A community-based cluster randomised controlled trial to evaluate the effectiveness of different bundles of nutrition-specific interventions in improving mean length-for-age z score among children at 24 months of age in rural Bangladesh: study protocol

Although unaddressed by the Millennium Development Goals, reduction of stunting prevalence became the focus of several high-profile nutrition initiatives such as Scaling Up Nutrition (SUN, 2010) [14], World Health Assembly 2012 [15] and Nutrition for Growth 2013 [16]. The World Health Assembly 2012 has a target of 40% reduction of U-5 stunting by 2025. Hence, the recent Sustainable Development Goals (SDGs) adopted by the UN in 2015 included stunting reduction as one of their goals (SDG 2.2) [17].

Childhood stunting is considered to have in utero origins [18]. The first 1000 days of life (conception to postpartum two years) is regarded as a critical ‘window of opportunity’ for growth and development [19]. The long-term interactions of both proximal (i.e. biological and environmental) and distal (i.e. socioeconomic) factors play significant roles during this period in influencing the linear growth of children [20]. The proximal factors of childhood stunting in South Asia include undernourished mothers, limited dietary diversity during pregnancy, poor infant and young child feeding practices (IYCF) including suboptimal breastfeeding and/or complementary feeding, and inappropriate sanitation and hygiene [21]. However, even short-term improvements in nutritional status (in utero and postnatal) can result in substantial mean height gain of the child even within a single generation [22].

In the Lancet Maternal and Child Nutrition Series 2013, Bhutta et al. (2013) statistically modelled the effect of 10 nutrition-specific interventions on lives saved and economic costs in 34 countries containing the majority of the children with stunted growth [23]. They suggested that at an estimated additional cost of $9.6 billion per annum for 90% coverage of these interventions, one-fifth of stunting among the U-5 children can be averted. However, there is a need for primary research to test and clarify the combined impact of nutrition-specific interventions as packages. They agreed about this gap and that further trials are needed.

This study aims to test the effectiveness of the intervention bundle(s) on improving length-for-age Z score (LAZ) of children at two years of age. We hypothesised that our intervention bundle(s) would cause a change of at least 0.4 in mean LAZ of children, translating to at least 30% reduction in stunting in that arm(s), compared to that in the comparison arm.

This cRCT evaluates the effectiveness of 4 different combinations of the selected pre- and post-natal nutrition-specific interventions for improving children’s linear growth, compared to the comparison arm with usual practices. The three types of BCC are combined into one ‘BCC’ intervention and delivered universally in all four intervention arms. Profile of the four intervention arms are – intervention arm 1: BCC with PNS and CFS; intervention arm 2: BCC with PNS; intervention arm 3: BCC with CFS; intervention arm 4: BCC alone. (Fig. 1 ). The cohort of the mother-child dyads will be followed-up over the intervention period of ~36 months starting from recruitment to 24 months of child’s age.

The study is being conducted in the Habiganj district of Sylhet division in Bangladesh where stunting prevalence has remained stagnant over a decade [11]. Further, the prevalence of household food insecurity was highest (77%) and the prevalence of household food deficit was second highest (33%) in Sylhet [24].

We purposively selected two adjacent sub-districts (Bahubal and Nabiganj) of Habiganj district. These two sub-districts have a total of 20 unions (lowest administrative unit). We excluded 8 unions, 3 being in hard-to-reach areas (seasonal flooding and poor road transport system) and 5 having intensive nutrition intervention projects implemented by other non-government organisations (NGOs). The remaining 12 unions had enough population to cover the desired sample size. Each union was then divided into clusters, each cluster comprising of ~450 households or ~2000 population (average 2–3 villages). We selected five clusters or multiples of 5 clusters from each union, 125 clusters in total following the cluster exclusion criteria detailed in Table 1. We then applied block randomisation by generating random sequence of 1–5 (each number representing a specific study arm) of 25 blocks with the analytical software Stata SE 14.2. Clusters were then allocated to arms, five at a time, based on the random sequence (Fig. 2). Taking equal numbers of clusters per arm from each union will ensure neutralizing the effect of variations, if any, of background characteristics between the unions. Allocation concealment was impractical due to community-based study design and nature of BCC intervention. However, care is taken to keep the outcome assessors blinded to the interventions allocated to specific clusters. The intervention and the evaluation teams have minimal contact to minimise information bias.

Each study cluster comprised of an average population of 2000 and necessary adjustments were considered for cluster randomization. Crude birth rate (CBR) in rural Bangladesh in 2011 was 23 live births per 1000 population [25]. Therefore, taking six months of recruitment window for pregnant women, each cluster would have yielded ~23 new pregnancies on average.

Several assumptions were considered for the sample size calculation [26]. The first assumption was that mean LAZ (our primary outcome variable) for 18–23 months old children in the comparison arm will be −2.0, and standard deviation of mean LAZ in both intervention and comparison arms will be 1, based on the mean HAZ reported in this age group in the BDHS 2011. These BDHS 2011 [25] parameters were the latest available during this protocol development. The estimates were still valid in 2015 since the latest BDHS (2014) showed similar stunting rates in Sylhet division [11]. Second, a 0.4 difference in mean LAZ score was expected in intervention arms compared to comparison. As the third assumption, we considered an 80% power and 5% alpha; the inter-cluster correlation coefficient of 0.06, derived from the design effect for HAZ in BDHS 2011, was used for adjustment of clustering effects. Fourth, the ratio between 4 intervention and 1 comparison arms was set at 1:1:1:1:1. Taking all the above assumptions into account, the Stata SE 13 software calculated the sample size to be 175 children per arm, which can be yielded from 25 clusters. However, to examine the secondary outcome variables between the comparison arm and individual intervention arm with a higher statistical power, the sample size in the comparison arm was doubled, thus the ratio between participant numbers in each intervention arm and comparison arm was raised to 1:2, respectively. The final allocation ratio of study participants was thus 1:1:1:1:2 for intervention arm 1: intervention arm 2: intervention arm 3: intervention arm 4: comparison arm 5. Thus, the final sample size was 175 children in each intervention arm and 350 children in the comparison arm. Considering the characteristics of the intervention, however, enrolment was during pregnancy. We therefore considered the following losses- 11% pregnancy loss due to abortion and still births (personal communication from Health and Demographic Surveillance System run by icddr,b at Mirzapur, Tangail), a 4% mortality in the first two years of life [25], 15% losses due to non-consent to participate by eligible women and loss-to-follow up. Loss-to-follow up includes refusal and migration out of study area. Thus, final sample size considering participant recruitment during pregnancy was 250 pregnant women in each intervention arm and 500 pregnant women in the comparison arm which is expected to yield the required number of children at 24 months from the enrolled cohort.

We enrolled 10 pregnant women in each intervention cluster and 20 in each comparison cluster, following the inclusion and exclusion criteria for participant eligibility (Table 1). Data collectors systematically conducted door-to-door surveys of all households in the study area to identify newly pregnant women, which were repeated in subsequent monthly cycles. The woman’s pregnancy status was assessed by enquiring about the first date of her last menstrual period (LMP), and possible pregnancy was confirmed with a sensitive pregnancy urine test kit (Excel®). Women tested positive were invited to voluntarily participate in the study with appropriate informed written consent. Recruitment of eligible pregnant women started in November 2015 and was completed in May 2016. Follow up of the cohort is ongoing.

In the comparison arm, study participants are receiving usual nutrition interventions available from routine health services only. Pregnant women are eligible to receive nutrition counselling on maternal and child nutrition during antenatal check-up (ANC) visits and caregivers of children under two years of age receives IYCF counselling during postnatal care visits within 42 days and sick child management visits at health facilities and by NGO workers (if available) [31]. However, contact coverage and content quality of these platforms and interventions is very low and does not reach those who do not seek the services. Antenatal iron-folic acid supplements are available during ANC contacts at public health facilities. Children aged 6–59 months receive vitamin A supplements during biannual national vitamin A campaigns, although coverage is unsatisfactory [11].

Counselling and supplement sachets are delivered through our specially trained Community Health Workers (CHWs). Their tasks are supervised and monitored by two layers of field supervisors – a team of Field Supervisors (FSs), and a Field Research Assistant (FRA). Minimum educational qualification of the CHWs is Higher Secondary School Certificate (HSC) completed, while the FSs and the FRA have completed their Bachelor’s degree. The intervention team was given intense counselling training in three phases, and monthly meetings are held for field-related troubleshooting and as refresher courses. Supervisors provide daily feedback to associated CHWs during on-site counselling supervision.

At least 4 antenatal counselling sessions are delivered during pregnancy. Following live birth notification, 8 counselling sessions are delivered within 48 h and 7–14 days of birth, and at completion of every month till 6 months. Thereafter, counselling is delivered every third month till postnatal 23 completed months (Table 1). The CHWs also distribute supplement sachets every month, and provide counselling on the importance and instructions on supplement taking and storage.

A range of quantitative data collection methods, including structured questionnaire and anthropometric measurements, have been adopted for assessments of primary and secondary outcomes. A timeline depicts the schedule of data collection in Table 2.

All questionnaire-based data and anthropometric measurements are collected by trained data collectors. Anthropometric measurements are taken following measurement-specific standard operating procedure [38]. Measuring tools are calibrated regularly. All data collectors are female and locally recruited. Before initiation of participant recruitment, the data collection team was intensely trained by the central team on all aspects of data collection, including consent taking, different data collection methods, anthropometric measurements, question throwing techniques, use of the tablet and cultural aspects to consider during data collection. Monthly meetings are held for troubleshooting and refresher trainings are conducted as needed. Following the initial training, all data collectors were standardised for anthropometric measurements against two gold standard trainers from icddr,b. Subsequently, refreshers trainings are conducted every two months to retain measurement standards. The data collectors are directly supervised by two FRAs who monitor data collection quality and provide on-site feedback as needed. One field manager is responsible for overall field coordination.

A bespoke automated Project Management Information System (MIS) has been developed linking data collection and intervention delivery through android-based tablet (handheld)-PC with a central database where progress can be monitored by the field supervisors and central staff. On pregnancy enrolment through the tab, an android platform-based application automatically generates subsequent prenatal visit plans and tools at pre-specified schedules and tasks of CHWs and data collectors (intervention modules for CHWs and questionnaire for data collectors). Similarly, birth is registered in the tab immediately upon notification. To ensure post-partum data collection within 24 h of delivery, a mobile phone-based birth-notification system has been established. Enrolled pregnant women and their family members are continuously encouraged to notify birth by text message or voice calls as early as possible, preferably within six hours of birth. Registration of live births then generates appropriate plans and tools for postpartum visits by CHWs and data collectors. The web-linked desktop dashboard system is used to monitor activities/performance of CHWs and data collectors in real time by field supervisors and the central team.

Data analysis to detect change of the primary outcome (mean LAZ) will be conducted with length measured at 1, 3, 6, 9, 12, 15, 18, 21 and 24 months of age using t-tests assuming LAZ will satisfy the normality assumption. Longitudinal assessment of significance using lag time of stunting (mean length > −2SD below median of WHO reference population) will be done by the log rank test, followed by fitting a Cox proportional hazards model for each intervention arm to determine the change in risk of stunting compared to the comparison arm. Multiple linear regression will be employed to quantify the effect of intervention in each intervention arm compared to the comparison arm in terms of mean LAZ change and some secondary continuous outcome measures including changes in mother’s weight, birth weight, diarrhoea morbidity. Logistic regression will be used on binary outcome variables which comprise most of our secondary outcomes measures (e.g. – EBF and breastfeeding initiation within 1 h of birth). All these models will be controlled for possible confounders (household food security, educational background and SES) if randomisation of the clusters does not control for differences in these factors at baseline. For regression models, we shall consider random effects of clustering. All analysis will be intention-to-treat. Quantitative data analysis will be done using Stata SE 14.2 (StataCorp LP, College Station, Texas, USA).