Formal maternal employment is associated with lower odds of exclusive breastfeeding by 14 weeks postpartum: A cross-sectional survey in Naivasha, Kenya

Background: In many low-and middle-income countries, improvements in exclusive breastfeeding (EBF) have stalled, delaying reductions in child mortality. Maternal employment is a potential barrier to EBF. Objectives: We evaluated associations between maternal employment and breastfeeding (BF) status. We compared formally and non-formally employed mothers in Naivasha, Kenya, where commercial floriculture and hospitality industries employ many women. Methods: We conducted a cross-sectional survey among mothers (n = 1186) from September 2018 to October 2019 at 4 postpartum time points: At hospital discharge (n = 296) and at 6 wk (n = 298), 14 wk (n = 295), and 36 wk (to estimate BF at 24 wk; n = 297) postpartum. Mothers reported their BF status and reasons for EBF cessation. We used multivariable logistic regression models to test the association between formal maternal employment and 3 outcomes: early BF initiation (within 1 h of birth), EBF at each time point, and continued BF at 9 mo. Models were informed by a directed acyclic graph: A causal diagram used to characterize the relationship among variables that influence the independent (employment) and dependent (BF status) variables. Results: EBF did not differ by employment status at hospital discharge or at 6 wk postpartum. However, formally employed mothers were less likely than those not formally employed to report EBF at 14 wk (59.0% compared with 95.4%, respectively; AOR: 0.19; 95% CI: 0.10, 0.34) and at 24 wk (19.0% compared with 49.6%, respectively; AOR: 0.25; 95% CI: 0.14, 0.44). The prevalence of continued BF at 36 wk did not differ by group (98.1% for formally employed compared with 98.5% for non-formally employed women; AOR: 0.80; 95% CI: 0.10, 6.08). The primary reasons reported for early EBF cessation were returning to work (46.5%), introducing other foods based on the child’s age (33.5%), or perceived milk insufficiency (13.7%). Conclusions: As more women engage in formal employment in low-and middle-income countries, additional supports to help prolong the period of EBF may be beneficial for formally employed mothers and their children. Between September 2018 and October 2019, we repeatedly conducted cross-sectional surveys at 4 postpartum time points to investigate the associations between maternal employment and BF practices in Naivasha, Kenya. We hypothesized that formal employment would be associated with a lower prevalence and reduced odds of BF at each time point and across each indicator (early initiation of BF at 0 wk; EBF at 0, 6, 14, and 24 wk; predominant BF at 0, 6, 14, and 24 wk; and continued BF at 36 wk). Naivasha is a peri-urban city in Nakuru County, located 100 km north of Nairobi, with a population of ∼355,000 (24). This area contains the largest concentration of commercial flower farms in the country, which are the primary sources of employment in this region. The majority of flower farm employees reside in several densely populated peri-urban informal settlements, with varying access to electricity and sanitation services. Employment within the floriculture industry is characterized by long commuting distances and separation from children, compared to employment in the informal sector (e.g., tailoring, subsistence farming, self-employment, trading), where women have more flexible schedules and BF opportunities (25). Mothers were recruited 1–4 d postpartum and at routine infant immunization visits at 6, 14, and 36 wk postpartum at 2 public facilities (the Naivasha Sub-County Referral Hospital and the Karagita Dispensary) and a private facility subsidized by a local floriculture company that serves farmworkers (the South Lake Medical Center). All postpartum women with a live birth admitted to the maternity wards or presenting for immunizations to the health facilities on recruitment days were screened for eligibility. Mothers who were 1–4 d, 5–7 wk, 13–15 wk, or 9 mo (± 1 wk) postpartum were eligible, regardless of past or present child morbidity. Screening did not include evaluating mental health status. Health center staff introduced mothers to the research team. The research team explained the study purpose to all mothers present for immunizations and within the maternity wards through group announcements. Upon recruitment, we obtained written, informed consent from all participants. Surveys were administered verbally by a team of 5 trained research staff in Swahili, English, or another preferred language of the mother, using paper questionnaires. Mothers were eligible to participate once in the cross-sectional survey. The survey collected information on 5 domains: 1) household assets and demographics; 2) employment status and benefits; 3) IYCF practices; 4) access to reproductive and other health services; and 5) health status of the child and the mother. We queried participants about their household assets (house material composition, vehicle, television, mobile phone) and demographics (educational attainment, household income, marital status, household size, parity, religion, and tribe) using questions from the Demographic and Health Survey (7). We asked mothers about whether they were employed, the type of employment and responsibilities, the name of the employer, the availability of a formal contract, the hours worked, and the availability of maternity leave and other policies to support BF. IYCF practices were assessed using standardized questions from the Demographic Health Survey and WHO Indicators (7, 26). Using a 24-h list-based recall method, we recorded the types of liquid, semisolid, and solid foods given during the previous day and the number of BF and other feeding episodes. Demographic Health Survey questions were used to assess antenatal care utilization, the delivery setting, the type of delivery, and the presence of a skilled attendant at childbirth (7). We assessed child morbidity in the prior 2 wk according to 5 common illnesses and symptoms: diarrhea, pneumonia, fever, malaria, and cough. The HIV status of the mother was self-reported. The primary anticipated reason for EBF cessation was assessed using a single question with multiple response options (going to work, child refusal, uncomfortable/did not want to, perceived milk insufficiency, pregnant again, baby cries after being breastfed, child age, other). Upon cessation of EBF, we also queried mothers regarding actual reasons why they introduced mixed feeding, defined as the initiation of feeding other liquids and foods along with breastmilk (26). The same multiple response options were provided. The survey also assessed the availability of workplace supports for BF through questions on maternity leave benefits and the availability and use of employer-supported lactation rooms, childcare, and housing. The full survey is available in Supplementary Methods 1. We examined missing data and replaced missing values by recontacting mothers by phone. To identify a 20% difference in EBF rates at each time point (1–4 d, 6 wk, 14 wk, and 24 wk postpartum), with 80% power and an alpha <0.05, we aimed to recruit a minimum of 124 mothers at each point. We successfully enrolled 296 women at 1–4 d postpartum, 298 women at 6 wk postpartum, 295 women at 14 wk postpartum, and 297 women at 36 wk postpartum who reported retrospectively to 24 wk postpartum. This higher sample size increased power to 96.4% for EBF, 96.4% for early initiation, and 99.9% for detecting a 20% difference in these outcomes between employment groups. Our initial recruitment approach did not use employment as a screening factor, which resulted in a higher proportion of non–formally employed mothers. Thus, we focused later recruitment on formally employed mothers at each time point, resulting in higher recruitment numbers than initially planned. Maternal employment status was ascertained by self-report. We first asked women about their current employment or, if mothers were not working and were within 6 wk postpartum, about their recent employment status. Mothers were then asked about the type of occupation, the number of hours worked per week, and the existence of a contract to be further classified as formally, informally, or self-employed. We classified employment as formal if women worked for a registered employer (e.g., a commercial farm, business, company, school, health-care facility), worked ≥20 h/wk, and received regular compensation. For mothers currently on maternity leave (recruited at 0 and 6 wk postpartum), employment type was classified based on the type of work before delivery. Mothers were classified as not employed if they indicated that they reported no current employment. If mothers were employed during their pregnancies, but indicated that they did not intend to return to work and were not receiving maternity leave benefits, they were classified as not employed. In the study context, women usually work through 36 wk of pregnancy. Our dependent variables encompassed 3 BF indicators: 1) early initiation of BF; 2) EBF or predominant BF at 6, 14, and 24 wk postpartum (coded as separate variables in the analysis); and 3) continued BF at 9 mo postpartum. Early initiation of BF was defined as BF within 1 h of childbirth (26). EBF was defined as feeding breastmilk only with no other liquids or solids (26) from childbirth through the time point of data collection. EBF duration, defined as the number of weeks of reported EBF, was measured retrospectively by mothers who were recruited at 36 wk. These mothers were considered to have been EBF at 6 mo if only breastmilk was fed through 24 wk of age. To establish conservative estimates of BF status, we applied a modified approach to the WHO method to determine BF status. BF status was determined by the IYCF practices during the previous 24 h, as well as by assessing the last week when mothers gave breastmilk exclusively. Thus, a mother who was EBF during the previous day, but who fed infant formula in the previous week, was classified as not EBF. We classified children as predominantly breastfed if they were given breastmilk along with juice, water, or other liquids, including medicines and vitamins/minerals, but not milk or semisolid or other foods through the specified time (26). Continued BF at 9 mo was defined as any BF among children in this age group (26). We attempted to minimize social desirability or recall bias in the assessment of BF outcomes by including multiple questions about BF duration and exclusivity. We assessed EBF duration, the week at which other foods and drinks were first consumed, and the child's diet in the 24 h preceding the survey, using the validated IYCF feeding practices tool (26). Vitamin or mineral supplements were not assessed. We sought to minimize temporal changes in BF practices by completing study recruitment over 13 mo. To reduce the likelihood of selection bias, we restricted our recruitment to 3 health facilities that represented 3 levels of the health-care system and we recruited women without prior knowledge of their socio-demographic factors. The research team received training on responsible conduct in research, survey procedures, and anthropometric assessments. Reporting bias related to maternal employment was minimized by describing the purpose of the study as seeking to identify opportunities to better support mothers by understanding the challenges they encounter in feeding their infants. The study was not promoted as an employment study. For questions related to reasons for cessation of EBF, self-reported answers were coded into categories. “Returning to work” was indicated if a participant responded that this was a reason for EBF cessation; it was not provided as an option before the respondent offered a response. To identify confounding variables, we constructed a directed acyclic graph (Supplemental Figure 1): a causal diagram used to characterize the relationships among variables that influence the primary independent variable (employment) and the dependent variable (BF status) and are not on the causal pathway (27). Informed by existing literature on the employment–BF relationship and influences of BF in the study context, the same set of confounders were used for all models (13, 16, 28). These include maternal age (years), marital status (married or not), maternal education (some secondary or higher versus less than secondary), tribal affiliation (Kikuyu, Luhya, Kiisi, or other), child morbidity (presence of diarrhea, fever, malaria, cough, or fever in the previous 2 wk), cesarean delivery (versus vaginal birth), and HIV status (positive or negative). We assessed the linearity of continuous variables (maternal age, maternal education) with outcomes by specifying disjointed indicator variables. Household income was hypothesized as a mediator of the association a priori. Maternal employment would likely result in increased individual-level income, thus increasing household wealth and, subsequently, influencing BF practices. However, wealth prior to workforce entry could be a plausible confounder of the employment–EBF association (i.e., household wealth could influence whether a woman works or not). In these cross-sectional data, we cannot establish the temporality of wealth and workforce entry. Therefore, wealth was considered a mediator and was not included as a covariate in our models (29). We employed separate multivariable logistic regression models to test the association between maternal employment (formally employed versus informally employed, self-employed, and not employed) and each BF variable (early initiation of BF and EBF, predominant BF, and continued BF at 9 mo). In sensitivity analyses, we compared BF outcomes by employment status when restricting formal employment to only mothers employed at flower farms or in other commercial agricultural farms (n = 398). In addition, we adjusted for multiple comparisons using the Holm-Bonferroni sequential correction, given that we assessed associations for multiple outcomes across 4 time points (30). First, results for both chi-square tests and linear regressions were ordered from the smallest P value to the largest P value. Second, the second-smallest P value was corrected with a Bonferroni approach [(number of tests − order of test + 1) × P value]. The correction procedure stops when the first nonsignificant test is obtained (30). In our main analyses, the alpha was set to 0.05. STATA version 14.1 (StataCorp LP) was used to conduct all analyses. The study adhered to the Strengthening Research for Observational Studies in Epidemiology (STROBE) guidance (Supplementary Methods 2) (31). All study procedures were approved by the Kenya Medical Research Institute Scientific Ethical Review Unit (study number KEMRI/SERU/CCR/0112/3712) and the Wheaton College Institutional Review Board (study number 3712).