Study setting and sample size
The study was carried out using a decision-analytic Markov cohort model and depicted the Nigerian scenario. The population used in the study was an estimated size of 34.6 million Nigerian children under five years. This figure was based on the 2018 population report [10]. This population was used as the starting population in the model (well state) but at risk of having childhood pneumonia.
Study perspective
The study used retrospective data related to Nigeria to estimate the benefit and cost of promoting iCCM through the PPMVs. The cost was estimated from the payers’ perspective (government, donor agencies or third-party payer).
Interventions
- Amoxicillin DT: In this scenario (current scenario), amoxicillin DT is given to patients with non-severe pneumonia (moderate pneumonia) where the child has fast breathing pneumonia at a dose of 80mg/kg/day in two divided doses for 5 days.
- Amoxicillin DT plus promotion: In addition to amoxicillin DT for 5 days as in the current scenario above, this scenario (promotion scenario) involves the training of the PPMVs on signs of pneumonia including danger signs, use of respiratory rate timers, how to dose the drug and when to refer patients in complicated cases to healthcare facilities. It also includes the free distribution of one respiratory rate timer to each PPMV shop.
Choice of model and assumptions
We used a simulation-based decision-analytic Markov model in our analysis with a yearly cycle for 5 years. The states in the model were: well state, moderate (clinical) pneumonia, severe pneumonia and death [11,12]. The starting age in the model was under-one year. The model was developed using Nigerian-specific data. The transition probabilities of moving to moderate and severe states were estimated from the national incidence rates from a global systematic review [1,11]. Yearly incidence rates were converted to yearly probabilities using the formula, P = 1 - where P = probability, r = rate and t = time (year). The mortality rate from all-cause of disease and from pneumonia were obtained from the 2017 Institute of Health Metrics and Evaluation report for Nigeria [13]. In the model, children will start from the well state and with each cycle, they may remain well, move to moderate or severe pneumonia or may die as shown in Figure 1.
Figure 1: Model figure
Based on recent education programs to PPMVs conducted by the MCSP, the USAID and the PCN described earlier and elsewhere [9], we estimated that one health educator will train 10 PPMVs per day (in two sessions); 4 training sessions (2 days) will be held per week per health educator; 16 training sessions will be held per month per health educator and 176 training sessions will be held per annum (11 months) per health educator. The experience from the recent outreach revealed that at least a day-spacing is necessary for trainers to prepare for and travel to the next training venue. We assumed that for each PPMV shop, one representative will be trained. The number of PPMV shops was obtained from a recent survey in Nigeria as 24.58 per 100,000 population [14]. Thus, 55 health educators will be required based on this estimate. We included two extra educators in case of unforeseen shortfall of educators which brings the total to 57 educators. Based on Nigeria’s 6 geopolitical zones, 8 educators were assigned to each of four geopolitical zones (north-central; northeast; south-east; and south-south). A total of 12 and 11 educators were assigned to north-west and south-west zone respectively due to their relatively high population with north-west having the highest in Nigeria. One educator will be stationed in the northern region in case of any surge or demand while the other in the southern region. Two vehicles; two computers and two electronic projectors will be allocated to each geopolitical zone. The outreach in 2018 showed that at least five educators were needed for each training to carter for the number of the trainees.
The PPMV shops and community pharmacies are the major providers of health service at the community level for non-severe health conditions in Nigeria. The survey by the MCSP and the USAID showed that care-seeking at PPMVs shops for fever, diarrhoea, cough or pneumonia was 43% in Nigeria and was used in this study [9]. Overall care-seeking at health facilities was found to be about 40% [2,9]. Details of the assumptions are shown in Table 1 and Additional file 1.
Table 1: Input data and assumptions in the model
Time horizon and discount rate
The model simulated cost and outcome within the period of 0 to 5 years for a population of 34.6 million at risk of having pneumonia. Cost and benefit (health outcome) were discounted at a rate of 5% [17].
The measure of effectiveness (relative risk ratio)
The effectiveness of promoting iCCM for pneumonia was modelled as relative risk. The relative risk ratio was obtained from a recent study in Uganda that compared the effect of promotion (through PPMVs education and support with a diagnostic kit and amoxicillin DT) to no promotion [8]. Details of parameters inputs and distribution are shown in Table 2.
Health outcome
The primary benefit of the promotion was measured in terms of disability-adjusted life years (DALY) averted. The DALY calculation was based on 2017 global burden of disease study and we used recently updated disability weights for moderate and severe pneumonia [18]. The DALY was calculated as the sum of the years of life lived with disability (YLD) from morbidity and the years of life lost (YLL) from mortality. The DALYs were calculated for each cycle and accumulated over the time horizon of five years and averaged to obtain the mean DALY per patient. In calculating the monetary value of a DALY, we used the Harvard-led guideline for conducting a benefit-cost analysis project [19]. The valuation was based on ‘‘value of statistical life year’’ (VSLY) with one DALY averted valued at 1.3 times the GNI per capita of a country in sub-Saharan Africa. The secondary outcome was measured as the cost of hospitalisation due to severe pneumonia averted.
Table 2: Parameters and distribution in the Markov model
Determination of cost
The cost was estimated from the payers’ perspective. The cost component for the current scenario (amoxicillin DT) includes the cost of amoxicillin DT (1*10 pack) for children ≤ 1 year (2 – 12 months) or 2*10 pack (for children ≥ 1≤ 5 years) which are the recommended pack sizes [22]. In the promotion scenario, the cost components include amoxicillin DT as in the current scenario above plus the cost of the promotion. The administrative cost components for promotion were obtained from WHO-CHOICE [15] for AFRO D region. The components include a program director, a program coordinator, health educators, an administrative officer, a data entry clerk, a finance officer, vehicles, vehicle drivers, a logistic officer, office, and an external consultant. The cost values were discounted to 2018 USD. The costs were converted using the price level ratio of Naira to US dollar. The cost of other components like electrical utility was obtained from the Nigerian electricity regulatory commission rate [16]. The cost of telephone calls, electronic projectors, hotel accommodation, travel allowance and computers, were estimated based on the recent education outreach conducted in 2018 in Nigeria and from surveys. The costs used in the model were annual costs since the model runs in a yearly cycle. The cost of vehicles was annualised based on useful life years of 8 years [15] while 1year useful year was used for the electronics. The cost of promotion was distributed across the under-five non-severe pneumonia incidence and based on the caregivers’ PPMV care-seeking (43%). All costs were expressed in 2018 US dollars. Gamma distribution was used to capture the uncertainty inherent in the cost parameters. Details of the costs are shown in Table 2 and Additional file 1.
Data analysis
We estimated the cost per treatment course. Based on the report of 0.29 episodes of pneumonia per child year in developing countries [24], we then estimated the cost per annum which was used in the model. The cost data, transition probabilities, relative risk and utilities were made probabilistic using the appropriate distributions as shown in Table 2. The DALY was calculated by combining the YLD and YLL for each cycle year. YLD = number of cases duration till remission or death disability weight [25,26]. YLL = number of deaths due to pneumonia life expectancy at the age of death [26]. The standard life expectancy of < 1 year (54.7) and 1 – 4 years of (57.9) was obtained from the Nigerian life table [27]. The DALYs across each cycle was summed and averaged to obtain the standard DALYs which was used in the probabilistic sensitivity analysis (PSA). The DALYs averted was calculated as the difference between the DALYs lost in the current scenario (amoxicillin DT) and the promotion scenario (amoxicillin DT plus promotion). Half cycle correction using the life table method was employed in the model [28]. The PSA was used to assess simultaneous uncertainty in the variables. This approach is well suited to express overall parameters uncertainty [29]. To assess how simultaneous change of several variables affects the cost and benefit, a Monte-Carlo simulation (1000 iterations) was performed (a type of multivariate sensitivity analysis). This technique runs many simulations by repeatedly drawing samples from probability distributions of input variables. Data were analysed using Microsoft 365 Excel.