Prevalence of Double Burden of Malnutrition among Indian Pre-School Children: an Analysis of Cross-Sectional DLHS-4 Data from 23 States
Antti Kukka ______Master Degree Project in International Heath, 30 credits. Spring 2018 International Maternal and Child Health Department of Women’s and Children’s Health Supervisor: Prof. Anita Kar, Savitribai Phule Pune University, Pune, India Word Count: 10672
UPPSALA UNIVERSITET 2 (58)
Abstract Background: Double burden of malnutrition means co-existence of under- and overnutrition in a community, household or an individual. It is observed in countries undergoing nutrition transition. Research on double burden among Indian children aged under five has been scarce. I therefore aim to study the prevalence of double burden of malnutrition in this age group.
Methods: Cross-sectional population-based data from District Level Household and Facility Survey 4 conducted in 2012-2013 in 23 Indian States and Union Territories were used for the analysis. Prevalence of overweight, stunting and anaemia were examined as main outcomes at the community level. Simultaneous overweight and stunting or anaemia in same child were the main outcomes for individual level prevalence. Households with stunted child and overweight adult were also examined.
Results: In this sample, 40.1 % of children were stunted, 74.2 % anaemic and 8.2 % overweight. Simultaneous overweight and stunting as well as overweight and anaemia both affected about 5.5 % of children. Stunted children had a higher prevalence of overweight than non-stunted children. No such association was seen for anaemia. At household level, 26.5 % of households had an overweight adult and a stunted child.
Discussion: Double burden of malnutrition is emerging among Indian children aged under five and should be included in national nutrition policy. Undernutrition is still bigger concern than overnutrition, but overweight is spreading also to lower wealth quintiles. Further studies should be done to examine this phenomenon at the national level. UPPSALA UNIVERSITET 3 (58)
Contents Abstract ...... 2 Figures and Tables ...... 5 Tables ...... 5 Figures ...... 5 Supplementary Tables ...... 5 Supplementary Figures ...... 5 Abbreviations ...... 6 Introduction ...... 7 Nutrition Transition ...... 7 Double Burden of Malnutrition ...... 7 Individual Level ...... 8 Household Level ...... 10 Community Level ...... 11 Rationale for the Study, Research Questions and Objectives ...... 12 Methods ...... 14 Study Design and Population ...... 14 Outcome Variables ...... 15 Participant Selection ...... 16 Estimation of Potential Bias in Outcome Variables...... 19 Explanatory Variables ...... 20 Individual Level Double Burden ...... 20 Household Double Burden ...... 20 Wealth Index ...... 21 Statistical Methods ...... 24 Ethical Approval...... 25 Results ...... 26 Prevalence of Community Level Double Burden of Childhood Malnutrition ...... 26 Prevalence of Individual Level Double Burden of Childhood Malnutrition...... 27 Prevalence of Household Double Burden ...... 29 Discussion ...... 31 Limitations...... 34 Conclusion ...... 35 References ...... 36 UPPSALA UNIVERSITET 4 (58)
Annex ...... 48 Details on Wealth Index Generation ...... 48 Supplementary Tables ...... 51 Supplementary Figures ...... 57
UPPSALA UNIVERSITET 5 (58)
Figures and Tables Tables Table 1: Definitions of Study Outcome Variables Table 2: Prevalence and Predictors of Different forms of Malnutrition Table 3: Prevalence Ratios for Association between Different Forms of Malnutrition Table 4: Distribution of Double Burden Households Figures Figure 1: Drivers of Double Burden of Malnutrition Figure 2: Flowchart of Participant Selection for Main Outcomes
Supplementary Tables Supplementary Table 1: Options Available in Dataset for Age Determination Supplementary Table 2: Prevalence Estimates of Malnutrition Using Two Different Cut-Off Criteria Supplementary Table 3: Distribution of Observed Anthropometric Indices Before Data Cleaning Supplementary Table 4: Classification of Water and Sanitation Types into Improved and Unimproved Forms Supplementary Table 5: Distribution of Potential Household Wealth Variables by Household Location Supplementary Table 6: Weightings of Household Features and Assets Used to Generate Household Wealth Index Supplementary Table 7: State Distribution of Different Forms of Malnutrition
Supplementary Table 8: Bivariate Analysis of Predictors of Dual Burden Households Supplementary Table 9: Distribution of Households into Wealth Quintiles Using Wealth Scores Unadjusted (Row) and Adjusted by Household Locality (Column) Supplementary Table 10: Distribution of Households into Wealth Quintiles in Rural and Urban Areas Using Two Different Wealth Index Scores Supplementary Figures Supplementary Figure 1: Zone Division of States Included in the Study Supplementary Figure 2: Histograms of Anthropometric Indices Before Data Cleaning Supplementary Figure 3: Histogram of Distribution of Unadjusted Household Wealth Scores for All, Rural and Urban Household Supplementary Figure 4: Stacked Bar Chart of Division of Households into Wealth Quintiles in Each non- EAG State and Union Territory of India According to the Unadjusted Wealth Index Scores
UPPSALA UNIVERSITET 6 (58)
Abbreviations BAZ = BMI-for-Age Z-score BMI = Body Mass Index CAB = Clinical, Anthropometric and Biochemical component CI = Confidence Interval DOHaD = Developmental Origins of Health and Disease DHS = Demographic and Health Survey DLHS = District Level Household and Facility Survey EAG = Empowered Action Group HAZ = Height-for-Age Z-score ICDS = Integrated Child Development Services IIPS = International Institute for Population Sciences JMP = WHO and UNICEF Joint Monitoring Programme LMIC = Low- and Middle-Income Country NCD = Non-Communicable disease NFHS = National Family Health Survey OAC = Overweight-Anaemic Child OBC = Other Backward Classes OSC = Overweight-Stunted Child PCA = Principal Component Analysis PR = Prevalence Ratio PSU = Primary Sampling Unit SC = Scheduled Caste SCOA = Stunted Child – Overweight Adult SCOWT = Stunted Child - Overweight Mother ST = Scheduled Tribe SD = Standard Deviation SSU = Secondary Sampling Unit UNICEF = United Nations Children’s Fund WAZ = Weight-for-Age Z-score WHO = World Health Organization WHZ = Weight-for-Height Z-score UPPSALA UNIVERSITET 7 (58)
Introduction
Nutrition Transition Higher prevalence of childhood undernutrition in South Asia than in Sub-Saharan Africa despite equal or better economic status is known as the Asian Enigma (1). The reasons of undernutrition in India have been extensively studied (2–4) and its negative health and economic consequences are acknowledged by players like the World Bank (5). Over the period of United Nations’ (UN) Millennium Development Goals a significant decline in rates of childhood undernutrition was seen in South Asia (6,7), but still today India alone is home to about a third of all the stunted and wasted children aged under five in the world (8). Simultaneously, the problem of childhood overnutrition has emerged both globally (9) as well as in the region (10–12). Limiting the spread of childhood overweight was included in the new Sustainable Development Goals (13) and its significance to the public health in India is being increasingly recognized (14–17).
The shift in pattern of malnutrition from under- to overnutrition is termed nutrition transition. It is associated with economic, epidemiologic and demographic transitions. Nutrition transition was originally conceptualized by Popkin in 1990’s based on historical trajectories from High-Income Countries (18), but has since been shown to apply also to Low- and Middle-Income Countries (LMIC) with some modifications (19). For example in China, rapid urbanization of the last couple of decades has led to an increase in obesogenic diets and decreased physical activity among children causing an increase in overweight (20). Initially overweight affects wealthier parts of the society, but later the transition spreads to lower income groups (21,22). In India the economy has reached a level where the burden of overweight among adults typically starts to shift to the poorer classes thus increasing health disparities (23). Nutrition transition in India was reviewed by UN’s Food and Agriculture Organization in 2006 with strong evidence for on-going transition among the wealthier segments of the society (19). Another review from 2011 pointed out that urban Indians now have access to obesogenic diets (24) and the burden of disease across the country is rapidly changing towards non-communicable diseases (NCD) (25).
Double Burden of Malnutrition Co-existence of under- and overnutrition is known as double or dual burden of malnutrition and is a well- recognized phenomenon affecting children in LMICs (21,26). It encompasses micro- and macro-nutrient malnutrition and the resulting anthropometric changes as well as diet related NCDs (27). The concept of double burden was not originally included in Popkin’s model on nutrition transition, but due to observed UPPSALA UNIVERSITET 8 (58)
persistence of undernutrition in LMICs has since become an essential part of it. Popkin explains that a cause for the difference between the current trajectories of LMICs and the historical transitions in High- Income Countries is the increased speed of nutrition transition, which fuelled by globalization, urbanization and media exposure among other factors, takes place at earlier stages of economic development (28). Other drivers of double burden include biological adaptability, environmental and social factors related to food security and supply (Figure 1) (27). Already in 2006 UN Standing Committee on Nutrition published a Global Agenda on Tackling the Double Burden of Malnutrition that identified priority actions to prevent the spread of double burden (29). Varela-Silva et. al. described the phenomenon at three levels – individual, household and community – and drew important attention its definitions (30). In 2014 Tzioumis and Adair reviewed double burden in particular reference to children (21).
Figure 1: Drivers of Double Burden of Malnutrition. Adapted by Antti Kukka from WHO, UNICEF and World Bank (31)
Individual Level Most research on double burden of malnutrition at individual level has concentrated on examining the effects of nutrition transition and genetic adaptation over the life-course rather than at a single point of time (32,33). It is generally believed that the golden window for securing optimal linear growth are the first 1000 days after conception, meaning that children that are stunted during the first two years of life are likely to grow into short adults (26,34). Stunted women on the other hand are more prone to experiencing intra-uterine growth restriction during pregnancy (26) and up to 46.9 % of Indian children are born small UPPSALA UNIVERSITET 9 (58)
for gestational age (35). Moreover, at six months’ age around one in five children in India is stunted (36). According to the Barker’s hypothesis of Developmental Origins of Health and Disease (DOHaD), undernourishment during foetal period not only leads to short stature, but also increases risk for adult obesity, NCDs and mortality (37). The mechanisms of this adaptation sometimes termed thrifty phenotype have been studied extensively (38) and it seems that factors affecting child growth during different parts of development can have differential effects to adult body composition (39). In the New Delhi Birth Cohort higher birth weight and BMI during first 2 years were correlated with higher adult lean mass whereas weight gain after two years of age led to more adiposity (40). An examination of intergenerational change in this urban middle-class cohort showed a significant increase in height and weight of children when compared to their parents born around 1970 with a corresponding decrease in stunting and increase in overweight (41).
Paradoxically, both over- and undernutrition can exist simultaneously in same child with “short and plump” phenotype recognized in Latin America already in 1970’s (42,43). Since then it has been studied also in Africa (44). Popkin associated co-existence of overweight and stunting in same child with rapid shift in diet and physical activity as well as the Barker’s DOHaD theory (45). Later studies have tried to quantify these changes and the associated biological adaptation (46–48). A doctorate thesis in demography by Bates in 2014 pointed out that children who are both stunted and overweight might be counted twice when examining the prevalence of stunting and overweight, which could impact the way public health priorities are set (49). The latest World Health Organization (WHO), United Nations Children’s Fund (UNICEF) and World Bank joint malnutrition estimates acknowledge the phenomenon of individual double burden, but provide no estimation on its prevalence (31). As also the clinical implications of this phenomenon have been understudied, interventions for these children are lacking (49). In 2005 the prevalence of children simultaneously overweight and stunted (OSC) in India was only 1.0 %, but this still meant that about two thirds of overweight children were also stunted (49). Babies in India are often born with lower level of muscle mass but relatively more abdominal fat than their European counterparts (50). This could be interpreted as a sign of the thrifty phenotype already present at birth and genetic adaptation might thus explain not only predisposition to overweight and NCDs in adulthood, but also co-existence of under- and overnutrition in the same child during childhood.
Individual double burden of malnutrition includes not only anthropometric changes but a micronutrient component too. Studies from both High- (51) and Low-Income settings including India (52) have found overweight children to be at an increased risk of anaemia. This could be explained by higher iron need UPPSALA UNIVERSITET 10 (58)
because of increased body mass as well as decreased absorption of iron from the gut because of chronic inflammation (53). A study among South Indian adults, however, noted that the risk for double burden of anaemia and overweight was not statistically independent from prevalence of the two (54).
Research by Golden in the 1990’s helped to identify nutrients that have little storage space in the body but are essential to growth (55). These so-called type II nutrients include for example protein and zinc (56). Protein intake and quality among children in India appears to be inadequate (57–59) and zinc deficiency is estimated to affect up to 30 % of the total population (60). Zinc deficiency suppresses appetite (56) and Zinc supplementation is an important component of treatment for childhood diarrhoea that causes impaired absorption of nutrients (61). At national level diets in India are improving, but median intake of energy remains below recommendations and has even declined since 1970’s both among children and adults (58,59), which however could be balanced by decreased physical activity. It is conceivable that at least in the wealthier segments of population a change from energy-low, nutrient-high to energy-dense but nutrient-low diets (19) could have contributed to co-double burden of malnutrition in children.
Wasting commonly results from acute illness, but diarrhoea and sub-clinical chronic inflammation of the gut due to poor sanitary conditions known as environmental or tropical enteropathy are also some of the potential causes of stunting (62–66). However, the impact of infections on nutrition goes beyond anthropometric failure. The association between childhood enteric infections and metabolic syndrome in adulthood has been recently reviewed with tentative support for a link between the two (67). For example, a study from Guatemala found that diarrhoea during first two years of life might increase odds for overweight in adulthood (68). Recent studies from High-Income Countries have associated changes in gut microbiota induced by antibiotic treatment during infancy with overweight in later childhood (69,70). This finding further highlights the interplay between microbiology, inflammation and nutrition that is likely to have a role in the formation of double burden of malnutrition.
Household Level At household level, double burden of malnutrition is said to exist when both under- and overweight persons are living in the same home (71). A typical example of this is a pair of stunted child - overweight mother (SCOWT) (72). Doak et.al. hypothesized that as nutrition transition occurs first among adults, an underweight child - overweight adult is a sign of on-going nutrition transition (71) and showed that in a study of 7 countries, double burden households are more common in Middle-Income than Low- or High- Income Countries (73). This finding was later confirmed in a larger ecological study (72). UPPSALA UNIVERSITET 11 (58)
Potential reasons behind the co-existence of over- and undernutrition in the same household are closely related to those appearing at the individual level too. According to the DOHaD-theory, undernourished children are likely to develop low levels of lean body mass but high fat mass as adults (32). Coupled with increased availability of energy-dense, nutrient-low foods and sedentary lifestyles brought by nutrition transition, youth and adults will get fatter, but their children might miss-out on nutrients required for ensuring linear growth (21). Another factor of potential importance is the differential energy needs of children and adults with children requiring proportionally more energy than grown-ups (73). It has also been suggested that double burden at household level is a sign of an increasing individualization in diets and activity patterns among household members (72).
Cross-sectional studies have identified further risk factors for occurrence of SCOWT pairs such as higher number of siblings, lack of maternal education and old age of mother (74). However, as these are also risk factors for underweight in children and overweight in adults, it has been questioned whether they contribute to SCOWT as an independent phenomenon or whether SCOWT is simply a reflection of overall prevalence of under- and overnutrition in the society (74,75). Looking at expected count of undernourished child – overweight mother based on the prevalence of its two components in Demographic and Health Survey (DHS) data, two large studies found that SCOWT households are less common than expected purely on the basis of prevalence of childhood undernutrition and maternal overnutrition, and that it is especially the prevalence of maternal overweight that determines the prevalence of SCOWT households (74,75). These findings speak against need for developing interventions to tackle double burden households as an entity of its own rather than targeting the traditional underlying factors of overweight and undernutrition separately (75).
Two previous studies have examined household double burden in India. In 1998, the national prevalence of SCOWT was 1.2 % – one of the lowest among 42 countries with DHS data between 1992 to 2001 (72). A non-peer reviewed report on the Young Lives cohort study from Andhra Pradesh found SCOWT- prevalence in 2006 to be 1.0 % in rural and 3.4 % in the urban sample (76). Overweight mother – underweight child -pairs were slightly more common than SCOWT in the latter study (76).
Community Level Double burden of malnutrition is defined at community level as co-existence of high prevalence of stunting or underweight, and overweight (30,77). What exactly constitutes a high burden is less clear. Like UPPSALA UNIVERSITET 12 (58)
household double burden, community double burden is also considered to be a sign of nutrition transition at national or sub-national level (21). India is struggling with persistent high burden of childhood undernutrition with simultaneous increase in overweight and obesity among both adults and children (11,78,79). However, inter- and intra-state differences are large in the vast country (25,80–82). Among adults in India, a study using data from second National Family Health Survey (NFHS-2) found that at village and block level communities with high burden of under- and overweight women are to a great extent segregated from each other (83). This finding was later confirmed by another paper using similar methods with data from both NFHS-2 and 3 (84). It is interesting to observe that during the six years between rounds 2 and 3 of the NFHS, the negative correlation between underweight and overweight grew larger suggesting an increased separation of neighbourhoods with dominant burden of overweight from those with burden of underweight. Two further papers examining Indian women found that overweight and underweight are also highly segregated by income class, and that double burden does not occur to a significant extent in any of the socio-economic classes (85,86). Such findings have led to questioning the concept of double burden at community level (84).
Studies conducted among students of an urban Indian middle-class primary school (87) and adolescent children of urban industrial workers found some degree of co-existence of under- and overnutrition in these communities (88). Both anaemia and overweight were also commonly found among affluent school children in a review of Indian studies from early 2000’s (89). Notably though, none of the above papers looked at predictors of double burden per se and I was not able to find articles examining double burden at community level among younger Indian children.
Rationale for the Study, Research Questions and Objectives India is experiencing economic growth and urbanization (90) coupled with demographic (91), epidemiologic (25,92) and nutrition transitions (59). Double burden of malnutrition is expected to emerge in such transitioning societies. The concept of double burden of malnutrition helps to identify progress of nutrition transition in a country and to inform policy priorities. Tackling double burden promises double gains (27) and avoids the pitfall of decreasing one form of malnutrition at the expense of increasing another (72,73). The cost of current underinvestment in nutrition in India is dear as childhood malnutrition has implications to adult productivity and health with risk of NCDs increased both from over- (37) and undernutrition (93).
UPPSALA UNIVERSITET 13 (58)
Rising childhood overnutrition (12) and the persistence of undernutrition (4,36) have been consistent findings in Indian studies made during the last decades. Despite of this, research on double burden among pre-school children has been scarce. At the time of literature review in spring 2017, I was not able to find any papers with data on double burden among pre-school children during the last 10 years and thus a knowledge gap regarding the prevalence of double burden in this age group remains.
Previous studies have shown that household wealth is a major determinant of childhood undernutrition in India with gap in prevalence of stunting between richest and poorest wealth quintiles one of the largest in the world (94). This gap increased during the years of fast economic growth at the turn of the millennium when wealthier households experienced fast improvement in their living standard (95). It seems, however, that some degree of undernutrition persists even in the highest socio-economic classes, where prevalence of overweight has increased (11,12). This could hypothetically lead to double burden of malnutrition in children of wealthy households.
This study aims to examine what is the prevalence of double burden of malnutrition among children under five years of age in India at community, household and individual levels. Individual level double burden of malnutrition is defined in two ways: first, a child being simultaneously overweight and stunted and secondly overweight and anaemic. Household double burden is defined as a household with at least one overweight adult and at least one stunted child. Lastly, community level double burden is explored by finding out the prevalence of different forms of malnutrition among children under five years of age in various geographic locations and socio-economic strata.
UPPSALA UNIVERSITET 14 (58)
Methods The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines were used for writing this paper (96).
Study Design and Population This is a prevalence study that uses data from cross-sectional District Level Household and Facility Survey (DLHS) round 4 from 2012-2013. DLHS is a population based survey that was conducted in 26 non- Empowered Action Group (EAG) states and Union Territories by the International Institute for Population Sciences (IIPS) under the designation of The Ministry of Health and Family Welfare, Government of India (97,98). In difference to the NFHS, the DLHS is representative at district level. In addition to the usual household and individual level data on maternal and child health and nutrition, it also collected data on village features and health care facilities (97). Clinical, Anthropometric and Biochemistry (CAB) component was introduced to this round of the DLHS (98). Despite these benefits, the DLHS has been underutilized in research in comparison to the NFHS (99).
The DLHS-4 dataset available in public domain does not include the state of Jammu and Kashmir nor the Union Territories of Dadra and Nagar Haveli, and Lakshadweep (100). Nor are the eight EAG states and Assam included as they were covered by a different survey called the Annual Health Survey (101). The available dataset covers over half of the total Indian population (102) living in states with wide range of economic development (103). In terms of malnutrition estimates too, these states have had very diverse progress (80,82). However, it should be stressed that the survey is not nationally representative and that it excluded states deemed by the Indian Government to be the poorest performing in terms of developmental indicators (101). For a map of included states see Supplementary Figure 1, Annex.
DLHS-4 used weighted multi-stage stratified sampling to collect a statistically representative sample of population at district level. Sampling method is explained in detail elsewhere (97), but in short, in urban areas two or three-stage sampling was used with blocks as primary sampling units (PSU) and households as secondary sampling units (SSU). In rural areas villages were used as PSUs and households as SSUs. Sampling weights available in the dataset were used for analysis and the method of calculation is explained elsewhere (97). Data collection took place between December 2012 and March 2014 (100).
UPPSALA UNIVERSITET 15 (58)
Health facilities within the sampled PSUs were all included in the facility survey alongside all the community health centres, and sub-divisional and district hospitals in the chosen districts. Trained data collectors worked in a team of three female and male to visit all the sampled households. One supervisor and two investigators were tasked to collect the CAB information (97). This study utilized data from household questionnaire, which was completed by 391772 households and included the CAB measurements of 1593609 individuals.
Outcome Variables Outcome variables for different levels of double burden are presented in Table 1. Stunting was selected as the main proxy for undernutrition as it reflects long term undernutrition and is less prone to effects of acute illness than wasting, whereas underweight is considered a combination of the two and thus not a direct opposite of overweight in children (104). Anaemia was used as a proxy for micronutrient deficiencies due to the availability of data. Household double burden was defined as stunted child – overweight adult residing in same household as it was not possible to link mother-child pairs across the household and women’s questionnaire in the dataset.
Table 1: Definitions of Study Outcome Variables Level of Double Burden Examined Prevalence Definition of Outcome Variable (Outcome Variable) Individual Level Overweight and Stunted Child a Child with BAZ >2SD and HAZ <-2SD Overweight and Anaemic Child b Child with BAZ >2SD and Hb <11,0g/dl Household Level Stunted Child a and Overweight Adult c Child with HAZ <-2SD and in same Household adult with BMI ≥23.0kg/m2 Community Level Overweight Child a BAZ >2SD Stunted Child a HAZ <-2SD Wasted Child a WHZ <-2SD Underweight Child a WAZ <-2SD Anaemic Child b Hb <11,0g/dl a = age 30-1825 days; b = 180-1825 days; c = ≥18 years BAZ = Age-adjusted Body-Mass Index Z-score; BMI = Body-Mass Index; HAZ = Height for Age Z-score; Hb = Haemoglobin concentration in blood; SD = Standard Deviation; WHZ = Weight for Height Z-score; WAZ = Weight for Age Z-score Reference for all Z-scores is WHO Multicentre Growth Reference Study 2006 (105)
Stunting was defined as Height-for-Age Z-score (HAZ) <-2 Standard Deviations (SD) of the median of WHO growth chart and overweight as age-adjusted Body-Mass Index (BAZ) >+2SD (105). For overweight, the WHO recommended cut-off was chosen instead of the International Obesity Task Force UPPSALA UNIVERSITET 16 (58)
option (106) due to the former being currently recommended by the Indian Academy of Pediatrics (107). In order to enable global comparison, an alternative definition of overweight (Weight for Height Z-score [WHZ] >2 SD) was also calculated, but this was not used in further analyses. Also other definitions of undernutrition were explored, namely wasting (WHZ <-2SD), underweight (Weight-for-Age Z-score [WAZ] <-2SD) and Composite Index of Anthropometric Failure (CIAF) (all children with at least one of the three forms of anthropometric failure) (108). All the above indices were calculated for age group 30- 1825 days (1.0 to 59.9 months). To determine prevalence of anaemia, a slightly different age group was targeted due to data being available only for children aged 180 to 1825 days (6.0 to 59.9 months). The WHO-recommended cut-off of haemoglobin concentration in blood <11.0 g/dl was defined as anaemia (109). Lastly, for household double burden adult overweight was determined using both the cut-off of BMI ≥23.0 kg/m2 recommended for Asian populations and ≥25.0 kg/m2 for global comparability in age group 18 and above (110). As no major difference was seen in the results, reporting is based on the local reference of BMI ≥23.0 kg/m2. Exact details of the DLHS-4 anthropometric data collection and standardization have not been published.
The dataset used for this study (111) did not include any anthropometric indices so they were calculated by the author. Inclusion criteria for the anthropometric analysis was set as all children aged 30 to 1825 days with height or weight measured and for anaemia all children aged 180 to 1825 days with haemoglobin measured. For adults, all persons aged ≥18 years with valid measurement of height and weight were included. Cases with unknown age, implausible haemoglobin, height or weight or implausible anthropometric indices as per WHO definition (105) were excluded pairwise meaning that for each index a different number of cases was available.
Participant Selection Flowchart of participant selection is presented in Figure 2. The complete CAB dataset included 1610571 cases, out of which 1.1 % were removed as duplicates. Age of adults in years was taken straight from the dataset with 1117249 cases reportedly ≥18 years old. Out of adults, 80.6 % (900470) agreed to participate to height and weight measurements, out of whom 878857 had both measurements done. As there are no uniformly adopted data cleaning criteria for identifying valid adult measurements, I used arbitrary cut-offs of height <1.25m or >2.2m and weight <35kg or >250kg as biologically implausible and therefore deleted 2.7 % of the records. Further 187 records were deleted because of an unlikely BMI <13.5kg/m2 or >76.0 kg/m2 making total share of deleted cases 23.5 % of the original dataset and the number of adults with valid UPPSALA UNIVERSITET 17 (58)
measurement 855090. These adults came from 52565 households, where also at least one child had valid height measurement. This was the final analytic sample size for household level analysis.
For creation of anthropometric indices, children’s ages needed to be verified. The dataset included four methods that could be potentially used for this (Supplementary Table 1, Annex). Exact age of the child was calculated by deducting the date of interview from date of birth, which was missing for 12.1 % of cases eligible by crude age in years. Despite missing this sizable proportion of children, I chose to stick to this method as age is considered to be the most difficult of anthropometric variables to quantify (112).
Out of the cases with valid age for anthropometric measurement (30 to 1825 days, n=98812), 72516 agreed to be measured and 71505 had their height or length measured while 71615 were measured for weight. I entered the data of 72483 eligible children with either height or weight measured to WHO Anthro -program (113) for calculation of anthropometric indices and Z-scores. Seven cases were removed because of definition of sex as other. Further 437 cases were removed from height measurement because of being considered biologically implausible by the Anthro-program (<38.0cm or >150.0cm). Similarly, 332 children were removed from weight measurements because of too low or high value (<0.9kg or >58.0 kg). These removed cases are likely to represent errors in either measurement or data entry. Altogether 943 children were missing for either weight or height and thus were not included in the BAZ calculation.
To exert further control on the reliability of the results, WHO Anthro also uses cut-offs for flagging created anthropometric indices as biologically implausible according to WHO criteria from 2006 (105). International consensus on appropriate methodology is lacking (114), but I chose the WHO option due to their widespread use in previous nutrition studies in India (3,36,115–117) as well as globally in the DHS datasets (118). Supplementary Table 2, Annex shows the number of cases excluded by the chosen data cleaning method in comparison to another popular alternative (119). Final analytic sample size for individual level double burden analysis with HAZ was 66314 children and for BAZ 63155 cases. These correspond to 58.3 % and 55.7 % respectively of the original sample of 113483 individuals aged under 5 by the crude age.
UPPSALA UNIVERSITET 18 (58)
Figure 2: Flowchart of Participant Selection for Main Outcomes UPPSALA UNIVERSITET 19 (58)
For haemoglobin test, only children aged 180 to 1825 days were included (n=89174). Out of the children with appropriate age, 45643 (51.2 %) underwent the test. Further 137 cases were excluded for having biologically implausible haemoglobin values determined by WHO (<2.5g/dl or >20.0g/dl) (120), which brought the final analytic sample size to 45506 cases or only 40.1 % of all children aged under 5 by the reported crude age.
Estimation of Potential Bias in Outcome Variables Children included in the final analyses were significantly older than those with age in complete years (mean age in completed years 2.04 vs. 1.98 years, summary t-test p<0.001). This difference is probably mostly explained by the exclusion of children under 30 days of age from the anthropometric analysis group and those under 180 days from the haemoglobin group. The final analytical group was also slightly wealthier (mean wealth index score -0.023 [one-sample t-test 95 % CI -0.030…-0.016] vs. -0.063 [95 % CI -0.069…- 0.057]). No difference was seen in level of urbanicity (37.8 % [95 % CI 37.4-38.1]) vs. 38.1 % [95 % CI 37.8-38.3]). Clear digit preference was observed when checking the frequencies of reported weight and height measurements, which predisposes the study to systemic error (119). The resulting anthropometric measurements were all skewed to the left and had a platykurtic distribution in comparison to the standard distribution.
The dataset included additional information on the method of measuring tallness of the child, namely whether measure was taken standing or recumbent. This variable was missing for 3.3 % of cases and had generally a reasonable direction of length for younger and height for older children. The WHO Anthro program, however, overrode this variable with an age-based definition and presumed all children under 24 months to be measured recumbent as is generally guided (113). This caused discrepancy in 7084 (9.8 %) of records. As it is not possible to know which proportion of children were measured and recorded correctly, I settled with the automatic definition. The difference in tallness between standing and lying child is about 0.7 cm and this might have impacted the final outcome of the study by causing random error (105).
The distribution of anthropometric indices before data cleaning are presented in Supplementary Table 3, Annex and Supplementary Figure 2, Annex. They show that the sample is skewed to left with mean Z- score for BAZ -0.7 and for HAZ -1.0. Additionally, the sample distribution is spread out as exemplified by large SDs ranging from of 3.6 in both indices used for primary analysis. SD of Z-scores has been suggested to be a marker of data quality in nutrition surveys (121) and in a recent article Grellety and UPPSALA UNIVERSITET 20 (58)
Golden proposed that surveys with standard deviations above 1.2 should be rejected due to high likelihood of systemic and random errors (112). Measurement error and digit preference are some of the errors that increase the SD (112), but it can be also influenced by true heterogeneity in the population (118) , which is expected to be great in such diverse country as India.
Comparing the undernutrition prevalence figures presented in this paper with those available in DLHS-4 state factsheets shows that my estimates are generally higher presumably because the state reports seem to have been generated by total number of children under five including those with no measurements done as the denominator (100).
Explanatory Variables Individual Level Double Burden Selection of variables was guided by the aim of localizing the potential double burden in India rather than exhaustingly identifying its underlying causes. Age of child was considered as a potential explanatory variable with divisions 1.0 to 5.9 months, 6.0 to 23.9 months and 24.0 to 59.9 months due to differential patterns of malnutrition in these age groups (122). Sex was treated as a dichotomous variable male/female. States were divided into 6 geographical zones as per administrative definition (123) that approximately follow cultural division within the country (See Supplementary Figure 1, Annex) (124). None of the states in the Central Zone were part of the dataset cutting down the number of zones into 5. Andaman and Nicobar was included in the Southern Zone and Sikkim in the North-Eastern Zone (123). Rural-urban residence is determined in the DLHS dataset according to Census definitions (125). Wealth quintiles were calculated from household assets as explained in separate section. Residence in a household with an overweight adult was calculated as explained in the outcome variable section for household double burden. Residence in household with underweight adult was done similarly with BMI cut-off of <18.5kg/m2. For adult anaemia, haemoglobin was measured from 804880 adults, out of whom 4265 cases were removed for implausible values (120).
Household Double Burden In addition to the factors named above, some household characteristics that were thought to potentially distinguish double burden households from others were considered. Number of household members is an obvious confounder in this kind of analysis and was thus considered as a scale variable. DLHS classifies household head as Scheduled Cast (SC)/Scheduled Tribe (ST)/Other Backward Classes (OBC)/None UPPSALA UNIVERSITET 21 (58)
according to the Indian Constitution (126). This stratification has been questioned but is commonly used in research (127). Household cooking facilities were combined into a variable with three categories similar to previous studies on the sub-continent: no use of solid fuels; use of solid fuels in separate kitchen and solid fuels without a separate kitchen (128). Access to drinking water is defined as a binary variable according to WHO and UNICEF Joint Monitoring Programme (JMP) classification (Supplementary Table 4, Annex) (129). In case of bottled water, no information on other sources of potable water was available and these 1191 (1.6 %) cases were set as missing. Because of strong evidence pointing out that also so called improved sources of water can be contaminated (130), water treatment by electronic purification, filter, boiling, or using alum or chlorine was also considered as a variable. For sanitation, binary JMP categories (Supplementary Table 4, Annex) were used. Safe disposal of infant faeces was defined as disposal to toilet (129).
Availability of health services in the village of the child was asked in DLHS village dataset and defined three ways: availability of Integrated Child Development Services (ICDS), of public health provider and of private provider. Additionally, the existence of Village Health, Sanitation and Nutrition Committee was considered (131).
Wealth Index The DLHS-4 dataset used in this study (111) did not include a ready measure of household wealth and it was therefore necessary to create one for comparing the children from rich and poor households. No data on household income were collected in the survey, but instead the DLHS-4 has a range of household assets that are commonly used as a proxy for household wealth (132). Several methods have been used to apply weights for these assets to derive a wealth index score for each household (133). Principal Component Analysis (PCA) was first proposed by Filmer and Pritchett in India in the turn of the Millennium (134) and since then it has been adopted as method of analysis both for DHS and UNICEF Multiple Indicator Cluster Surveys among others (135). It is generally thought to represent long-term accumulated wealth rather than current income (133). Instructions for generating the index from assets commonly encountered in the DHS and other surveys have been published by Vyas and Kumaranayake (136) as well as Rutstein (137).
Selection of Variables for the Wealth Index There is no single optimal list of indicators to be selected for wealth index and indeed some authors emphasize the need for local adaptation of the set of indicator variables (133). As an overall rule, indicators that have a direct relationship with the main outcome of interest of the study should not be used in UPPSALA UNIVERSITET 22 (58)
constructing the wealth index (136). Filmer and Pritchett used water and sanitation of the household as an indicator of wealth in India (134), but because of their association with undernutrition in cross-sectional studies (3,138,139), I did not include them in my calculations. In 2014 Bassani et.al. published household wealth indices for each round of DLHS and NFHS done prior to DLHS-4 and chose different assets for each round depending on data availability (140). They also strived to choose as few variables as possible to keep the index simple (140) whereas Rutstein and Johnson recommend to use as many variables as possible to decrease the risk of clumping of households into same few scores (132).
It is generally recommended to convert qualitative categorical variables into binary form as it is often difficult to assign categories a clear order in relation to wealth (132,136). Housing type was therefore made into three separate variables: pucca, semi-pucca or kaccha with pucca meaning complete permanent housing and kaccha indicating a house made of rudimentary materials. Similarly, I converted the main type of lighting in the household to electricity or no electricity as the combined prevalence of other forms of lighting was meagre 4.2 %. Supplementary Table 5, Annex presents the prevalence of potential variables disaggregated by rural and urban location of the households.
Ownership of Below Poverty Line card was also asked in the questionnaire, but I chose not to include it as it is a construct of other household assets. Ownership of computer with and without internet was combined into one variable as the frequencies for both were quite low (5.7 % of all households) and the distinction between them unclear. Similarly, three types of carts were collated into variable with overall prevalence of 3.1 %. For scale variables, number of persons per room was created and land owned in acres included.
No missing values were reported in any of the household assets. In other variables, their share was negligible (< 1 %) in all states except in building material of house in Mizoram, where 10.5 % of households were categorized to belong to type other. As no further information on the qualities of these households was available, the “other” was re-classified as missing. This small number of households (0.002 % of total) should not make a difference to the overall wealth index.
Raw number of households and rural-urban location were found to match the available state reports (100). State weights included in the DLHS-4 household dataset were applied for the household features selected as wealth index variables and the weighted prevalence were found to match the state reports within one percentage point error of margin except in Arunachal Pradesh, where 207 households had no state weights in the original dataset. Also in Daman and Diu, 629 households missed weights, but as no state factsheets UPPSALA UNIVERSITET 23 (58)
were available for Union Territories for verification, I could not check their correctness. As the total number of households affected is small (0.002 %), this should not make a significant difference to the overall wealth index. Additionally, variable on household ownership and the two scale variables created by myself were not included in the state reports and thus could not be verified.
Performing Factor Analysis PCA was done to unweighted cases as suggested by the two instructions (136,137). As the data were not standardized, correlation matrix for components with Eigenvalue >1 was chosen (136,137). Missing values were replaced by mean of all values as per recommendations (136,137). Considering the low number of missing values this should not affect the outcome. For details on the mathematical calculations involved in PCA, please see Filmer and Prichett’s article (134).
All analyses were first performed to the complete dataset and then separately to rural and urban households as proposed by both Rutstein (137) and Vyas and Kumaranayake (136). The reason that for example DHS does separate indices for rural and urban areas is the way that some service-related variables like availability of water and sanitation facilities might bias the results towards urban areas and fail to differentiate poor rural households from one another (135). However, only such service related variable selected for my wealth index was electricity as the main source of lighting. Additionally, I found that there was generally little difference between the levels of asset ownership in rural and urban areas (Supplementary Table 5, Annex) and that the household wealth index scores generated by common list of variables had very high agreement with those created by separate lists of variables (Pearson’s correlation co-efficient >0.95 and p<0.05 for all comparisons). I therefore chose to keep a common list of indicators for both rural and urban areas.
The first component of PCA for all the households explained 19.3 % of variation in wealth. While this is a low number, it is par with previous studies and including of additional components is generally not recommended as their association with wealth is unclear (136). Supplementary Table 6, Annex, shows the included variables and their weightings. All the directions of the associations were logical: owning more land, a house or any of the assets increased wealth whereas more people per room and living in semi-pucca or kaccha house all had negative associations. Range of wealth scores was 7.3 with standard deviation of 1.0 and mean of 0.0 because of the use of correlation matrix. Applying PCA to the same list of variables in urban and rural areas only explained 18.5 % and 18.4 % of variation in wealth respectively. All directions of associations were logical and weightings similar to the common index (data not shown). UPPSALA UNIVERSITET 24 (58)
Supplementary Figure 3, Annex shows three histograms of households’ wealth scores applied to all households followed by urban and rural households separately. The common index scores are skewed to the poorer households on the left and has some evidence of clumping, but is nonetheless seemingly able to differentiate households from each other on a wide continuum. This skewing to the left is mainly explained by the rural households, whereas the distribution of the urban households’ wealth index scores follows approximately the normal curve.
Despite choosing separate variables for rural and urban areas, Rutstein advocates for creating a final composite wealth index score that makes rural and urban areas comparable (135,137). I followed the instructions provided (137), compared the results between adjusted and unadjusted scores and ended up using the unadjusted wealth quintiles as I felt they were easier to understand and fairer to the true observed division of assets. The whole process of adjustment is explained in detail in Annex.
In summary, I have created household wealth quintiles mainly following the same principles that are used by the DHS (137) and thus also the Indian NFHS. There are two major exceptions to the rules proposed by Rutstein (137). I did not adjust for household size as this was advised against as potentially misleading by two other instructions (133,136). Nor did I adjust for rural-urban differences as the quintile divisions produced by Rutstein’s method did not seem plausible.
Statistical Methods IBM SPSS version 24 was used for data analysis (141). Descriptive statics were used to define the prevalence of different forms of double burden across demographic strata. In univariate analysis, Wilson’s procedure without continuity correction was used to calculate the 95 % Confidence Intervals (CI) for the proportions derived (142). For bivariate analysis, Pearson’s chi-square test was used for categorical data and independent samples t-test for continuous variables.
To compare prevalences between different groups, prevalence ratio (PR) was calculated as (143,144): PR = (a/(a+b))/(c/c+d)
Based on the individual components of double burden, expected prevalence of individual double burden was calculated as: P (%) = (n children stunted * n children overweight)/(N total)2 * 100 UPPSALA UNIVERSITET 25 (58)
A similar calculation was done for double burden of anaemia and overweight.
Expected prevalence of household double burden was calculated as: P (%) = (n household with stunted child * n household with overweight adult)/(N total)2 * 100
State household weights were applied to analysis to account for the multi-phase sampling method (97).
Ethical Approval Participating in the DLHS-4 survey was voluntary and no reward was given to participants. All DLHS-4 questionnaires begin with obtaining informed participant consent. All information given was confidential and participants had the right to withdraw at any time (145). This study is based on data available in public domain and all identifiable information has been removed (111). Institutional ethical approval for this study was waived by as it uses secondary data available in the public domain.
UPPSALA UNIVERSITET 26 (58)
Results
Details of the participant selection are presented in methods section and in Figure 2. The dataset included 113483 children aged under 5 years. A valid BMI measurement was available for 63155 children and a valid HAZ score for 66134 children aged 1.0 to 59.9 months. Haemoglobin was measured from 45506 children aged 6.0 to 59.9 months. These children came from 52565 households where at least one adult had a valid BMI measurement. The most common reason for not being included in the study was refusal to participate in measurement followed by lack of precise date of birth for calculating a correct age.
Prevalence of Community Level Double Burden of Childhood Malnutrition In the 23 states included in the analysis, childhood undernutrition was far more prevalent than overnutrition (Table 2). Proportion of children stunted was 40.1 % followed by underweight (31.3 %) and wasting (20.4 %). Combined Index of Anthropometric Failure showed that 59.0 % of children had at least one of these three forms of growth faltering. Additionally, 74.2 % of children with measured haemoglobin were anaemic. Overweight on the other hand affected 8.2 % of the children in the sample when defined as BAZ >2 SD and 5.4 % when defined as WHZ >2 SD.
Results of bivariate analysis are presented in Table 2. Disaggregation be gender showed that male children were more commonly affected by all forms of malnutrition except anaemia. All forms of undernutrition were more commonly encountered in the rural areas whereas overweight had higher prevalence in the cities. Region with the highest prevalence of all forms of undernutrition was the Eastern Zone. The lowest prevalence of stunting was found in West (37.1 %) while anaemia was least common in North (70.3 %). North-East had the lowest figures for both underweight (23.9 %) and wasting. Overweight was the biggest problem in North-East (10.0 %) and of least concern in West and North (7.2 % and 7.3 % respectively). There was a wide variation in malnutrition across states too distribution shown in Supplementary Table 7, Annex. Out of the 23 states examined, the highest prevalence of overweight was seen in Andaman and Nicobar (13.8 %) and the lowest in Tamil Nadu (5.7 %). Stunting on the other hand was most common in Meghalaya (58.7 %) and lowest in Goa (22.5 %) while anaemia prevalence ranged between 88.7 % in Sikkim and 54.4 % in Kerala. All forms of malnutrition were most common in age group 6.0 to 23.9 months except for stunting, where a slight additional increase was seen in age group 24.0 to 59.9 months. All forms of malnutrition had a logical direction in relation to household wealth with level of all types of undernutrition steeply decreasing in wealthier households. However, even in the wealthiest quintile 31.8 % of children were still stunted and 69.1 % anaemic. Overweight on the other hand increased as wealth UPPSALA UNIVERSITET 27 (58)
grew, but the observed gradient was not as steep as with undernutrition and showed only one percentage point difference between the poorest and the richest quintiles (7.9 % vs. 8.9 %). Table 2 shows the results for each form of malnutrition except wasting, which had similar patterns to those of underweight.
In bivariate model, prevalence of undernutrition in children was slightly higher in households with an underweight adult and lower in a household with an overweight adult. Correspondingly, overweight was more common among children who lived with at least one overweight adult (PR 1.19 [95 % CI 1.12-1.26]) and less common in those living with an underweight adult (PR 0.83 [95 % CI 0.79-0.88]). Children living with an anaemic adult had anaemia about 40% more often than those living without an anaemic adult in the same household (PR 1.41 [95 % CI 1.38-1.44]).
Table 2: Prevalence and Predictors of Different forms of Malnutrition Over- Under- OSC χ²-test OAC χ²-test χ²-test Stunted χ²-test Anaemic χ²-test χ²-test weight weight N=61328 p-value N=40295 p-value p-value N=66134 p-value N=45506 p-value p-value N=63155 N=70223 Total % 5.4 5.5 8.2 40.1 74.2 31.3 Male 5.9 6.0 8.9 41.6 74.1 32.2 Sex, % Female 4.8 <0.001 5.0 <0.001 7.4 <0.001 38.6 <0.001 74.4 0.353 30.4 <0.001 0 to 5.9m 4.7 5.2 8.6 19.4 not applicable 24.3 Age, % 6 to 23.9m 6.8 7.7 10.6 38.8 78.3 27.5 ≥24m 4.7 <0.001 4.7 <0.001 7.1 <0.001 42.9 <0.001 73.0 <0.001 34.0 <0.001 Rural 5.3 5.5 7.9 42.5 75.6 33.9 Location, % Urban 5.5 0.191 5.6 0.502 8.6 0.001 36.3 <0.001 71.9 <0.001 27.1 <0.001 North 4.9 5.0 7.3 41.2 70.3 31.3 Nort-East 6.6 6.3 10.0 44.4 72.3 23.9 Zone, % East 6.2 7.0 8.6 45.1 87.9 38.0 West 4.4 5.2 7.2 37.1 82.0 37.6 South 5.3 <0.001 5.2 <0.001 8.2 <0.001 37.7 <0.001 72.9 <0.001 30.8 <0.001 1st 5.2 5.7 7.9 48.1 79.1 41.5 2nd 5.4 5.2 7.9 44.3 75.9 35.6 Wealth Quintile, % 3rd 5.4 5.6 8.0 39.9 74.7 30.8 4th 5.2 5.2 8.2 36.0 71.8 26.2 5th 5.5 0.791 6.0 0.121 8.9 0.017 31.8 <0.001 69.1 <0.001 21.6 <0.001 Over- Under- OSC PR (95% CI) OAC PR (95% CI) PR (95% CI) Stunted PR (95% CI) Anaemic PR (95% CI) PR (95% CI) Adults in Household weight weight None 5.0 1.12 5.2 1.01 7.3 1.19 44.9 0.83 77.0 0.94 38.7 0.69 Overweight BMI ≥23.0, % ≥1 5.6 (1.04-1.20) 5.7 (1.00-1.19) 8.7 (1.12-1.26) 37.1 (0.81-0.84) 72.5 (0.93-0.95) 26.6 (0.67-0.70) None 5.6 0.87 5.7 0.91 8.7 0.83 39.3 1.05 72.2 1.08 27.8 1.34 Underweight <18.5, % ≥1 4.9 (0.81-0.93) 5.2 (0.83-0.99) 7.2 (0.79-0.88) 41.3 (1.03-1.07) 77.9 (1.07-1.09) 37.3 (1.31-1.37) None 5.4 1.00 4.0 1.50 8.0 1.03 40.9 0.98 56.4 1.41 30.8 1.02 Anaemia Hb > 11.0g/dl, % ≥1 5.4 (0.92-1.08) 6.0 (1.34-1.67) 8.2 (0.97-1.10) 39.9 (0.95-1.00) 79.5 (1.38-1.44) 31.4 (0.99-1.05) OSC = Overweight and Stunted Child, OAC = Overweight and Anaemic Child, PR = Prevalence Ratio for Outcome ≥1 (reference category = "None") 1st wealth quintile = poorest Cases are weighted by state household weight
Prevalence of Individual Level Double Burden of Childhood Malnutrition Individual level double burden, defined as co-existence of overweight and stunting in same child, was encountered in 5.4 % (95 % CI 5.2-5.5) of the children aged under 5 years in the sample. It was more common than expected purely by the prevalence estimates of its individual components (expected prevalence 3.1 % [95 % CI 2.9-3.1]). Bivariate analysis (Table 3) showed that stunted children were almost UPPSALA UNIVERSITET 28 (58)
four times more likely to be overweight than non-stunted children (12.8 % vs. 3.3 %, PR 3.92 [95 % CI 3.68-4.19]). Among overweight children, stunting was found in 73.8 % of cases, which was almost double the prevalence of stunting among non-overweight (Prevalence Ratio 1.88 [95 % CI 1.84-1.92]). If OSC were dealt as a separate entity and removed from the prevalence estimates of overweight and stunting, the prevalence of stunting dropped to 35.2 % and prevalence of overweight to 3.0 %.
Table 3: Prevalence Ratios for Association between Different Forms of Malnutrition
Overweight Stunted Anaemic No 39.3 74.3 Overweight, % Yes Not applicable 73.8 73.5 Prevalence Ratio (95% CI) 1.88 (1.84-1.92) 0.97 (0.91-1.03) No 3.3 73.5 Stunted, % Yes 12.8 Not applicable 75.4 Prevalence Ratio (95% CI) 3.92 (3.68-4.19) 1.08 (1.05-1.12) No 7.7 39.6 Anaemic, % Yes 7.4 42.0 Not applicable Prevalence Ratio (95% CI) 0.97 (0.90-1.04) 1.06 (1.04-1.09) Reference category for PR is "No"
Bivariate associations also showed that OSC was more common among males than females (Table 2). Age- wise OSC peaked at 6.8 % among children aged 6 to 23.9 months whereas both the younger and older age group had prevalence of 4.7 %. OSC seemed to occur most commonly in North-Eastern part of the country (6.6 %), whereas the Western Zone had the lowest prevalence (4.4 %). Meghalaya had the highest state prevalence of OSC (9.2 %) and Goa the lowest (2.9 %) (Supplementary Table 7, Annex). No difference was seen between rural and urban areas or across household wealth quintiles. Living in a household with an overweight adult slightly increased the prevalence for OSC (PR 1.12 [95 % CI 1.04-1.20) whereas children residing in a household with an underweight adult were less commonly OSC (PR 0.87 [95 % CI 0.81-0.93]).
When individual double burden was defined as simultaneous overweight and anaemia in a child (OAC), its prevalence was 5.5 % (95 % CI 5.3-5.8). Out of all anaemic children (74.2 %), only 7.4 % were also overweight, whereas among the 8.2% of children that were overweight and had haemoglobin measured, 73.5 % were also anaemic. These figures are similar to the overall prevalence estimates of the two conditions and thus, unlike with overweight and stunting, anaemia and overweight were not associated with each other in a bivariate model. When OAC were removed from the proportion of overweight and anaemic, the prevalence estimates of the two dropped to 4.6 % and 69.3 % respectively. UPPSALA UNIVERSITET 29 (58)
Males were more commonly affected by OAC than females and the age group with highest prevalence was 6 to 23.9 months. OAC showed evidence of regional patterning with East having the highest prevalence (7.0 %) and North the lowest (5.0 %). Out of all states, Andaman and Nicobar had the highest (10.1 %) and Himachal Pradesh the lowest prevalence (2.3 %) (Supplementary Table 7, Annex). No significant difference was seen between rural and urban areas or wealth quintiles in a chi-square test. OAC children were about 50 % more common in households with an anaemic adult than in those without one [PR 95 % CI 1.34-1.67] whereas adult underweight or overweight did not have a significant effect to the OAC prevalence.
Prevalence of Household Double Burden Household double burden of malnutrition was defined as a household with a stunted child and an overweight adult (SCOA). Prevalence of such households in the sample was 25.6 % (95 % CI 25.2-25.9). Notably, only 18.9 % of households had neither a stunted child nor an overweight adult. Expected prevalence of SCOA-households based on its components was 27.5 % (95 % CI 27.1-27.9). This is statistically significantly higher than the observed prevalence implying a separation of households with stunted children from those with overweight adults. Another type of household double burden structure with overweight child and underweight adult was also examined, but this had a low prevalence of 3.2 %.
In comparison to households without both a stunted child and an overweight adult, SCOA-households were significantly more likely to be found in urban than rural areas (Pearson’s chi-square p<0.001) (Table 4). Northern Zone had by far the highest prevalence of SCOA-households (30.6 %) whereas least common this phenomenon was in Eastern Zone (17.5 %). Household wealth had strong effect on the prevalence of household double burden with lowest prevalence in the poorest quintile 18.7 % and highest in the richest 29.4 % (Pearson’s chi-square p<0.001). Fitting to this wealth distribution, in a bivariate analysis SCOA- households were statistically significantly more likely to have access to safe water and improved toilet facilities (Supplementary Table 8, Annex). They would also more often practice safe disposal of infant faeces and cook without solid fuels. Anganwadi, VHNSC and health providers were more commonly encountered in the villages of SCOA than non-SCOA-households. Important confounder for these results was the fact that SCOA-households were also likely to have more members than non-SCOA-households (mean number of members 6.5 vs. 5.8, independent samples t-test p<0.001). UPPSALA UNIVERSITET 30 (58)
Table 4: Distribution of Double Burden Households
SCOA χ²-test n=13430 p-value Rural 26.8 Location, % Urban 24.8 <0.001 North 30.6 Nort-East 22.5 Zone, % East 17.5 West 22.0 South 26.8 <0.001 1st 18.7 2nd 23.7 Wealth 3rd 27.5 Quintile, % 4th 28.2 5th 29.4 <0.001 SCOA = Stunted Child - Overweight Adult household Cases are weighted by state household weight
UPPSALA UNIVERSITET 31 (58)
Discussion This study looked at the phenomenon of double burden of malnutrition among children aged under 5 years using cross-sectional data from, at the time of writing, the latest representative survey of 23 states and 284 districts in India. The findings indicate that double burden of malnutrition in age group below 5 years is becoming a real issue in the country. Undernutrition remains the most urgent concern with Composite Index of Anthropometric Failure showing that 59.0 % of children were either stunted, wasted or underweight. Anaemia affected 74.2 % of children measured. However, the prevalence of overweight was also quite high 8.2 %. Analysis on individual level double burden showed that 5.4 % of children were simultaneously overweight and stunted while 5.5 % were both overweight and anaemic. At household level, 25.6 % of households had at least one stunted child and one overweight adult living under the same roof.
It is not possible to precisely define community level double burden using the method of prevalence only as a commonly agreed definition of what constitutes a high burden of over- and undernutrition is lacking (84). However, the results of this study show that Indian children have more than double the global prevalence of different forms of undernutrition and approximately global average of overweight (31). As such it should be fair to say that India is facing a double burden of childhood malnutrition at the national level.
Previous surveys from India have found prevalence of overweight in this age group to lay between 1-3 % (16,146), which would indicate a significant rise in this study, although the methodology of defining overweight differs between the studies. I also explored an alternative definition of overweight (WHZ >2 SD) and at 5.4 % found it to be above the previous figures. The level of undernutrition, namely stunting (40.1 %), wasting (20.4 %) and underweight (31.3 %), corresponds with a slow decline seen in the NFHS studies over the last decades (16,78) but remains alarmingly high. Zone and state level analysis shows that in no part of the country has the burden of overnutrition taken over that of undernutrition despite the dataset including the wealthiest part of the country.
Disaggregating the results further, all forms of undernutrition decreased steeply by wealth quintiles, but the prevalence of overweight increased only slightly as wealth grew. This moderate increase goes against my original hypothesis that double burden would be mainly and issue for the wealthier households. The finding is also in disagreement with results presented in previous reviews on childhood overnutrition in UPPSALA UNIVERSITET 32 (58)
India (12,147), but matches with the global situation (26) and fits with the observation that as nutrition transition progresses, the burden of overweight starts to shift to lower socio-economic classes who have less access to healthier food choices (22,148). Many of the societal factors that currently lead to undernutrition in children might cause overnutrition under improved conditions of household food security (149). For example, the Government of India has been subsidising sugar, oil and cereals to poor families via the Public Distribution Scheme (150) and while this program is needed to ensure food security, its current formulation is not adept for tackling the issue of double burden of malnutrition.
Another major government scheme, the ICDS has taken big steps towards a more comprehensive approach to nutrition and child health over its 40 years of existence, but faces shortages in both human and material resources that affect the quality of services provided (151). Should the ICDS function the way its intended to and provide growth monitoring and supplementary nutrition, health check-ups and immunization as well as pre-school and nutrition education, it could be an effective tool for tackling the double burden, but the way the programme is currently set-up is very much concentrated around undernutrition and in no way acknowledges the rising issue of overnutrition and double burden (152). In this study, 99 % of households lived in villages with Anganwadi centres that provide the ICDS services and in bivariate analysis having access to Anganwadi was associated with higher prevalence of household double burden (Supplementary Table 8, Annex). While few conclusions can be drawn from this simple analysis, it seems obvious that the ICDS has not been able to achieve its target of undernutrition removal and is now facing an altogether new challenge in the form overnutrition that calls for change in approach.
A staggering 74.2 % of children with measured haemoglobin in this study were anaemic and anaemia was equally common among overweight as non-overweight children. This finding is in agreement with previous study of double burden among South Indian adults and reminds us that positive energy balance is not a guarantee of adequate micronutrient utilization (54). In comparison to recent NHFS-surveys, the observed prevalence of anaemia was approximately five percent points higher than in NFHS-3 from 2005- 2006 and goes against the positive progress seen in the most recent NFHS-4 from 2015-2016, where the national childhood anaemia prevalence was estimated to be 58.4 % (78). In this study, household wealth had only a modest effect on the prevalence of anaemia with 69.1 % of children from wealthiest households being anaemic. So far the main target of anaemia reduction policy in India has been iron supplementation, but only about half of the burden is in fact caused by iron deficiency while the rest is explained by other nutritional deficiencies, congenital conditions, infections and inflammation among others (153). The life- time costs of anaemia in this age group in India have been estimated to reach 8.3 Million Disability UPPSALA UNIVERSITET 33 (58)
Adjusted Life Years and 24 Billion United States Dollars (154) and there is an urgent need for action to comprehensively tackle the underlying causes of anaemia in India.
Combination of overweight and stunting in same child was affecting 5.4 % of the sample. The only previous report with data on this type of double burden in India showed a prevalence of 1.0 % in the NFHS-3 data from 2005-2006 (49). In difference to this study, that paper used WHZ-based definition of overweight which gives a lower prevalence estimate. Nonetheless, the increase seems substantial. In comparison to 77 LMICs, the prevalence observed in this study is above 75th percentile though the same caveat for defining overweight remains (49). Overwhelming majority (73.9 %) of overweight children were also stunted, and the prevalence of OSC was higher than expected purely by the prevalence of its components. Bivariate analysis showed that overweight children were more likely to be stunted than non- overweight children were and vice versa. Perhaps because overweight was more common in urban areas and stunting in rural, no rural-urban difference was seen in OSC. Nor was there a clear household wealth gradient in prevalence of OSC, which again suggests that double burden is not an issue limited to only certain pockets of the society. However, no multivariate analysis was done to account for potential confounders.
Further studies should be undertaken first to verify whether OSC indeed is its own entity with predictive factors separate from those of its components. Studies elsewhere in LMICs have shown that reduced fat oxidation and lower insulin-like growth factor might be some of the physiological mechanisms that predispose stunted children to overweight (47,48), but these findings are so far inconclusive (46). Second task would be to find out whether the prognosis of OSC children differs from that of peers that are only stunted or overweight, but this objective is complicated by the challenge of defining OSC in a meaningful manner. In this study, the high co-occurrence of stunting and overweight is at least partly explained by the fact that I chose to use BMI to define overweight, which takes height to second power (BMI = weight/height²) and thus increases the effect of shortness on measure of overnutrition. Considering that the linear relationship between short stature and risk for poor health makes cut-offs for stunting arbitrary (30), and that the debate on the definition of childhood obesity is ongoing especially in reference to Asian children (155), any longitudinal study would surely find it difficult to distinguish the effects of combined stunting and overweight from those brought by growth faltering and changes in body composition alone.
Household level double burden, defined as at least one stunted child and one overweight adult, was encountered in around one fourth of households of the study sample. Wealth gradient in SCOA-households UPPSALA UNIVERSITET 34 (58)
indicated that, surprisingly, the households most burdened by this type of double malnutrition were poor rather than rich. As the differences between SCOA and non-SCOA households was small and no multivariate analysis was done, these results should be considered inconclusive. Previous studies in India have only examined double burden in mother-child pairs so direct comparison is not possible, but as earlier results have put the estimates at around one percent (72,76), it could be that household double burden is also increasing. Similar to other researchers (74,75), I also found that the observed count of double burden households was lower than could have been expected purely by its components indicating that households with at least one overweight adult are to some degree segregated from those with a stunted child. This finding argues against the need of own separate interventions for double burden households. Considering the DOHaD-theory and life-cycle perspective, interventions like exclusive breastfeeding and improving nutrition of women in fertile age that reduce childhood stunting should also decrease adult overweight and thus cut down household double burden (33).
Limitations This study has several significant limitations. Firstly, the DLHS-4 dataset used for analysis (111) is not nationally representative and in fact it excludes many of the poorest performing states in India leading to impression that nutrition transition among Indian children would have progressed further than it really has when it comes to national level. The results are thus not generalizable to India as whole.
Furthermore, anthropometric data in DLHS-4 showed clear signs of digit preference and distorted sample distribution, which implies low data quality considered even non-usable by some authors in the field (112). While standard data cleaning methods were applied, this form of random error is likely to decrease the internal validity of the results. Data cleaning combined with refusals and missing values especially in haemoglobin measures made the final analytical sample slightly older and wealthier than the original sample, which is another potential source of bias, but as the difference between the groups was small, this is not expected to have impacted the outcome in a systematic manner. Seemingly in difference to published state reports (100), prevalence estimates presented in this paper were generated using only measured and cleaned cases as denominator and thus my figures are higher than those officially available. At the same time, apart from anaemia the prevalence estimates match well with the latest national survey data available.
This study was cross-sectional in nature and thus the progression of different forms of malnutrition could not be studied. Therefore, it is not possible to infer causal links between different forms of malnutrition for UPPSALA UNIVERSITET 35 (58)
example in the case of stunted children seeming to be prone to overweight. Additionally, analysis about underlying predictors of double burden was limited as it was not possible to link children across household and women’s datasets, which meant that maternal characteristics and feeding habits of the child could not be examined as potential explanatory factors.
As the wealth quintiles used in the study were not adjusted for states or urbanicity, it could be that children living in rural areas and poorer states ended up in the lowest wealth quintile despite being from households relatively wealthy to their local peers. Similarly, children from urban households in rich states might have been categorised as rich despite having low living standard in comparison to their neighbours and perhaps disposable income that is too low for providing adequate diet. This discrepancy could explain the unexpectedly small wealth and urbanicity gradient in prevalence of overweight, OSC and OAC children.
Conclusion Nutrition transition in India is increasingly affecting also children and this study is first of its kind to explicitly look at double burden of malnutrition in age group below 5 years. The results show that while undernutrition remains the key issue in the wealthier states of the country, overnutrition is also increasing and both are not uncommonly found in the same child or same household. Alarmingly, the issue of overnutrition and double burden does not seem to be limited to the wealthiest households but is pervading through all the income groups as predicted by the nutrition transition theory. Due to limitations in data quality and analytical methods, the findings of this study should be considered preliminary and require confirmation by more research. Further studies should also be undertaken to find out what lays behind the circa 5% of children affected by simultaneous overweight and stunting as well as overweight and anaemia, and what is their prognosis. In the meanwhile, nutrition policy planners in India should ensure that efforts to decrease childhood undernutrition do not increase the burden of overweight by including the concept of double burden of malnutrition in the programs aiming to improve childhood nutrition status in the country.
I declare no conflict of interest.
UPPSALA UNIVERSITET 36 (58)
References 1. Ramalingaswami V, Jonson U, Rohde J. The Asian Enigma. In: The progress of nations [Internet]. New York: UNICEF; 1996. Available from: http://www.popline.org/node/307827
2. Haddad L, Nisbett N, Barnett I, Valli E. Maharashtra’s Child Stunting Declines: What is Driving Them? Findings of a Multidisciplinary Analysis [Internet]. 2014 [cited 2016 Oct 12]. Available from: https://opendocs.ids.ac.uk/opendocs/handle/123456789/4254
3. Corsi DJ, Mejía-Guevara I, Subramanian SV. Risk factors for chronic undernutrition among children in India: Estimating relative importance, population attributable risk and fractions. Soc Sci Med 1982. 2016 May;157:165–85.
4. Fenske N, Burns J, Hothorn T, Rehfuess EA. Understanding Child Stunting in India: A Comprehensive Analysis of Socio-Economic, Nutritional and Environmental Determinants Using Additive Quantile Regression. PLOS ONE. 2013 Nov 4;8(11):e78692.
5. Shekar M, Heaver R, Lee Y-K. Repositioning nutrition as central to development: A strategy for large scale action. World Bank Publications; 2006.
6. Headey D, Hoddinott J, Ali D, Tesfaye R, Dereje M. The Other Asian Enigma: Explaining the Rapid Reduction of Undernutrition in Bangladesh. World Dev. 2015 Feb;66:749–61.
7. Requejo J, Victora C, Bryce J. Countdown to 2015 Final Report [Internet]. UNICEF and World Health Organization; [cited 2016 May 3]. Report No.: ISBN: 978-92-806-4814-0. Available from: http://countdown2015mnch.org/reports-and-articles/2015-final-report
8. UNICEF-WHO-The World Bank Group. Joint child malnutrition estimates - Levels and trends (2015 edition) [Internet]. [cited 2016 Feb 7]. Available from: http://www.who.int/nutgrowthdb/estimates2014/en/
9. de Onis M, Blössner M, Borghi E. Global prevalence and trends of overweight and obesity among preschool children. Am J Clin Nutr. 2010 Nov;92(5):1257–64.
10. Rahman S, Islam MT, Alam DS. Obesity and overweight in Bangladeshi children and adolescents: a scoping review. BMC Public Health. 2014 Jan 22;14:70.
11. Hoque ME, Doi SA, Mannan M, Long K, Niessen LW, Mamun AA. Prevalence of overweight and obesity among children and adolescents of the Indian subcontinent: a meta-analysis. Nutr Rev. 2014 Aug 1;72(8):541–50.
12. Ranjani H, Mehreen T, Pradeepa R, Anjana R, Garg R, Anand K, et al. Epidemiology of childhood overweight & obesity in India: A systematic review. Indian J Med Res. 2016;143(2):160.
13. United Nations General Assembly. Resolution 70/1: Transforming Our World: The 2030 Agenda for Sustainable Development [Internet]. 2015 [cited 2018 May 14]. Available from: http://www.un.org/en/development/desa/population/migration/generalassembly/docs/globalcompa ct/A_RES_70_1_E.pdf UPPSALA UNIVERSITET 37 (58)
14. Khandelwal S, Reddy KS. Eliciting a policy response for the rising epidemic of overweight- obesity in India. Obes Rev. 2013 Nov 1;14:114–25.
15. Kapil U, Sachdev HPS. Urgent need to orient public health response to rapid nutrition transition. Indian J Community Med Off Publ Indian Assoc Prev Soc Med. 2012 Oct;37(4):207–10.
16. Wang Y, Chen H-J, Shaikh S, Mathur P. Is obesity becoming a public health problem in India? Examine the shift from under- to overnutrition problems over time. Obes Rev Off J Int Assoc Study Obes. 2009 Jul;10(4):456–74.
17. Roth CL, Jain V. Rising Obesity in Children: A Serious Public Health Concern. Indian J Pediatr. 2018 Feb 17;1–2.
18. Popkin BM. Nutritional patterns and transitions. Popul Dev Rev. 1993;138–157.
19. Kennedy G, Nantel G, Shetty P. Assessment of the double burden of malnutrition in six case study countries. FAO Food Nutr Pap FAO [Internet]. 2006 [cited 2017 Apr 28]; Available from: http://agris.fao.org/agris-search/search.do?recordID=XF2006427739
20. Dong F, Howard AG, Herring AH, Thompson AL, Adair LS, Popkin BM, et al. Parent-child associations for changes in diet, screen time, and physical activity across two decades in modernizing China: China Health and Nutrition Survey 1991-2009. Int J Behav Nutr Phys Act. 2016 Nov 11;13(1):118.
21. Tzioumis E, Adair LS. Childhood dual burden of under- and over-nutrition in low- and middle- income countries: a critical review. Food Nutr Bull. 2014 Jun;35(2):230–43.
22. Jones-Smith JC, Gordon-Larsen P, Siddiqi A, Popkin BM. Is the burden of overweight shifting to the poor across the globe? Time trends among women in 39 low- and middle-income countries (1991-2008). Int J Obes 2005. 2012 Aug;36(8):1114–20.
23. Monteiro CA, Conde WL, Lu B, Popkin BM. Obesity and inequities in health in the developing world. Int J Obes Relat Metab Disord J Int Assoc Study Obes. 2004 Sep;28(9):1181–6.
24. Misra A, Singhal N, Sivakumar B, Bhagat N, Jaiswal A, Khurana L. Nutrition transition in India: Secular trends in dietary intake and their relationship to diet-related non-communicable diseases. J Diabetes. 2011;3(4):278–292.
25. India State-Level Disease Burden Initiative Collaborators. Nations within a nation: variations in epidemiological transition across the states of India, 1990-2016 in the Global Burden of Disease Study. Lancet Lond Engl. 2017 Dec 2;390(10111):2437–60.
26. Black RE, Victora CG, Walker SP, Bhutta ZA, Christian P, De Onis M, et al. Maternal and child undernutrition and overweight in low-income and middle-income countries. The lancet. 2013;382(9890):427–451.
27. World Health Organization. The double burden of malnutrition: policy brief [Internet]. World Health Organization; [cited 2018 Mar 4]. Available from: http://www.who.int/nutrition/publications/doubleburdenmalnutrition-policybrief/en/ UPPSALA UNIVERSITET 38 (58)
28. Popkin BM. The shift in stages of the nutrition transition in the developing world differs from past experiences! Public Health Nutr. 2002;5(1A):205–214.
29. United Nations Standing Committee on Nutrition. Tackling Double Burden of Malnutrition: Global Agenda. SCN News. 2006 mid;32.
30. Varela-Silva MI, Dickinson F, Wilson H, Azcorra H, Griffiths PL, Bogin B. The nutritional dual- burden in developing countries–how is it assessed and what are the health implications? Coll Antropol. 2012;36(1):39–45.
31. UNICEF-WHO-The World Bank Group. Joint child malnutrition estimates - Levels and trends (2017 edition) [Internet]. WHO. [cited 2018 Mar 10]. Available from: http://www.who.int/nutgrowthdb/estimates2016/en/
32. Kulkarni B, Hills AP, Byrne NM. Nutritional influences over the life course on lean body mass of individuals in developing countries. Nutr Rev. 2014;72(3):190–204.
33. Caballero B. A nutrition paradox--underweight and obesity in developing countries. N Engl J Med. 2005 Apr 14;352(15):1514–6.
34. Victora CG, Onis M de, Hallal PC, Blössner M, Shrimpton R. Worldwide Timing of Growth Faltering: Revisiting Implications for Interventions. Pediatrics. 2010 Mar 1;125(3):e473–80.
35. Black RE. Global Prevalence of Small for Gestational Age Births. Nestle Nutr Inst Workshop Ser. 2015;81:1–7.
36. Patwari AK, Kumar S, Beard J. Undernutrition among infants less than 6 months of age: an underestimated public health problem in India. Matern Child Nutr. 2015 Jan;11(1):119–26.
37. Barker DJP. The developmental origins of adult disease. J Am Coll Nutr. 2004;23(sup6):588S– 595S.
38. Parlee SD, MacDougald OA. Maternal nutrition and risk of obesity in offspring: The Trojan horse of developmental plasticity. Biochim Biophys Acta BBA - Mol Basis Dis. 2014 Mar;1842(3):495– 506.
39. Corvalán C, Gregory CO, Ramirez-Zea M, Martorell R, Stein AD. Size at birth, infant, early and later childhood growth and adult body composition: a prospective study in a stunted population. Int J Epidemiol. 2007 Jun 1;36(3):550–7.
40. Sachdev HS, Fall CHD, Osmond C, Lakshmy R, Dey Biswas SK, Leary SD, et al. Anthropometric indicators of body composition in young adults: relation to size at birth and serial measurements of body mass index in childhood in the New Delhi birth cohort. Am J Clin Nutr. 2005 Aug;82(2):456–66.
41. Sinha S, Aggarwal AR, Osmond C, Fall CH, Bhargava SK, Sachdev HS. Intergenerational change in anthropometric indices and their predictors among children in New Delhi Birth Cohort. Indian Pediatr. 2017;54:185–92. UPPSALA UNIVERSITET 39 (58)
42. Adrianzen B, Baertl JM, Graham GG. Growth of children from extremely poor families. Am J Clin Nutr. 1973 Sep;26(9):926–30.
43. Martorell R, Mendoza FS, Castillo RO, Pawson IG, Budge CC. Short and plump physique of Mexican-American children. Am J Phys Anthropol. 1987 Aug;73(4):475–87.
44. Said-Mohamed R, Allirot X, Sobgui M, Pasquet DP. Determinants of overweight associated with stunting in preschool children of Yaoundé, Cameroon. Ann Hum Biol. 2009 Jan 1;36(2):146–61.
45. Popkin BM, Richards MK, Montiero CA. Stunting is Associated with Overweight in Children of Four Nations That Are Undergoing the Nutrition Transition. J Nutr. 1996 Dec;126:3009–3016.
46. Said-Mohamed R, Bernard JY, Ndzana A-C, Pasquet P. Is overweight in stunted preschool children in Cameroon related to reductions in fat oxidation, resting energy expenditure and physical activity? PloS One. 2012;7(6):e39007.
47. Hoffman DJ, Sawaya AL, Verreschi I, Tucker KL, Roberts SB. Why are nutritionally stunted children at increased risk of obesity? Studies of metabolic rate and fat oxidation in shantytown children from São Paulo, Brazil. Am J Clin Nutr. 2000 Sep 1;72(3):702–7.
48. Sawaya AL, Martins PA, Grillo LP, Florencio TT. Long-term effects of early malnutrition on body weight regulation. Nutr Rev. 2004;62(suppl 2):S127–S133.
49. Bates K. Double or divergent? Stuntingoverweightness among children and the ‘burden’of malnutrition: a study of Albania [Internet]. The London School of Economics and Political Science (LSE); 2014 [cited 2017 Apr 28]. Available from: http://etheses.lse.ac.uk/id/eprint/998
50. Yajnik CS, Fall CHD, Coyaji KJ, Hirve SS, Rao S, Barker DJP, et al. Neonatal anthropometry: the thin–fat Indian baby. The Pune maternal nutrition study. Int J Obes. 2003;27(2):173–180.
51. Nead KG, Halterman JS, Kaczorowski JM, Auinger P, Weitzman M. Overweight children and adolescents: a risk group for iron deficiency. Pediatrics. 2004;114(1):104–108.
52. Zimmermann MB, Zeder C, Muthayya S, Winichagoon P, Chaouki N, Aeberli I, et al. Adiposity in women and children from transition countries predicts decreased iron absorption, iron deficiency and a reduced response to iron fortification. Int J Obes. 2008;32(7):1098–1104.
53. Yanoff LB, Menzie CM, Denkinger B, Sebring NG, McHugh T, Remaley AT, et al. Inflammation and iron deficiency in the hypoferremia of obesity. Int J Obes. 2007;31(9):1412–1419.
54. Jones AD, Hayter AKM, Baker CP, Prabhakaran P, Gupta V, Kulkarni B, et al. The co-occurrence of anemia and cardiometabolic disease risk demonstrates sex-specific sociodemographic patterning in an urbanizing rural region of southern India. Eur J Clin Nutr. 2016 Mar;70(3):364–72.
55. Golden MH. Specific deficiencies versus growth failure: type I and type II nutrients. SCN News. 1995;(12):10–4.
56. Millward DJ. Nutrition, infection and stunting: the roles of deficiencies of individual nutrients and foods, and of inflammation, as determinants of reduced linear growth of children. Nutr Res Rev. 2017;30(1):50–72. UPPSALA UNIVERSITET 40 (58)
57. Swaminathan S, Vaz M, Kurpad AV. Protein intakes in India. Br J Nutr. 2012 Aug;108 Suppl 2:S50-58.
58. Varadharajan KS, Thomas T, Kurpad AV. Poverty and the state of nutrition in India. Asia Pac J Clin Nutr. 2013;22(3):326–339.
59. Shankar B, Agrawal S, Beaudreault AR, Avula L, Martorell R, Osendarp S, et al. Dietary and nutritional change in India: implications for strategies, policies, and interventions. Ann N Y Acad Sci. 2017 Mar 17;
60. Wessells KR, Brown KH. Estimating the global prevalence of zinc deficiency: results based on zinc availability in national food supplies and the prevalence of stunting. PloS One. 2012;7(11):e50568.
61. Lazzerini M, Wanzira H. Oral zinc for treating diarrhoea in children. In: Cochrane Database of Systematic Reviews [Internet]. John Wiley & Sons, Ltd; 2016 [cited 2017 Apr 25]. Available from: http://cochranelibrary-wiley.com/doi/10.1002/14651858.CD005436.pub5/full
62. Checkley W, Buckley G, Gilman RH, Assis AM, Guerrant RL, Morris SS, et al. Multi-country analysis of the effects of diarrhoea on childhood stunting. Int J Epidemiol. 2008 Aug;37(4):816– 30.
63. Lin A, Arnold BF, Afreen S, Goto R, Huda TMN, Haque R, et al. Household environmental conditions are associated with enteropathy and impaired growth in rural Bangladesh. Am J Trop Med Hyg. 2013 Jul;89(1):130–7.
64. Humphrey JH. Child undernutrition, tropical enteropathy, toilets, and handwashing. The Lancet. 2009 Sep 19;374(9694):1032–5.
65. Prendergast AJ, Humphrey JH. The stunting syndrome in developing countries. Paediatr Int Child Health. 2014 Nov;34(4):250–65.
66. Guerrant RL, Oriá RB, Moore SR, Oriá MOB, Lima AAM. Malnutrition as an enteric infectious disease with long-term effects on child development. Nutr Rev. 2008 Sep;66(9):487–505.
67. DeBoer MD, Lima AAM, Oría RB, Scharf RJ, Moore SR, Luna MA, et al. Early childhood growth failure and the developmental origins of adult disease: do enteric infections and malnutrition increase risk for the metabolic syndrome? Nutr Rev. 2012 Nov;70(11):642–53.
68. Margolis R. Childhood Morbidity and Health in Early Adulthood: Life course linkages in a high morbidity context. Adv Life Course Res. 2010 Dec 1;15(4):132.
69. Bailey LC, Forrest CB, Zhang P, Richards TM, Livshits A, DeRusso PA. Association of antibiotics in infancy with early childhood obesity. JAMA Pediatr. 2014 Nov;168(11):1063–9.
70. Turta O, Rautava S. Antibiotics, obesity and the link to microbes - what are we doing to our children? BMC Med. 2016 Apr 19;14(1):57. UPPSALA UNIVERSITET 41 (58)
71. Doak C, Adair L, Bentley M, Fengying Z, Popkin B. The underweight/overweight household: an exploration of household sociodemographic and dietary factors in China. Public Health Nutr. 2002;5(1a):215–221.
72. Garrett JL, Ruel MT. Stunted child–overweight mother pairs: Prevalence and association with economic development and urbanization. Food Nutr Bull. 2005;26(2):209–221.
73. Doak CM, Adair LS, Bentley M, Monteiro C, Popkin BM. The dual burden household and the nutrition transition paradox. Int J Obes 2005. 2005 Jan;29(1):129–36.
74. Jehn M, Brewis A. Paradoxical malnutrition in mother-child pairs: untangling the phenomenon of over- and under-nutrition in underdeveloped economies. Econ Hum Biol. 2009 Mar;7(1):28–35.
75. Dieffenbach S, Stein AD. Stunted Child/Overweight Mother Pairs Represent a Statistical Artifact, Not a Distinct Entity. J Nutr. 2012 Apr 1;142(4):771–3.
76. Barnett I. Is the Dual Burden of Over-and Under-nutrition a Concern for Poor Households in Ethiopia, India, Peru and Vietnam? [Internet]. Young Lives; 2011 [cited 2017 Mar 23]. Available from: http://repositorio.minedu.gob.pe/handle/123456789/768
77. Abdullah A. The Double Burden of Undernutrition and Overnutrition in Developing Countries: an Update. Curr Obes Rep. 2015 Sep 1;4(3):337–49.
78. International Institute for Population Sciences. National Family Health Survey 4: 2015-2016 India Factsheet [Internet]. IIPS, Mumbai; [cited 2017 Mar 23]. Available from: http://rchiips.org/nfhs/pdf/NFHS4/India.pdf
79. Ng M, Fleming T, Robinson M, Thomson B, Graetz N, Margono C, et al. Global, regional, and national prevalence of overweight and obesity in children and adults during 1980–2013: a systematic analysis for the Global Burden of Disease Study 2013. The Lancet. 2014 Sep 5;384(9945):766–81.
80. International Food Policy Research Institute. Global Nutrition Report 2015: Actions and accountability to advance nutrition and sustainable development. Intl Food Policy Res Inst; 2015. 201 p.
81. Sengupta A, Angeli F, Syamala TS, Dagnelie PC, van Schayck CP. Overweight and obesity prevalence among Indian women by place of residence and socio-economic status: Contrasting patterns from “underweight states” and “overweight states” of India. Soc Sci Med 1982. 2015 Aug;138:161–9.
82. International Institute for Population Sciences. National Family Health Survey 4: 2015-2016 State Factsheets [Internet]. IIPS, Mumbai; [cited 2013 Mar 23]. Available from: http://rchiips.org/nfhs/factsheet_NFHS-4.shtml
83. Ackerson LK, Kawachi I, Barbeau EM, Subramanian SV. Geography of underweight and overweight among women in India: A multilevel analysis of 3204 neighborhoods in 26 states. Econ Hum Biol. 2008 Jul;6(2):264. UPPSALA UNIVERSITET 42 (58)
84. Corsi DJ, Finlay JE, Subramanian SV. Global Burden of Double Malnutrition: Has Anyone Seen It? PLoS ONE 2011;6(9):e25120.
85. SV Subramanian, Jessica M Perkins, Kashif T Khan. Do burdens of underweight and overweight coexist among lower socioeconomic groups in India? Am J Clin Nutr. 2009;(90):369–76.
86. Kulkarni VS, Kulkarni VS, Gaiha R. “Double Burden of Malnutrition”: Reexamining the Coexistence of Undernutrition and Overweight Among Women in India. Int J Health Serv Plan Adm Eval. 2017 Jan;47(1):108–33.
87. Dasgupta P, Saha R, Nubé M. Changes in body size, shape and nutritional status of middle-class Bengali boys of Kolkata, India, 1982--2002. Econ Hum Biol. 2008 Mar;6(1):75–94.
88. Jeemon P, Prabhakaran D, Mohan V, Thankappan KR, Joshi PP, Ahmed F, et al. Double burden of underweight and overweight among children (10-19 years of age) of employees working in Indian industrial units. Natl Med J India. 2009 Aug;22(4):172–6.
89. Srihari G, Eilander A, Muthayya S, Kurpad AV, Seshadri S. Nutritional status of affluent Indian school children: what and how much do we know? Indian Pediatr. 2007 Mar;44(3):204–13.
90. Leveraging Urbanization in South Asia [Internet]. World Bank. [cited 2017 Mar 23]. Available from: http://www.worldbank.org/en/region/sar/publication/urbanization-south-asia-cities
91. Kumar U. India’s Demographic Transition: Boon or Bane? Asia Pac Policy Stud. 2014 Jan 1;1(1):186–203.
92. Government of India. Situation Analyses: Backdrop to the National Health Policy 2017 [Internet]. March2017 [cited 2017 Apr 4]. Available from: http://www.mohfw.nic.in/showfile.php?lid=4276
93. Daniels SR. The Consequences of Childhood Overweight and Obesity. Future Child. 2006;16(1):47–67.
94. International Food Policy Research Institute. Global Nutrition Report 2016: From Promise to Impact - Ending Malnutrition by 2030. International Food Policy Research Institute, Washington DC; 2016.
95. Subramanyam MA, Kawachi I, Berkman LF, Subramanian SV. Socioeconomic Inequalities in Childhood Undernutrition in India: Analyzing Trends between 1992 and 2005. PLOS ONE. 2010 Jun 30;5(6):e11392.
96. Vandenbroucke JP, Elm E von, Altman DG, Gøtzsche PC, Mulrow CD, Pocock SJ, et al. Strengthening the Reporting of Observational Studies in Epidemiology (STROBE): Explanation and Elaboration. PLOS Med. 2007 Oct 16;4(10):e297.
97. International Institute for Population Sciences. District Level Household and Facility Survey (DLHS-4) 2012-2013: Maharashtra [Internet]. 2014 [cited 2017 Jan 23]. Available from: https://nrhm-mis.nic.in/DLHS4/State%20Reports/Maharashtra.pdf
98. Das J. DISTRICT LEVEL HOUSEHOLD AND FACILITY SURVEY-4 (DLHS-4): Integration of Clinical, Anthropometric and Biochemical (CAB) component [Internet]. International Institute for UPPSALA UNIVERSITET 43 (58)
Population Sciences; 2014 [cited 2017 Jan 23]. Available from: http://www.nihfw.org/Doc/ongoing%20special%20activities/DISTRICT%20LEVEL%20HOUSE HOLD%20AND%20FACILITY%20SURVEY.pdf
99. Dandona R, Pandey A, Dandona L. A review of national health surveys in India. Bull World Health Organ. 2016 Apr 1;94(4):286-296A.
100. International Institute for Population Sciences. District Level Household and Facility Survey 2012- 13: State Reports [Internet]. [cited 2018 Mar 18]. Available from: http://rchiips.org/DLHS-4.html
101. Government of India, Ministry of Home Affairs. Press Release: Annual Health Survey (AHS) in 8 EAG States and Assam [Internet]. 2011 [cited 2017 Apr 7]. Available from: http://censusindia.gov.in/vital_statistics/AHSBulletins/files/AHSpr.pdf
102. Government of India. Census 2011: List of states with Population, Sex Ratio and Literacy [Internet]. [cited 2017 Jan 24]. Available from: http://www.census2011.co.in/states.php
103. NITI Ayog, Government of India. State Ranks according to NSDP Per Capita as on 01.08.2014 [Internet]. [cited 2017 Jan 24]. Available from: https://community.data.gov.in/state-ranks- according-to-nsdp-per-capita-as-on-01-08-2014/
104. World Health Organization. Nutrition Landscape Information System (NLIS) country profile indicators: interpretation guide [Internet]. 2010 [cited 2017 May 2]. Available from: http://www.who.int/nutrition/nlis_interpretation_guide.pdf
105. World Health Organization. WHO child growth standards: length/height for age, weight-for-age, weight-for-length, weight-for-height and body mass index-for-age, methods and development. [Internet]. World Health Organization; 2006 [cited 2017 Jan 31]. Available from: https://www.cabdirect.org/cabdirect/abstract/20063123347
106. Cole TJ, Lobstein T. Extended international (IOTF) body mass index cut-offs for thinness, overweight and obesity. Pediatr Obes. 2012 Aug;7(4):284–94.
107. Parthasarathy A. IAP Textbook of Pediatrics. JP Medical Ltd; 2016. 1409 p.
108. Svedberg P. Poverty and Undernutrition: Theory, Measurement, and Policy. Oxford University Press; 2000. 394 p.
109. WHO | Iron deficiency anaemia: assessment, prevention and control [Internet]. WHO. [cited 2017 Jan 24]. Available from: http://www.who.int/nutrition/publications/micronutrients/anaemia_iron_deficiency/WHO_NHD_0 1.3/en/
110. Barba C, Cavalli-Sforza T, Cutter J, Darnton-Hill I, others. Appropriate body-mass index for Asian populations and its implications for policy and intervention strategies. The lancet. 2004;363(9403):157.
111. Government of India, Ministry of Health and Family Welfare. Health Management Information System: DLHS-4 [Internet]. [cited 2017 Apr 7]. Available from: https://nrhm- mis.nic.in/SitePages/DLHS-4.aspx UPPSALA UNIVERSITET 44 (58)
112. Grellety E, Golden MH. The Effect of Random Error on Diagnostic Accuracy Illustrated with the Anthropometric Diagnosis of Malnutrition. PLOS ONE. 2016 Dec 28;11(12):e0168585.
113. World Health Organization. WHO Anthro for personal computers, version 3.1, 2010: Software for assessing growth and development of the world’s children. [Internet]. 2010 [cited 2017 Mar 30]. Available from: http://www.who.int/childgrowth/software/en/
114. Crowe S, Seal A, Grijalva-Eternod C, Kerac M. Effect of nutrition survey ‘cleaning criteria’on estimates of malnutrition prevalence and disease burden: secondary data analysis. PeerJ. 2014;2:e380.
115. Aguayo VM, Nair R, Badgaiyan N, Krishna V. Determinants of stunting and poor linear growth in children under 2 years of age in India: an in-depth analysis of Maharashtra’s comprehensive nutrition survey. Matern Child Nutr. 2016 May;12 Suppl 1:121–40.
116. Jayachandran S, Pande R. Why are Indian children so short? [Internet]. National Bureau of Economic Research; 2015 [cited 2016 Nov 6]. Available from: http://www.nber.org.ezproxy.its.uu.se/papers/w21036
117. Johri M, Subramanian SV, Koné GK, Dudeja S, Chandra D, Minoyan N, et al. Maternal Health Literacy Is Associated with Early Childhood Nutritional Status in India. J Nutr. 2016 Jul;146(7):1402–10.
118. Assaf S, Kothari MT, Pullum TW. An Assessment of the Quality of DHS Anthropometric Data, 2005-2014. ICF International; 2015.
119. Standardized Monitoring and Assessment of Relief and Transitions SMART Initiative. Measuring Mortality, Nutritional Status, and Food Security in Crisis Situations: SMART Methodology Version 1 [Internet]. 2006 [cited 2017 Apr 3]. Available from: http://smartmethodology.org/survey-planning-tools/smart-methodology/smart-methodology- manual/
120. World Health Organization. The global prevalence of anaemia in 2011 [Internet]. WHO Geneva; 2015 [cited 2017 May 11]. Available from: http://apps.who.int/iris/bitstream/10665/177094/1/9789241564960_eng.pdf
121. Mei Z, Grummer-Strawn LM. Standard deviation of anthropometric Z-scores as a data quality assessment tool using the 2006 WHO growth standards: a cross country analysis. Bull World Health Organ. 2007;85(6):441–448.
122. Young MF, Martorell R. The public health challenge of early growth failure in India. Eur J Clin Nutr. 2013 May;67(5):496–500.
123. Government of India, Ministry of Home Affairds. Zonal Councils [Internet]. [cited 2017 May 1]. Available from: http://mha.nic.in/zonal_council
124. Government of India, Ministry of Culture. Cultural Zones of India [Internet]. [cited 2017 May 1]. Available from: www.indiaculture.nic.in/about-us/autonomus-bodies/zonal-cultural-centers UPPSALA UNIVERSITET 45 (58)
125. Government of India. Census of India 2011: Provisional Population Totals Urban Agglomerations and Cities [Internet]. [cited 2017 Apr 10]. Available from: http://censusindia.gov.in/2011-prov- results/paper2/data_files/India2/1.%20Data%20Highlight.pdf
126. President of India. The Constitution of India: Scheduled Casts Order 1950/1 [Internet]. 1950 [cited 2017 Apr 8]. Available from: http://lawmin.nic.in/ld/subord/rule3a.htm
127. Chitnis S, Pai Panandiker VA. Definition of the terms scheduled castes and scheduled tribes: a crisis of ambivalence. Polit Backwardness Reserv Policy India New Delhi India Cent Policy Res. 1997;
128. Bassani DG, Jha P, Dhingra N, Kumar R. Child mortality from solid-fuel use in India: a nationally-representative case-control study. BMC Public Health. 2010;10:491.
129. World Health Organization, United Nations Children’s Fund. Core questions on drinking-water and sanitation for household surveys [Internet]. World Health Organization; 2006. Available from: http://apps.who.int/iris/bitstream/10665/43489/1/9789241563260_eng.pdf
130. Bain R, Cronk R, Wright J, Yang H, Slaymaker T, Bartram J. Fecal Contamination of Drinking- Water in Low- and Middle-Income Countries: A Systematic Review and Meta-Analysis. PLoS Med 2014 May 6;11(5):e1001644
131. Governnment of India. National Health Mission: Village Health Sanitation & Nutrition Committee (VHSNC) [Internet]. [cited 2017 Apr 13]. Available from: http://nrhm.gov.in/communitisation/village-health-sanitation-nutrition-committee.html
132. Rutstein SO, Johnson K. The DHS wealth index. 2004 [cited 2017 Feb 22]; Available from: http://www.dhsprogram.com/publications/publication-CR6-Comparative-Reports.cfm
133. Howe LD, Hargreaves JR, Huttly SR. Issues in the construction of wealth indices for the measurement of socio-economic position in low-income countries. Emerg Themes Epidemiol. 2008;5:3.
134. Filmer D, Pritchett LH. Estimating wealth effects without expenditure data--or tears: an application to educational enrollments in states of India. Demography. 2001 Feb;38(1):115–32.
135. Rutstein SO. The DHS wealth index: Approaches for rural and urban areas. 2008 [cited 2017 Feb 22]; Available from: http://www.dhsprogram.com/publications/publication-WP60-Working- Papers.cfm
136. Vyas S, Kumaranayake L. Constructing socio-economic status indices: how to use principal components analysis. Health Policy Plan. 2006;21(6):459–468.
137. Rutstein SO. Steps to constructing the new DHS wealth index [Internet]. Rockville, MD: ICF International; 2015 [cited 2017 Feb 23]. Available from: http://www.dhsprogram.com/programming/wealth%20index/Steps_to_constructing_the_new_DH S_Wealth_Index.pdf
138. Sahu S, Kumar Sg, Bhat Bv, Premarajan K, Sarkar S, Roy G, et al. Malnutrition among under-five children in India and strategies for control. J Nat Sci Biol Med. 2015;6(1):18. UPPSALA UNIVERSITET 46 (58)
139. Spears D, Ghosh A, Cumming O. Open Defecation and Childhood Stunting in India: An Ecological Analysis of New Data from 112 Districts. PLoS ONE [Internet]. 2013 Sep 16 [cited 2017 Jan 24];8(9). Available from: http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3774764/
140. Bassani DG, Corsi DJ, Gaffey MF, Barros AJ. Local Distributions of Wealth to Describe Health Inequalities in India: A New Approach for Analyzing Nationally Representative Household Survey Data, 1992–2008. PloS One. 2014;9(10):e110694.
141. IBM SPSS Statistics Version 24. IBM; 2016.
142. Wilson EB. Probable inference, the law of succession, and statistical inference. J Am Stat Assoc. 1927;22(158):209–212.
143. Tamhane AR, Westfall AO, Burkholder GA, Cutter GR. Prevalence odds ratio versus prevalence ratio: choice comes with consequences. Stat Med. 2016 Dec 30;35(30):5730–5.
144. Thompson ML, Myers JE, Kriebel D. Prevalence odds ratio or prevalence ratio in the analysis of cross sectional data: what is to be done? Occup Environ Med. 1998 Apr;55(4):272–7.
145. International Institute for Population Sciences. DISTRICT LEVEL HOUSEHOLD AND FACILITYSURVEY (DLHS - 4): HOUSEHOLD QUESTIONNAIRE.
146. National Institute of Nutrition. Diet & Nutritional Status of Population and Prevalence of Hypertension among Adults in Rural Areas: National Nutrition Monitoring Bureau Technical Report 24 [Internet]. National Institute of Nutrition, Hyderabad India; [cited 2018 Mar 10]. Available from: http://nnmbindia.org/NNMBReport06Nov20.pdf
147. Mistry SK, Puthussery S. Risk factors of overweight and obesity in childhood and adolescence in South Asian countries: a systematic review of the evidence. Public Health. 2015 Mar;129(3):200– 9.
148. Hawkes C. Uneven dietary development: linking the policies and processes of globalization with the nutrition transition, obesity and diet-related chronic diseases. Glob Health. 2006;2:4.
149. Chaturvedi S, Ramji S, Arora NK, Rewal S, Dasgupta R, Deshmukh V, et al. Time-constrained mother and expanding market: emerging model of under-nutrition in India. BMC Public Health. 2016;16:632.
150. GOVERNMENT OF INDIA. SIGNIFICANT INITIATIVES AND ACHIEVEMENTS OF DEPARTMENT OF FOOD & PUBLIC DISTRIBUTION ( January, 2016 to December , 2016) [Internet]. [cited 2018 Mar 11]. Available from: http://dfpd.nic.in/1sGbO2W68mUlunCgKmpnLF5WHm/ebook17Janc2017english.pdf
151. Gupta A, Gupta SK, Nongkynrih B. Integrated Child Development Services (ICDS) Scheme: A Journey of 37 years. Indian J Community Health. 2013;25(1):77–81.
152. Government of India, Ministry of Women and Child Welfare. ICDS Mission: The Broad Framework for Implementation [Internet]. 2012 [cited 2018 Mar 11]. Available from: https://www.bpni.org/WBW/2013/Broad-Framework-of-Implementation-ICDS-Mission.pdf UPPSALA UNIVERSITET 47 (58)
153. Sachdev HPS, Gera T. Preventing childhood anemia in India: iron supplementation and beyond. Eur J Clin Nutr. 2013 May;67(5):475–80.
154. Plessow R, Arora NK, Brunner B, Tzogiou C, Eichler K, Brügger U, et al. Social Costs of Iron Deficiency Anemia in 6–59-Month-Old Children in India. PLoS ONE [Internet]. 2015 Aug 27 [cited 2018 May 17];10(8). Available from: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4552473/
155. Viner RM, Cole TJ, Fry T, Gupta S, Kinra S, McCarthy D, et al. Insufficient evidence to support separate BMI definitions for obesity in children and adolescents from south Asian ethnic groups in the UK. Int J Obes 2005. 2010 Apr;34(4):656–8.
156. Economic Statistical Organisation Punjab. State Wise Data [Internet]. Central Statistical Organisation, New Delhi; 2016. Available from: http://www.esopb.gov.in/static/PDF/GSDP/Statewise-Data/state%20wise%20data.pdf
157. Rutstein SO, Staveteig S. Making the Demographic and Health Surveys wealth index comparable. In ICF Internat.; 2014 [cited 2017 Feb 23]. Available from: http://iussp.org/sites/default/files/event_call_for_papers/ComparativeWealth-DRAFT-IUSSP.pdf
UPPSALA UNIVERSITET 48 (58)
Annex
Details on Wealth Index Generation I followed the instructions of Rutstein (135) for adjusting wealth index by urbanicity and first did a regression model for urban households only, where the dependant variable was household wealth index scores based on the first PCA for all households and the independent variable was the score of second PCA done only for urban households. Unsurprisingly, as the base variables of the PCA scores were same and only their distribution changed, the resulting R2 was high 0.993 with a highly significant p-value of <0.001.
The constant α1 was 0.381 whereas the unstandardized co-efficient B1 was 0.992. Similar regression for 2 rural households resulted in R = 0.993, p = 0.000, α2 = -0.273 and B2 = 0.907. Based on constants and co- efficients of these linear regression models, a new combined wealth score variable was calculated with the following formula combined from Rutstein (135):
WSn = α1 + B1 * WSu + α2 + B2 * WSr, where WSn = new combined wealth score
α1 = constant of linear regression 1 for urban households
B1 = coefficient of linear regression 1 for urban households WSu = household wealth score for urban households
α2 = constant of linear regression 2 for rural households
B2 = coefficient of linear regression 2 for rural households WSr = household wealth score for rural households
Rutstein additionally proposes adjustment to household size (137) justified by the fact that most analyses concern poor people rather than poor households (132). This is however rejected by other authors (133,136) and as my outcome variables are both at household and individual level, I chose not to make this adjustment.
Creating Wealth Quintiles from the Index Scores A cross-tabulation between both created wealth quintiles and all indicator variables finding good internal coherence with proportion of households owning each asset higher in the wealthier quintiles. Relationship between construction material of the house and wealth was also as expected. Acres of land owned increased by wealth and number of people per room decreased. Only exception to the rule was association between UPPSALA UNIVERSITET 49 (58)
household ownership and wealth quintiles, which showed little change by increasing wealth. This was explained by almost universal house-ownership in the rural areas whereas the urban sub-group showed expected direction of association (data not shown). Both the adjusted and unadjusted scores provided similar results. To further examine the effect of using the unadjusted versus weighted wealth index I did a cross-tabulation of quintiles produced by the two scores to find out how much changed occurred in classification of households between the two options (Supplementary Table 9, Annex). It revealed that in the extreme ends of wealth only 69.0% and 85.0% of households kept their position at the poorest and richest quintile respectively.
Looking at frequency tables of distribution of households in quintiles separately in rural and urban locations (Supplementary Table 10, Annex), the unadjusted model is more likely to categorize urban people into the richest and rural people into the poorest quintile. While in the unadjusted model 27.5% of rural and only 8.9% of urban households were categorized into the poorest quintile, after the adjustment this relationship was opposite with 17.1% and 24.3% respective shares. Similarly, in the richest end the share of rural households went up from 12.2% to 17.2% and the urban share dropped from 31.5% to 24.1%. I feel that the adjustment done by the manner that was originally designed for mapping wealth scores created with different set of assets makes perhaps too drastic changes to the distribution of households into quintiles in rural and urban locations and feel therefore that the original unadjusted model is both easier to understand and fairer to the true division of asset-based wealth.
Further Discussion on Wealth Quintiles Another level of complexity can be added if the state level differences are to be considered in the creation of the wealth quintiles. The Gross Domestic State Product per capita of non-EAG states in India varies almost six-fold between the lowest Manipur and the highest Goa (156). Supplementary Figure 4, Annex shows stacked bar charts of the division of households into wealth quintiles in each state using the unadjusted wealth index and household state weights. It demonstrates how un-equally wealth is distributed in different non-EAG states and Union Territories of India per my wealth index, and how overrepresented the wealthier states are in the richest wealth quintile and the poorest in the poorest wealth quintile.
These differences suggest that it could be beneficial to compute wealth indices separately for each state, which was done by Bassani et.al. in their paper on the previous rounds of DLHS and NFHS (140). A major drawback of creating separate indices for each state is the non-comparability of wealth scores between the states this adjustment was therefore not done. However, the observed differential distribution of wealth in UPPSALA UNIVERSITET 50 (58)
states could explain some of the undernutrition among the richest quintile and overnutrition among the poorest.
Rutstein and Staveteig of the DHS have already proposed making asset-based PCA wealth scores comparable across countries (157). In my case, linear regression could be used to adjust for state wealth in a similar fashion as was done to rural-urban difference. To test this option, I ran the PCA of common set of variables for all households separately for each state. This resulted in some implausible weightings of the components such as household ownership turning negative in several states. To my knowledge such adjustment for state wealth has not been done in studies dealing with previous rounds of DLHS and NFHS and would thus affect the comparability of my results with them. UPPSALA UNIVERSITET 51 (58)
Supplementary Tables
Supplementary Table 1: Options Available in Dataset for Age Determination
Method Strengths Weaknesses Calculating age by date of interview minus date of birth Gives the exact age Missing for 12.1% (household questionnaire) Age in completed years No missing values Only completed years (household questionnaire) Age at test (CAB No missing values Only completed years questionnaire) Age in years, months and days (household Gives the exact age Missing for 37% questionnaire) CAB = Clinical, Anthropometric and Biochemical component
Supplementary Table 2: Prevalence Estimates of Malnutrition Using Two Different Cut-Off Criteria WHO 2006 (103) SMART 2006 (117) Observed stunted cases 26544 21725 Stunted N of children excluded 4898 15188 Prevalence of stunting % 40.1 38.9 Observed wasted cases 12910 9867 Wasted N of children excluded 4982 11402 Prevalence of wasting % 20.4 17.4 Observed overweight cases 5157 1188 Overweight N of children excluded 7185 14472 Prevalence of overweight % 8.2 2.1 Observed underweight cases 21988 20345 Underweight N of children excluded 1060 4374 Prevalence of underweight % 31.3 30.4
Supplementary Table 3: Distribution of Observed Anthropometric Indices Before Data Cleaning
WHZ HAZ WAZ BAZ N Valid 68181 71032 71283 70340 Missing 5969 3118 2868 3811 Mean -.82107 -1.03226 -1.25793 -.69910 Std. Deviation 2.999049 3.593523 2.311586 3.634674
UPPSALA UNIVERSITET 52 (58)
Supplementary Table 4: Classification of Water and Sanitation Types into Improved and Unimproved Forms (129)
DRINKING-WATER SOURCE CATEGORIES SANITATION CATEGORIES "Improved" sources of drinking-water: "Improved" sanitation: Piped water into dwelling Flush toilet Piped water to yard/plot Piped sewer system Public tap or standpipe Septic tank Tubewell or borehole Flush/pour flush to pit latrine Protected dug well Ventilated improved pit latrine (VIP) Protected spring Pit latrine with slab Rainwater Composting toilet "Unimproved" sources of drinking-water: "Unimproved" sanitation: Unprotected spring Flush/pour flush to elsewhere Unprotected dug well Pit latrine without slab Cart with small tank/drum Bucket Tanker-truck Hanging toilet or hanging latrine Surface water Shared toilet Bottled water No facilities or bush or field
Supplementary Table 5: Distribution of Potential Household Wealth Variables by Household Location
Household Locality Rural, n=234675 Urban, n=159594 Total, n=394269 59.5% of households 40.5% of households 100% of households
n n n % of Rural % of Urban % of Total Household owning any land Yes 107731 45.9% 27011 16.9% 134742 34.2% Household owning their house Yes 218694 93.2% 113900 71.5% 332594 84.4% Main source of lighting Electricity 218949 93.3% 157052 98.5% 376001 95.4% Kerosene 10632 4.5% 1490 0.9% 12122 3.1% Solar 2645 1.1% 313 0.2% 2958 0.8% Other oils 483 0.2% 102 0.1% 585 0.1% Other 752 0.3% 206 0.1% 957 0.2% Missing 1214 0.5% 341 0.2% 1555 0.4% Below Poverty Line card Yes 81335 34.7% 54855 34.4% 136190 34.6% Do not know 2172 0.9% 1404 0.9% 3576 0.9% Missing 85 <0.1% 24 <0.1% 109 <0.1% House type: Pucca house Yes 82084 35.1% 100771 63.5% 182854 46.6% Missing 1138 0.5% 925 0.6% 2063 0.5% Semi- Pucca house Yes 91716 39.3% 44694 28.2% 136410 34.8% UPPSALA UNIVERSITET 53 (58)
Missing 1138 0.5% 925 0.6% 2063 0.5% Kaccha house Yes 59737 25.6% 13114 8.3% 72851 18.6% Missing 1138 0.5% 925 0.6% 2063 0.5% Asset ownership: Radio/transistor Yes 26682 11.4% 18694 11.7% 45377 11.5% Television Yes 167863 71.5% 142832 89.5% 310695 78.8% Computer/laptop without internet Yes 6581 2.8% 15867 9.9% 22448 5.7% Computer /laptop with internet yes 5970 2.5% 16338 10.2% 22308 5.7% Any Computer yes 11833 5.0% 30301 19.0% 42133 10.7% Telephone yes 13672 5.8% 19609 12.3% 33281 8.4% Mobile phone yes 190578 81.2% 145123 91.0% 335701 85.2% Refrigerator yes 56527 24.1% 81276 51.0% 137803 35.0% Watch or clock yes 182360 77.7% 138255 86.7% 320615 81.3% Water pump /tube well yes 19334 8.2% 15776 9.9% 35110 8.9% Cooler or air-conditioner yes 28817 12.3% 40965 25.7% 69782 17.7% Washing machine yes 24861 10.6% 47436 29.7% 72297 18.3% Sewing machine yes 53931 23.0% 47948 30.1% 101879 25.8% Bicycle yes 91116 38.8% 61258 38.4% 152373 38.7% Motorcycle, scooter or moped yes 63753 27.2% 72549 45.5% 136302 34.6% Car, jeep or a van yes 11543 4.9% 20124 12.6% 31667 8.0% Tractor yes 6728 2.9% 1033 0.6% 7762 2.0% Cart driven by animal yes 7733 3.3% 777 0.5% 8509 2.2% Cart driven by machine yes 2317 1.0% 698 0.4% 3015 0.8% Other cart yes 1400 0.6% 1017 0.6% 2418 0.6% Any cart yes 10203 4.3% 2206 1.4% 12410 3.1% Adjusted by household state weights No missing values in household assets
Supplementary Table 6: Weightings of Household Features and Assets Used to Generate Household Wealth Index
General features: Assets: Land ownership in acres .046 Radio or transistor .080 Watch or clock .370 Household owning their house .018 Television .533 Bicycle .154 Persons per room -.310 Computer .528 Motorcycle, scooter or moped .616 Electricity as main lighting .305 Telephone .407 Car, jeep or van .486 House type: Mobile phone .401 Tractor .221 Pucca house .668 Washing machine .690 Water pump or tube well .314 Semi-Pucca house -.306 Refrigerator .748 Cooler or air-conditioner .577 Kaccha house -.484 Sewing machine .522 Any cart .127 UPPSALA UNIVERSITET 54 (58)
Supplementary Table 7: State Distribution of Different Forms of Malnutrition
OSC OAC Overweight Stunted Anaemic Underweight Zone State Count Row % Count Row % Count Row % Count Row % Count Row % Count Row % Himachal Prad. 31 5.5% 4 2.3% 51 8.8% 231 38.0% 123 67.0% 149 22.8% Punjab 325 4.4% 261 4.6% 480 6.4% 3079 40.2% 4266 70.3% 2067 26.6% Chandigarh 16 6.8% 7 4.6% 19 7.6% 109 43.9% 105 62.1% 70 27.8% North Haryana 330 5.4% 226 5.5% 490 7.8% 2826 41.6% 3353 69.9% 2547 35.9% New Delhi 42 5.3% 46 7.2% 84 9.8% 425 48.6% 558 74.7% 390 39.2% Zone Total: 745 4.9% 544 5.0% 1123 7.3% 6669 41.2% 8405 70.3% 5223 31.3% Sikkim 40 5.2% 36 6.4% 59 7.5% 313 38.8% 543 88.7% 134 16.2% Arunachal Prad 112 6.1% 94 7.5% 207 10.8% 897 45.4% 937 64.7% 640 29.3% Nagaland 45 5.3% 25 5.1% 72 8.3% 388 43.9% 337 63.1% 232 24.6% Manipur 89 6.6% 59 8.1% 148 10.6% 609 42.8% 612 75.5% 401 26.6% North-East Mizoram 221 6.5% 150 6.2% 343 9.9% 1412 40.6% 1995 78.8% 630 17.4% Tripura 48 5.6% 34 4.9% 76 8.8% 346 39.5% 386 52.4% 240 26.7% Meghalaya 141 9.2% 32 5.0% 180 11.4% 938 58.7% 509 73.9% 511 30.7% Zone Total: 696 6.6% 430 6.3% 1085 10.0% 4902 44.4% 5318 72.2% 2788 23.9% West Bengal 284 6.2% 152 7.0% 403 8.6% 2209 45.1% 2127 87.9% 1972 38.0% East Zone Total: 284 6.2% 152 7.0% 403 8.6% 2209 45.1% 2127 87.9% 1972 38.0% Daman & Diu 0 0.0% 0 0.0% 5 12.9% 17 45.7% 0 0.0% 21 45.4% Goa 9 2.9% 11 8.0% 22 7.0% 78 22.5% 130 78.0% 78 21.3% West Maharashtra 371 4.5% 273 5.1% 612 7.1% 3515 37.6% 5400 82.1% 3891 38.1% Zone Total: 380 4.4% 284 5.2% 639 7.2% 3610 37.1% 5530 82.0% 3990 37.6% Andhra Prad. 140 5.9% 126 7.3% 238 9.5% 1093 41.7% 1695 82.0% 984 33.4% Karnataka 593 6.4% 399 6.2% 860 8.9% 4507 45.1% 5636 78.9% 3667 35.2% Kerala 59 5.4% 20 4.6% 108 9.5% 314 26.4% 293 54.4% 273 21.1% Tamil Nadu 229 3.1% 144 3.0% 423 5.7% 2333 30.3% 3324 63.9% 2110 26.1% South Pundicherry 43 5.2% 31 5.2% 76 8.8% 235 26.3% 429 61.4% 201 20.5% Andaman & Nic. 20 7.8% 15 10.1% 37 13.8% 82 29.5% 132 77.1% 60 19.5% Telangana 99 7.3% 79 8.9% 165 11.4% 589 36.7% 894 74.6% 721 36.4% Zone Total: 1183 5.3% 814 5.4% 1906 8.2% 9154 37.7% 12401 72.9% 8015 30.8% TOTAL 3288 5.4% 2224 5.5% 5157 8.2% 26544 40.1% 33781 74.2% 21988 31.3% OSC = Overweight and Stunted Child, OAC = Overweight and Anaemic Child
UPPSALA UNIVERSITET 55 (58)
Supplementary Table 8: Bivariate Analysis of Predictors of Dual Burden Households
SCOA nSCOA n=13430 n=39135 Row % (share of all households) 25.6% 74.4% Unpaired t- 6.5 5.8 Mean number of household members test p<0.001 χ²-test Household Characteristics Column % Column % p-value SC 26.6 25.5 ST 17.7 20.4 Household Head Caste OBC 33.5 33.6 other 22.2 20.4 <0.001 Safe Water 92.9 91.4 <0.001 Treated water 32.7 33.1 0.347 Improved Toilet 71.9 67.6 <0.001 Safe infant feces 41.9 39.2 <0.001 No solid fuel 48.9 45.0 Cooking Facilities Solid fuel in separate kitchen 19.4 23.7 Solid fuel, no separate kitchen 31.7 31.2 <0.001 Village Characteristics Anganwadi 99.3 98.9 0.002 VHNSC 59.6 57.4 0.001 Private Health Provider 38.9 35.0 <0.001 Public Health Provider 64.6 63.3 0.047 Any health provider 73.7 71.0 <0.001 SCOA = Stunted Child-Overweight Adult household nSCOA= non-SCOA household (can still have either one of components) SC = Scheduled Caste; ST = Scheduled Tribe; OBC = Other Backward Classes VHNSC = Village Health, Nutrition and Sanitation Council Cases are weighted by state household weight
Supplementary Table 9: Distribution of Households into Wealth Quintiles Using Wealth Scores Unadjusted (Row) and Adjusted by Household Locality (Column)
Wealth Quintiles of Adjusted Score Wealth Quintiles of Unadjusted Score 1st 2nd 3rd 4th 5th Total 1st % within unadjusted score 69.0% 31.0% 0.0% 0.0% 0.0% 100.0% 2nd % within unadjusted score 28.6% 38.2% 33.2% 0.0% 0.0% 100.0% 3rd % within unadjusted score 2.5% 30.2% 43.0% 24.3% 0.0% 100.0% 4th % within unadjusted score 0.0% 0.7% 23.5% 60.8% 15.0% 100.0% 5th % within unadjusted score 0.0% 0.0% 0.0% 15.0% 85.0% 100.0% 20.0% 20.0% 19.9% 20.0% 20.0% 100.0%
UPPSALA UNIVERSITET 56 (58)
Supplementary Table 10: Distribution of Households into Wealth Quintiles in Rural and Urban Areas Using Two Different Wealth Index Scores
Original score Adjusted score Household Wealth Locality Quintile n % n % Rural 1 64629 27.5 40164 17.1 2 54981 23.4 53277 22.7 3 48312 20.6 55276 23.6 4 38112 16.2 45524 19.4 5 28642 12.2 40434 17.2 Total 234675 100.0 234675 100.0 Urban 1 14208 8.9 38682 24.3 2 23854 15.0 25750 16.1 3 30653 19.2 23360 14.6 4 40589 25.4 33311 20.9 5 50199 31.5 38402 24.1 Total 159504 100.0 159504 100.0 Adjusted by state household weights.
UPPSALA UNIVERSITET 57 (58)
Supplementary Figures
Supplementary Figure 1: Zone Division of States Included in the Study UPPSALA UNIVERSITET 58 (58)
Supplementary Figure 2: Histograms of Anthropometric Indices Before Data Cleaning
Supplementary Figure 3: Histogram of Distribution of Unadjusted Household Wealth Scores for All, Rural and Urban Household
100.0%
80.0%
60.0%
40.0%
20.0%
0.0%
tr
kl
dl
nl
tg
tn
ar
cg
sk
hr
ga
ka
ap an
pb ml
hp
dd pd
mz
wb
mh
mn Total
1st Quintile 2nd Quintile 3rd Quintile 4th Quintile 5th Quintile
Supplementary Figure 4: Stacked Bar Chart of Division of Households into Wealth Quintiles in Each non-EAG State and Union Territory of India According to the Unadjusted Wealth Index Scores. *(1st Quintile = Richest) **For abbreviations of state names please refer to Supplementary Figure 1, Annex.