Program description
The program focused on addressing both key supply and demand challenges that hamper availability, affordability and uptake of ORS and zinc products for childhood diarrhoea treatment. On the supply side, the program collected and analysed market data to generate and share insights such as potential market size, competitive landscape, and return on investment for ORS and zinc with manufacturers and importers in order to stimulate and guide their investment decisions regarding production, promotion, and sale of ORS and zinc. The program also shared the latest evidence on low-osmolarity ORS and zinc with relevant agencies in order to update regulatory guidelines and facilitate product registration and market entry of the optimal products. Furthermore, suppliers were provided with technical assistance on cost reduction, for example through cost-of-goods-sold (COGS) analysis to identify lower cost inputs and packaging optimization. To improve rural availability, suppliers were incentivised through time-limited financial and technical support to develop distribution models that would sustainably reach underserved areas. For example, the program signed agreements with suppliers to share a proportion of the costs for establishing a rural sales force if suppliers met targets on volume and rural availability of ORS and zinc. They were also supported with the development and implementation of innovative marketing and sales strategies. To engender competition and ensure a vibrant market, multiple suppliers were concurrently supported to enter the market. The number of low-osmolarity ORS products competing for market share increased to 35 from just one, and zinc DT increased to 12 and co-packs increased to 10 from zero during the program.
To increase demand, the program conducted trainings and mentoring sessions with health care professionals and patent and proprietary medicines vendors (PPMVs). The trainings were conducted once but three to four cycles of one-on-one mentoring sessions were conducted through follow-up visits to individual providers by specially trained peers to reinforce knowledge and entrench the right practices. PPMVs are largely informal private medicine retailers but they are a dominant source of treatment for childhood illnesses, especially in underserved communities [20]. Also, key influencers and various community networks were engaged to reach mothers and caregivers with messages on the right steps to take for children with diarrhoea.
Study design
The study design used baseline and endline household surveys to measure changes in the coverage of ORS and zinc for the treatment of diarrhoea in children under-five. In addition, the surveys also collected data on household assets in order to measure household wealth. The household surveys were modelled using standardized questions from the DHS and Multiple Indicator Cluster Survey (MICS). Further details on the study design have been published elsewhere [17]. The study was reported according to the Strengthening the Reporting of Observational Studies in Epidemiology Statement [21].
Setting
The study was conducted in the eight Nigerian program states. Nigeria is the most populous country in Africa with population estimated at nearly 200 million [22]. Children less than 5 years account for 17.6% of the population and 51% are from rural areas. There are wide wealth disparities across the country with the northern part of the country being the poorest. On the average, 54% of the population live below international poverty line of US$1.25 per day [23]. Nigeria operates a mixed system with a three-tiered public health system co-existing with a heterogenous private sector. The private sector provides about 60% of health service delivery [24]. Nearly two-thirds of children are brought to PPMVs for treatment. The eights states where the program was implemented account for 40% of the burden of childhood diarrhoea [25]. Three of the states are in the South and five are in the North of the country.
Study population
The study population were children under five who resided in one of the eight states within Nigeria where the program was implemented and had diarrhoea within the past 2 weeks preceding the survey. The study used a stratified, multi-stage cluster randomized sampling design for identifying and selecting study participants which is described elsewhere [17].
Data sources
Data were collected using population-based household surveys among caregivers of children under five. The baseline survey was conducted between December 2013 and November 2014 and the endline survey was conducted between April 2016 and May 2017. The surveys collected information on household characteristics including asset ownership, caregiver knowledge, care seeking behaviour, geographical location, caregiver and child sex and age, prevalence of diarrhoea, and treatment practices.
Variables
The primary study outcome was treatment of diarrhoea with ORS and zinc by children who had diarrhoea within the last two weeks of the survey. The survey used standardized questions adopted from the DHS and MICS to measure ORS and zinc coverage. The primary caregiver of children in each household were asked whether any child had diarrhoea in the last 2 weeks preceding the survey. Diarrhoea was defined as passing three or more loose stools in a 24 h period. For children who did have diarrhoea within the last 2 weeks preceding the survey, caregivers were asked if anything was used to treat the diarrhoea, including a fluid made from a packet called ORS, zinc tablets, or zinc syrup.
The survey also collected information on household assets, such as ownership of land, vehicles, farm animals, household goods, and structure of living accommodations. The complete list of household asset questions is presented in the supplementary appendix S1. These household asset questions were also adopted from model DHS and MICS surveys.
Households were designated as living in urban or rural areas based on their census enumeration block. We obtained census enumeration maps from Nigeria’s Bureau of Statistics (NBOS) and National Population Commission (NPC). Each of the sampled census enumeration areas were pre-assigned as urban or rural by NBOS and NPC.
Sample size
The sample size calculation for the study was designed to detect change in ORS and zinc coverage between the baseline and endline surveys. We used the following formula to calculate the sample size for the study:
$$ N= Deft\ast \frac{{\left({z}_{a/2}-{z}_b\right)}^2\left[{p}_1\left(1-{p}_1\right)+{p}_2\left(1-{p}_2\right)\right]}{{\left({p}_1-{p}_2\right)}^2} $$
where N is the desired sample sizes of children with diarrhoea in each state (assuming one child per household), Deft is the design effect due to clustering which we assumed to be 1.5 [16], p1 is the 2011 state coverage estimates (the most recent coverage data available at the time) [26], and p2 is the endline estimates necessary to see a 25% difference over time [27]. We used a two-sided t test with 95% confidence, 80% power, and equal variances. We took into account the prevalence of diarrhoea found and allowed for a 5% non-response rate [26]. To simplify training and field work management, assuring better data quality, we used the largest sample size required (ie, Lagos) as the sample size for all states. In each state, the sample size was 940 households with a child under five. Based on the lower density of population within Cross Rivers and Rivers, the sample size was reduced to 930 households with children under five.
Statistical analysis
Stata version 14 (Stata Corp, College Station TX, USA) software was used in the first step of our analyses. We constructed wealth quintiles (poorest, second, middle, fourth, and richest) using principal component analysis of the household assets. Wealth quintiles were used as a measure of SES. Treatment coverage was defined as the percentage of children with diarrhoea in the 2 weeks preceding survey who were treated with ORS and zinc. We conducted descriptive statistics for characteristics of the child, caregiver, and household for each survey period. Standard errors and 95% confidence intervals were estimated using Taylor linearized methods. Pearson’s chi-squared tests were used to compare characteristics and outcomes between baseline and endline surveys and estimate p-values. To account for potential confounding due to differences in population characteristics between baseline and endline and adjust for sampling design, we conducted multi-level, mixed-effects logistic regression modelling with ORS, zinc, and combined ORS and zinc use as the dependent outcome and survey period and population characteristics as independent predictors. Using the model results, we constructed predictive probabilities of ORS, zinc, and combined ORS and zinc coverage at baseline and endline and for all sub-populations at both survey periods. Analyses were stratified by geographical location and SES. All analyses incorporated sampling weights and took into account clustering at the enumeration area and household levels.
Microsoft Excel was used to compute simple absolute and relative disparities for pairwise comparisons and summary measures of inequality [28, 29]. As SES is comprised of ordered subgroups, a single comparison between the richest quintile and the poorest quintile was done, with the former being the reference group (28). Absolute disparities (AD) and relative disparities (RD) were determined as follows:
AD = y2 – y1, where y2 represent coverage in the reference group and y1 refers to coverage in the comparison group. RD = y2/y1, where y2 represent coverage in the reference group and y1 represent coverage in the comparison group.
Concentration curves and indices were employed where the objective was to provide a summary measure across multiple subgroups, which was the case with SES [30]. Concentration curves provide a visual representation of inequality. Concentration curves above the hypothetical line of equality (45–degree line) implies that coverage is concentrated among the poor, below the line of equality mean that coverage is concentrated among the rich and along the line of equality implies equality between groups [29]. Concentration curve was computed with the cumulative percentage of children treated with each of ORS and zinc plotted on y-axis and the cumulative percentage of the population of children with diarrhoea ranked by SES, beginning with the poorest and ending with the richest plotted on the x-axis.
According to Wagstaff et al. (1991), the concentration index is the most appropriate measure of health inequality since it reflects the experiences of the entire population and is sensitive to changes in the distribution of the population across subgroups [31]. As against Erreygers concentration index which has been proposed for ordinal health indicators such as self-reported health [32], Wagstaff indices were used here since the indicator of interest (treatment of childhood diarrhoea with ORS and zinc) were not ordinal and required no scaling.
Defined as twice the area between the concentration curve and the line of equality, the concentration index (C.Index) was computed using the following formula (Fuller and Lury, 1977):
$$ \mathrm{C}.\mathrm{Index}=\left({\mathrm{P}}_1{\mathrm{L}}_2-{\mathrm{P}}_2{\mathrm{L}}_1\right)+\left({\mathrm{P}}_2{\mathrm{L}}_3-\mathrm{P}3{\mathrm{L}}_2\right)+\dots +\left({\mathrm{P}}_{\mathrm{T}-1}{\mathrm{L}}_{\mathrm{T}}-{\mathrm{P}}_{\mathrm{T}}{\mathrm{L}}_{\mathrm{T}-1}\right) $$
where Pt is the cumulative percentage of the sample ranked by economic status in group t, and Lt is the corresponding concentration curve ordinate. T is the number of socio-economic groups or wealth quintiles [29].
Concentration Index takes on values between − 1 and + 1 with 0 representing equality. The index quantifies the degree of relative inequality among subgroups and indicate the extent to which coverage is concentrated among the advantaged or disadvantaged. The larger the absolute value, the greater the disparity. A positive index is obtained when the curve lies below the diagonal (C.Index > 0) indicating that coverage is higher among the richer groups while a negative index is obtained when the curve lies above the diagonal (C.Index < 0) indicating that coverage is higher among the poor [29].