Prevalence of low birth weight and its associated factor at birth in Sub-Saharan Africa: A generalized linear mixed model

Background Low birth weight (LBW) is one of the major determinants of perinatal survival, infant morbidity, and mortality, as well as the risk of developmental disabilities and illnesses in future lives. Though studies were conducted to assess the magnitude and associated factors of low birth weight, most of the studies were at a single center and little information on the regional level. Hence, this study assessed the prevalence and associated factors of low birth weight in Sub-Saharan countries. Method This study was based on secondary data sources from 35 Sub-Saharan countries’ Demography and Health Survey (DHS). For this study, we used the Kids Record (KR file) data set. In the KR file, all under-five children who were born in the last five years preceding the survey in the selected enumeration area who had birth weight data were included for the study. To identify determinants of low birth weight multivariable mixed-effect logistic regression model fitted. Adjusted Odds Ratios (AOR) with a 95% Confidence Interval (CI) and p-value ≤0.05 in the multivariable model were used to declare significant factors associated with low birth weight at birth. Result The pooled prevalence of newborn babies’ low birth weight measured at birth in Sub-Saharan Africa was 9.76% with (95% CI: 9.63% to 9.89%). Female child, women not participated in healthcare decision making, and wider birth intervals, divorced/ separated women, and twin pregnancies associated with increased occurrences of low birth weight, while some level of woman and husband education, antenatal care visits, older maternal age, and multiparity associated with reduced occurrence low birth weight. Conclusion This study revealed that the magnitude of low birth weight was high in sub-Saharan Africa countries. Therefore, the finding suggests that more emphasis is important for women with a lack of support, multiples, and healthcare decision-making problems.


Preface
The Demographic and Health Surveys (DHS) Program is one of the principal sources of international data on fertility, family planning, maternal and child health, nutrition, mortality, environmental health, HIV/AIDS, malaria, and provision of health services.
One of the objectives of The DHS Program is to continually assess and improve the methodology and procedures used to carry out national-level surveys as well as to offer additional tools for analysis. Improvements in methods used will enhance the accuracy and depth of information collected by The DHS Program and relied on by policymakers and program managers in low-and middle-income countries.
While data quality is a main topic of the DHS Methodological Reports series, the reports also examine issues of sampling, questionnaire comparability, survey procedures, and methodological approaches. The topics explored in this series are selected by The DHS Program in consultation with the U.S. Agency for International Development.
It is hoped that the DHS Methodological Reports will be useful to researchers, policymakers, and survey specialists, particularly those engaged in work in low-and middle-income countries, and will be used to enhance the quality and analysis of survey data.

Sunita Kishor
Director, The DHS Program ix Abstract This paper investigates potential biases in analysis of data for only last-born children or non-last-born children instead of all children, and in the analysis of succeeding birth or pregnancy intervals, in both the five-year period preceding the Demographic and Health Surveys (DHS) surveys and for longer periods of time. Beyond the usual considerations of omission of births and transference of births across questionnaire age boundaries, using data for only last-born children or non-last-born children instead of for all children born in the five years preceding the survey can result in biased research findings. Unfortunately, for certain child health outcomes, some DHS data are only collected for last-born children, and other data only for non-last-born. The correction of one bias may create another bias, as in the estimation of mortality risks by succeeding interval. Given the likely negative effects on the health and well-being of children whose birth is followed quickly by another birth, substantial efforts should be made to try to overcome these biases. xi

Executive Summary
Many studies have shown the deleterious relationships between the length of preceding inter-birth or inter-pregnancy intervals and the pregnancy, birth, and child following the interval. The conception and birth of another child after a short period of time would also be expected to negatively affect the earlierborn child. These considerations have led researchers to attempt to use data from The Demographic and Health Surveys (DHS) Program to investigate health outcomes linked to the subsequent birth or pregnancy interval. Some of the outcomes of interest have data collected only for the five years preceding the survey (e.g. current nutritional status and vaccination rates), while others can use earlier data (for example, child mortality). However, the nature of the problem and the structure of DHS data can lead to serious biases in such analyses. It is the purpose of this Methodological Report to describe and illustrate some of these biases.
In 45 recent DHS surveys, of the nearly half a million children born in the five years preceding the survey, over 70 percent are the most recently born (last birth). The demographic and social characteristics of last-born children differ from those of all children born in the five-year year period preceding the survey. Last-born children are slightly less likely than non-last-born children to be of low birth order. Forty-seven percent of last-born children are of birth orders 1 or 2 compared with 51 percent of non-lastborn children. Last-born children are more likely to be male than the non-last-born, 51.6 versus 50.0 percent, respectively. Last-born living children tend to be younger than non-last-born children: about 24 percent of last-born children are age 3-4 compared with 70 percent of the non-last-born.
Mother's age at birth is about two years older for last-born children compared with non-last-born children. Surprisingly, preceding birth-to-birth intervals are much longer for last-born children, with a mean 9.5 months greater (for children of birth order 2 or higher). Comparing last-born children with nonlast-born children for social characteristics, there are also surprising differences. Non-last-born children are more likely to live in rural areas, to have been born at home, to have mothers with less education, and to come from somewhat poorer households.
The study of the effects of succeeding inter-pregnancy intervals on children's nutritional status requires that the children be born within the five years preceding the survey, and the next child (if any) would need to be conceived between the date of the index child's birth and the date of the interview. For an index child age 54 months at the time of the survey, for example, the maximum subsequent interval would be approximately 43 months; for an index child age 18 months the maximum subsequent interval would be about 7 months. Thus there is a strong relationship between the age of the index child and the possible length of the subsequent interval.
The older the index child, the longer the inter-pregnancy interval can be, with the mean length rising from 10 to 32 months. On the other hand, chronic nutritional status and underweight increase with the age of children, up to about age 24 months. Therefore, younger children have shorter subsequent intervals on average and also are less likely to have chronic undernutrition, creating a strong bias in the analysis of the relationship between subsequent interval length and chronic nutritional status.
The study of under-5 mortality in relation to the length of the succeeding intergenesic interval is fraught with potential bias from a shortened duration of breastfeeding and the desire to replace a child who died. Excluding all deaths of children who die before the next child is conceived can control for these biases. However, this control could introduce another source of bias: the longer the succeeding birth-toconception interval, the more deaths and deaths at older ages are excluded. The percent of under-5 deaths prior to a succeeding conception that are excluded from analyses increases as the interval lengthens.
xii Slightly more than half of deaths are excluded when the interval is less than six months, increasing to 83 percent for reference intervals of 36-47 months, and up to 96 percent for intervals of eight or more years.
Beyond the usual considerations of omission of births and transference of births across questionnaire age boundaries, using data for last-born children or non-last-born children instead of data for all children born in the five years preceding the survey can result in biased research findings. Unfortunately, some DHS data are only collected for last-born children, and other data only for non-last-born children. The correction of one bias may create another bias, as in the estimation of mortality risks by succeeding interval.
Further thought and research are needed on how to overcome these biases for analysis of last children and succeeding intervals. Given the likely negative effects on the health and well-being of children whose birth is followed quickly by another, substantial efforts should be made to try to overcome these biases.

Introduction and Purpose
Many studies have shown the deleterious relationships between the length of preceding inter-birth or inter-pregnancy intervals and the pregnancy, birth, and child that follows the interval. Outcomes that have been investigated include miscarriage and stillbirth; complications of pregnancy and delivery; low birth weight; neonatal; infant; and child mortality; and poor nutritional status.
In addition to the effects of preceding birth interval on child survival, studies have documented a strong relationship between the length of the preceding birth interval and chronic and general undernutrition such that nutrition outcomes are poorer for children with shorter preceding birth intervals (Dewey and Cohen 2007;Rutstein 2005). A recent study by Rutstein and Winter (2014) demonstrated the deleterious effects of combinations of preceding intergenesic intervals, birth order, and mother's age at birth on childhood mortality and nutritional status.
The conception and birth of another child after a short period of time would also seem to negatively affect the earlier-born child at the start of the intergenesic interval. The earlier child's breastfeeding could be cut short, the mother may not have a chance to recover from the earlier pregnancy, birth, and breastfeeding, and care and attention given to the earlier child may be reduced. These considerations have led researchers to attempt to use data from the Demographic and Health Surveys (DHS) to investigate health outcomes linked to the subsequent birth or pregnancy interval. However, the nature of the problem and the structure of DHS data can lead to serious biases in these studies. It is the purpose of this paper to describe and illustrate some of these biases, which are hard to overcome in these kinds of data sets. Appendix Table A1 describes the types of data that are available for children and their restriction by age (or time since birth), birth order, survival status, and other data type.
First, the use of information pertaining to the most-recently-born (last-born) child in the five years preceding the survey is compared with that of all children born in the most recent five-year period, and with children born in the fifteen-year period preceding the survey. Also, children born in the five years preceding the survey with a younger sibling or with the conception of a younger sibling (i.e., those with a succeeding interval) are compared with all children born in the most recent five-year period. Second, factors affecting the nutritional status of children under age 5 that are related to succeeding intervals are studied to evaluate their possibilities of causing bias in studies relating succeeding intervals to the nutritional status of children under age 5. Third, biases that may be present in the relationship between mortality of children under age 5 and succeeding intervals are examined.

Characteristics of Most-Recently-Born Children and Non-Most-Recently-Born Children Compared with All Children Born in the Most Recent Five-Year Period
The demographic and social characteristics of most-recently-born children differ from those of all children born in the five-year year period preceding the survey. Studied here are birth order, sex, age at time of survey, mother's age at birth, length of the preceding birth interval, survival of the preceding child, area of residence, delivery in a health facility, mother's level of education, and household wealth. Table 1 presents selected demographic and social characteristics of children born in the five years preceding the survey and alive at the time of the survey by whether the child is the most recently born. Of the near half a million children born in the preceding five years in 45 recent DHS surveys, 1 over 70 percent are the most recently born (the last birth). These children differ somewhat in many ways from under-5 children who are not the last birth. However, due to the large representation of last-born children among all under-5 children, differences between the last-born and all children are lower. Therefore, the discussion will focus on the differences between last-born and non-last-born children.
Last-born children are slightly less likely than non-last-born children to be of low birth order. Forty-seven percent of last-born children are of birth orders 1 or 2 compared with 51 percent of non-last-born children.
Mean birth order is 0.2 children higher among the last-born. Last-born children are more likely to be male than the non-last-born, 51.6 versus 50.0 percent, respectively. This result is surprising in that the sex ratio of non-last-born children is 100. Somewhat less surprising is that there are more males among the lastborn (sex ratio of 107). This difference may be due to a preference for male children among couples who stop having children. For all children under age 5, the sex ratio is 105, the expected sex ratio at birth.
Last-born living children tend to be younger than non-last-born children. While there is an approximate balance of all children under age 5, at around 20 percent for each single year of age, only about 24 percent of last-born children are age 3-4 compared with 70 percent of non-last-born children. This striking difference may cause substantial bias when analyses select just last-born or non-last-born children instead of all children. The time since birth also shows that non-last-born children were born about 1.7 years earlier than last-born children. The difference between the current age and time since birth distributions is due to the greater percentage of deaths among non-last-born children, at 11 percent versus 4 percent for last-born children.
The age at birth of the mother is somewhat older for last-born children, by about two years, compared with that of the mothers of non-last-born children. Forty-three percent of last-born children were born to mothers under age 25 compared with 55 percent of non-last-born children.
Preceding birth-to-birth intervals are surprisingly much longer for last-born children, with a mean 9.5 months greater (for children of birth order 2 or higher). This difference may also bias results of analyses if only one of the two groups of children is taken to represent all children under age 5. 2 However, the percentage of children whose next older sibling died is not much different between the two groups.
Comparing last-born children with non-last-born children for social characteristics, there are surprising differences. Non-last-born children are more likely to live in rural areas, are more likely to have been born at home instead of in a health facility, have mothers with lower levels of education, and come from somewhat poorer households. The differences between last-born children and non-last-born children may be partly due to differences in fertility levels between countries and areas within countries, as in higher fertility areas (such as rural areas) women are more likely to have had more than one child in the five years preceding the survey.
It is therefore seen that selecting just last-born children or just non-last-born children among the children under age 5 may result in several sources of bias if it is assumed that the results would be the same as for all children. Due to the preponderance of last-born children among all children, the biases in using lastborn children are less than those in using non-last-born children. However, the type of analysis undertaken and data available may require using one group or the other instead of all children under age 5. For example, DHS now collects many prenatal and postnatal care indicators only for last-born children.
Given that far fewer births of non-last-born children are delivered in a health facility, biased conclusions may result. Dietary information is collected for almost no non-last-born children. If these children were weaned early, their nutritional status could be different from those of last-born children.

Nutritional Status of Children under Age 5
The study of the effects of succeeding inter-pregnancy intervals looks at children's nutritional status according to the length of time between the index child's birth and the conception of the next live-born child, if any. Given that children studied need to have been born within the five years preceding the survey, the next child (if any) would need to be conceived between the date of the index child's birth and the date of interview. For an index child age 54 months, the maximum subsequent interval would be approximately 43 months; for an index child age 18 months, the maximum subsequent interval would be about 7 months. Thus there is a very strong relationship between the age of the index child and the length of the subsequent interval. Table 2 demonstrates this relationship. The older the index child, the longer can be the inter-pregnancy interval, with the mean length rising from 10 to 32 months. On the other hand, chronic nutritional status and underweight increase with the age of children up to about age 24 months. In Figure 1, children's height-for-age z-score decreases with increasing age until about age 20 months. In Figure 2 the percentage of children who are stunted increases with the child's age up to about 24 months. Therefore, younger children have shorter subsequent intervals on average and also have less chronic undernutrition, creating a strong bias in the analysis of the relationship between subsequent interval length and chronic nutritional status.  Table 3 below summarizes the confounding that occurs between succeeding interval and chronic undernutrition by age of the index child. Children younger than 18 months have no or almost no succeeding births. For children 18 months or older, the proportion with a succeeding birth increases rapidly, as do the proportions for children with a succeeding birth interval between 12 and 23 months and between 24 and 35 months. However, there are no birth intervals 48 months or longer. The mean interval length also increases rapidly with the child's age. As noted above, the average height-for-age z-score decreases and the percentage stunted increases with the child's age, creating a serious confounder for examining the relationship between succeeding interval and chronic nutritional status. Another serious confounder is breastfeeding. Unless they are postpartum abstinent or using postpartum contraception, mothers who stop breastfeeding are more likely to have a short succeeding interval. The nutritional impact will depend on what age the weaning occurred and what foods the child received after weaning. On the other hand, women who become pregnant while breastfeeding are very likely to stop, and so short succeeding intervals may cause early weaning. Most DHS surveys do not possess data on the duration of breastfeeding of the index child with a succeeding interval, and the early DHS surveys that asked about duration of breastfeeding of non-last-born children had data that were severely heaped, making them useless for analysis.

Under-5 Mortality by Succeeding Birth-to-Conception Interval
The study of under-5 mortality in relation to the length of the succeeding intergenesic interval is fraught with potential for bias. One such bias relates to breastfeeding. Children who stop breastfeeding early in life face additional mortality risks due to malnutrition and increased prevalence of disease, including diarrhea and acute respiratory infection, especially in low-income settings where milk substitutes are inadequately given and quality health services may not be frequently used. Shortened durations of breastfeeding lead to the early return of postpartum fecundity and are a cause of a short time until the next pregnancy where adequate postpartum contraception and/or abstinence are not practiced. Children who die at young ages also lead to shorter periods of postpartum lactational amenorrhea and thus short succeeding intervals. Another potentially confounding factor is the desire to replace the child who died.
Families may want to have another child as soon as possible when a young child dies and thus may not use contraception or prolong periods of postpartum abstinence.
One way to control for these breastfeeding and replacement-desire effects is to exclude all deaths of children who die before the next child is conceived. A child who dies before the next is conceived cannot have been affected by that conception. However, this control could be introducing an additional source of bias: the longer the succeeding birth-to-conception interval, the more deaths and deaths at older ages are excluded. Given that the risk of mortality declines as age at exposure increases, there is a double lowering of mortality of longer intervals compared with shorter intervals. This phenomenon may be the cause of the following results. Figure 3 shows the adjusted risk of child death by birth-to-conception interval relative to that of the reference interval (36-47 months), using pooled data from 52 DHS surveys for the 15-year period preceding the survey. The relative risk ratio has been adjusted by including a host of characteristics in a Cox hazard regression. 3 Due to the large selection bias that underestimates children's deaths in the reference interval and longer intervals, the relative risks of dying for children with shorter intervals are too high when compared with those of the reference period, giving a misleading impression of the mortality effects of short succeeding intervals. Table 4 shows the percentage of under-five deaths prior to conception of the succeeding child that are excluded from analyses, by succeeding interval for children born in the 15-year period preceding the survey. When the interval is less than six months, slightly more than half of under-5 deaths that occurred prior to conception of the succeeding child are excluded (column 4). These deaths would be mainly neonatal deaths. As the interval lengthens, the percent of under-5 deaths that are excluded increases, reaching 83 percent for the 36-47 month reference interval, and up to 96 percent for intervals of eight years (96 months) or more.
Another way of viewing this bias is the ratio of deaths by interval length to that of the reference category. For an interval of less than 6 months between birth and the succeeding conception, the proportion of deaths included is almost three times higher compared with the reference category, and twice as high for intervals of 12 to 17 months (column 5).

Conclusions and Further Research Needed
When researchers use the DHS surveys for the analysis of children born in the five years preceding the survey, they must be very careful to consider and avoid inherent biases that derive from both the data structure and the nature of the analysis. Beyond the usual considerations of omission of births and transference of births across questionnaire age boundaries, particularly the health section and calendar age/date boundaries, using data for last-born children or non-last-born children instead of all children born in the five years preceding the survey will probably result in biased research findings. Unfortunately, some DHS data are only collected for last-born children, for example, children's dietary intake and several child health questions. Other types of research call for using non-last-born children, such as the analysis of current nutritional status by succeeding birth or pregnancy interval, because survey-measured nutritional status of children under age 5 is only relevant for those born in the last five years who have a succeeding interval-that is, non-last-born children under age 5. Also presented is the case that the necessary correction of one bias may incur the creation of another type of bias, as in the estimation of mortality risks by succeeding interval.
Further thought and research are needed on how to overcome these biases for analysis of last-born children and succeeding intervals. Possible solutions are to obtain information on all children instead of just the last-born, collect past nutritional status information, as the Peru Continuous DHS survey does from health cards, and collect follow-up information using cohort panels. Other analytical techniques may also be helpful such as differential weighting, instrumental variable and two-stage regressions, 4 and timevarying covariation for death and next conception. Given the possibly great effects on the health and wellbeing of children whose birth is followed quickly by that of another, substantial efforts should be made in trying to overcome these biases. Preceding and succeeding birth interval, age of mother at birth All