Costing approach
A micro-costing approach was used which included all fixed and variable costs in the supply chain. Document review and semi-structured interviews were carried out at the central, intermediate, and facility storage levels. Quantitative information came from financial reports, KEMSA and CAME databases, distribution records, and health facility records. Additional information was obtained from logistics reports, transport schedules, expenditure and audit reports. When possible, data on actual rather than budgeted expenditures were used. Extrapolations and inferences were drawn from existing data when detailed information was not available. Labour costs were calculated on the basis of self-reported hours. Figure 3 illustrates the approach used.
An application for ethical approval was made to the University of Michigan Institutional Review Board on July 10, 2013. A determination of “not regulated status” was made (eResearch ID: HUM00078395) and on the basis of this determination, ethical approval was waived by the appropriate bodies in Benin and Kenya.
Survey guides were developed to facilitate collection of data on costs of each supply chain function including procurement, product quality control, storage, and transportation. Cost inputs were identified and included labour, utilities, security, maintenance, insurance and depreciation [8]. All costs are reported in 2013 US dollars (USD).
Data collection
Data collection took place between September and November 2013. A free, open-source software application, the CommCare® HQ Platform, was used to develop and administer the electronic survey guides to individuals at the central, regional and facility levels and aggregate the data collected.
Sampling
The sites surveyed were selected from Ministry of Health facility lists. The sampling methodology aimed to reflect geographic variety, malaria endemicity, rural and urban locations, accessibility of health facilities, and other factors that can affect supply chain costs. In addition, sites from each level of the distribution system were surveyed. The methodology was not intended to be statistically representative of the entire country, but rather to isolate the most relevant elements to this costing exercise.
In Benin, 22 health establishments were selected (six distribution depots, five hospitals, and 11 health centres) in three departments of Benin: Atacora, Borgou, and Littoral-Atlantique. Facilities were chosen within a two-hour radius of three city centres: Cotonou, Natitingou, and Parakou. In addition, data were obtained from CAME and PNLP. Specialized health facilities, national hospitals, and provincial facilities which did not carry ACT or RDTs were excluded from the sample. A total of 26 people were interviewed at the central (1), regional (3), depot (6) and facility (16) levels.
In Kenya, 22 health facilities were selected (six hospitals, seven dispensaries, and nine health centres) within 18 km from a main road in three provinces: Coastal, Nairobi, and Nyanza. In addition, data were collected at the three KEMSA warehouses and DoMC. One respondent was interviewed at each of the facilities and two at the PPB and NQCL level. At the KEMSA level, a team of respondents provided the information required for the costing. In both countries respondents were pharmacists, pharmacy technicians, nurses, stock managers or clinicians.
Annual throughput
Data on the total volume of products moving through the entire supply chain over the course of one year or annual throughput were collected from KEMSA and CAME using shipment history and procurement status reports from July 2012 through June 2013. This included donor procurements. The annual volume and value of products was estimated by multiplying the number of each stock-keeping unit (SKU) by the volume of each unit in cubic metres and the unit price and summing across all products. SKUs of the four ACT packs were included in the analysis (blister packs of 6, 12, 18, and 24 tablets) in addition to RDT kits. In all cases, “ACT” refers to AL in Kenya and Benin.
Cost categories
Costs for management and administration including any maintenance and information technology (IT) costs were integrated into the functions and not considered as a separate category.
Procurement costs
Procurement expenses included costs of forecasting, tender development, management, and award. Inputs included labour costs for developing the forecast, bid evaluation, and award as well as costs for advertising tenders in the local media. Annual procurement costs in both Benin and Kenya were calculated by dividing the ACT or RDT tenders by the total number of tenders. The resulting figure was multiplied by the annual cost of the tendering process and divided by the total units of ACT and RDTs, respectively. When products were financed by PMI, they were procured directly through USAID | DELIVER and the government did not incur any procurement cost.
Product quality control costs
Quality control costs were mainly associated with the collection and testing of products. In Benin, samples from each product batch of ACT were tested upon arrival in the country. LNCQ charges PNLP a fee for testing each batch. Costs were, therefore, calculated according to the number of batches arriving in the country.
In Kenya, the supplier conducts pre-shipment inspection and this cost was, therefore, not included in the estimates. The only costs for quality testing of ACT were from post-marketing surveillance activities carried out at select sites using a Minilab®. The total cost for this activity was readily available from PPB. Secondary testing, when necessary was performed by NQCL which charged a standard price per sample.
Storage costs
Storage costs included fixed (e.g. infrastructure equipment) and recurrent costs (e.g. utilities, rent, equipment maintenance, stock management, labour, insurance, security, and other administrative expenses). It also included the costs associated with depreciation of equipment, IT, and warehouses and their potential replacement value. Storage costs per unit were calculated according to the total cost figures provided by CAME and KEMSA divided by the total volume distributed that year.
Transportation costs
Transportation costs included labour costs for staff, depreciation of vehicles, fuel, repairs, maintenance, and insurance. Costs were classified as transportation if the cost was associated with the delivery or pick-up from a storage facility. In Benin, CAME Cotonou made quarterly shipments to its depots in Parakou and Natitingou and staff used health facility vehicles or public transport to obtain commodities from these central tiers. ACT and RDTs were shipped separately and associated costs were easily calculated using the above inputs. In Kenya, transport was subcontracted and these costs were readily available from KEMSA. Shipments were made quarterly and the costs attributable to ACT and RDTs were estimated according to the volume of space occupied in a standard truck.
Labour costs
Labour costs for supply chain-related tasks were calculated based on self-reported hours during the interview process by staff responsible for managing malaria commodities. Each staff member’s civil service grade was matched to the public service management salary guide to obtain labour costs.
Data analysis
Our primary objective was to estimate the supply chain costs per unit of ACT or RDT distributed. In all cases a unit was ACT was one treatment course packaged in a single unit pack and a single RDT. This cost was calculated by dividing the annual PSM costs by the annual throughput of all commodities. Throughput was estimated by function, input type, and level of the supply chain. Cost and throughput were expressed in 2013 USD, using an average exchange rate of 494.11 Benin Central African francs and 84.76 Kenyan shillings [18].
The aggregate supply chain cost per unit from the national store to service delivery level was calculated as follows:
Aggregate supply chain cost per unit = [cost per unit at the Central Medical Store (CMS)] + [average cost per dose at intermediate tier(s)] + [average cost per unit for the health centres and hospitals].
Unit costs of ACT and RDTs at health facilities were also calculated. From these, an average cost was derived by taking each facility cost and weighting it to the number of ACT and RDTs procured by that health facility. This cost was then divided by the median \( \left(M\overline{X}\right) \) volume procured and multiplied by the volume of a treatment or test (v
ACT
) to obtain per unit cost for ACT or RDTs. For example, for labour costs per ACT, for every health centre or hospital “i” the formula is represented as follows:
$$ \frac{\Sigma \left( Labor\kern0.5em Cos{t}_i\kern0.5em \times \kern0.5em Total\kern0.5em ACT\kern0.5em Units\kern0.5em Procure{d}_i\right)/\varSigma \left( Total\kern0.5em ACT\kern0.5em Units\kern0.5em Procure{d}_i\right)}{\left(M\overline{X}\left( Total\kern0.5em Volume\kern0.5em Procure{d}_i\right)\right)}\times {v}_{ACT} $$
This calculation was made for each of the other costs: utilities, labour, transportation, security, and depreciation of IT, equipment, and warehouses.
Volumes in cubic metres for each SKU were obtained from KEMSA. Equivalent data were not available in Benin, and therefore KEMSA volumes were used for identical products. For products distributed in Benin that were not stocked by KEMSA, standard volume data for products from the USAID | DELIVER project were used [19]. When data were not otherwise available, proxy volumes from products with similar volumes were used.
The replacement costs for the government-owned storage facilities, associated buildings, and fixed assets (e.g., vehicles and equipment) were estimated using straight-line depreciation. Staff and the portion of other resources allocated specifically to ACT and RDTs were identified. Where this information was not available, weighted averages were used to apportion the costs specifically to these products.
Most health facilities perform mainly clinical activities; therefore the administrative function of a facility covered several other non–logistics-related functions. The cost of administrative supply chain activities at the health facility level was estimated to be 10% of total administrative operating costs. For clinics, 50% of operating costs were apportioned to storage activities. In Benin, zonal depots were dedicated to storage activities and, therefore, incurred 100% of operating costs. Similarly, CAME warehouses had spaces dedicated to anti-malarials and hence also incurred 100% of the related expenditures.
The various cost drivers in each of the four functions were assessed to determine if one function accounted for a large proportion of costs relative to the others, and how changes in the quantity or volume of goods passing through the system affected each function. Depending on the nature of the supply chain function, either volume-proportional or value-proportional cost allocations were used. In the volume-proportional cost allocation method, the fixed costs and overhead costs were allocated to the products in proportion to the relative volume in quantity or cubic metre. In the value-proportional cost allocation method, the fixed costs and overhead costs were allocated to malaria commodities in proportion to the monetary value of the products. Volume-proportional cost allocation, as measured by the quantity of product, was used to allocate costs related to procurement (tender management) and quality assurance. For example, each ACT unit incurs a cost for quality assurance activities. To determine this per unit cost, the total yearly cost of testing ACT was divided by the total number of units shipped from KEMSA for that same year normalized to USD1. Storage and transport functions were analyzed using volume-proportional cost allocation in cubic metres. Value-proportional cost allocation was used only to estimate insurance costs.
Given the uncertainty in the cost estimates, the standard deviation (SD) from the mean weighted costs was calculated [see Additional files 1 and 2].