Do childhood growth indicators in developing countries cluster? Implications for intervention strategies

Abstract

Objective: The effectiveness of geographic targeting in nutrition programmes depends largely on the degree to which malnutrition clusters within particular areas. This study investigates the extent to which the childhood nutrition indicators, stunting (heightfor-age Z-score ,22) and wasting (weight-for-height Z-score ,22), are spatially clustered; this information is used to determine the implications of spatial clustering for the effectiveness of geographic targeting. Design: Analysis of data from Demographic and Health Survey (DHS) results. Clustering is assessed by calculating intra-cluster correlation coefficients (ICCs). Estimating the proportion of malnourished children covered by a programme successfully targeting 10% of clusters with the highest malnutrition prevalences allows an assessment of the effectiveness of geographic targeting. Setting: Fifty-eight DHS III (1992III ( -1997 and DHS IV (1998 reports from 46 developing countries. Subjects: Pre-school children of mothers interviewed by DHS. Main results: The extent of clustering of nutritional status was surprisingly low (median ICC for national samples: stunting ¼ 0.054, wasting ¼ 0.032) and most countries were characterised by having an ICC , 0.1 -i.e. low clustering -for childhood undernutrition (91% of countries for wasting and 78% for stunting). Our assessment of the effectiveness of geographic targeting showed that coverage was better for wasting than for stunting; for wasting, 23% of countries would achieve less than 20% coverage, compared with 76% of countries achieving less than 20% coverage for stunting. Coverage is dependent on the overall prevalence of malnutrition and the ICC. Conclusions: Childhood nutritional status is determined at the household, or even individual, level; nutrition programmes that are geographically targeted may result in high levels of under-coverage and leakage, thereby compromising their costeffectiveness; the lack of clustering questions the appropriateness of current nutrition interventions.

Local spatial clustering of stunting and wasting among children under the age of 5 years: implications for intervention strategies

“…Our study builds upon the findings on the spatial structuring of undernutrition measured by anthropometric indices on different scales, including the household and community level (17)(18)(19) . Little area-level clustering of undernutrition among children under 5 years has been reported elsewhere (20,21) . Based on these findings, some hypothesized that the basis for the differences in nutritional status within a country is determined entirely by individual-or householdlevel characteristics (21) .…”

“…Some indicated that the distribution of undernutrition is heterogeneous within a geographic area, varies between villages and households (17) and is spatially structured (18,19) . Others argued against the evidence for clustering at the community level (20) and identified the geographic extent of clustering only at a household or individual level (21) .In the present paper, we attempted to investigate the local spatial structure of stunting and wasting among children under the age of 5 years. More specifically, we aimed to: (i) evaluate whether undernutrition indicators (stunting and wasting) have a tendency to cluster; (ii) determine the physical location and scale of clustering; and (iii) determine the risk factors for the observed clustering.…”

The Clustering Effects of Surfaces within the Tooth and Teeth within Individuals

“…There are various methods available to calculate ICC for a binary outcome, as discussed by Wu et al (2012) for randomized controlled trials and Fenn et al (2004) for crosssectional survey data. The ICC estimation from the random intercept logistic method suggested by Wu et al has some limitations.…”

“…Equation 2 (Donald and Donner 1987;Fenn et al 2004) where σ 2 b is the between-cluster variance of the outcome variable. For binary outcome measures, σ π π…”

“…The relationship between DE, cluster size, and ICC is represented in the following equation: (Fenn et al 2004) where DE is the design effect, and m is the average number of respondents per cluster, or average cluster size.…”