Determinants of very low birth weight in India: The National Family Health Survey – 4 [version 2; peer review: 2 approved]

Background Low birth weight (LBW) is susceptible to neonatal complications, chronic medical conditions, and neurodevelopmental disabilities. We aim to describe the determinants of very low birth weight (VLBW) in India and compare it with the determinants of LBW based on the National Family Health Survey – 4 (NHFS-4) Methods Data from the NFHS-4 on birthweight and other socio-demographic characteristics for the youngest child born in the family during the five years preceding the survey were used. Data of 147,762 infant–mother pairs were included. Multiple logistic regression models were employed to delineate the independent predictors of VLBW (birth weight<1500 g) or LBW (birth weight: 1500-2499 g). Of the 147,762 children included in the study, VLBW and LBW were observed in 1.2% and 15.8% of children, respectively. The odds of VLBW were higher in female children (aOR: 1.36, 95% CI: 1.15–1.60), among mothers aged 13–19 years (aOR: 1.58, 95% CI: 1.22–2.07), This study was aimed to describe the determinants of very low birth weight (VLBW) in India based on the National Family Health Survey. However, the authors present the results of determinants of both VLBW (bw<1500 gm) and LBW (BW between 1500 and 2499 gm) infants. The results of LBW are out of this study's aim.


Introduction
Low birth weight (LBW), defined as birth weight less than 2500 g, is a significant public health problem globally, and remains as a major health issue in India 1,2 . Very low birth weight (VLBW), a sub-group with birth weight <1500 g, is a high-risk group with considerable mortality and morbidity [3][4][5] . Advances in medical care, treatment facilities, and progress in availability of these services over the last several decades including the establishment of level three nurseries for neonates across India, improved the survival of both LBW and VLBW babies [6][7][8] . However, the survived babies with LBW are susceptible to neonatal complications, recurrent hospitalisations, chronic medical conditions and neurodevelopmental disabilities like intellectual disabilities, and learning disabilities 9,10 . It also increases the future risk of chronic diseases and other comorbidities. For example, adult diseases such as hypertension, dyslipidaemia and insulin resistance are closely related to a LBW, leading to markedly increased rates of cardiovascular, metabolic and renal diseases in later life 11 . Understanding the determinants of VLBW among infants is critical for planning further interventions in reducing the associated morbidity and mortality. We describe the socio economic and maternal determinants of VLBW infants and compare it with the determinants of Low Birth Weight (LBW) in India based on the National Family Health Survey -4 (NHFS-4) data.

Ethical statement
Our study is based on a secondary analysis of existing data from NFHS-4, which is an anonymous and publicly available dataset. The dataset has no identifiers of the survey participants. At the beginning of the survey, the interviewer took informed consent from each participant after explaining the purpose of the study. The informed consent explained that the participation was voluntary, and participants had the right to refuse or stop the interview at any point. The NFHS-4 obtained ethical clearance from the Ethical Review Board of the International Institute for Population Science (IIPS), which performed these surveys. We registered at the DHS site as data users and submitted a research proposal to study the determinants of VLBW. The Demographic Health Survey (DHS) program gave access to the data after reviewing the submitted proposal (10.6084/m9.figshare.16606787). We downloaded the required data from https://www.dhsprogram.com/data/available-datasets.cfm. We accepted all terms and conditions attached with the data sharing policy of DHS.

Data source
We used data from the NFHS-4 which was conducted during the year 2015-2016. The first NHFS survey began during early 1990s. The NFHS presents nationally representative data on population, health, and nutrition for India including its states as well as union territories. The survey also intended to offer state and national-level estimates of fertility, mortality, family planning, adolescent reproductive health, high-risk sexual behavior, HIV-related knowledge and use of healthcare services in the country.
Using a multi-stage sample design, NFHS-4 covered sample households all over India. A stratified two-stage sampling design was adopted for the NFHS-4 survey. For all districts surveyed, a uniform sampling design was used considering rural and urban areas as strata. To select primary sampling units (PSUs), the Census of India 2011 served as the sampling frame. The PSUs in rural areas were villages whereas it was census enumeration blocks (CEBs) in urban areas. The villages and CEBs were selected from the sampling frame with probability proportional to size (PPS) sampling. A household mapping and listing operation was performed at every selected PSU before the main survey. In the second stage, a random selection of 22 households from each PSU was done. The details on study design, sampling, and data collection schedule of the NFHS have been published elsewhere (http://rchiips.org/Nfhs/NFHS-3%20Data/VOL-1/India_vol-ume_I_corrected_17oct08.pdf, https://dhsprogram.com/pubs/ pdf/FR339/FR339.pdf). The fourth round (NFHS-4) collected data from 30 states and six union territories from India. The NFHS-4 survey gathered information from 699,686 women, and 112,122 men.

Study participants
We used the data on birthweight and other socio-demographic characteristics for the youngest child born in the family during the past five years preceding the survey (n=190,898 children). Data of 37,306 children were reported as not weighed. Additionally, data from 5729 children were reported as special answers or do not know ( Figure 1). Children with a birth weight of less than 1500 g and birth weight between 1500-2499 g were considered as VLBW and LBW, respectively. We included 147,762 infant-mother pairs meeting the inclusion criteria in the study (Figure 1).

Study variables
We grouped the study variables into three blocks representing distal, intermediate and proximal determinants, using a conceptual hierarchy-based approach 12 i.e., socioeconomic characteristics, use of the healthcare services or the programmatic factors including antenatal care (ANC), and maternal and new-born characteristics, respectively ( Figure 2).
The key study variables were individual and household sociodemographic characteristics including age and education of Any further responses from the reviewers can be found at the end of the article REVISED the mother, wealth index, marital status, religious background, and place of residence (Table 1). Reproductive characteristics of the mother included age at birth of the index child, birth order, birth interval, the type of complications during pregnancy and general health behaviours including smoking and alcohol status. The antenatal check-up (ANC) status included the timing of the first ANC visit, number of ANC visits, tetanus injection during pregnancy, place of delivery, and service accessibility. Anthropometric measures included height and body mass index of the mother. We also included the anaemia status of the mother as a study variable.

Data analysis
We used STATA Version 16.1 STATA Corp (RRID:SCR_012763) for the data analysis. We explored the bivariate associations between socio-demographic and maternal variables and low birth-weight phenotypes (VLBW and LBW). The statistically significant predictors (P<0.10) from the bivariate model were further analysed using multiple logistic regression models to establish the independent association between these variables and LBW phenotypes. A correlation matrix was employed to check multicollinearity. In the final multivariable regression model, we excluded BMI, type of delivery, place of delivery, pregnancy duration, first ANC visit and religion to avoid multicollinearity. We generated adjusted odds ratio (aOR) with their 95% confidence intervals (CI). A Poisson regression model was also generated to explore associations of socio-demographic and maternal variables and VLBW.

General characteristics
Of the 147,762 children included in the study, 1722 (1.2%) were with VLBW. In total 23,308 (15.8%) children had LBW. More than half (54.5%) of the children were boys ( Table 2). Nearly two-thirds (64%) of the mothers reported height greater than 150 cm. The body mass index was more than 18.5 kg/m 2 in four-fifths (80.0%) of the mothers. 87% of the mothers belonged to the 20-34 years at the time of childbirth. About 19% each belonged to scheduled caste and scheduled tribes. Nearly two-thirds of the mothers (64.8%) reported secondary or higher education. 40% belonged to poorer or poorest wealth quintiles. One-third of the mothers reported severe or moderate anaemia. More than two-thirds (72.4%) of the mothers reported their first antenatal care (ANC) visits during the first trimester itself.

Factors associated with very low birth weight
In the bivariate analysis, the child's gender, height, BMI, birth order, age of the mother, anaemia level, tobacco and alcohol use, thyroid disease, antenatal visits, place of delivery, multiple pregnancy, caste, religion, educational status, wealth quintile, geographic region, and pregnancy duration, were associated with VLBW (Supplementary Table 1

Factors associated with low birth weight
In the bi-variate analysis, the child's gender, height, BMI, birth order, age of the mother, anaemia level, tobacco and alcohol use, antenatal visits, place of delivery, multiple pregnancy, place of residence, caste, religion, educational status, wealth quintile, geographic region, timing of first ANC visits and appropriate ANC use were associated with LBW (Supplementary Table 2: 10.6084/m9.figshare.18393758). In the multivariable logistic regression model ( Table 4), odds of LBW were higher in girl children when compared to boys (aOR: 1.21, 95% CI: 1.15-1.26). Children with birth order greater than four were having lower odds for LBW than children with birth order one to three (aOR: 0.86, 95% CI: 0.80-0.92). Mothers aged 13-19 years had higher odds for VLBW when compared to mothers aged 20-24 years (aOR: 1.17, 95% CI: 1.06-1.26). Mothers with no education (aOR: 1.08, 95% CI: 1.02-1.15) and those with primary education (aOR: 1.16, 95% CI: 1.08-1.25) had higher odds of LBW as compared to those in the secondary education category. Children who belonged to scheduled tribe had 1.13-times higher odds for LBW versus children from other forward caste (aOR: 1.13, 95% CI: 1.03-1.24). Children from poorest/poorer wealth quintiles had higher odds of LBW versus those from rich or richest wealth quintiles (aOR: 1.11, 95% CI: 1.03-1.19). When compared with children from Western states, those from Eastern states (aOR: 0.75, 95% CI: 0.68-0.82), North-Eastern states (aOR: 0.61, 95% CI: 0.55-0.69) and Southern states (aOR: 0.90, 95% CI: 0.82-0.99) had lower odds for LBW. The odds of LBW were 1.20-times higher in mothers with severe or moderate anaemia versus non-anaemic mothers (95% CI: 1.13-1.26). Mothers who followed recommended ANC had lower odds of LBW compared with the reference group of mothers who did not follow ANC recommendations (aOR: 0.78, 95% CI: 0.73-0.83). The odds of having LBW were eight-times higher (aOR: 8.68, 95% CI: 7.05-10.68) among children of mothers with the multiple pregnancy versus singleton pregnancy. Mothers whose height was less than 150 cm had 36% higher odds of LBW compared to mothers with height greater than 150 cm (aOR: 1.36, 95% CI 1.29-1.43).

Discussion
The programmatic factors included in the conceptual model as intermediate factors and the proximal factors were significant predictors of VLBW in India. Although the distal determinants such as the social and economic predictors were not independently associated with VLBW, they may directly influence the intermediate determinants and therefore influence      In our study girl children reported higher odds of presenting with LBW phonotypes as compared with boys. This is consistent with findings from other studies 12,13 . Male children in general have a tendency for higher birth weights and they are about 150 g heavier when compared to a female child and this difference in weight occurs often after 28 weeks of gestation 14,15 . Stunting in mothers is a significant predictor of both the LBW phenotypes. Comparison of our findings with those from other studies confirms that stunted mothers give birth to LBW child more often 16,17 and it could be related to the growth restriction of the fetus in the smaller uterus of stunted mothers 15 .
Our study showed association between birth order and LBW and age of the mother with both LBW and VLBW. Similar studies conducted elsewhere showed consistent findings related to the influence of maternal age and birth order on the birth weight of the child 18-20 . Maternal undernourishment and anaemia may have reflective effects on maternal weight gain and thereby birth weight of the child 21,22 . In our study, moderate to severe anaemia was associated with higher propensity for VLBW.
We demonstrate that educational status is an independent predictor of LBW. The odds of LBW were higher among mothers with "no education or primary level education" when compared with mothers with secondary level education. Educational level of the mother is one of the predictors of LBW in low-income countries 12,23,24 . However, we could not determine consistent association between educational status of mother with VLBW. Similarly, our study did not establish relationship between child's wealth quintile and VLBW. In contrast, a previous study from Brazil suggested an inverse association between family income with prevalence rates of VLBW 25 . Further, belonging to a Scheduled Tribe increased the odds for LBW in our study. However, no evidence for increased risk of VLBW was detected for Scheduled Tribe population in our study and this is in contrast to the previous findings 19,26 .
Similar to the results of previous studies, our study demonstrates the association between lack of appropriate ANC and LBW phonotypes [27][28][29] . Evidence suggest that social determinants of health play a major role in access to health care, especially maternal health care in India. In India, the most pertinent social determinants influencing maternal health service utilization include socio-economic status, caste/ethnicity, education, gender, and religion 30-33 . Along with the above determinants, reports from NFHS-4 also points towards the influence of lack of husband's participation in ANC and unintended pregnancies on lowering the odds for ANC utilization 33 . Furthermore, the interaction between wealth and literacy is found to have a very strong role in maternal health care utilization indicators in India 34 . The utilization of ANC and their determinants need to be explored in detail to recognize the barriers and opportunities to advance maternal health services in India.
Multiple pregnancies increased the odds of LBW and VLBW in our study. In India, there has been a progressive increase in availability of assisted reproductive technology (ART) services along with the advances in ART 35-37 . ART facilities like in vitro

Acknowledgement
We acknowledge the Demographic and Health Survey program for providing access to the data. fertilization has raised the incidence of multiple pregnancy in the country due to preference for multiple embryo transfer, which increases the chance of a pregnancy 38,39 . Additionally, maternal parity is known to influence the incidence of LBW and VLBW infants 40-43 .
Our study has some limitations. Firstly, birth weight was missing for more than 30,000 deliveries. The missing data were more from mothers who were from the marginalized communities. Mothers from lower socio-economic strata and disadvantaged population are known to have higher occurrence of LBW. Thus, our analysis could underestimate the various socio-economic factors associated with LBW in India. Secondly, information collected from the mothers on the antenatal and natal factors were from the past five years. Hence, the data quality is likely to be affected by recall bias.

Conclusion
Despite having several common risk factors with the phenotypes of LBW and VLBW, the relationship is different in both the groups. For example, the social and economic determinants are unique to LBW. The VLBW is prominently associated with several genetic, nutritional, and demographic factors. The increasing trend in rate of multiple pregnancy and its association with VLBW poses a public health concern. Taken together, our results suggest that interventions geared towards improvements in antenatal care access, maternal health and nutritional status may reduce the number of VLBW infants in India. Interventions focused on reducing the number of VLBW infants can ultimately reduce infant mortality. Further, it may reduce the future burden of cardiovascular and metabolic disease conditions that are associated with VLBW.

Data availability
Our study used data from the from individual recode file IAIR74DT, of the Demographic and Health Survey of India. The file mainly includes information on women in reproductive age group. Access to the data from DHS could be done using the

Open Peer Review
Reviewer Report 07 March 2022 https://doi.org/10.21956/wellcomeopenres.19311.r48520 © 2022 Apte A. This is an open access peer review report distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Aditi Apte
Vadu Rural Health Program, KEM Hospital Research Centre, Pune, Maharashtra, India Title -Informative and balanced summary of what was done but does not mention about study design.