Report: How can we produce a time series of country level childhood wasting estimates, accounting for seasonality: exploring the impact of survey timing
This project explored the seasonal effects of wasting scores with the goal to establish if it is possible to answer the following question: “what would the wasting score have been had it been measured in a different month of that year?”.
-
Undernutrition in children is assessed from measurements of growth, primarily weight and height. Wasting, or low weight for height is characterised by a loss or deficit of soft tissue, particularly body fat and skeletal muscle. In 2022, wasting was estimated to affect 45 million (6.8%) children under 5 years of age, and 13% of under-five child deaths are attributed to wasting each year. The Joint Malnutrition Estimation Group (JME) hold a large amount of data on child wasting. However, the timing of surveys is not focused on capturing the main issues surrounding child nutrition; rather, quite understandably, it is focused on minimising the costs of surveys. This results in inconsistent survey periods, and the timing of the surveys can have an impact on wasting measures. This project explored the seasonal effects of wasting scores with the goal to establish if it is possible to answer the following question: “what would the wasting score have been had it been measured in a different month of that year?”. Results indicated the following:
1. Wasting does vary seasonally in each country and that there appears to be a ‘wasting season’;
2. Controlling for wealth and education, the wasting scores still varied monthly;
3. The multi-level logistic regression model provided a list of variables that are correlated with wasting, many of which vary seasonally;
4. The multi-level logistic regression model was able to accurately estimate monthly wasting values using a year and month of survey.
The results indicate that prediction is more accurate when data are available from multiple months and years. The study used data only from the Demographic and Health Survey (DHS) which limited the amount of data available and limited the ability to establish seasonal patterns in some countries. It is recommended that this study be repeated by combining SMART surveys with the DHS data to establish a longer and deeper time series for each country.
Geospatial data had a limited but significant impact on the models. It is recommended that future work should establish proxy metrics for specific issues affected child wasting from geospatial sources. For example, the remotely sensed normalised difference vegetation index (NDVI) is the most commonly used geospatial variable in these types of models across the literature, however, in our project, it was often not significantly related with wasting. NDVI is an artificially created index, so it cannot be mechanistically linked to wasting or food production. It could be used to generate additional metrics that are more directly related to wasting or food production and availability. For example, converting an NDVI time series into the number of growing days in the year and then linked this to a cropping calendar to see if these days were above or below a threshold for particular crops. Physical access to markets is also important for food supply/access. Furthermore, geospatial variables only focused on food production (temperature, rainfall, drought, and soil moisture) and did not consider food access. In future, estimating travel time/access to markets should also be included.
Overall, the results of this project indicate that there are seasonal patterns and that statistical models can establish these patterns and estimate monthly wasting values using a range of covariates. Therefore, adding further data on wasting to build the time series/seasonal patterns along with further work on the geospatial covariate design should lead to a more accurate set of estimations. Ultimately, this could allow for monthly wasting correction factors/adjustment factors to be created for specific countries.
