Study area
The MMP LSM/HI trial was part of an operational malaria control project, the Majete Malaria Project (MMP), implemented from 2014 to 2019. MMP was a collaboration between academic institutions (University of Malawi, College of Medicine; Wageningen University and Research; Academic Medical Center, University of Amsterdam; and Liverpool School of Tropical Medicine), non-governmental organisations (African Parks-Majete (AP) and The Hunger Project (THP)) and the Malawi Government Ministry of Health and Chikwawa District Health Office (DHO). The study site and design are reported in detail elsewhere [21]. Briefly, the Majete Wildlife Reserve (MWR) is a wildlife conservation area in Chikwawa district, about 60 kilometres south of Blantyre city, in Malawi. The MMP study area covered an estimated 260 km2, which included three ‘focal areas’ surrounding the MWR (Figure 1). In 2015, the MMP study area catchment population was estimated at 25,000 people. From April 2015 to April 2016, malaria parasite prevalence in children aged 6-59 months was 33.8% (95% confidence interval 30.8-36.9%). (McCann et al, in press) The main source of income for the communities surrounding MWR is subsistence farming.[22]
In Malawi, malaria transmission is perennial, peaking roughly with the rainy season from November to April.[23] The main malaria vectors are Anopheles gambiae s.s., Anopheles arabiensis and Anopheles funestus.[24] Malaria control relies on long-lasting ITNs (LLINs), IPTp, prompt diagnosis and effective case management with RDTs and ACT artemether lumefantrine, respectively. IRS is currently being implemented in selected districts. Both LSM and house screening were used for malaria control in the 1900s but only LSM, specifically in targeted communities, is included in the 2017-2022 NMCP strategic plan, although it is yet to be implemented to date.[21,23]
The main objective of MMP was to reduce malaria transmission through implementation of community-led interventions including LSM and HI, in addition to standard NMCP interventions.[21] Community engagement was a core objective of the project.(15–18) Community volunteers, called health animators, facilitated intensive community engagement workshops throughout the MMP catchment area as well as LSM and HI intervention implementation by community members in their respective villages.[21,25] Health animators facilitated the ‘epicentre approach’ initiated and managed by THP in communities around MWR. An epicentre is a grouping of villages through which communities are supported to create solutions for their own socio-economic problems, with a view towards self-reliance.[26] The MMP was embedded into this epicentre approach with particular focus on malaria control as an additional means towards socio-economic development.[21]
Trial design
The MMP LSM/HI trial used a 2x2 factorial design to assess the effect of LSM and HI on malaria transmission when added to standard national malaria control programme (NMCP) interventions at scale-up for impact targets, over a two-year period. The four trial arms were: 1) NMCP+HI; 2) NMCP+LSM; 3) NMCP+HI+LSM; 4) NMCP only (control arm); hereinafter HI, LSM, HI+LSM and control arms, respectively. NMCP interventions comprised LLINs; IPTp; and prompt diagnosis and effective case management with RDTs and ACTs, respectively. In addition, as part of the community engagement programme, all trial arms implemented ‘malaria village workshops’ aimed at increasing awareness and uptake of the NMCP interventions.[25] Each workshop, attended by residents from one or more villages and facilitated by a health animator, involved discussing a malaria-related topic. These workshops started in all MMP catchment area villages one year before the LSM/HI trial and continued throughout the trial. Additional workshops in HI and/or LSM villages focused on HI and/or LSM, as per trial arm.[22] No IRS was implemented in the study area during the trial period.[21]
Trial interventions
MMP project management and technical staff from the academic institutions provided oversight and coordinated logistics relating to intervention implementation. THP and AP facilitated community engagement and intervention uptake. The district health office granted permission and facilitated integration of trial interventions into the existing health system structures. Detailed staff roles are included in Additional File S1. Following discussions with village leaders and trainings, community members implemented all interventions, as below.
LSM consisted of draining, filling and bacterial larviciding (using Bacillus thuringiensis israelensis, abbreviated Bti, AM625 strain, commercial name: VectoBac WDG [Valent Biosciences, Libertyville IL, USA]) of standing water bodies. Following an initial training of trainers, health animators were trained on LSM rationale and methods and cascaded training to other community members. An LSM committee (up to 12 people per one or two villages) was chosen to oversee LSM, including the organization of draining and filling activities. The LSM committees were also directly responsible for Bti application. LSM activities occurred all year round: LSM-specific malaria workshops once every 2 weeks; Bti application weekly; the majority of draining and filling was conducted once off initially, with maintenance to prevent standing water as needed thereafter.[25] Bti, protective clothing (face masks and rubber boots) and spraying equipment were provided by the project but the communities conducted all mapping of water bodies, draining, filling and spraying, as well as pre- and post-spray larval sampling as a method for communities to monitor the programme. (see Figure 2) Apart from Bti, imported from the USA, all materials required were locally acquired.
HI entailed structural modification of houses to prevent mosquito entry, specifically closing of eaves using locally available materials used in constructing the houses (mostly brick and mud) and screening holes/spaces used for ventilation, including windows, with wire mesh. Except for wire mesh; heavy-duty scissors for cutting wire mesh; and measuring tape, all other materials, e.g. nails, brick and mud for closing eaves and door frame modifications for implementing HI were provided by communities. (see Figure 2) Similar to LSM, communities were responsible for carrying out all HI activities, with HI committees providing community-level oversight of activities and monitoring progress.
Costing approach
This is a retrospective cost analysis from a societal perspective using a combination of ‘top-down’ and ‘bottom-up’ approaches.[27] The societal perspective includes both provider (health system) and community costs and is considered the gold standard in economic evaluation. Top-down costing involves allocating programme level costs to component activities while bottom-up costing involves estimating resource use at a micro level and then summing up to estimate total programme level costs. We also used the ‘ingredients approach’ which involves identifying quantities of inputs (ingredients) used and their unit costs. Analysis was conducted for financial and economic costs.[28] Financial costs are those where money changes hands, whereas economic costs also include non-financial (e.g. donated) resources in addition to financial costs; and thus accounts for all resources consumed.[28,29]
Resource identification and valuation
Trial protocol and operating manuals were reviewed to populate a list of all activities conducted in each of the three intervention arms (i.e. the control arm was excluded), and the activities were categorised into pre-implementation and implementation phases. Activities were categorised as programmatic, i.e. necessary for routine, intervention implementation, based on field operating manuals. Research activities, defined as those not necessary for routine implementation, were excluded from the analysis. For each programmatic activity, requisite resources (‘ingredients’) and their quantities were identified from protocols, operating manuals, inventories, activity logbooks, progress reports and socio-economic sub-studies; and clarified with project staff.[22,25] The proportion of full time employment (%FTE) was used to estimate staff time spent on trial implementation (as opposed to research or non-trial related activities) based on discussions with relevant staff members; estimated staff cost (%FTE x salary) were then allocated to the relevant activity. (Additional File S1) Programme management and overhead costs were directly allocated to trial arms according to estimated proportions of use.[29] Unit prices of purchased inputs were valued based on purchase prices extracted from financial records (excluding taxes). Donated resources, and purchased inputs for which unit prices were missing in financial records, were valued using market prices, e.g. local supplier price catalogues for physical items or services; or Malawian minimum wage rates (for non-skilled domestic labour) for community members' time donation, assuming the Ministry of Health would pay communities in programmatic implementation.
All cost data was entered into a Microsoft Excel spreadsheet and summarised into cost categories i.e. staff, training, donated labour, consumables, transport, equipment and malaria workshops, etc. Equipment with a value of more than US$100 and a useful life of more than one year were defined as capital costs and treated separately in the analysis (below). Thus the final spreadsheet captured quantities and unit costs of each resource input (purchased and donated) used in implementing LSM and HI in the MMP LSM/HI trial.
Cost allocation to trial arms
Given the factorial trial design, all resources were shared by at least two arms. Thus, suitable proxies were used to allocate the estimated total cost of each input to the respective consuming trial arms. Briefly, 1) purchase costs for capital items which would only be purchased in full assuming each arm were implemented independently, e.g. vehicles, were allocated in full to each arm 2) HI-specific resources required for implementing HI-related activities, e.g. wire mesh, were allocated proportionately to HI-containing arms, i.e. HI only and HI+LSM arms, but not LSM only arm; 3) LSM-specific resources required for implementing LSM-related activities, e.g. larvicide, were proportionately allocated to LSM-containing arms, i.e. LSM only and HI+LSM arms, but not HI only arm; 4) non-intervention specific resources, e.g. stationery, were allocated proportionately to all arms. Proxies were weighted for the number of households for HI-related items; habitat size for LSM-related resources; and number of people per arm for non-intervention specific resources. (Figure 3, Additional File S2)
Financial and Economic cost analysis
Resources were separated into financial (where money changed hands) and non-financial if donated, including pre-existing resources (e.g. project administration offices), community members’ time and materials. The financial costing included financial inputs only; while the economic costing considered both financial and non-financial costs. For all resource inputs, the original (consumption) currency, i.e. Malawi Kwacha (MWK), Euro or USD, and year was recorded for all costs. Costs were inflated to 2017 values and then converted to equivalent 2017 USD values using year-average International Financial Statistics inflation and exchange rates (https://data.imf.org/regular.aspx?key=61545862), as described in Turner et al.[30] (see Additional File S3) Capital item costs were annualised in the financial costing by dividing the purchase price by the useful life and similarly, with discounting at 3% in the economic costing, so that only the value of the capital item used during the project lifetime was included in the analysis. For each resource, unit cost was multiplied by quantity consumed to estimate the total financial or economic cost.
The total cost (financial or economic) of each intervention arm is aggregated as total programme cost; total average annual cost; and per household and person per year costs.
The total programme cost for each intervention, i.e. total cost from start to completion of the project, was the sum of allocated costs to that intervention. The total average annual cost is the average of implementation years 1 and 2 costs, which excluded preparatory phase costs. Per household and per person costs are the total average annual costs divided by the number of households in that intervention arm; and the total average annual cost divided by the number of people in the intervention arm, respectively. To aid in planning scale-up, costs are also presented as pre-implementation and implementation phase; by major cost category; and capital and non-capital. Since we aimed to guide implementation scale-up, we allocated shared costs so that each intervention arm had the minimum resources (including administration and village-level resources) required to be implemented independently of the other intervention arms. Consequently, the grand sum of estimated total costs of implementing each arm as standalone exceeds total program costs, but more accurately represents the costs of delivering the intervention arms individually.
For each intervention arm, results are presented in 2017 US$ for the total, annual, and per person and household financial and economic costs. Costs presented below are economic, unless otherwise specified. Financial and economic costs are presented in Tables 1-3.
Sensitivity analysis
The robustness to and effect of structural assumptions and parameter uncertainty on estimated total and per person costs was explored in one- and multi-way probabilistic sensitivity analyses using @RISK software® v 7.6 (Palisade Incorporated, USA); an add-in to Microsoft Excel. Cost categories were included in sensitivity analysis if they accounted for ≥25% of total annual average economic costs. We predetermined to explore the effect of changes in population covered as it is a key driver in cost per person calculations.[28] Thus staff costs and population covered were included in sensitivity analysis simulations; all other input categories were excluded. (See Additional File S4) We simulated (100,000 iterations per simulation) the cost per person for each arm when values of staff costs and population covered were varied. For each iteration, input values were randomly sampled from a triangular distribution (defined by minimum, most likely, maximum values) of possible values of staff costs and population covered to calculate the total/per person costs, holding other input costs constant. For population covered, -/+ 20% changes (as minimum and maximum values, respectively) from the trial baseline mean estimates were assumed. For staff costs, the minimum was defined as total staff costs when all salaries were paid using local salary scales without changing the staff structure; i.e. international staff paid as nationals. The trial estimate was taken as maximum value. It was assumed that staff salaries in a trial represent highest possible personnel costs compared to routine implementation where government programme staff are usually used. For both staff and population covered, the trial mean estimates were taken as the ‘most likely’ value. For each simulation, the cost per person was summarised as a frequency distribution (summarised as mean, 5th and 95th percentile limits) of the 100,000 iteration estimates.
Microsoft Excel ® 2016 was used for all data management and analysis.