Population attributable risk of key modifiable risk factors associated with non-exclusive breastfeeding in Nigeria

Background: Non-exclusive breastfeeding (non-EBF) is a risk factor for many of the 2300 under-five deaths occurring daily in Nigeria – a developing country with approximately 40 million children. This study aimed to quantify and compare the attributable burden of key modifiable risk factors associated with non-EBF in Nigeria to inform strategic policy responses and initiatives. Methods: Relative risk and exposure prevalence for selected modifiable risk factors were used to calculate population attributable fractions based on Nigeria Demographic and Health Surveys data for the period (1999-2013). Scenarios based on feasible impact of community-based interventions in reducing exposure prevalence were also considered to calculate comparative potential impact fractions. Results: In Nigeria, an estimated 22.8% (95% Confidence Interval, CI: 9.2-37.0%) of non-EBF was attributable to primary and no maternal education; 24.7% (95% CI: 9.5-39.5%) to middle and poor household wealth, 9.7% (1.7-18.1%) to lower number (1-3) and no antenatal care visits; 18.8% (95% CI: 6.9-30.8%) to home delivery and 16.6% (95% CI: 3.0-31.3%) to delivery assisted by a non-health professional. In combination, more than half of all cases of non-EBF (64.5%; 95% CI: 50.0-76.4%) could be attributed to those modifiable risk factors. Scenarios based on feasible impacts of community-based approaches to improve health service access and human capacity suggest that an avoidable burden of non-EBF practice of approximately 11% (95% CI: -5.4; 24.7) is achievable. Conclusion: Key modifiable risk factors contribute significantly to non-EBF in Nigerian women. Community-based initiatives and appropriate socio-economic government policies that specifically consider those modifiable risk factors could substantially reduce non-EBF practice in Nigeria.

Relative risks and prevalence data used to calculate the PAF for selected modifiable risk factors were based on an analysis of nationally representative data, the Nigeria Demographic and Health Survey (NDHS) for the years 1999 (N = 8199), 2003 (N = 7620), 2008 (N = 33,385) and 2013 (N = 38,948). The surveys were conducted by the National Population Commission (NPC) and Inner City Fund (ICF) International using a multi-stage stratified sampling technique. Sample sizes were selected from the 1991 (1999 and 2003 NDHS) and 2006 (2008 and 2013 NDHS) census frames. The increase in sample size in 2008 and 2013 reflects growth in the Nigerian population and an inclusion of additional sets of survey questions. Additional information on the data sources (including data collection techniques) have been described elsewhere [28–30]. We have provided estimates for the period spanning (1999–2013) because (i) the high prevalence of non-EBF has remained unchanged between 1999 and 2013, despite the introduction and implementation of infant and young child feeding policies [6]; (ii) the period captures a time of stable political environment in Nigeria, after more than a decade of authoritarian regimes; and (iii) this study provides additional evidence on strategic initiatives to reduce rate of non-EBF in Nigeria, consistent with previous reports [31, 32]. The stable political system in Nigeria was associated with an increase in health care financing and heightened maternal and child health programmes [7]. Exclusive breastfeeding was defined as infants aged 0–5 months who received breast milk as the only source of nourishment, but allowed oral rehydration solution, drops or syrups of vitamins and medicines [33]. In the analysis, non-exclusive breastfeeding (non-EBF) was the main outcome and was expressed as a dichotomous outcome (that is, respondents who exclusively breastfed were coded as ‘0’ and those who did not were coded as ‘1’). The exposures considered in the analysis included a range of socio-economic and health service factors, and the selection of these factors was based on evidence from previously published studies from Nigeria [5, 10]. Socio-economic risk factors included the mother’s highest educational level (categorized as no education, primary or secondary and above education) and household wealth index (categorized as poor, middle or rich). The household wealth index was calculated as a measure of household assets such as ownership of transportation devices, ownership of durable goods and household facilities, which was derived from a principal components analysis conducted by the National Population Commission (NPC) and ICF Macro [28, 29] and was used in previous studies [5, 13]. Health service factors included the number of antenatal clinic visits (categorized as no antenatal visit, one to three antenatal visit or four and above antenatal visits, reflecting the WHO four-visit ANC model for focused antenatal care) [34] and the place of delivery (categorized as health facility or home). The type of delivery assistance received was assessed, and was categorized as either a health professional, traditional birth attendant or untrained personnel. A traditional birth attendant is usually a woman who assists the mother during childbirth and who initially acquired her skills by delivering babies herself, or by working with other traditional birth attendants [35]. The prevalence estimates of these key modifiable risk factors were used to calculate the PAF. The total number of non-exclusive breastfeeding cases were examined over the study period (1999–2013), stratified by socio-economic and health service variables to determine the absolute number of cases of non-EBF within the study population. Prevalence estimates, and calculation of standard errors (for calculation of 95% confidence intervals) were adjusted using sampling weights to account for the cluster sampling design used in the NDHS [28, 29]. Using multi-level regression models to estimate relative risk, relative differences between study factors were investigated over the study period (1999–2013) [Table 1]. Study variables included socioeconomic factors (maternal education and household wealth index) and health service factors (place of delivery, frequency of antenatal visits and delivery assistance). Multi-level regression models adjusted for the potential confounding factors of geopolitical region, maternal age, birth interval and sex of the baby as employed in previous studies [5, 10]. Period was specified in the analyses as a categorical variable, and to reduce recall bias, the analyses were restricted to the youngest living children aged less than 24 months, living with mothers (aged 15–49 years). Multi-level adjustments to estimate the relative risk were conducted using generalized linear latent and mixed models (gllamm) method to account for clustering of individuals within geographic areas [36]. All analyses were carried out in Stata version 13.0 (Stata Inc., College Station, TX, USA) with weighted prevalence estimates calculated using the ‘svy’ function to allow for cluster sampling. Relative risk, exposure prevalence for PAF, PIF and estimated cases of non-EBF due to the exposures in Nigeria, 1999–2013 Cases: cases of non-EBF; non-cases (control): cases of EBF (a) Proportion is percent of exposed non-cases, an estimate of the exposure in the population. That is, non-cases as a proportion of total non-cases PAF: Population attributable fraction; PIF: Potential impact fraction For the calculation of the PIF: assumption of continued improvements in high-school completion rates in women [39] was made; impact fractions for maternal education were estimated assuming a 5% relative decrease in the proportion of women not completing high school from 23% to 22%. For antenatal visits, impact fractions were estimated assuming a reduction of 17% in women having no antenatal care (from 22% to 5%) [31, 32]. For delivery assistance and place of delivery, impact fractions were estimated assuming a relative reduction of 15% based on community-based interventions to improve exclusive breastfeeding practice [31]. No alternative scenario was defined for household wealth because of a lack of data relating to interventions resulting in income re-distribution In this study, the population attributable fraction (PAF) describes the proportion of non-EBF in Nigeria that could hypothetically be prevented if the modifiable risk factors were reduced in the population. Information on risk factor prevalence and relative risk (RR) associated with non-EBF was used to calculate the PAF using the formula [19]: Where P is the prevalence of the exposure in the population for a given exposure category, and RR the corresponding relative risk for the exposure category, calculated from the NDHS datasets (1999–2013). Risk factors used in this analysis were low maternal education, poor household wealth, no antenatal clinic visits, home delivery and delivery assistance from untrained personnel. A joint PAF across all risk factors was also calculated using the formula [37]: Where r represents each exposure variable. The assumption that exposures are independent and uncorrelated has been diminished with the use of relative risks that have been adjusted for potential confounders. Avoidable burden refers to the potential reduction in future burden of disease or health outcome that could be attained by changing the current distribution of risk factors to an alternative distribution of risk factors. Consistent with the classification of the counterfactual distribution of exposure suggested by Murray and Lopez [38], a series of scenarios based on evidence from community interventions to improve exclusive breastfeeding [31, 32] and trends in maternal educational achievement [39] were also used to estimate a feasible minimum risk for each exposure to calculate potential impact fractions [40, 41]. Murray and Lopez [38] classified counterfactual exposure distribution into: (i) theoretical minimum risk; (ii) plausible minimum risk; (iii) feasible minimum risk; and (iv) cost-effective minimum risk. Theoretical minimum risk refers to the exposure distribution that would result in the lowest population-level risk, regardless of whether currently achievable. Plausible minimum risk is the imaginable distribution of risks that would reduce the risk in the population of interest, if attained. Feasible minimum risk refers to a distribution of risk that has been achieved in some population, while cost-effective minimum risk reflects the exposure reduction using a range of cost-effective strategic interventions [38, 42]. As noted above, the PAF provides estimates based on the unrealistic counterfactual scenario of the elimination of the exposure from a given population. In this study, potential impact fractions employed estimation of the avoidable burden of non-EBF using a feasible minimum risk distribution, that is, a scenario that could be achieved in a Nigeria based on previous evidence. Accordingly, the present study also estimates potential impact fractions for comparisons with the PAF estimates and to provide evidence on the potential impact of different interventions for policy makers in Nigeria using the formula [43]: Where Pc is the proportion of the population in a given exposure category c, RRc is the relative risk for that exposure category calculated from the NDHS dataset relative to the unexposed category, and Pc* is the estimated proportion of the population in a given exposure c after the intervention. In this analysis, the estimated effect of interventions was based on previously published studies. For antenatal visits, impact fractions were estimated assuming a reduction of 17% in women having no antenatal care (from 22% to 5%) [31, 32]. For delivery assistance and place of delivery, impact fractions were estimated assuming a relative reduction of 15% based on community-based interventions to improve exclusive breastfeeding practice [31]. Finally, assuming continued improvements in high-school completion rates in women [39], impact fractions for maternal education were estimated assuming a 5% relative decrease in the proportion of women not completing high school from 23% to 22%. No alternative scenario was defined for household wealth because of a lack of data relating to interventions resulting in income re-distribution. Monte-Carlo simulation models using Ersatz Software 1.31 [44] were used to estimate the 95% confidence intervals for population attributable fraction and potential impact fraction estimates, to account for the uncertainty around the exposure prevalence and relative risk estimates. A beta probability distribution was used for exposure prevalence estimates (based on cases and non-cases) using the ErBeta function, and a normal distribution was used for relative risk estimates (based on a normal distribution for the natural logarithm of the RR) using the ErRelative Risk function, to estimate 95% confidence intervals for estimates after 10,000 iterations to ensure model convergence. To obtain the number of non-exclusive breastfeeding cases attributable to each exposure, attributable fractions were multiplied by the estimated total number of non-EBF cases. The DHS project obtained the required ethical approvals from the National Health Research Ethic Committee (NHREC) in Nigeria before the surveys were conducted (Assigned Number NHREC/01/01/2007). Participants were informed of the rationale for the surveys, confidentiality of their responses, and that respondents did not need to answer the questions if they do not feel comfortable doing so. Participants provided written informed consent before they participated in the surveys. The data used in this study were anonymous and publicly available to apply for online. Approval was sought from MEASURE DHS/ICF International and permission was granted for this use.

N/A

• ogbo, felix akpojene
• page, andrew j.
• idoko, john
• agho, kingsley e.