Our paper is one outcome of a much broader project, which has documented trends in risk factors for non-communicable diseases, or NCDs. These factors include overweight and underweight, diabetes, high blood pressure, and suboptimal cholesterol profiles. When the group started to study these variables in the late 2000s, fragmentary data had been collected by various researchers, but there were no coherent estimates of trends in countries around the world. The group collated as much of the available data as possible, and then developed a statistical model to estimate these trends. However, the sparsity of information available across countries and time meant that the model fit could not be based simply on the data. Instead, covariates related to NCDs were also included in the model; this helped to produce realistic trend estimates.
The authors involved at the time selected national-level covariates known to be related to NCDs. This included GDP, the proportion of people living in urban areas, and mean education level. There is also an established relationship between diet and NCD risk factors, but there was not an obvious single variable available. However, they were aware that the UN Food and Agriculture Organization has published food balance sheets for member states over many decades. These data describe the availability of food types for human consumption, with time series starting in 1961. To include all the 18 relevant food types as covariates seemed excessive, so there was a need to summarise the data. This was the key part of the analysis: when they applied principal component analysis to the data, they were surprised and pleased to find that the loadings for the first four principal components described coherent diet types, and that the components explained much of the information in the data. The four principal component scores were included as covariates in the statistical model, and the estimates of NCD risk factors were subsequently published.
We realised that these results were of interest in themselves and should be published. We also extended the analysis: additional work includes an index that describes overall change in national-level food supply over the past 50 years, and an examination of the relationships between the four scores and mean BMI and mean attained height in adults. One of the key results in the paper is that food supply has changed most in East Asia, particularly in South Korea, China and Taiwan, and least in sub-Saharan Africa. Also, the animal source and sugar diet, which might be seen as typically western, is now common across much of the world, and there has been a general increase in the proportion of diet made up of vegetables. We anticipate that our results will be used by other researchers to examine how variables such as changing trade patterns affect diets, and how changes in diet affect the environment and health.