In 2022, the Observatory partnered with the Data for Children Collaborative in an impact collaboration process to investigate how we can best produce a time series of childhood wasting estimates that account for seasonality, survey timings and climate impacts. Here, team members* reflect on the SEASNUT project’s initial finding and insights, highlighting some of the challenges around the available data and the potential use of geospatial data to help explain the seasonality often observed in the data.

The problem

The Joint Malnutrition Estimation (JME) group (UNICEF, World Health Organisation and the World Bank) releases annual estimates on the status of child stunting, overweight, underweight, wasting and severe wasting.

Since child wasting varies inter-annually and can fluctuate rapidly [1], the timings of the surveys, especially in relation to hunger periods and agricultural calendars, that drive the estimates can have a big impact on measures of the prevalence of wasting.

The SEASNUT project is examining if secondary data can be used to account for fluctuations in wasting measurement, by correcting temporal inconsistencies in sampling to allow partial historical trends to be adjusted for seasonal fluctuations to produce and a more accurate measurement of wasting prevalence.

What is child wasting?

Wasting can have serious impacts on the health, development and life of a child [2]. It occurs when nutrient intake does not meet the demands for physiological and biochemical functions, growth and capacity to respond to illness. When the body is deprived of enough food and nutrients it uses body fat, muscle and other nutrients to maintain essential metabolic processes [3, 4], resulting in weight loss and can lead to a failure to grow. Wasting can occur at any stage of development, including in utero [4]. Globally an estimated 45 million children under-five are wasted (JME) and 13% of under-five child deaths are attributed to wasting each year [5]. Furthermore, for those children that survive severe wasting, each occurrence increases the risk of stunting [6], which is associated with a range of further problems related to development and future economic power.

Work so far

The project so far has examined the JME data on wasting and stunting to establish the data available for each country, the timings of surveys and wasting prevalence and importantly flagging the limitations with the data for this sort of exploratory study. Summaries of national levels of wasting and stunting have been collated on this website, showing, overall that the trend in child stunting has been decreasing in most countries (Figure 1).

Figure 1 Child stunting estimates constructed from the JME database (black lines show stunting prevalance in regions; the red line is the national average) showing that the trends in most countries are decreasing over time.

For each country in the JME database, we calculated the percentage of children under 5 years of age falling below -2 standard deviations (moderate and severe) from the median weight-for-height of the reference population (Figure 2). In Bangladesh, for example, 25 separate datasets can be used to estimate national level wasting prevalence between 1986 and 2019 and there are nine surveys allowing for sub-national wasting estimate. The child wasting prevalence for each country is also trending downwards, but this is less clear than for child stunting, partly because of the challenges of collecting child wasting statistics.

Figure 2 Prevalence of child wasting (%) constructed from the JME database. This shows the number of surveys conducted, the wasting percentage for regions (black line) and the national average values (red line).

We are trying to examine if these fluctuations (Figure 2) are due to survey timing or whether they reflect seasonal cycles of wasting.To date, exploratory modelling has used Demographic and Health Survey (DHS) data in Bangladesh from 1996 to 2018. We focused on estimating monthly wasting prevalence for each survey period. This is an important step because some of the surveys were conducted over 3, 6 or even 9 months, usually spanning key periods in the cropping calendars.

After creating a database to provide monthly estimates of child wasting, the constructed monthly wasting prevalence data were plotted for each surveyed year to establish if there were any seasonal patterns (Figure 3). In Bangladesh, data shows that the prevalence of child wasting can range from 8% to above 20%. Moreover, the number of surveyed months per year range from two (1996, 1999) to six (2007, 2011), highlighting the difficulties in exploring seasonal patterns. For more information on why this is such a challenge see [1]. Ideally, to establish a trend/seasonal pattern we would see monthly wasting prevalence rates every year in which the survey was conducted. When only partial trends are possible from the data, we used statistical methods to establish if trends may exist in the data.  

Figure 3 Monthly wasting prevalence in Bangladesh constructed from Demographic Health Survey (DHS) data between 1996 and 2018.

This exploratory analysis in Bangladesh tried to establish if wasting prevalence rates followed episodic/seasonal patterns. One of our initial investigations led us to examine if a sine wave approach could be used to predict wasting prevalence rates in Bangladesh (using a Binomial Generalized Linear Model). Despite fluctuations in the wasting prevalence rates (Figure 3) observed data showed an overall declining trend in the wasting prevalence which appears to be partially captured in the model outputs in Chittagong (Figure 4). 

Figure 4: Predicted rural wasting prevalence (%) in Chittagong, Bangladesh. Red points are observed wasting values from the survey data, black points are predicted wasting values. Note, this graph is just for illustration purposes, we do not imply that the seasonality of wasting can be predicted with a sine wave approach.

The initial model predictions (black lines and clear black points) appear to capture an overall downward trend while also identifying a seasonal fluctuation. This illustrates what we mean by seasonal/episodic patterns. The sine wave approach is an over-simplification of the patterns because the frequency and amplitude of the model will remain fixed which does not reflect seasonal cycles in wasting as these often follow climatic patterns – several studies show that wasting is higher during the rainy season in South Asia, but the rainy seasons can vary in length as well as timing (onset and end). Thus, we also seek to identify if geospatial data collected on a recurring monthly basis can help explain some of the seasonality that is often seen in the wasting data.

Next steps

The team is building more detailed models that account for the drivers of child wasting identified in the literature. The priority is to identify particular drivers that could be estimated using geospatial datasets such as earth observation satellite data.

Many years of research in child nutrition indicate that household poverty is the main driver of child wasting. Even when there is plenty of food grown within a region, much food is exported out of the local system to national or global markets. Thus, if families are affected by poverty, they often cannot afford to buy enough food, resulting in children losing weight. Seasonality therefore may have a limited impact on child wasting prevalence rates. However, since poorer households are net purchasers of food, if in a particular system there is a poor harvest this is likely to push up the prices of staple foods, which will have an impact on the amounts and types of food a poor household can purchase.

Thus, even if the main driver of child wasting is poverty, it may be possible to partially identify when wasting increases by using geospatial data that captures seasonality in particular domains.  Then, we can perhaps offer the JME an explainable model that allows them to account for some of the variability observed in their nutrition surveys.


This post was drafted by team members: Ian Waldock, Duncan Hornby, Victor Odipo, Surbhi Agarwal, Richard Kumapley, Robert Johnston, Simon Sovoe, Craig Hutton and Gary Watmough.


[1] Johnston, R., et al. 021.  Methods for assessing seasonal and annual trends in wasting in Indian Surveys (NFHS-3, 4, RSOC & CNNS), PLoS ONE 16(11): e0260301.

[2] Brown K.H., et al 1982. Seasonal changes in nutritional status and the prevalence of malnutrition in a longitudinal study of young children in rural Bangladesh. American Journal of Clinical Nutrition 36(2):303–13.

[3] Cahill GF. 2006. Fuel Metabolism in Starvation. Annual Review of Nutrition 26(1):1–22.

[4] Bhutta, Z.A., et al. 2017. Severe childhood malnutrition. Nature Reviews Disease Primers 3(1):17067.

[5] Black, R.E., et al. 2013. Maternal and child undernutrition and overweight in low-income and middle-income countries. The Lancet 382(9890):427–51.

[6] Schoenbuchner, S.M., et al. 2016. The relationship between wasting and stunting: a retrospective cohort analysis of longitudinal data in Gambian children from 1976 to 2016. American Journal of Clinical Nutrition 110(2):498–507.

Contact us for more information or to partner with us: