Does postnatal care have a role in improving newborn feeding? A study in 15 sub–Saharan African countries

Background Breastfeeding is known as a key intervention to improve newborn health and survival while prelacteal feeds (liquids other than breastmilk within 3 days of birth) represents a departure from optimal feeding practices. Recent programmatic guidelines from the WHO and UNICEF outline the need to improve newborn feeding and points to postnatal care (PNC) as a potential mechanism to do so. This study examines if PNC and type of PNC provider are associated with key newborn feeding practices: breastfeeding within 1 day and prelacteal feeds. Methods We use data from the Demographic and Health Surveys for 15 sub–Saharan African countries to estimate 4 separate pooled, multilevel, logistic regression models to predict the newborn feeding outcomes. Findings PNC is significantly associated with increased breastfeeding within 1day (OR = 1.35, P < 0.001) but is not associated with PLFs (OR = 1.04, P = 0.195). PNC provided by nurses, midwives and untrained health workers is also associated with higher odds of breastfeeding within 1 day of birth (OR = 1.39, P < 0.001, (OR = 1.95, P < 0.001) while PNC provided by untrained health workers is associated with increased odds of PLFs (OR = 1.20, P = 0.017). Conclusions PNC delivered through customary care may be an effective strategy to improve the breastfeeding within 1 day but not to discourage PLFs. Further analysis should be done to examine how these variables operate at the country level to produce finer programmatic insight.

Does postnatal care have a role in improving newborn feeding? A study in 15 sub-Saharan African countries Background Breastfeeding is known as a key intervention to improve newborn health and survival while prelacteal feeds (liquids other than breastmilk within 3 days of birth) represents a departure from optimal feeding practices. Recent programmatic guidelines from the WHO and UNICEF outline the need to improve newborn feeding and points to postnatal care (PNC) as a potential mechanism to do so. This study examines if PNC and type of PNC provider are associated with key newborn feeding practices: breastfeeding within 1 day and prelacteal feeds.

Methods:
We use data from the Demographic and Health Surveys for 15 sub-Saharan African countries to estimate 4 separate pooled, multilevel, logistic regression models to predict the newborn feeding outcomes.
Findings: PNC is significantly associated with increased breastfeeding within 1day (OR = 1.35, P < 0.001) but is not associated with PLFs (OR = 1.04, P = 0.195). PNC provided by nurses, midwives and untrained health workers is also associated with higher odds of breastfeeding within 1 day of birth (OR = 1.39, P < 0.001, (OR = 1.95, P < 0.001) while PNC provided by untrained health workers is associated with increased odds of PLFs (OR = 1.20, P = 0.017).
Conclusions: PNC delivered through customary care may be an effective strategy to improve the breastfeeding within 1 day but not to discourage PLFs. Further analysis should be done to examine how these variables operate at the country level to produce finer programmatic insight.
Breastfeeding is recognized as a key intervention to improve the health and survival of children and the use of optimal breastfeeding practices such as exclusive breastfeeding is one of the most effective means to reduce undernutrition, an underlying cause of under-five mortality [1]. The World Health Organization (WHO) and the United National Children' s Fund (UNICEF) recommend early initiation of breastfeeding [2] which refers to breastfeeding of a newborn within an hour of birth. Global monitoring efforts by UNICEF also include initiation of breastfeeding within one day of birth which provides additional information on the feeding patterns of newborns and the behaviors of women. Early initiation of breastfeeding has a number of health benefits, one of which is to reduce neonatal mortality [3][4][5]. The early ingestion of breastmilk can have positive effects on a newborn' s immune systems such as the provision of immunoglobulins and lymphocytes [6][7][8], priming of the gastrointestinal tract and decreasing the permeability of the tract to pathogens, including HIV [9,10]. Another health benefit of early initiation of breastfeeding is reduced rates of diarrhea among infants, as demonstrated in Egypt and Pakistan [11,12].
Early initiation of breastfeeding is also associated with a number of factors. One such factor is skin-toskin contact with the mother [13,14], a form of thermal care which is a recommended means to reduce neonatal mortality [15]. Early breastfeeding is also associated with a number of factors related to contact with the health system. For example, in Brazil, early initiation is associated with vaginal delivery as well as other factors such as antenatal guidance on breastfeeding and having a full term pregnancy [16]. Other studies point out that breastfeeding within an hour of birth is less likely to occur when women have caesarian sections, even in the presence of hospital practices that favor breastfeeding [17,18]. In a review article, authors find that higher socio-economic status is associated with lower odds of breastfeeding initiation but this pattern is only seen in developing countries [19].
Prelacteal feeds (PLFs) represent a departure from optimal newborn feeding practices. PLFs are any liquid other than breast milk that is given to the newborn before breastfeeding is established between the mother and newborn. The WHO and UNICEF outline that for successful breastfeeding, PLFs should be avoided and PLFs should not be encouraged unless medically indicated [20]. These feeds usually occur within the first few days of life and are associated with a number of negative health outcomes for the newborn and mother. These include insufficient maternal milk production, newborn diarrhea and reduced length of breastfeeding duration [21,22]. PLFs can also expose newborns to infections through the ingestion of contaminated food and liquids which can act on the GI tract to increase permeability to pathogens, and hence, increase newborn infections [9,11].
A number of studies have shown factors related to PLFs. For example, PLFs are negatively associated with early initiation of breastfeeding (within an hour of birth) [23]. In India, PLFS were associated with lower maternal education among hospital-delivered infants [24]. However, in rural, Western Uganda, more educated women were more prone to provide PLFs to newborns [25]. In low socio-economic settlements in Karachi, Pakistan, PLFs were associated with having a birth attendant [26]. In a national study in Nepal, women without education, who were not working, who had no antenatal care and were first time mothers were more likely to provide PLFs [27]. Both in India and Vietnam, newborns of women with a cesarean section were more likely to ingest PLFs [24,28].
In a recent joint statement, the WHO and UNICEF recommend that all newborns, regardless of place of birth (whether in a facility or not), should receive a basic package of care, including postnatal care which includes the promotion and support of exclusive breastfeeding and the early initiation of breastfeeding [29]. Interventions such as thermal care, hygienic cord care, examination for danger signs and improving parental knowledge of care seeking are also recommended. The evidence on the importance of PNC from developing countries comes mainly from South Asian countries (India, Bangladesh and Pakistan) and are from interventions and trials at sub-national levels (such as districts, villages and communities) [30][31][32].
Currently, there is a gap in the literature on how interventions such as PNC are associated with newborn feeding practices at the national level, when delivered through usual services of the government and nongovernmental sources of care ie, outside of an intervention setting. The literature is especially sparse for sub-Saharan Africa. The only study we found was a small, cross-sectional study in Ethiopia [33] where PNC was associated with increased odds of timely initiation of breastfeeding. Apart from the issue of generalizability of PNC interventions, we currently do not know which type of provider of PNC is best suited to improve the newborn feeding outcomes. The WHO-UNICEF PNC recommendation acknowledges that skilled and unskilled health workers can provide PNC though skilled providers are better suited [29]. However, in the literature on newborn feeding, we find varying opinions on if skilled or unskilled care can improve breastfeeding. In Bangladesh, for example, specially trained peer counselors can improve initiation and duration of exclusive breastfeeding [34]. However, a literature review finds that trained health care workers (physicians, nurses etc.) were found to be a barrier to providing quality information, counseling and care to women on early breastfeeding [19].
The main objective of this paper is to examine the association between PNC within 1 day and two key newborn feeding practices: breastfeeding within 1 day and prelacteal feeds. Given that WHO-UNICEF recommends both skilled and unskilled health workers to provide PNC and that there are mixed results regarding the association of provider type on newborn feeding, we also examine if the type of provider of PNC is important for the two stated outcomes. We use data from nationally representative surveys in 15 sub-Saharan African countries in a pooled, multi-level analysis, controlling for a number of individual and country-level variables. The results of this paper can provide indications on which types of providers are best suited for the delivery of PNC as it relates to newborn feeding.

Data and variables
Data for this study are from the USAID-supported Demographic and Health Surveys (DHS). DHS surveys collect data from nationally-representative probability samples of households. Households are selected using a two-stage sample design where census enumeration areas are first selected and then a random sample of households is selected in the second stage. Within selected households, all women ages 15-49 are interviewed and provide information on themselves and their children on various health, population and nutrition issues. Women also provide informed consent to the survey prior to the start of questions. All data are anonymized. This analysis focuses on the last birth in the last two years before the surveys for which information on PNC is provided. We include There are two outcome variables. The first is the percentage of newborns who were breastfed within 1 day of birth among all newborns. The second outcome variable is the percentage of newborns who received a PLF ie, a feed that occurs within 3 days of births that is not breastmilk. The measure of PLFs is based on asking the mother if, within the first 3 days after delivery, the newborn was given anything to drink, other than breast milk. This is only asked for newborns who were ever breastfed.
The key independent variable is PNC within 1 day which refers to any check within 1 day to a newborn following birth. The question also provides examples of what a check may entail (checking temperature, cord etc.). We exclude a check by 'others' (such as friends or relatives as these are not likely to be medical). Women were also asked, if for the last birth in the 2 years before the survey, what provider or traditional birth attendant performed the check on the newborn' s health. Qualitative work confirms that women are able to tell coherent narratives about the moments around birth and recognize checks on the health of a child [35]. Given that PLFs can occur anytime within 3 days, we attempt to establish PNC preceding PLFs by defining PNC as a check within 1 day of birth instead of 3 days. Both of these outcomes are binary. To investigate if PNC provider is associated with the outcomes, we create a variable for PNC provided by three categories of caregivers: physicians, nurses/midwives/auxiliary midwives and finally, traditional birth attendants/community health workers/other.
In our models, we introduce a number of statistical controls based on the literature, classified as individuallevel controls or country-level controls. We include: age of the mother, previous birth interval, parity, caesarian section of birth, use of antenatal care (ANC), receipt of tetanus toxoid vaccination, skilled delivery, educational level of the woman, marital status, media access (regular access to print and mass media), place of residence and a wealth index of household goods and assets (provided in the DHS data files), constructed using Principal Component Analysis of household-level ownership of goods and assets.
We include 4 binary, country-level variables to account for the variation in the supply of PNC. The five country-level variables are: Gross Domestic Product (GDP) per capita ("high" when US$ 1000 or greater per capita or "low" when below US$ 1000 per capita), per capita government expenditure on health ("high" when US$ 100 or greater per capita and "low" when below US$ 100 per capita), number of physicians per 1000 population ("high" when the value is 0.1 or greater and "low" when the value is below 0.1) and finally, the number of nurses per 1000 population ("high" when the value is 1 and greater and "low" when the value is less than 1). Finally, since there are prominent recommendations on newborn feeding practices in areas of high HIV prevalence, we included a dummy variable for HIV prevalence ("high" when 5% or greater and "low" when less than 5%) as an explanatory variable in the models.

Statistical analysis
We use descriptive statistics and multivariate models to examine the association between the main predictors and the outcomes. First, we describe the sample using frequencies of the variables and then produce cross-tabulations of key variables by the outcome variables using chi-square tests. Finally, we model the outcome variables on the key variables (in separate models), with a number of statistical controls. Univariate analysis is done at the country level to provide an indication of the contribution of each country to overall sample but as the aim of the analysis is cross-country, the remainder of the analysis is done at the aggregate level.
As breastfeeding within 1 day and PLFs are binary outcomes, a logistic regression model can be used, assuming that the error term follows a logistic distribution. However, as we study individual-level data from different countries, this suggest that these data are clustered and as a consequence, a multilevel model may be required (MLM). To verify if MLM is needed, we compared all MLMs to single level logistic regressions using a Likliehood-ratio (LR) test. These results should that the data are clustered at the country level and that MLMs perform better than the single-level logistic regressions. In our models, country-level variance was between 4 to 11 percent. Multilevel models and bivariate table are run without sample weights while univariate are weighted using DHS sample weights provided in datafiles.

RESULTS
Breastfeeding within a day of birth is high (81 percent) and varies considerably across the countries, ranging from 66 percent in Cote d'Ivoire to 94 percent in Mali ( Table 1). Levels of prelacteal feeds are lower (39 percent overall), ranging from 11 percent in Namibia to 65 percent in Cote d'Ivoire. PNC is low overall; only 15 percent of the sample received PNC within a day, of which the vast majority was provided by a nurse (12 percent) and only 2 and 1 percent provided by physicians and by traditional birth attendants/ community health workers/others (TBA/CHWs/others) respectively. In the sample, about half of the women had 3 or fewer children. Caesarian sections are uncommon (4 percent). More than half of the women had contact with the health system through ANC care (52 percent), receipt of tetanus toxoid (56 percent) and had a skilled delivery (62 percent). The majority of the sample is married, has no education, no regular access to media and about 40 percent is classified into the poorest or second lowest wealth quintiles.
In 7 of the 15 countries, newborns who receive PNC were more likely to be breastfed within 1 day compared with newborns who did not receive PNC but in several countries (eg, Comoros, Congo (Brazzaville), Uganda), the opposite occurs (Figure 1). Figure 2 shows that while overall newborns receiving          PNC are significantly less likely to receive a prelacteal feed, patterns by country vary considerably; 5 countries show a statistically significant relationship but 4 show the opposite pattern.
In the bivariate analysis, newborns receiving PNC within 1 day are significantly more likely to initiate breastfeeding within a day and less likely to receive a prelacteal feed ( Table 2). Women receiving antena- tal care, tetanus toxoid and skilled delivery are significantly more likely to breastfeed within a day and less likely to provide a prelacteal feed to the newborn. A caesarian birth is significantly associated with breastfeeding within 1 day but not with PLFs. Women with no education are less likely to breastfeed early and more likely to provide a prelacteal feed. While household wealth is positively associated with breastfeeding within 1 day, the association is negative with prelacetal feeds. Women in urban areas are more likely than rural women to initiate breastfeeding within 1 day and less likely to give a prelacetal feed. Bivariate analysis of the country-level variables also shows lower levels of GDP, expenditures, and physician and nurse density are associated with greater initiation of breastfeeding within 1 day and lower levels of prelacteal feeds. In countries with higher HIV prevalence, breastfeeding within 1 day is higher and prelacteal feeds are lower.             Table 3 shows that after controlling for individual and country-level variables, PNC within 1 day is not significantly associated with prelacteal feeds (OR = 1.04, 95% CI 0.98-1.09). Age is significantly associated with the outcome in the model with older women tending to have lower odds of providing prelacteal feeds to newborn while birth spacing and parity were not associated with prelacteal feeds. Newborns who had a Caesarian section delivery are significantly more likely to have PLFs (OR = 1.60, 95% CI 1.46-1.76). Contact with the health care system through ANC, tetanus toxoid vaccination and skilled delivery are significantly associated with lower odds of prelacteal feeds (see Table 3). For example, skilled delivery is associated with a 42% reduction in odds of prelacteal feeding (OR = 0.58, 95% CI 0.56-0.61). Education shows a clear gradient with prelacteal feeds; as the educational level of the woman increases, the odds of prelacteal feeding decreases (see Table 3). Of the country-level characteristics, only the density of physicians is significantly associated with prelacteal feeds in the models: higher density of physicians is associated with higher odds of prelacteal feeds (OR = 2.26, CI 1. 22-4.17). In model 4 of Table 3, the type of provider of PNC is not associated with prelacteal feeds. Other results remain similar to model 2 of the second panel of Table 3.

DISCUSSION
PNC is one of the current strategies recommended for scale-up and implementation in many developing countries to improve health outcomes for newborns and mothers. While several trials and intervention studies show that PNC can improve newborn feeding patterns [30][31][32], this is the first study to demonstrate this association using national-level data for multiple countries in sub-Saharan Africa.
The major findings are that PNC is associated with breastfeeding within 1 day though not with prelacteal feeds. These findings are important as they suggest that PNC when delivered through customary care (as opposed to intervention and trial conditions) can be a useful strategy to improve breastfeeding (within 1 day) but not to reduce PLFs. These findings highlight the need to strengthen clinical practice so that providers of PNC can move beyond promoting timely initiation of breastfeeding to providing more emphasis on the avoidance of PLFs, which by definition would improve exclusive breastfeeding rates in these countries.
Our findings also indicate that both trained medical personal (nurses, midwives and auxiliary midwives) and untrained providers of PNC are associated with increased odds of breastfeeding within 1 day though the type of provider of PNC is not associated with PLFs. Given that all of the countries that we studied are developing countries, use of untrained persons for this type of intervention may be a useful implementation approach as the promotion of optimal newborn feeding does not require high levels of specialized training A third important finding from this study is that, with the exception of caesarian section, contact with the formal health care system is associated with improved newborn feeding practices. This is seen in other studies eg, Nepal [27] and India [24]. This underscores the utility of the continuum of care and reinforces the need to implement around this framework. Delivery mode by caesarian section, however, is associated with poorer newborn feeding outcomes, a finding that is reflected in a number of other studies [16,24,28,36,37], even in the presence of baby-friendly policies [17].
Our study has a number of limitations. DHS data do not include any information on what procedures were done during a check and therefore cannot control for content of care. We also use cross-sectional data where PNC was not randomly assigned to individuals. As such, we are not able to provide causal linkages between PNC and the outcomes though we are able to examine associations. One of the more studied variables on breastfeeding initiation is breastfeeding within 1 hour of birth. With our data, we could study the association of PNC within an hour and breastfeeding within the same time period. However, we considered that a short time period of 1 hour does not provide sufficient time for PNC to be provided (given that in these settings, even PNC within 1 day is low). The literature also identifies a number of additional factors that predict early initiation of breastfeeding and PLFs which were not available for analysis. For example, intention to breastfeed [38] is an important predictor of initiation and duration of breastfeeding but was not available in DHS data. Dealing with sample weights is a challenge for analysis of this kind. Different countries contribute varying proportions of the overall sample and do not reflect the relative population size of the country. Appropriate sample weight can be constructed though the sample weights must be de-normalized. However, the appropriate sampling fraction for each country and their population sizes used to create these weights are not publicly available.
Despite these limitations, our findings are consistent with trials and intervention studies, and overall, PNC policy and practice can be further tailored to reduce PLFs rates. Further research at a country-level is needed to understand if the results of this aggregate, multi-country study are reflected within each of these countries.

Disclaimer:
The findings and conclusions in this report are those of the authors and do not necessarily represent the official position of their respective organizations.