Effectiveness of community health workers in improving early initiation and exclusive breastfeeding rates in a low‐resource setting: A cluster‐randomized longitudinal study

Abstract Little evidence exists in Kenya on the potential of community health workers (CHWs) in promoting exclusive breastfeeding (EBF) and early breastfeeding initiation (EBI) in resource‐restricted settings where very low EBF rates (2% to 12%) have been documented. The study utilized CHWs and assessed their effectiveness in promoting EBF and EBI. The cluster‐randomized longitudinal design was used and sixteen villages from Kiandutu Slum in Thika randomly assigned into either intervention group (IG) or comparison group (CG). Pregnant women attending Maternal Child Health (MCH) clinic were recruited. The IG received nutrition education sessions conducted by CHWs at home, two prenatally and six postnatally, plus the routine MCH care. The CG went through routine MCH care only. Infants feeding data were collected at 6, 10, 14, and 24 weeks postpartum by research assistants blinded to the intervention allocation. Differences in EBF and EBI in the two groups were tested using χ 2 tests, Kaplan–Meier survival analysis and generalized estimating equations. Of the 526 recruited in the study, 431 remained and were included in the analysis (IG = 176) and CG (225). The prevalence of EBF at 24 weeks was 45.3% in the IG compared with 15.0% in the CG, revealing a statistically significant difference log rank = 20.277, (1, n = 314) p < .001. The difference was not statistically significant in EBI prevalence between the IG (58.2%) and CG (60.3%; χ 2 = 0.008, p = .928). The CHWs have potential effectiveness in promoting EBF but not EBI. The link between the health center and CHWs should be strengthened to promote EBF.

of the first stool. Globally, rates of EBI and EBF are generally low (Black et al., 2013). EBI rates are 62% (KNBS & ICF Macro, 2014) similar to the rates of exclusive breastfeeding which stands 61%.
Rates of EBF in the resource-restricted settings are very low at 2% (Kimani-Murage et al., 2011). Under nutrition including suboptimal breastfeeding is associated with 3.1 million global child deaths annually (Black et al., 2013). Breastfeeding promotion is therefore pivotal to infant health promotion and survival especially in the low-and middle-income countries (LMICS). Many health promotion interventions have utilized Community Health Workers (CHWs) in their programs. The CHWs were identified as a distinguishing feature for the provision of primary health care for people in resource-restricted settings in the Alma-Ata Declaration (Lehmann & Sanders, 2007).
They act as a mitigating factor where healthcare access is limited due to limited numbers of healthcare workers by providing essential Maternal Child Health (MCH) care at the household and community level. They also provide education, preventive health services, and play the role of liaison between the community and facility-based services (Bhutta, Lassi, Pariyo, & Huicho, 2010;UNFPA, 2011). The use of CHWs, compared to usual healthcare services, has been found to be effective in increasing rates of early initiation of breastfeeding (within one hour); rates of breastfeeding and EBF (Lewin et al., 2010). Findings from interventions that were implemented to promote EBF using CHWs indicate effectiveness (Balaluka et al., 2012;Gilmore & McAuliffe, 2013;Tylleskar et al., 2011), although Kimani-Murage et al. (2015) found no difference in the intervention (that utilized CHWs to promote EBF) and control groups (that received usual care). Evidence of the effectiveness of CHWs in Kenya is therefore scanty. The present study examines the efficiency of CHWs in promoting early initiation and EBF among mothers in the context of a cluster-randomized controlled nutrition education intervention in a low-resource urban setting. Rates of EBF in these settings have been found to be extremely low (Kimani-Murage et al., 2011;Ochola, Labadarios, & Nduati, 2012).

| Study site
The study was conducted in Kiandutu slum in Kiambu County-Thika West District. Kiandutu slum is one (1) kilometer from Thika town, which is about 40 km Northeast of Nairobi. The informal settlement has 17 villages each with a village elder (Kariuki, 2012) and is served by two health centers, Kiandutu and Makongeni.

| Study design and sample size calculation
The study was a cluster-randomized longitudinal study conducted between September 2013 and October 2014. Cluster randomization design is used to evaluate interventions implemented at the community level, for logistical convenience and to avoid infiltration (Hayes & Bennet, 1999) hence the design was found suitable for this study.
The study entailed two study groups the IG and the CG. To obtain a representative sample for the two groups, a level of precision of 5% (for a two-sided t test) and power of 80% was set. The sample was multiplied by a design effect of 2 and by 20% for general attrition and 6% loss from low birth weight (LBW). The formula gave a requirement of 157 per arm but this figure was inflated by 60% to cater for high mobility characteristic of the slum observed during the pilot study. A total of 526 mothers were recruited after informed consent.

| Randomization and sampling
The Microsoft ® excel function was used to randomize sixteen out of the seventeen villages into either intervention or comparison groups.
The principle investigator recruited pregnant women attending prenatal clinic in Kiandutu and Makongeni health centers based on the village of residence. A two-stage screening process was applied in sampling. Initial screening of pregnant women took place during the first prenatal visit. The second screening took place during the first postnatal visit. The inclusion criteria for the initial screening included being over 18 years less than six (<6) months gestation; without a history of chronic disorders such as hypertension, diabetes, HIV, and tuberculosis. Postnatal inclusion criteria included; term delivery, singleton births, and a birth outcome more than 2.5 kg. The recruitment of the women took place between September 2013 and December 2013 and was based on informed written consent (by signature or thump print). The recruitment took place on a rolling basis until the desired sample size of 263 per arm was attained. Figure 1 is the schematic representation of the randomization procedure. Standardization of the data collecting tools and procedures for both the researcher and the enumerators was accomplished through group sessions by mimicking the respondent researcher situation.

| Selection and training of the research team
The study used validated semi-structured questionnaires.

| Description of the intervention
The intervention emphasized the (a) importance of EBF and EBI (b) conditions infants were susceptible to with early complementary feeding (c) developing breastfeeding confidence (

| Data collection
To ensure content, semantic and technical equivalence were ascertained, the English version of the semi-structured questionnaire was translated into Kiswahili with the help of a Linguist from Egerton University. The Kiswahili questionnaires were then translated into English through blind back-translation by a neutral person. Adjustments were made by the research team. The questionnaire was pilot tested as documented in Mituki et al. (2017).
The primary outcome of the study was the prevalence of EBF and was calculated as the ratio of infants at a particular postpartum visit (6 weeks, 10 weeks 14 weeks & 6 months) who were fed on breast milk alone in the 24-hr preceding the interview to the total number of infants in the specific groups (intervention or comparison) and the total number of infants under the study. Infants who did not receive any other food or liquids other than breast milk were deemed to be exclusively breastfed. The cumulative prevalence of EBF was also determined at 6 months. The secondary outcome of the study EBI was determined by asking mothers how much time had elapsed before breastfeeding was initiated for the index infant. The prevalence of timely initiation of breastfeeding was then calculated as the ratio of infants' breastfeed within 1 hr of delivery to the total number of infants.

| Statistical analysis
Data were analyzed using the Statistical Package for Social Sciences (SPSS) computer program version 20 (IBM SPSS Statistics, IBM Corp.). Continuous variables were described by means and ranges.
Categorical data were described in terms of numbers and percentages. Descriptive statistics, χ 2 test (adjusted for cluster randomization), generalized estimating equations (GEE), and Kaplan-Meier survival analysis were used to compare variables in the two study groups. All tests were two-sided and were considered statistically significant at p < .05.

| RE SULTS
A total of the 628 women who met the study inclusion criteria were invited to participate in the study, 526 consented to participate and were enrolled into the study. Ninety-two declined, 80% (74) without any reason, twelve (12) cited intention to return to their rural homes, while 6 cited lack of time. The 526 study participants were randomized into either the IG (263) or the CG (263). Eighty-seven (33%) from the IG were lost to follow-up while only eight (3%) in the CG.

| Prevalence of early initiation of breastfeeding
Prevalence of EBI was determined by categorizing the mothers who initiated breastfeeding in less than an hour and over one hour after delivery. Thus, the prevalence of EBI (<1 hr) was 58.2% among mothers in the IG and 60.3% for those in CG, the difference was statistically insignificant with adjusted χ 2 test (χ 2 = 0.008, p = .928).

| Discussion
Primary health care utilizes CHWs in the provision of basic services to people at the community level. The efforts are geared toward entrenching the community (level one) in the health delivery system, a strategy that provides a mechanism through which households and communities can take an active role in health-related development issues. The community-based approach is supposed to complement the formal health system.
The study therefore presents evidence-based information for the implementation of community-based EBF promotion strategies while strengthening already existing community-based strategies for health promotion.  (Mituki et al., 2017). It might mean that context-specific barriers to EBI should be considered in the design of an intervention seeking EBI promotion.
The prevalence of EBF was also low at six months in the two groups in comparison with the national figures. The IG, however, had higher prevalence, but only statistically and significantly different at the 6th month.
It is worthwhile noting that clinic visits which provide an avenue for growth monitoring and EBF promotion for majority of Kenyan Mothers, with children below 9 months are modeled in the pattern of the National Immunization Schedule. The schedule consists of five contacts between birth and 9 months, (at birth, 6 weeks, 10 weeks, 14 weeks, and the 9th month). After the 14th week, up to the 9th month a breastfeeding mother may not visit the clinic unless the infant is unwell and conversely may not receive EBF promotion messages. This fact may be a potential reason why there were no statistically significant differences in EBF rates of mothers in the IG and CG, at 6, 10, and 14 weeks after adjusting for cluster random- opposed to the group that utilized BFHI alone at 13%. These studies reveal that CHWs can offer breastfeeding support which can promote EBF for the first six months.
The study had strengths. First, we used cluster randomization which was successful since there were no statistically significant differences between the IG and CG groups in their baseline characteristics. Secondly, this was a longitudinal study and mother-infant pairs were followed up to six months, the results therefore may reflect the true picture of rates of EBF since national data (from the Kenya Demographic Health Survey) uses a single 24-hr recall cross-sectionally to compute rates. The study, however, had several limitations, first, in the collection of EBI and EBF data, we relied on maternal recall, and there may have been courtesy bias which could have led to overestimating the effect of the intervention. To minimize this bias, we used research assistants who were blinded to the study group mothers belonged to collect data while the CHWs implemented the intervention. Secondly, there might have been infiltration since the proximity of the Health Centres and the different villages were close. To mitigate against this factor, analysis was adjusted for cluster randomization.

| CON CLUS I ON AND RECOMMENDATIONS
This study revealed low rates of EBI and EBF among the mothers, the prevalence of EBF was, however, higher in the IG with a statistically significant difference than the CG at the 6th month.