy foods in formal outlets. Conclusion Our study presents one of the most detailed geospatial exploration of the urban food environment in Africa. The high exposure of sugar-sweetened beverages and alcohol both available and advertised represent changing urban food environments. The concentration of unhealthy foods and beverages in formal outlets and advertisements of unhealthy products may offer important policy opportunities for regulation and action.


Appendix A -Location of study sites
Low income neighbourhoods were randomly selected in each city. In Accra, the selection of a neighbourhood was informed by the Accra Poverty Mapping Exercise (CHF International, 2010; https://www.globalcommunities.org/publications/2010-accra-poverty-map.pdf). Four areas were identified as being poverty endemic. Amongst these, Ga Mashie which is made up of James Town and Ussher Town, was purposively selected. A simple random sampling exercise was then applied and James Town was selected as the neighbourhood of interest. In the city of Ho, the United Nations Human Settlements Programme (UN-HABITAT) urban profiling report informed the selection of the study site. The report highlighted that 36% of the population lived in four poor areas within the city: Bankoe, Hliha, Ahoe and Dome (UNHABITAT 2009; https://uni.unhabitat.org/wp-content/uploads/2014/07/Ghana-Ho-City-Profile.pdf). Amongst these four areas, Dome was then randomly selected. In Nairobi, we used data from the Kenya National Bureau of Statistics (KNBS) to identify the deprivation level of locations (wards), and randomly selected Makadara Constituency. Jericho, Bahati, Maringo, Hamza, Makongeni and Mbotela communities in Makadara were purposively selected as these were areas we could feasibly work in.
Figures A1-A3 visualise the data collected within our study sites. Maps were created using R and the 'ggmap' library (1). Base maps were taken from 'stamen' maps. Map tiles by Stamen Design, under CC BY 3.0. Data by OpenStreetMap, under ODbL.

Appendix B -Description of outlet, advert, food and beverage types
Outlet type descriptions Outlet and advertisement types were defined following a project meeting involving all of the international project partners, researchers and local field workers representing both Ghana and Kenya. The aim was to find consensus over our definitions based on individual expertise and evidence from the wider literature (summarised during discussions), as well as local subject knowledge. Definitions were then validated and refined during the pilot phase of our tool. We opted against using other existing classifications since they were often derived for other settings or countries that were not always relevant to the contexts we were collecting data in. Through designing our own classification, we developed a new system that was relevant to both urban Ghana and Kenya, as well as simple and efficient for data collection. We did not record mobile local vendors since they did not have a fixed location, though accept that their non-trivial prevalence means our audit has undercounted the number of outlets. We initially included categories for bakeries (n = 1), chop bars and cold stores (Ghana only, n = 11 and 2 respectively), markets (we record each outlet within the market, not the overall market it itself, n = 1) and an 'other' category (n = 36). We did not include these categories in the analyses due to their low prevalence and excluded the data from all analyses.

Food and beverage type descriptions
Individual items were classified into broader groups to help improve their interpretation and minimise small number issues. We also grouped items (based on expert opinions and evidence across the literature) into whether we would expect them to increase or decrease in consumption following the nutrition transition to situate our data within broader nutritional trends in African countries. Peppers, onions Yes * Condiments could include products that are commercially processed from multinational companies, as well as those prepared at home or with local small-scale production. ** The classification of foods and beverages into "healthy" and "unhealthy" was informed by our related work on dietary intake (as part of the broader project), which used a nutrient profiling classification based on recognised methods (2) of foods and beverages consumed in the same cities. We did not classify items during data collection, however applied the classification after to aid the interpretation of our results.  There were some noticeable differences in the number of advertisements by outlet type (Table C2). You most likely to encounter an advertisement within a supermarket (with the mean count also being three times higher than any other outlet). Bars/pubs also had high proportion of adverts within them, as did restaurants and shops. There was low prevalence in table tops and local vendors. Formal outlets were more likely to contain adverts.   We compared the foods and beverages being sold (Table C3) and advertised (Table C4) in formal and informal outlets. The items sold between formal and informal outlets were somewhat different. Most products were more common in formal outlets, reflecting that they sold a greater number of items (see Table C1). Noticeably, unhealthy foods were more common in formal outlets (e.g. fats/oils 46.9% vs 16.0%; sugar sweetened spreads 45.2% vs 12.6%), with similar patterns for drinks as well (i.e. sugar sweetened beverages 75.7% vs 22.8%; alcohol 40.0% vs 3.5%). Not all items were more common in formal outlets; vegetables and fruits were more common in informal outlets. Fresh meat, poultry and fish were similar between outlet type. There was little difference in the foods and beverages being advertised. Only alcohol displayed a large difference, with it being uncommon in informal outlets than compared to formal outlets where it was more prevalent.

Appendix C -Descriptive statistics on items sold by outlet type
We repeated these summary statistics by specific outlet type as well (see Tables C5 to C7). In summary, they present similar patterns to those described above. Foods and beverages sold by local vendors and stands/table tops tended to be dominated by healthier foods and raw ingredients (Table C5). Supermarkets and shops had a greater diversity of items sold. Pubs and bars mostly sold alcohol and sugar sweetened beverages, with a few selling snacks or fresh meat/poultry. The foods and beverages sold by outlets differed to those advertised (Table C6). There was less diversity with lower values reported (bar supermarkets). Many of the foods and beverages advertised represented the most common items sold (i.e. the higher values in Table C5). For advert types, drinks were most commonly advertised. These were particularly common in posters, with onsite and painting having a greater diversity of foods and beverages advertised (reflecting outlet fronts advertising what they sold).   The AIC and G 2 models produced similar patterns (Figures D1 and D3). For the BIC the pattern was clearer ( Figure D2). An increasing number of groups produces better fitting models, but a decreasing rate of improvement. There is no clear knee point though. Following a 5 class solution there is little improvement with subsequent additional class added. Since we are looking for the parsmonious solution, a 5 class solution might work best given that additional classes are not associated with large improvements in model fit. The results for BIC are clearer ( Figure D2). We broadly see an improving solution upto 5 classes, whereby model performance is flat onwards. Based on these metrics, a 5 class solution was selected as the final model. Exploring model interpretation of additonal class solutions does not reveal any distinct classes, merely splitting up established classes into less similar classes that differ on small characteristic.