Identification of infants at risk of child undernutrition in India: building a predictive algorithm with data from a nationally representative survey. (March 2019)

• Record Type: Journal Article
• Authors: Soni, Apurv; Allison, Jeroan
• Is Part Of: Lancet. Volume 7(2019)Supplement 1
• Page Start: S24
• Publication Date: 2019-03
• Subjects: World health -- Periodicals; 362.105
• DOI: 10.1016/S2214-109X(19)30109-3
• Languages: English
• ISSNs: 2214-109X

Abstract

Abstract: Background: India is at the centre of global child undernutrition, with a burden of malnourished children nearly twice that of all sub-Saharan African countries. The Indian government has established many national-level initiatives to address this public health crisis. However, there is substantial variation between and within states in the prevalence of child undernutrition, and strategies to identify at-risk populations are needed in the context of limited resources. Here, we describe the development of an algorithm that can be deployed at the time of delivery to characterise newborns' risk of undernutrition in the first 5 years of life. Methods: We extracted data on 232 440 children younger than 5 years from the 2015–16 National Family Health Survey. A child was considered undernourished if either height-for-age, weight-for-height, or weight-for-age was more than 2 SD below median WHO child growth standards. We used predictor variables identified in published studies if they could be measured at the time of delivery and used multilevel logistic regression to model the outcome. Model calibration was assessed using the Hosmer-Lemeshow test. We tested internal validity of the model using 200 bootstrapped samples to derive an optimism-adjusted c-statistic. All analyses were done with svy command in STATA to account for complex clustered sampling. Findings: In 2016, 54·7% of Indian children younger than 5 years were either stunted (38·4%), underweight (35·8%), or wasted (21·0%). The predictive model for overall undernutrition included maternal factors (height, education, reproductive history, number of antenatal visits), child factors (sex and birthweight), and household characteristics (district of residence, caste, rural residence, toilet availability, presence of a separate kitchen). The model demonstrated good discrimination ability (c-statistic: 0·688, optimism-adjusted c: 0·686). The prevalence of child undernutrition in the lowest decile risk group was 24·0%, and in the highest decile risk group it was 77·9%. Interpretation: We are developing a mobile-based app to collect the information and categorise children into a risk decile at the time of delivery. Since more than 80% of births in India are registered by community health workers, who are also responsible for implementing child nutrition programmes, this tool could help workers identify infants at risk of malnutrition and prioritise their interventions. Funding: AS: NIH-NCATS (TL1-TR001454) and NIH-NICHD (1F30HD091975-01A1) [AS] and NIH-NIMH (P60-MD006912-05) [JA].