The Impact of Retail-Sector Delivery of Artemether–Lumefantrine on Malaria Treatment of Children under Five in Kenya: A Cluster Randomized Controlled Trial

In a cluster randomized trial, Beth Kangwana and colleagues find that provision of subsidized packs of the malaria therapy artemether-lumefantrine to shops more than doubled the proportion of children with fever who received drugs promptly.

after one year of policy implementation in the public sector, AL was only prescribed for 22% of febrile children [17][18][19][20][21][22]. According to a recent evaluation of the Kenyan public sector, weak product supply chains, poor training and supervision as well as a lack in health care workers' prescribing confidence are all thought to have contributed to poor prescribing practices [23]. Weaknesses within the public sector have been acknowledged by the government who are working in collaboration with both local and international organisations to improve performance [24].
Even if a majority of children accessing care within the public sector received AL, the treatment seeking behaviour patterns in Kenya are such that a significant proportion of healthcare is first sought through the private sector [25][26][27][28] (see typology of health care providers in Table 1). A household survey carried out in four endemic districts in Kenya revealed that 90% of caregivers took some action to treat a child's fever within 48 hours of symptom onset. Of these, 47% first sought treatment in the private retail sector and only 35% went to public or not for profit health facilities [27]. A small proportion, 23.3% of all these fevers were treated with an anti-malarial within 48 hours, of which 61% were obtained from the public sector, 28% from the retail sector and only 10% by self administration of medicines available in the household. The proportion of febrile children who received first line recommended AL within 48 hours was only 10.2%. As expected, the majority of AL (95%) was dispensed from public health facilities [27]. What this demonstrates is that health care for malaria is heavily sourced from the private retail sector; however, the services received tend to be poor. Care provided from this sector for the treatment of malaria is mainly based on ineffective medications [25][26][27]. Since a high proportion of individuals seek treatment within the retail sector [29][30][31], encouraging AL distribution within this sector at an affordable price, along with improving the quality of health care services offered is expected to significantly expand the coverage of effective malaria treatment within the community [32].

d) Improving Delivery of Anti Malarials through Retailers
Home Management of Malaria (HMM) is a strategy that has been supported by RBM with the aim of increasing prompt and effective treatment of malaria within the community. This strategy exploits the strengths and improves the services offered by providers outside the public facilities.
HMM in the retail sector can be broken down into various intervention components which include training and capacity building; demand generation/ consumer information; quality assurance and the creation of an enabling environment. Previous pilots carried out on HMM interventions within the retail sector show that they are more successful if they begin with a comprehensive situation analysis to understand the legal and market environment of the retail sector; if they involve all the necessary stakeholders from the retailers to public health officials; if the strategy consists of a combination of approaches; and if the interventions are on-going to ensure sustainability of outcomes [6,31,[33][34][35][36][37][38][39][40][41][42][43][44][45][46][47][48][49][50][51].
Studies have been carried out to evaluate the value of HMM within the retail sector however, the evidence remains limited. A number of HMM interventions based on ACTs are currently underway in Tanzania, Uganda, Rwanda, Cambodia, Madagascar and Nigeria. Preliminary results from the ACT pilot in Tanzania indicate that delivery of ACT through retailers increased uptake of the medication, but access of the medication to those in lower socio economic quintiles remained poor (unpublished data from the Clinton Foundation Tanzania Pilot ACT Study, 2007). However, no evaluations have yet been completed and none are yet underway in Kenya.
Previous HMM studies have focused on other anti-malarial therapies such as SP, however the nature of the dosing and cost of ACTs may alter the outcomes of the interventions. Moreover, many evaluations focus on intermediate outcomes such as provider knowledge and behaviour rather than more health related outcomes such as medicine use within the community. In addition, the outcomes measured vary between studies making comparability difficult [32,52,53]. No studies have used a cluster randomised approach, and most studies do not even include a control group, relying instead on pre and post data only [32,53]. This may have exposed studies to possible confounders. Since very few studies have evaluated outcomes by socio-economic status, there is little information on how equitable these interventions are [32]. Also, many studies have evaluated outcomes soon after implementation of the interventions (3-4 months) so the sustainability of these programs and their long term effectiveness remain unknown [32,53]. There has only been one comprehensive cost-effectiveness analysis of HMM carried out in Kenya [54], therefore little is known on how much the government should budget to roll out HMM interventions on a national scale and the relative value for money of these strategies. Lastly, very few pilots have been published in peer reviewed journals, thus the quality of the data available may be questionable [32,53].

Justification for Study
Kenya has planned to roll out AL into the private sector in 2009 [14] however this decision should be based on sound evidence. Even with the information available from the previous monotherapy retail sector interventions, many key questions regarding HMM interventions and the distribution of AL through the private retail sector remain unanswered. These include a) can retailers provide AL in an appropriate manner, and b) will HMM interventions within the private retail sector significantly improve access to AL in Kenya, and in an equitable way. It is hoped that this and other HMM pilots, based on common outcome measures will be carried out in Kenya to address these questions and augment the data that are already available. All these data can then be collated to provide substantial evidence that can be used by policy makers to guide their decisions on if and how to distribute AL outside of public facilities.

Null Hypothesis
The provision of pre-packaged, subsidized, AL delivered through private sector retailers will have no effect on improving the coverage of prompt effective anti-malarial treatment.

7a) General Objectives
To evaluate to what extent the provision of pre packaged, subsidized, AL delivered through private sector retailers will improve the coverage of prompt effective anti-malarial treatment.

1)
To determine the impact on the proportion of children under five with fever being treated promptly with appropriate anti-malarial treatment, and adhering to the correct dose (accessibility and utilisation).
2) To determine if private sector retailers can deliver AL to appropriate standards of quality for the treatment of fever in children under five years (provision).
3) To determine distribution of benefits of retail sector delivery of AL by socio-economic status (equitable coverage).

4)
To explore reasons for the impact observed and identify any challenges in the implementation process (explanation of experience).

General Study Design
The study design employs a pre-post randomised cluster controlled design, with clusters (sub locations) randomly allocated to intervention and control groups 1 . 9 intervention and 9 control sub locations will be located across 3 districts in Western province. In the intervention areas, prepacked, subsidized paediatric AL will be introduced in September 2008 through selected private sector retailers serving the sub location population, together with a range of supportive activities (see below). Public sector delivery will continue as normal. In the control sub locations, routine delivery of AL through public sector outlets will also continue as normal. Control and intervention areas will be selected with consideration of geographic proximity to ensure that the control areas are not 'contaminated' by the intervention. Baseline data collection will take place in July-August 2008 and the post intervention data collection will occur in July-August 2009. These survey dates are considered to coincide with peak malaria season and therefore have been selected to maximise the number of fevers identified.

The Intervention Package
The intervention package will be designed and implemented by the DOMC in collaboration with Population Services International (PSI), Ministry of Health (MOH) staff at the province and district level and other key stakeholders. The role of KEMRI/ Wellcome Trust Research Program (KWTRP) will be limited to evaluation. PSI is a non profit organisation that promotes healthy behaviour through the use of health marketing programs. Its main focus is on malaria, family planning, HIV/AIDS/STI prevention, diarrhoeal diseases, micronutrient deficiencies and waterborne disease. PSI will draw from their previous experiences from the delivery of retail sector ACT in a number of countries including Tanzania, Rwanda, Cambodia, Madagascar and Nigeria. A brief summary of the intervention is described below. A more detailed intervention plan will be developed in collaboration with the above stakeholders.
The Product -Pre-packaged AL for children under 5: PSI and the DOMC will develop a branded pre-packaged AL product for the treatment of malaria in children. In line with dosing recommendations, two doses will be developed; one for 6 months to 3 year olds (5-15kg) and one for 4 to 8 year olds (15-25kg). The target group for this intervention will be children under five years of age. The product's outer packaging and insert will use locally developed low literacy instructions to improve appropriate dose recognition by caregivers and shopkeepers and promote adherence to the full regimen. The process of product development will be based on extensive formative research, pre-testing and modification in consultation with the case management team of the DOMC. The product's instructions will also include details on the Integrated Management of Childhood Illness (IMCI) danger signs and the need to refer to the public health service severe conditions and children under one year 2 .
Price: It is proposed that the consumer price of the pre-packaged AL will be competitive with the currently available SP and amodiaquine (AQ) brands of malaria treatment. This price is envisaged to be approximately 20 Kenya Shillings (Ksh) per treatment course for all under five doses. This approach will ensure that price is not a significant barrier to accessing effective antimalarial treatment and that there is no significant price incentive for consumers to choose an inappropriate malaria treatment. This price will also undercut artemisinin monotherapies substantially meaning that we can expect demand for them to be reduced in the intervention sites. The price will be clearly shown on the packaging and will be widely promoted in communications to encourage adherence.
Place/Distribution: The pre-packaged product will be distributed directly to selected retail outlets on a routine basis. Outlets selected will be those that are perceived to be well established, respected businesses by the local community. These outlets should already stock anti-malarials or anti-pyretics and have been functioning for a specified minimum amount of time (to be defined following initial retail census). Selected outlets will have a notice visible to consumers informing them that the pre-packaged AL can be purchased from the store, which will tie into promotional activities.
Shopkeeper training: All selected shopkeepers will be given brief focused training. Shopkeepers will be trained to ask key questions related to symptoms and age of the child which will form the basis of their response in terms of referral to a public health facility or supply of an appropriate prepackaged dose of AL, combined with key messages on administering the treatment and adhering to the regimen. The content of the shopkeeper training will be based upon the experiences of the extensive shopkeeper training studies already undertaken in Kenya by the Ministry of Health (MOH) and KEMRI/ Wellcome Trust Research Programme (KWTRP) [55,56]. Shopkeepers will be trained on appropriate storage conditions of AL as well as identifying IMCI danger signs, and the need to refer specific conditions to public health facilities. The trainings will integrate local nurses as technical advisors. PSI will develop training, point of sale materials, and job aids including simple treatment algorithms with guidance from the DOMC and the drug policy technical working group (DPTWG) to improve the quality and quantity of information provided to consumers.
Promotion: PSI will carry out a series of promotional activities in the intervention areas and related dominant market centres. Messages will target caregivers of children under five and will promote appropriate treatment seeking behaviour including the benefits of AL and its availability both in public sector facilities and identified private sector outlets. Messages will be delivered through a range of interpersonal communications media including community drama, road shows, chief's barazas, and community group educational sessions. MOH public health technicians will also be briefed on the program and encouraged to conduct community awareness programs and integrate messages into health talks. In addition, where there are active district ITN advocacy/ information, education and communication (IEC) working groups, their support will be sought in conducting awareness raising in the targeted communities. These activities will be supported by point of sale materials, posters, leaflets and localized media such as wall paintings (mass media strategies will not be used in this pilot phase to avoid contamination between intervention and control areas).
Diagnosis: Diagnosis of malaria within the retail outlets will be based on a history of fever, as is standard practice in a majority of endemic rural public health facilities and stated in the World Health Organisation (WHO) guidelines [57,58]. Shopkeepers will be educated on this through training and IEC material.
Drug Regulation: AL is currently registered in Kenya as a prescription only medicine (POM). However, the transitional plan for implementation of the ACT Malaria Treatment Policy in Kenya states that AL should eventually be deregulated and that studies must be conducted to guide this process. PSI, KWTRP and the DOMC will seek exemption from the POM regulation for this pilot.
Pharmacovigilance: The Pharmacy and Poisons Board (PPB) have developed guidelines and tools for the collection of pharmacovigilance data on AL since its release in the public sector.
The intervention package will therefore be implemented in collaboration with the PPB to ensure that pharmacovigilance requirements are met. Shopkeepers will be educated on possible adverse effects and will advise caregivers to seek care from the nearest health facility for any suspected adverse drug reactions (ADRs). The shopkeeper will complete a CHW referral form for each such caregiver; one copy will be taken by the caregiver to the facility and the other copy will remain with the shopkeeper and later collected and sent to the PBB. All ADRs seen within health facilities will be reported back to the PPB. The PPB along with district investigation teams will be involved in following up any serious ADRs. The DOMC and PSI will work together with the PPB and provincial pharmacist to ensure staff in health care facilities within the province receive training on pharmacovigilance with regards to AL and to make them aware of the pilot.

8a) Study Sites
The study will take place in rural areas of 3 districts in Western Province. This province was selected for the following reasons:  It is malaria endemic (Figure 1),  PSI has a strong team of permanent sales staff on the ground 3 ,  There are no other planned HMM interventions 4 ,  Presence of a relatively active retail market.
Within Western province, Bungoma district was excluded because of the extent of previous malaria-related interventions which may make it atypical. Mount Elgon district was excluded because of the current political insecurity in that area. The districts chosen for the pilot were therefore Butere/ Mumias, Teso and Busia ( Figure 1). Butere and Mumias were officially divided into two districts in mid 2007, but because this split is recent, the two districts will be treated as one for the purposes of this study.
The pilot intervention will be implemented at the sub-location level. The sub-location was selected for the following reasons:  To provide a reasonable scale for implementation,  To ensure no contamination between intervention and control (which would be more likely if larger areas were used),  To contain the total medicine cost for the pilot phase.
The following inclusion criteria were used for selecting intervention and control sub-locations:  They are rural, as this is where access to health care is most limited and so HMM most needed. In addition retailers in urban and peri-urban sub-locations serve customers from a much wider area, which would lead to contamination;  The populations within the sub-locations are between 2,500 to 10,000; smaller sub-locations were excluded to ensure there was a reasonable scale for implementation and adequate sample sizes for the evaluation; larger sub-locations were excluded to contain the cost of the pilot.
Through random selection 18 sub-locations were identified, stratified across the 3 districts, each district having 3 controls and 3 interventions (Figure 2-4). To avoid contamination a buffer of at least 2 sub-locations was maintained between any selected sub-location. The sub locations shown below are the outcomes from our preliminary sample selection. The final selection will be confirmed after further visits to the areas by PSI to investigate feasibility.  N.B: 'Other' (see Legend) refers to all sub locations that do not fit the sub location criteria (e.g. urban or peri-urban and with populations <2,500 and > 10,000).  Table 2 gives an overview of the districts' demographics. Teso is the least poor district with 50% of the population living under the poverty line, followed by Butere/ Mumias, with Busia being the least well off. Butere/ Mumias is the district with the largest population, Teso is the least populated, containing around half the population of Butere/ Mumias. Butere/ Mumias is also the most densely populated district with 611 people per KM 2 . The number of health care facilities are greatest in Butere/ Mumias followed by Busia and then Teso. In all three districts just over 50% of household heads have completed primary school. The languages spoken in Busia and N.B: 'Other' (see Legend) refers to all sub locations that do not fit the sub location criteria (e.g. urban or peri-urban and with populations <2,500 and > 10,000).
Previous malaria control initiatives such as training of shopkeepers and community awareness programmes have been carried out in Busia by both the Ministry of Health and Non Governmental Organisations (NGOs). The DOMC and PSI will be intervening in all these districts to implement malaria community awareness programs and train public facility health care staff, funded by the President's Malaria Initiative (PMI). These activities will cover both pilot intervention and control areas, and take place after baseline data collection for this study has been completed. This allows us to assess the value added of retail sector provision once appropriate efforts to strengthen the public sector have been put in place. A more detailed context analysis will take place during the pilot to identify other health initiatives within the districts that may affect the outcomes of the pilot.  Table 3 shows the demographics for the selected sub-location populations, the unique ID represents the numbers of the individual sub-location on the district maps (Figures 2-4). These data reflect the experience at the sub location level. The populations within the selected Teso sub-locations tend to be less poor than the other two districts, with the percentage living below the poverty line ranging from 48-52%. In Busia the percentage living under the poverty line ranges from 64-69% while in Butere Mumias it is from 58 to 69%. Butere Mumias is the most densely populated with sub-location population densities ranging from 476 to 748 per KM 2 . Busia and Teso's population densities are quite similar with Teso ranging from 293-374 per KM 2 and Busia from 200 to 473 per KM 2 . Across the three districts, the average % poor and population densities between the control and intervention sub locations are similar ( Table 4). Retail outlets can either be found within a market centre where there is a group of shops together or outside of market centres as 'stand alone' or individual shops. There are on average 8 retail outlets in a market centre. Based on study visits and discussions with PSI and MOH staff, it is estimated that an average of 8-10 'stand alone' retail outlets and 3-4 market centres containing retail outlets sell medicines to a rural sub-location in Busia; in Butere/ Mumias around 15-18 'stand alone' retail outlets and 5-6 market centres, while in Teso it is about 12-14 'stand alone' retail outlets and 5-6 market centres. These details will be confirmed by further visits by PSI and through the initial retail census (see below) to be carried out by KWTRP.

8b) Study Populations
A total of six data collection activities will be conducted to evaluate the intervention. Specific study populations for each activity are listed below. and Global ACT subsidy monitoring and evaluation frameworks. The compulsory indicators are to be monitored by all pilots evaluating interventions to increase anti-malarial access outside the public sector within Kenya. Standardizing outcome measures between pilots will allow for data to be compared across studies, and will also ensure that the studies provide the DOMC with the relevant information they require for policy implications. The Red Cross will also run an HMM pilot based on these indicators, other future pilots are to be developed.
In this study we will focus on the compulsory indicators and some optional indicators that we consider to be relevant to monitoring the effect of the intervention. As previously mentioned, these indicators will be monitored through 6 data collection activities, each with its own data collection tool ( Figure 5). The tools for each activity have drawn on studies which have evaluated similar outcomes within the retail sector and the household level [27,[60][61][62]. The tools will be piloted before the studies to ensure questions are interpreted correctly by the respondents. The household survey, retail census, provider survey and mystery shopper survey will be administered both before and after the interventions in both the intervention and control areas. The focus group discussions will only be administered post intervention and only in the intervention areas. Data collection for the documentation of context will be on-going throughout the evaluation.

1) Retail Census Survey:
The retail census aims to identify all retail outlets selling medicines that serve the population of the study sub-locations. This will include retail outlets located both within the study sublocations and those in neighbouring sub-locations that are frequently used by the study population. For the purposes of this study, retail outlets include all shops selling medicines, including general retailers and registered pharmacies where these exist. Outlets such as bars, hardware shops and salons are excluded. Retail outlets will be identified initially from lists provided by PSI. This will be supplemented by visiting the areas and confirming the list with local leaders. In addition, a snowballing technique will be used where shopkeepers of known retail outlets will be asked to identify other retailers within the local area. All identified retail outlets being regularly accessed by the populations within the selected control and intervention sub locations will then be mapped. Field workers will visit each retail outlet and position it using a Global Positioning System (GPS) unit (Garmin etrex (Garmin Ltd, Kansas, USA) and Trimble 12 band). Three longitudinal and latitudinal readings will be taken for each outlet and the average reading noted to minimise errors in positioning. Current estimates indicate that the accuracy of GPS readings is within 15 meters of the true position [63]. Details of the type of anti-malarial medicines stocked within each outlet will also be recorded.
The database derived from these analyses will form the sampling frame for other retail outlet surveys and will also be used to monitor any leakage of public sector AL into retail outlets, any stocking of project AL by untrained private retail outlets, to monitor the coverage of the intervention and the distance of retail outlets to neighbouring households, hence accessibility.
The following indicators will therefore be monitored from this survey: A The retail census will take place before other baseline data collection activities and will be updated before the follow up data collection activities both in the control and intervention areas. Data will be collected on retail census questionnaires. It is estimated that the census will cover a maximum of 49 retail outlets per sub location; 882 retail outlets in total ( Table 5). Verbal consent to collect these data will be obtained from the available shopkeeper in each retail outlet.

2) Household Survey
The indicators to be measured include: These indicators aim to monitor any changes in access to effective anti-malarial treatment, adherence, treatment seeking behaviour and knowledge of malaria within the households. Indicators B, C and H will also be important as factors which potentially affect the development of resistance.
The household survey will be carried out both at baseline and post intervention, within the intervention and control sub-locations. Three enumeration areas (EAs) will be selected per sublocation (with probability proportional to size of EA). We will randomly sample 20 homesteads within each EA, equivalent to a total of 540 homesteads (roughly 1922 households) in each group. We will aim to visit the same 20 homesteads in the follow-up as in the baseline to optimise statistical power. Where it is not possible to visit the same homestead we will replace it with its nearest non-sampled neighbour. The sampling frame of homesteads in each EA will be created with the help of local village leaders. The list will be confirmed by visiting the EAs a month before the field work commences and recording GPS co-ordinates for each identified HH.
A household survey questionnaire will be administered to the household head and all care givers of children under 5, filling in one questionnaire per child. All indicators apart from indicator E will be focused on children under five years since they are the most vulnerable group. To monitor 'the proportion of non targeted household members receiving the intervention AL' (indicator E), a brief additional questionnaire will be administered to all members of the household aged 5 or over who had a fever within the past 2 weeks (care givers will be interviewed on behalf of children aged [5][6][7][8][9][10][11][12][13][14][15], to estimate what treatment was used and from which source. Indicator D will provide information on the leakage of project AL to untrained outlets. This will augment data collected for indicator C in the retail census i.e. 'availability of private sector project AL in untrained private retail outlets'.
The survey is restricted to fevers occurring in the previous two weeks since the recall period beyond this time point is questionable [64]. Although the RBM indicator states that treatment should be sought within 24 hours, 48 hours has also been included in indicators A and B as this could still be considered prompt. The denominator for 24 and 48 hours in these indicators includes all individuals reporting fever and visited by interviewers 2 to 3 days or more after the onset of symptoms. As the intervention aims that all under 1s should be referred to a health facility, and should not therefore receive AL from retailers except in an emergency, Indicator A will be calculated both with and without this group in the numerator. For indicator B, adherence will be defined as taking the quantity of medicine specified in the MOH treatment guidelines, and both under and over dosing will be considered as non-adherence. The timing of administration between doses within the 3 days will not be considered as recall of specific times may prove difficult. A photo-illustrated guide will be used to aid in the identification of antimalarial treatments [25,64].
Wealth indicators based on those used by the Demographic and Health Surveys (DHS) will be collected this survey to determine the socio-economic status of each household and gain better information on the equity impact of the intervention [65,66].Wealth indicators will include housing quality, sources of income, education status and ownership of livestock of the household head. Principal components analysis (PCA) will be used to construct a household wealth asset index from the information collected. The derived wealth indices will then be used to classify households into wealth quintiles [67]. In addition households can be linked to national socioeconomic quintiles using asset weights from nationwide surveys. Wealth indicators will be collected during both baseline and follow-up household surveys since they are subject to variation [67]. GPS co-ordinates will also be taken for each household interviewed to determine the distance of the household to the nearest retail outlet and facility.
Written consent will be obtained from the household head or their representative and verbal consent from the interviewee. The village elder or chief will be informed about the survey in advance. The village elder will be asked to aid the field team in identifying the households to be interviewed within the designated enumeration areas.

Sample Size Determination for Household Survey
The sample size calculation is based on a cluster randomized before and after study design, with 9 intervention and 9 control sub locations stratified across the three districts (Busia, Teso, Butere/ Mumias). The key outcome indicator is indicator A: The proportion of children under 5 years with fever in the past 2 weeks who started treatment with AL within 24 and 48 hours of fever onset, overall, by SEG and treatment source. The study design chosen for this study is a cluster randomised sample, for logistical reasons. The cluster randomisation will occur at the EA level with the primary sampling unit being the household.

Using a 'difference in proportion' sample size calculation and an estimated design effect the following was calculated (Appendix 3):
To detect a difference of 20 percentage points in the key outcome indicator, with a 5% significance and 80% power: n = [(z alpha/2 + z beta ) 2 (p1 (1-p1) + p2(1-p2)] / (p1 -p2) 2 Where: n = desired sample size (recent childhood fevers) in each group z alpha/2 = 1.96 (5% significance) z beta = 0.84 (80% power) p1 (control) = 0.2, the starting percentage (data derived from a fever survey carried out in June 2007 showed 11% of children under the age of five had access to an anti-malarial within 48 hours. This was increased to 20% to be conservative as a larger p1 will require a larger sample size).
p2 (intervention) = 0.4, the starting percentage plus a 20% point increase (assuming that a 20% point increase is the minimum necessary to justify the importance of the intervention in public health policy terms).

n = 79 childhood fevers required in each group (i.e. control and intervention arms)
Based on previous surveys carried out in Kisii, Kwale and Bondo [27,60], 3.4 homesteads (HS) need to be visited to find one recent childhood fever. Therefore a total of 269 HS will be required in each group.
Assuming a design effect of 2 to allow for clustering, a total of 538 HS will be required in each group, equivalent to 60 HS per sub-location (or around 214 households per sub-location [27,60]). If 20 HS were surveyed per EA, we would need to evaluate 3 EAs per sub-location. This would equate to a total of 27 EAs in the intervention group and 27 in the comparison 6 . This will be feasible for the majority of EAs, as data from Kisii, Kwale and Bondo show that the number of HS per EA ranges from 3-110 (mean=36, inter quartile range= 23-41). Where EAs with less than 20 HS are selected, they will be replaced with the nearest EA with at least 20 HS ( Table 5).
In addition, sample size calculations estimates were also made using the Hayes and Bennett Cluster Sampling Formula IJE 1999 7 (Appendix 2), which also showed that 27 EAs in each group would be adequate as a sampling frame.

3) Provider Survey
The indicators to be measured include: A. The median price charged by retail outlets for AL by age band B. The proportion of retail outlets with no expired AL available in stock C. The proportion of retail outlets reporting stock outs of AL within the past 2 weeks D. The proportion of retail outlets storing AL appropriately (see definition of 'appropriately' below) E. The proportion of retail outlets that have copies of the materials/ job aids required by the intervention (e.g. leaflets, posters, guidelines, pharmacovigilance reporting forms)

F. The proportion of retail outlet staff who know the correct information about malaria diagnosis and treatment practices
These indicators aim to monitor the affordability of anti-malarial therapies within the retail sector, whether any subsidies are passed onto the consumer and the nature of retail supplies of anti-malarials and source. In indicator C, the period of monitoring stock outs i.e. two weeks is chosen for recall purposes. In indicator D, 'appropriate' storage refers to keeping medicines off the floor, in a dry area, away from direct sunlight, and with the packaging intact. Also included in this survey will be questions to assess the knowledge of the provider including what he/ she knows about malaria symptoms, diagnosis, treatment and ADRs.
The indicators will be assessed through a provider survey questionnaire. Questions will be asked to the retailer present at the shop. Verbal consent to administer this questionnaire will be sought from the interviewees.
Sample Size for the Provider Survey: The survey will be conducted in outlets selected to deliver the subsidised ACT product. Depending on the total number of outlets included in the intervention, we may include all outlets in the provider survey, or take a random sample. It is estimated that around 150 outlets will be surveyed in each group (i.e. control and intervention groups ( Table 5).

4) Mystery Shopper Survey
The indicators to be measured include: A. The proportion of retail outlet staff that dispense AL to patients presenting with fever B. The proportion of retail outlet staff that dispense the correct dose of AL to patients presenting with fever C. The proportion of retail outlets that provide appropriate information to caregivers on how to give/ take AL D. The proportion of encounters where retail outlet staff asked for one or more IMCI danger signs to determine need for referral E. The proportion of retail outlet staff who told caregivers about any signs of progressive illness and recommended a referral visit to a facility or clinic if the signs appear F. The median price charged for an AL dose by age group The aim of this survey is to analyze the patient provider interaction to give better information on actual rather than self-reported provider behaviour. This survey will take place before the provider survey to avoid making retailers aware that they are being evaluated. It will augment and allow for triangulation with the provider survey.
The survey will take the form of a mystery shopper survey. Here, field workers will pose as ordinary care givers seeking care for a four year old child with fever. Questions will be administered to the staff selling medicines at that time, and any recommended medicines will be purchased by the field worker. Details of the interview will be filled in away from the outlet. Questions will be standardized between retail outlets to allow for comparisons of behaviour between retailers and over time. The indicators and data collection tools for this activity will be further refined once a detailed outline of the intervention activities have been developed. This method was chosen instead of direct observation or exit interviews because it minimises any potential bias that may occur through knowing one is being observed. In addition, achieving a reasonable sample size for exit interviews could be very time consuming in outlets which receive very few fever customers per day. This technique does however raise some ethical concerns as informed consent cannot be obtained from the medicine seller at the time of the interview [68][69][70]. Verbal consent will therefore be sought from all shopkeepers during the retail census for their willingness in principle to participate in the mystery shopper survey. They will be informed on what the survey involves, however neither whether their shop will be selected for the survey nor the date of the visitation will be revealed as this may affect the study outcomes.

Sample Size for Mystery Shopper Survey:
The mystery shopper survey will be conducted in the same outlets as the provider survey ( Table 5).

5) Focus Group Discussions (FGDs)
Focus Group Discussions will be carried out only after the implementation of the intervention and only within the intervention sub locations. The purpose is to ascertain the perceptions and opinions of the intervention from community members and shopkeepers. This may help confirm and explain observations in the quantitative data. Each group will contain 6 to 10 people. Two FGDs will take place per intervention sub location (i.e. a total of 18 FGDs, see Table 5). In each sub location a group of selected shopkeepers and a group of selected caregivers of children under five will be interviewed separately. To ensure that a range of experiences are presented, shopkeepers will be selected purposively, based on data from the provider survey to ensure that they include a mixture of those considered to be well and poorly performing. Care givers will be purposively selected based on treatment seeking behaviours, as revealed from the household survey.
Discussions will take the form of semi-structured interviews. Shopkeepers will be asked about the benefits and costs of being involved in the program as well as problems they may have experienced in the implementation process. Care givers will be asked about their opinions on treatment sources, availability of products, barriers faced in seeking treatment and any concerns they may have regarding the program. In addition the use of intervention AL in adults will also be explored (drug misuse).
A letter of invitation to the FGD sessions will be presented to the care-giver, his/ her household head and to the shopkeepers to be interviewed. It will explain the duration of the interview and that they may need to arrange for someone to carry out their daily activities for that period of time. A travel allowance will be given to each interviewee and drinks will be provided during the sessions. Verbal consent to carry out these discussions will be gained from the interviewees prior to initiation of the discussions. Discussions will be tape recorded and records supplemented by notes taken during the discussions by a field assistant. This will ensure the capture of verbal and non verbal cues and facilitate exploration of arising issues.

6) Documentation of Context
We will undertake careful documentation of context at national and district level and other factors which may have influenced the study outcomes in both the intervention and control areas.
A list of issues that may be of concern will be prepared to provide a framework for data collections. Throughout the study, a series of desk-work analyses of newspaper articles, minutes to meetings, draft proposals, budget allocations and memos, as well as in-depth discussions with the District Health Management Teams (DHMTS) at the local level, the DOMC at the national level and other relevant bodies will take place. From this, a chronology of events will be documented, as well as a summary of the events, the locations and the duration. These data will be taken into consideration during evaluation of the study outcomes. FOCUS GROUP DISCUSSIONS Care giver groups of 6-10 1 9 9

Estimating the Cost and Cost-effectiveness of the Intervention
Cost and cost-effectiveness data is considered important by the DOMC for assessing the affordability and value for money of all potential interventions. However, it is recognised that the unit costs of small scale pilots will not reflect those of nation-wide implementation, as they exclude certain activities (e.g. mass media) and do not reflect economies of scale. KWTRP will not therefore conduct a detailed costing of the pilot. However, the DOMC have decided to employ a costing consultant who will work together with PSI and other implementing agencies to provide comparable estimates of the costs of future nation-wide interventions. The indicators to be measured include: A. Implementation cost B. The cost-effectiveness in terms of the incremental cost per additional child receiving prompt AL treatment For the cost-effectiveness indicator, the 'effectiveness' measure will be obtained from the household survey. Potential differences in effectiveness between pilot and nation-wide implementation will be assessed using sensitivity analysis.

Feasibility of the Pilot in the Current Political Situation
In view of the current post election violence, certain measures will be undertaken to ensure the safety of the research study team:  The security and acceptability of the survey work will be discussed at length with local district ministry staff, police, community elders and KEMRI management before travelling to the field.  Recruitment of local staff will be sensitive to the ethnic tensions within the selected communities that will be surveyed.

ii) Provisions for data verification, and validation in the field
Field workers selected to administer the questionnaires will be chosen according to their level of field experience. They will be fluent in the local dialects spoken in the chosen districts. They will translate the tools into the local language then back translate them to ensure the questions have been translated correctly. Field workers will also undergo two days of training before baseline and intervention surveys to educate them on the purpose of the study, how to fill in the questionnaires and the quality of work expected from them.
As previously mentioned, the data collection tools (both qualitative and quantitative) have drawn on previous similar surveys [60,62,71], with results which have been published in peer reviewed journals. They have been adapted to suit the purposes of this study and will all be piloted prior to the surveys and further adjusted, if necessary. Data collected from the different tools can be compared to validate responses from different sources.
During the surveys, questionnaires will be reviewed at the end of each day by the team leaders to ensure all entries are completed appropriately. Any tools with incomplete or questionable entries will be sent back to be re-filled the next day. In 10% of the interviews in the retail census, the household survey and provider survey, field workers will be accompanied by the team leader to monitor their performance and see where any improvements can be made. Also, a random sample of 5% of interviewees in each area will be re-interviewed by team leaders (backchecking) to ensure concordance in responses. The focus group discussions will take place with the principal investigator (BB Kangwana) and an interpreter.

9) Data Management 9a) Data Storage
All data collected from the surveys will be cleaned before and after data entry and will be stored on PCs and USB drives at the KWTRP unit in Nairobi. All data on computers will be secured through the use of passwords. Tapes from the qualitative analyses, field notes and hard copies of the questionnaires will be stored in locked filing cabinets on site. All data entered into software will only have individual identification numbers. Access to any of these materials will be limited to investigators and when necessary clerks for data entry purposes.

9b) Data Management/ Data Analysis
Quantitative Data: Data collected from the retail census, household survey, provider and mystery shopper tools will be entered into individual customized data entry screens on MySQL (Version 5.1, California). These screens will be designed to be as similar as possible to the hard copies with consistency checks and built in ranges. Data will be double entered into the system by two different clerks. Microsoft Excel 2007 and STATA (version 9 StataCorp, College Station, Texas) will be used to analyse these data. To account for the different sampling weights of the EAs in the districts selected, estimates of proportions and means will be calculated using STATA's 'svy' commands. A wealth indicator will be calculated using PCA of household assets. From this, quintile distribution will be derived [67].
Qualitative Data: Data collected from the focus group discussions will be entered and transcribed in NVivo 7.0 software (QSR International). This software will allow data to be coded systematically into themes and sub-themes that can be evaluated to highlight the perspectives of households and retailers on the retail intervention package.

11) Ethical Consideration
This pilot was designed to provide the DOMC and MOH with the evidence they require to inform the decision on whether AL should be rolled out into the private retail sector, and how. The DOMC participated and approved in designing a list of key indicators which they feel are important for their decision making process. The key indicators also incorporate DOMC targets. All information collected from this study will be fed back to the DOMC to inform policy.
Community Engagement: Prior to baseline and post-intervention surveys, letters will be sent to the relevant DHMTs and District Officers (DOs). These letters will explain the study and its purpose. On arrival, the field team leader will meet with both parties, present a copy of the letter and verbally discuss the planned field work. During these discussions concerns or queries about the survey may be voiced. Once approval has been gained a representative from the DHMT will escort the team leader to the relevant village elders, chiefs and/ or local leaders. They will be given a copy of the letter given to the DHMTs together with a verbal explanation. Queries and concerns will be addressed during this time and verbal approval will be sought to go ahead with the survey. The DHMTs, DOs and community leaders will have access to the team throughout the survey so any concerns that arise during the survey can be addressed to the field team leader.
Individual consent: During the surveys, consent will be obtained from the necessary individuals. Informed sheets and consent forms will be translated into the local dialect. A copy of the information sheet will be left for the household heads and shopkeepers. The information sheet will include an introduction, the purpose of the study, how questions will be administered, the risks and benefits to those who participate, that the data collected will be confidential and that participation is purely on a voluntary basis. Written or verbal consent will be obtained prior to baseline and intervention surveys. All verbal consent will be witnessed by a study team member. For the household survey written consent will be gained from the household head or his/ her representative and verbal consent from all other interviewees. For the retail census and provider surveys verbal consent will be obtained from the interviewees; for the focus group discussions, verbal consent will be obtained from the caregivers and from the shopkeepers. Verbal consent for the mystery shopper survey will be obtained from the shopkeepers during the retail census survey. In all surveys interviewing shopkeepers, verbal as oppose to written consent will be obtained since previous experience shows that shopkeepers tend to be very cautious of inspectors and suspicious of having to sign documents that may be used to identify them or their shop.
Participants will be able to drop out at anytime within or between surveys and do not have to respond to questions they do not wish to. However, reasons for not participating or not answering questions will be obtained as this may be due to aspects of the surveys or intervention that are not acceptable to the community and that will therefore need to be acknowledged and possibly altered. Great efforts will be made to ensure that participant's confidentiality will be maintained at all times. This includes restricting access of all data to the investigators and data entry clerks when need be, destroying of tapes from FGDs after the research is completed, as well as using individual identification numbers on software programs for data analysis.
Training for those involved in administering consent: All fieldworkers will undergo training prior to both studies. Training will educate fieldworkers on the purpose of the study, the importance of consent and how to administer both the consent forms and questionnaires. While in the field, fieldworkers will have daily contact where possible through the use of a mobile phone with the team leaders and KWTRP headquarters in case of any queries.
One potential concern is that AL will be temporarily de-regulated in the study location from a POM to an over the counter medicine (OTC). AL has been used in public facilities for one year, however very little evidence from wide scale distribution is available on the range and impact of adverse effects of AL as an OTC medicine. The use of AL in this study will only be allowed after approval from the PPB. Retailers will be trained on when to refer patients to the public health sector; this information will also be printed on the pre-packaged AL leaflets.
The control areas will not have access to AL as an OTC medicine however, the government will continue providing it free of charge within the public sector in all districts. Therefore the existing access to AL in the control group will not be adversely affected.
During this pilot, retail sector AL will be heavily subsidized within the intervention group which could raise questions on sustainability. If, from all the available evidence the DOMC are willing to distribute AL within the community then there is a high possibility that funds to subsidise this medication at this level/ scale will become available through the implementation of Global Subsidy Funding for ACTs [72].

Feedback of information:
Please refer to item 12: expected application of results.

12) Expected Application of Results
The results from this study will be fed back to the DOMC, DHMTs in the study sites, and other interested stakeholders. This will occur together with other studies taking place within and beyond Kenya on retail sector ACT delivery and other strategies to improve access. This pilot will form part of the evidence required by the DOMC to inform the decision on whether distributing AL through the private retail sector will increase prompt access of effective antimalarial treatment to those who may benefit from it. The evidence may also be useful to other countries thinking of rolling out similar interventions and to national and international organizations who may be willing to fund such interventions at a national level.
The findings will be communicated to relevant and interested stakeholders through written full reports, summary briefing notes, group discussions and one on one meetings, published papers and presentations at international conferences.

15) Justification of Budget
This budget only covers the cost for the evaluation of the pilot. All intervention costs will be estimated and finance sort by PSI.

Personal Salaries and Benefits:
The principal investigator, (BB Kangwana) and an assistant research officer will be paid for a period of 24 months. The role of the assistant research officer will be to monitor the progress of the surveys in the field. Salaries for the research assistant and assistant research officer have been calculated using the appropriate KWTRP salary scales. Salaries also include 8 months for Catherine Goodman who will provide the majority of supervisory support. She will be paid according to the London School of Hygiene and Tropical Medicine salary scale.
Operating Expenses: This includes all costs required to carry out the field studies. The retail census, household survey, provider survey and mystery shopper survey will be carried out twice over a period of two years, at baseline and post intervention. The focus group discussions will only be carried out post intervention while documentation of context will be collected throughout the study period. The operating costs include travel costs, per diems and salaries of the field workers as per KWTRP guidelines. Training days will be held before each survey, one for the team leaders, one for the field workers on data collection, and one for all staff on the use of GPS units to map households and shops. Salaries and transportation costs will need to be paid to those attending the training days. Once in the field, village elders will be asked to help in locating the households and shops and will be paid 100ksh per day. The salaries and per diems are included for one field worker and a team leader to remain in the field for an extra three days after the surveys are complete to deal with any remaining errors that need to be corrected (call backs). Also included are costs for the production of the questionnaires, costs to pilot the questionnaires, the salaries for a software developer to design the data entry screens and clerks for data entry purposes and the costs to purchase 11 GPS units.
Supplies: This is the estimated cost of all the non-field work supplies required for the study at the KWTRP unit in Nairobi. It includes photocopying, stationary, communication and printing over the two year period. The estimates are based on routine costs charged at the KWTRP.
Travel and Accommodation: During the protocol development and study preparation an investigator will need to meet with a statistician and retail sector consultants in Kilifi. Travel and accommodation costs for these trips have been factored in here. Again, while the surveys are going on, an investigator will need to travel to the field sites for a couple of days to monitor the progress on the ground, check the quality of the data collection and deal with any major issues that may present. All costs for travel and accommodation have been estimated using standard KWTRP mileage and per diem charges.

16) Appendices 16a) Role of Each Investigator
Miss BB Kangwana: PI-Will be responsible for proposal development, supervision of training, data collection, analysis and report preparation and communication.

Dr C Goodman:
Will play a supervisory role in designing and co-ordinating proposal development and provide scientific guidance for the study.

Dr G Fegan:
Will provide statistical support in study design and analysis.
Dr AM Noor: Will be involved in developing GIS platforms and evaluation of socio-economic status.
Dr AJ Nyandigisi: Program pharmacist, Division of Malaria Control. Will be the principal representative/ liaison person for the DOMC. Will be responsible for coordinating planned HMM pilots for DOMC purposes and, in informing policy discussions on issues raised in the protocol and subsequent publications arising from the protocol.
Dr WS Akhwale: Head of Division of Malaria Control, Ministry of Health; will be responsible for informing policy discussions on issues raised in the protocol and subsequent publications arising from the protocol.

Pharmacy and Poisons Board:
Will act as an advisory body on pharmaceutical issues in the study, responsible for informing policy decisions on issues raised in the protocol and subsequent publications arising from the protocol. The PPB will also be responsible for the monitoring and evaluation of ADRs. This includes working with PSI and the DOMC in distribution of referral forms, training of shop keepers and health care workers on how to identify and record ADRs, and collection and analysis of data from referral forms and any other tools used to record ADRs..  Management information systems involve using retrospective data collected by the non-facility outlet to determine the prescribing and dispensing practices of the provider Data will not be strictly comparable between the above methods. However, no single method is possible in all studies for practical reasons. For example, it would not be efficient to do an exit interview where there are only a few customers per day; neither is it possible to use mystery shoppers at a clinic. Each study should therefore include the method most suitable to them.

Dr
Indicators for the patient provider encounters include:  The proportion of non-facility outlet staff that dispense an appropriate first line ACT to patients presenting with fever  The proportion of non-facility outlet staff that dispense the correct dose of first line ACT to patients presenting with fever  The proportion of non-facility outlets that provide appropriate information to patients/ caregivers on how to give/ take the first line ACT  The proportion of encounters where non-facility outlet staff asked one or more clinical questions to determine severity of malaria  The proportion of non facility outlet staff who told patients/ caregivers about any signs of progressive illness and recommended a referral visit to a doctor or clinic if the signs appear  The median price charged for an ACT dose by age group

B4: Qualitative Data Indicators
It is also recommended that studies include qualitative data collection activities such as focus group discussions and in-depth interviews with household members, providers and others involved in the distribution chain. These will complement, validate and help interpret the quantitative indicators, providing a richer understanding of the reasons behind achievements and constraints in implementation. Qualitative data may also be used to inform the design of quantitative instruments.

B5: Diagnostic Indicators
Diagnosis can be either clinical (presumptive) or parasitologically confirmed (microscopy or rapid diagnostic tests). It is likely that a mix of diagnostic approaches will be used across the studies. Where parasitologically confirmed diagnosis is used in an intervention, the following indicators should be evaluated:  The proportion of non-facility outlets staff able to confirm a malaria diagnosis according to national guidelines using the approved diagnostic  The proportion of non-facility outlet patients who undergo a malaria diagnostic test  The sensitivity and specificity of a diagnostic test. Diagnostics should be of high sensitivity since false negatives may result in failure to treat, and of high specificity to avoid over diagnosis and subsequent over treatment.  The proportion of non facility outlets reporting stock outs of the approved diagnostic or components required for its proper functioning, within the past 2 weeks

C1: Morbidity and Mortality Indicators
x Final health outcome measures such as number of severe malaria cases and number of malaria fatalities would clearly be desirable. However, they are difficult to obtain for 2 reasons:  They are relatively rare events and therefore require very large sample sizes;  Many severe cases and deaths are not seen at health facilities and therefore are difficult to identify; over 60% of deaths in Kenya and in the sub-region occur at home.
It is therefore not expected that most studies will include these indicators. Exceptions may be large scale studies which could consider facility reports of severe cases, or studies taking place within demographic surveillance areas where community-based malaria mortality rates can be assessed based on verbal autopsy.

C2: Drug Resistance Indicators
Pilots are unlikely to be able to measure the impact of their intervention on drug resistance directly through treatment failure rates or genetic resistance markers. Measuring such outcomes would require large sample sizes, long time frames, and considerably increase evaluation costs in terms of collection of blood samples and laboratory analysis.

C3: Drug Quality Indicators
All pilots will use ACT supplies from quality certified sources and will therefore not be at risk of poor quality from sub-standard or fraudulent manufacturing. However, it is possible that the storage and handling of ACT in the supply chain could negatively affect quality, and these intermediate outcomes are therefore included under the provider survey. Again, additional laboratory tests would be required to assess the impact of storage and handling on actual quality.

Summary of Compulsory Indicators
No. Indicator Source 1 The proportion of children under 5 years with fever in the past 2 weeks who started treatment with a first line ACT within 24 and 48 hours of fever onset, overall, by socio-economic group (SEG) and treatment source Household Survey The median price charged by non-facility outlets for a first line ACT by age band Provider Survey

6
The proportion of non-facility outlets with no expired first line ACT available in stock Provider Survey

7
The proportion of non-facility outlets reporting stock outs of the first line ACT within the past 2 weeks Provider Survey

8
The proportion of non-facility outlets storing first line ACT appropriately Provider Survey

9
The proportion of non-facility outlets that have copies of the materials/ job aids required by the intervention (e.g. leaflets, posters, guidelines) Provider Survey

10
Implementation cost per intervention area Costing Study 11 The cost-effectiveness in terms of the incremental cost per additional child receiving prompt ACT treatment