Using underweight for predicting stunting among children in India: Analysis from the Comprehensive National Nutrition Survey

Abstract Background: Stunting and underweight are the most commonly used indicators to assess the nutritional status of children. Prior research has highlighted the problems encountered while measuring the height of child. The current paper aims to assess the relationship between stunting and underweight and derive regression equations to predict stunting from underweight prevalence among children under five years of age.Method: Data was drawn from 38,060 and 219,796 nationally representative sample of children aged 0-4 years, from Comprehensive National Nutrition Survey (CNNS, 2016-18) and National Family Health Survey 4 (NFHS-4, 2015-16), respectively. Anthropometric indicators, stunting and underweight was calculated based on the 2006 WHO Child Growth Standards. Univariate and bivariate analysis was done to obtain estimates for stunting and underweight. A series of regression models were run to obtain an estimate of percent stunted as a function of percent underweight at the national and regional level. Predicted stunting prevalence was estimated from regression equation for selected states and compared with observed prevalence from other studies. Data were analysed using STATA V.16.0. Results: Nearly one out of four children under 5 years of age were stunted and underweight in CNNS and NFHS-4. Out of those stunted, 67% and 65% were underweight in CNNS and NFHS-4, respectively. At the national level, there was high correlation between the two indicators (r > 0.7) in both CNNS and NFHS-4, whereas at the regional level in NFHS-4, the correlation coefficient ranged from 0.32 for central region to 0.86 for southern region. At the national level the slope was 0.557 in CNNS and 0.610 in NFHS-4. At the regional level, it varied from 0.334 in the central region to 0.847 in the western region. Similarly, at the national level, the intercept (α) was almost same when we analyzed CNNS data or NFHS-4 data or both together (~15), however, wide variability was observed between different regions (4.61 in western region to 30.14 in central region).Conclusion: Our analysis shows that regression equations with child underweight prevalence may be used to predict stunting where the quality of length/ height measurement is poor or unavailable, in regions where high correlation between the two indicators was found.