Effect of nutrition survey ‘cleaning criteria’ on estimates of malnutrition prevalence and disease burden: secondary data analysis

Abstract

Tackling childhood malnutrition is a global health priority. A key indicator is the estimated prevalence of malnutrition, measured by nutrition surveys. Most aspects of survey design are standardised, but data ‘cleaning criteria’ are not. These aim to exclude extreme values which may represent measurement or data-entry errors. The effect of different cleaning criteria on malnutrition prevalence estimates was unknown. We applied five commonly used data cleaning criteria (WHO 2006; EPI-Info; WHO 1995 fixed; WHO 1995 flexible; SMART) to 21 national Demographic and Health Survey datasets. These included a total of 163,228 children, aged 6–59 months. We focused on wasting (low weight-for-height), a key indicator for treatment programmes. Choice of cleaning criteria had a marked effect: SMART were least inclusive, resulting in the lowest reported malnutrition prevalence, while WHO 2006 were most inclusive, resulting in the highest. Across the 21 countries, the proportion of records excluded was 3 to 5 times greater when using SMART compared to WHO 2006 criteria, resulting in differences in the estimated prevalence of total wasting of between 0.5 and 3.8%, and differences in severe wasting of 0.4–3.9%. The magnitude of difference was associated with the standard deviation of the survey sample, a statistic that can reflect both population heterogeneity and data quality. Using these results to estimate case-loads for treatment programmes resulted in large differences for all countries. Wasting prevalence and caseload estimations are strongly influenced by choice of cleaning criterion. Because key policy and programming decisions depend on these statistics, variations in analytical practice could lead to inconsistent and potentially inappropriate implementation of malnutrition treatment programmes. We therefore call for mandatory reporting of cleaning criteria use so that results can be compared and interpreted appropriately. International consensus is urgently needed regarding choice of criteria to improve the comparability of nutrition survey data.

Double burden or double counting of child malnutrition? The methodological and theoretical implications of stuntingoverweight in low and middle income countries

“…17 Although this type of issue has been documented elsewhere, 20 prevalence estimates is apparent in all LMICs (table 1). [17][18][19][20][21] In absolute numbers, using estimates of the population of children under-five for these countries' respective years, over 10 million children are found to be concurrently stunted and overweight. 17 18 19 DISCUSSION For the first time, the prevalence of stuntingoverweight across LMICs has been documented.…”

“…The AEs are presented in table 1, [17][18][19][20] next to the current rates published in the JMEs prevalence of stunting and overweight for all countries included in the study. SE and 95% CIs of the AEs are available in online supplementary table S1.…”

“…Excluding the stuntedoverweight, stunting rates are now 8.5% (≈13.1 minus 4.7) and overweight 8.5% (≈13.3 minus 4.7%) (table 1). [17][18][19][20][21] Thus, currently, when research involves using both stunting and overweight levels for a population, these data contain double-counted stuntedoverweight children.…”

“…20 For example, in the case of Kazakhstan, dropping biologically implausible cases for the analysis for AEs led to a sample size of 4985 children. In the JMEs, n=5015.…”

Parameters associated with design effect of child anthropometry indicators in small-scale field surveys

“…As expected, the median prevalence of wasting (10%) was lower than that for underweight (27%) or stunting (42%) [34]. The prevalences of underweight and stunting were closer to 50% than for wasting, which may in part explain the higher values of DEFF for underweight and stunting observed [2].…”