Browse the corpus

SUMMARY OF FINDINGS FOR THE MAIN COMPARISON [Explanation] Effects of retail sector ACT subsidy programmes on ACT use, availability, price and market share Population: Patients seeking treatment for suspected uncomplicated malaria Settings: Rural districts in East Africa (Kenya, Uganda, Tanzania) Intervention: Retail sector ACT price subsidies plus supportive interventions (retail outlet provider training, community awareness and mass media campaigns) Comparison: Standard practice (no subsidies) Outcomes Illustrative comparative risksa (95% CI) Absolute difference (95% CI) Number of participants (studies) GRADE certainty of the evidence Assumed risk Corresponding risk No ACT subsidy ACT subsidy ACT use (percentage of children under 5 years receiving ACT on the same day or following day of fever onset) Follow-up: 1 year 5.3%1 30.3% (19.4% to 41.2%)b,2 25% (14.1% to 35.9%)3,4,5 2,6626 (1 study) ⊕⊕⊕⊕ High ACT availability (percentage of outlets stocking ACTs for children under 5 years) Follow-up: 1 year <0.5% 32.4% (22.5% to 41.8%)7 31.9% (26.3% to 37.5%) 1 study (2 articles) ⊕⊕⊕⊕ High ACT availability (percentage of outlets stocking at least one ACT for patients of any age) Follow-up: 1 year 0.5% 72.7% (65.5% to 79.8% 72.2% (65.0% to 79.3%)8 1 study ⊕оסо Very low 9 ACT price (change in ACT prices for children under 5 years)c Follow-up: 1 year Median cost per ACT course: US$ 1.0810 Median cost per ACT course: US$ 0.2411-14 US$ 0.84 (IQR not estimable) 1 study ⊕⊕⊕⊕ High

ACT availability (percentage of outlets stocking at least one ACT for patients of any age) Follow-up: 1 year 0.5% 72.7% (65.5% to 79.8% 72.2% (65.0% to 79.3%)8 1 study ⊕оסо Very low 9 ACT price (change in ACT prices for children under 5 years)c Follow-up: 1 year Median cost per ACT course: US$ 1.0810 Median cost per ACT course: US$ 0.2411-14 US$ 0.84 (IQR not estimable) 1 study ⊕⊕⊕⊕ High ACT market share (volume of ACTs purchased as a proportion of all antimalarials purchased; all age groups) Follow-up period: 1 year Range: 0% to 1.0%15 Range: 25.4% to 65.0%16,17 Range 23.6% to 63.0% 1 study (2 articles) ⊕⊕⊕⊕ High Use of older antimalarials (amodiaquine, sulphadoxine-pyrimethamine; children under 5 years) Follow-up period: 1 year 34.4%18 24.0% (17.5% to 30.5%)19 -10.4% (-3.9% to -16.9%) 1 study ⊕⊕⊕⊕ High Adverse effects (such as the number of people receiving ACTs who do not have malaria) Not measured Not measured Not estimable 3 studies Not estimable

Use of older antimalarials (amodiaquine, sulphadoxine-pyrimethamine; children under 5 years) Follow-up period: 1 year 34.4%18 24.0% (17.5% to 30.5%)19 -10.4% (-3.9% to -16.9%) 1 study ⊕⊕⊕⊕ High Adverse effects (such as the number of people receiving ACTs who do not have malaria) Not measured Not measured Not estimable 3 studies Not estimable ACT = artemisinin-based combination therapy; CI = confidence interval; IQR = interquartile range aThe basis for the assumed risk (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). bThis suggests that among febrile children with an expected ACT usage rate of 5% without subsidy, subsidy programmes would increase usage by between 19% and 41% cCosts include only prices paid by patients to purchase ACTs. Costing based on US$-to-Kenyan shillings exchange rate for 1 November 2008.

ion (and its 95% CI). bThis suggests that among febrile children with an expected ACT usage rate of 5% without subsidy, subsidy programmes would increase usage by between 19% and 41% cCosts include only prices paid by patients to purchase ACTs. Costing based on US$-to-Kenyan shillings exchange rate for 1 November 2008. About the certainty of the evidence (GRADE)* High: This research provides a very good indication of the likely effect. The likelihood that the effect will be substantially different† is low. Moderate: This research provides a good indication of the likely effect. The likelihood that the effect will be substantially different† is moderate. Low: This research provides some indication of the likely effect. However, the likelihood that it will be substantially different† is high. Very low: This research does not provide a reliable indication of the likely effect. The likelihood that the effect will be substantially different† is very high *This is sometimes referred to as ‘quality of evidence’ or ‘confidence in the estimate’ †Substantially different = a large enough difference that it might affect a decision 1Based on Kangwana 2011 baseline values. 2Based on Kangwana 2011. The second study, Talisuna 2012, reported an absolute change of 16%, but with no baseline data or confidence intervals. Using baselines from Kangwana 2011 (5.3%) gives 20.3% ACT use in Talisuna 2012, which is consistent with the findings of Kangwana 2011.

About the certainty of the evidence (GRADE)* High: This research provides a very good indication of the likely effect. The likelihood that the effect will be substantially different† is low. Moderate: This research provides a good indication of the likely effect. The likelihood that the effect will be substantially different† is moderate. Low: This research provides some indication of the likely effect. However, the likelihood that it will be substantially different† is high. Very low: This research does not provide a reliable indication of the likely effect. The likelihood that the effect will be substantially different† is very high *This is sometimes referred to as ‘quality of evidence’ or ‘confidence in the estimate’ †Substantially different = a large enough difference that it might affect a decision 1Based on Kangwana 2011 baseline values. 2Based on Kangwana 2011. The second study, Talisuna 2012, reported an absolute change of 16%, but with no baseline data or confidence intervals. Using baselines from Kangwana 2011 (5.3%) gives 20.3% ACT use in Talisuna 2012, which is consistent with the findings of Kangwana 2011. 3Kangwana 2011: there was no correlation between socioeconomic status and use of artemether-lumefantrine (P = 0.875) or Tibamal (subsidised artemether-lumefantrine; P = 0.745).

2Based on Kangwana 2011. The second study, Talisuna 2012, reported an absolute change of 16%, but with no baseline data or confidence intervals. Using baselines from Kangwana 2011 (5.3%) gives 20.3% ACT use in Talisuna 2012, which is consistent with the findings of Kangwana 2011. 3Kangwana 2011: there was no correlation between socioeconomic status and use of artemether-lumefantrine (P = 0.875) or Tibamal (subsidised artemether-lumefantrine; P = 0.745). 4Talisuna 2012: Children under 5 years of age: odds ratio 10.0 (4.96 to 18.86); All age groups: patients in the intervention districts had a six-fold increase in ACT use relative to the control district (95% CI 4.22 to 8.44). Use of ACT was higher in the highest socioeconomic status stratum compared to the lowest stratum (odds ratio 2.4, 95% CI 1.72 to 3.35, p<0.001); the certainty of evidence was downgraded from low to very low due to likely selection bias (non-randomised design) and confounding by study site (only one control site included, results likely to be influenced by site-specific factors). 5Cohen 2015: Compared to an access rate of 19 percent in the control group, subsidies of 80 percent or more increased the likelihood that a malaria-like illness is treated with an ACT by 16 to 23 percentage points (i.e. 85 to 118 percent increase). However, subsidies were associated with overtreatment of malaria: only 56 percent of patients taking ACTs from the drug shop tested for malaria under the 92 percent subsidy.

more increased the likelihood that a malaria-like illness is treated with an ACT by 16 to 23 percentage points (i.e. 85 to 118 percent increase). However, subsidies were associated with overtreatment of malaria: only 56 percent of patients taking ACTs from the drug shop tested for malaria under the 92 percent subsidy. 6Total number of children surveyed at follow up at 12 months in Kangwana 2011. Data were collected on 2,749 children at baseline. 7Based on Kangwana 2011. The second publication Kangwana 2013 reported an absolute increase of 31.7% (22.0% to 41.3%). 8Based on Sabot 2009. Drug shops in population centres were more likely to stock ACTs than those in more remote areas (P < 0.001). 9Downgraded from low to very low certainty evidence due to high likelihood of selection bias (non-randomised design) and confounding by study site (only one control site included; results likely to be influenced by site-specific (contextual) factors). 10Based on Kangwana 2011 baseline values. ACT treatment course: six tablets (children aged 3-35 months) and 12 tablets (children aged 36-59 months). 11Based on Kangwana 2011 follow-up (intervention site) data. 95.3% (SD 5.9%) of caregivers in the intervention arm at follow-up who bought Tibamal said they purchased it at the recommended retail price of US$ 0.25. Of the eight not paying this price, three paid less than US$ 0.25 and five paid between US$ 0.31 and 1.23.

11Based on Kangwana 2011 follow-up (intervention site) data. 95.3% (SD 5.9%) of caregivers in the intervention arm at follow-up who bought Tibamal said they purchased it at the recommended retail price of US$ 0.25. Of the eight not paying this price, three paid less than US$ 0.25 and five paid between US$ 0.31 and 1.23. 12Kangwana 2013: In the mystery-shopper survey, at baseline there were only two doses of artemether-lumefantrine sold, at a cost of US$ 2.46 and 2.22. At follow-up, the 12 tab Tibamal was sold at a median price of US$ 0.25 (IQR 20–20), which was the recommended retail price. Of those not paying the recommended price, two paid US$ 0.37, another two paid US$ 0.49 because of buying two packs of the six tab to meet the required dose, and one paid US$ 0.74. 13Talisuna 2012: “Maximum recommended retail price was within 10% of the recommended ACT price”. The recommended retail price for an adult course of treatment – US$ 0.47 – was not adhered to (the median price at the endline survey was US$ 1.96). 14Sabot 2009: Children: the mean price paid for ACTs (US$ 0.19) was less than for both sulphadoxine-pyrimethamine (US$ 0.51, P = 0.001) and amodiaquine (US$ 0.86, P < 0.001); the price paid for ACTs did not vary by socioeconomic status or geographical location across all age groups; All age groups: the mean price for ACTs (US$ 0.58) did not differ from the price of sulphadoxine-pyrimethamine (US$ 0.67), but was higher than for amodiaquine (US$ 0.48, P < 0.001). 15Based on baseline data from Kangwana 2011 and Kangwana 2013.

14Sabot 2009: Children: the mean price paid for ACTs (US$ 0.19) was less than for both sulphadoxine-pyrimethamine (US$ 0.51, P = 0.001) and amodiaquine (US$ 0.86, P < 0.001); the price paid for ACTs did not vary by socioeconomic status or geographical location across all age groups; All age groups: the mean price for ACTs (US$ 0.58) did not differ from the price of sulphadoxine-pyrimethamine (US$ 0.67), but was higher than for amodiaquine (US$ 0.48, P < 0.001). 15Based on baseline data from Kangwana 2011 and Kangwana 2013. 16Sabot 2009 reported a market share of 8.9% (-0.5% to 18.2%) for children under 5 years of age, and 35.3% (29.8% to 40.7%) for patients ≥ 16 years. Children under 5 years of age: purchases of sulphadoxine-pyrimethamine in the intervention districts decreased from 7.0% to 4.0% and remained the same at 9.0% in the control district. Purchases of amodiaquine in the intervention districts declined from 91.0% to 36.0%, and from 91.0% to 36.0% in the control district. 17Talisuna 2012: All age groups: market share for ACTs in the intervention group was 43% at baseline and 69% at follow-up (control data not reported). 18Based on Kangwana 2011 baseline data. 19Talisuna 2012: At follow-up, use of quinine was 44% in the control group and 37% in the intervention group (odds ratio 0.76, 95% confidence interval 0.54 to 1.08; no baseline data were reported).

17Talisuna 2012: All age groups: market share for ACTs in the intervention group was 43% at baseline and 69% at follow-up (control data not reported). 18Based on Kangwana 2011 baseline data. 19Talisuna 2012: At follow-up, use of quinine was 44% in the control group and 37% in the intervention group (odds ratio 0.76, 95% confidence interval 0.54 to 1.08; no baseline data were reported). Background Malaria is a major cause of ill health and death in Africa (WHO 2012). Uncomplicated Plasmodium falciparum malaria is the commonest form of the disease and accounts for most of the malaria cases and deaths (WHO 2012). The World Health Organization (WHO) recommends artemisinin-based combination therapy (ACT) for the treatment of uncomplicated malaria (WHO 2012). These drugs are highly effective and have the potential to reduce the development of antimalarial resistance. Unfortunately, despite the WHO's recommendation and substantial donor funding, only one in five antimalarial drugs used in malaria treatment in malaria-endemic countries are ACTs (WHO 2009). Reasons for the low use include high prices of ACTs in the retail sector (drug shops and pharmacies) where most people seek treatment for fever or suspected malaria (Patouillard 2010; Talisuna 2009). The price of ACTs is typically more than 10 times the price of older, less effective antimalarial drugs such as amodiaquine (AQ) and sulphadoxine-pyrimethamine (SP) in the retail sector (Morris 2014).

hops and pharmacies) where most people seek treatment for fever or suspected malaria (Patouillard 2010; Talisuna 2009). The price of ACTs is typically more than 10 times the price of older, less effective antimalarial drugs such as amodiaquine (AQ) and sulphadoxine-pyrimethamine (SP) in the retail sector (Morris 2014). Description of the intervention From 2010 to 2012, the Global Fund established and operated a new global subsidy programme, termed the Affordable Medicines Facility-malaria (AMFm), funded by three donors (UNITAID, the UK government and the Bill & Melinda Gates Foundation). The programme operated in seven sub-Saharan African countries (Ghana, Kenya, Madagascar, Niger, Nigeria, mainland Tanzania and Zanzibar and Uganda). In November 2012, the Global Fund board decided to change the way in which the ACT subsidy scheme operated. Instead of being a ‘stand alone’ initiative at the Global Fund, the subsidy programme was incorporated into the Global Fund's core grant management and financial processes. The AMFm was renamed as the Private Sector Co-payment Mechanism. As of mid-2014, Ghana, Madagascar and Tanzania had integrated the Private Sector Co-payment Mechanism into existing Global Fund grants.

obal Fund, the subsidy programme was incorporated into the Global Fund's core grant management and financial processes. The AMFm was renamed as the Private Sector Co-payment Mechanism. As of mid-2014, Ghana, Madagascar and Tanzania had integrated the Private Sector Co-payment Mechanism into existing Global Fund grants. The aim of the AMFm subsidy is to reduce ACT retail prices to a level similar to older, less effective antimalarial drugs in order to increase demand and access for ACTs and displace artemisinin monotherapy and other sub-standard malaria treatments from the market, particularly among populations most vulnerable to malaria (children under five years of age and poor rural populations). Under the AMFm, ‘first-line buyers’ in the retail sector (those who buy ACTs directly from the manufacturer) pay about US$ 0.05 for a course of ACT rather than US$ 5.00 (the price paid before the AMFm; RBM 2007). The public sector can also purchase donor-subsidised ACTs, which may in turn broaden public sector ACT access. The subsidy programme is combined with supporting interventions (such as retail outlet provider training, community awareness and mass media campaigns) to facilitate effective delivery and appropriate use of ACTs.

public sector can also purchase donor-subsidised ACTs, which may in turn broaden public sector ACT access. The subsidy programme is combined with supporting interventions (such as retail outlet provider training, community awareness and mass media campaigns) to facilitate effective delivery and appropriate use of ACTs. How the intervention might work The AMFm subsidy programme is designed to increase access to ACTs in the retail and public sector by subsidising the prices of ACTs at the manufacturer level (Arrow 2004; Laxminarayan 2009). The programme aims to lower consumer prices of ACTs, compared to older and less effective antimalarial drugs, available through the retail sector via two mechanisms: (1) negotiating with manufacturers of ACTs to reduce ACT prices; and (2) co-paying a proportion (about 90%) of the reduced ACT price directly to participating manufacturers, hence further lowering prices to eligible wholesalers of ACTs (Global Fund; Laxminarayan 2009). The wholesalers thus pay a lower price for ACTs and prices fall all along the supply chain, increasing affordability for the final consumer, while at the same time undercutting the price of resistance-inducing artemisinin monotherapy and competing with the prices for chloroquine and SP.

Global Fund; Laxminarayan 2009). The wholesalers thus pay a lower price for ACTs and prices fall all along the supply chain, increasing affordability for the final consumer, while at the same time undercutting the price of resistance-inducing artemisinin monotherapy and competing with the prices for chloroquine and SP. The pre-specified benchmarks of success of the AMFm pilot by the AMFm included: an increase in ACT use of 10 to 15 percentage points; increase in ACT availability of 20 percentage points; increase in ACT market share of 10 to 15 percentage points; and quality-assured ACT price less than 300% of the dominant non-quality-assured ACTs (chloroquine or SP; Table 1). The AMFm process is illustrated in Figure 1. In Figure 2 we provide a logic framework for this review, showing the theory of impact of the subsidies on malaria burden, potential influences on these steps and the outcomes that can be measured to evaluate the subsidy programmes. Figure 1 Illustrative example of the AMFm process Figure 2 Logic framework for evaluating AMFm programme Table 1 Guidelines for success benchmarks at 1 and 2 years after effective start date of the AMFm Phase 1 at the country level Outcome Year 1 Year 2 Availability (The proportion of all facilities, private and public (including informal outlets), stocking QAACTs, among outlets with any antimalarials in stock at the time of the survey) Increase of 20 percentage points from baseline Increase of 40 percentage points from baseline

Table 1 Guidelines for success benchmarks at 1 and 2 years after effective start date of the AMFm Phase 1 at the country level Outcome Year 1 Year 2 Availability (The proportion of all facilities, private and public (including informal outlets), stocking QAACTs, among outlets with any antimalarials in stock at the time of the survey) Increase of 20 percentage points from baseline Increase of 40 percentage points from baseline Market share (Total volume of QAACTs sold or distributed as a proportion of the total volume of all antimalarials sold or distributed in the last 7 days via outlets that will be included in the Independent Evaluation's outlet surveys) Increase in ACT market share of 10 to 15 percentage points from baseline and Decrease in market share of AMT from baseline Increase in ACT market share of 15 to 20 percentage points from baseline and Decrease in market share of AMT from baseline Usea (Proportion of children under age 5 with fever who received a QAACT on the day that the fever started or on the following day) Increase of 5 to 10 percentage points from baseline Increase of 10 to 15 percentage points from baseline

Market share (Total volume of QAACTs sold or distributed as a proportion of the total volume of all antimalarials sold or distributed in the last 7 days via outlets that will be included in the Independent Evaluation's outlet surveys) Increase in ACT market share of 10 to 15 percentage points from baseline and Decrease in market share of AMT from baseline Increase in ACT market share of 15 to 20 percentage points from baseline and Decrease in market share of AMT from baseline Usea (Proportion of children under age 5 with fever who received a QAACT on the day that the fever started or on the following day) Increase of 5 to 10 percentage points from baseline Increase of 10 to 15 percentage points from baseline Price (Adult equivalent treatment dose) QAACT price < 300% of the price of the dominant non-QAACT (in most countries this is CQ or SP)b and Price of AMFm co-paid QAACT < price ofAMT (this is useful but not sufficient todetermine success) QAACT price < 150% of the price of the dominant non-QAACT (in most countries this is CQ or SP) and Price of AMFm co-paid QAACT < price of AMT (this is useful but not sufficient to determine success) ACT = artemisinin-based combination therapy; AMT = artemisinin monotherapy AMFm = Affordable Medicines Facility-malaria CQ = chloroquine QAACT = quality-assured artemisinin-based combination therapies SP = sulphadoxine-pyrimethamine

Price (Adult equivalent treatment dose) QAACT price < 300% of the price of the dominant non-QAACT (in most countries this is CQ or SP)b and Price of AMFm co-paid QAACT < price ofAMT (this is useful but not sufficient todetermine success) QAACT price < 150% of the price of the dominant non-QAACT (in most countries this is CQ or SP) and Price of AMFm co-paid QAACT < price of AMT (this is useful but not sufficient to determine success) ACT = artemisinin-based combination therapy; AMT = artemisinin monotherapy AMFm = Affordable Medicines Facility-malaria CQ = chloroquine QAACT = quality-assured artemisinin-based combination therapies SP = sulphadoxine-pyrimethamine aThe denominator for ACT use is ‘fever episodes in children under age 5’ (not ‘parasitologically confirmed malaria cases’). The Independent Evaluation relies on national surveys (e.g. demographic and health surveys, multiple indicator cluster surveys, malaria indicator surveys and ACTwatch surveys), which use the denominator ‘fever episodes in children under age 5’ due to a lack of proper malaria diagnosis in many countries bPrice change was the indicator with the weakest empirical basis for setting a 1-year expectation Source: Yamey 2012

aThe denominator for ACT use is ‘fever episodes in children under age 5’ (not ‘parasitologically confirmed malaria cases’). The Independent Evaluation relies on national surveys (e.g. demographic and health surveys, multiple indicator cluster surveys, malaria indicator surveys and ACTwatch surveys), which use the denominator ‘fever episodes in children under age 5’ due to a lack of proper malaria diagnosis in many countries bPrice change was the indicator with the weakest empirical basis for setting a 1-year expectation Source: Yamey 2012 Why it is important to do this review ACT subsidy programmes are expensive. The initial costs of the AMFm pilot that ran from 2010 to 2012 were estimated at US$ 343 million: US$ 216 million for the subsidy and US$ 127 million for the supportive interventions (Global Fund). It is therefore important to ensure that such programmes lead to intended outcomes: lower ACT prices; increased ACT availability and usage; crowding-out of older, less effective malaria drugs; and an end to the marketing of artemisinin monotherapy (such marketing might increase the development of Plasmodium falciparum resistance, rendering ACTs ineffective; Perkins 2008). It is also important to make sure these subsidies do not have unintended consequences (such as inappropriate use of ACTs among patients with non-malarial fevers leading to drug wastage and delay in care-seeking for appropriate treatment). Objectives To assess the effect of programmes that include ACT price subsidies for private retailers on ACT use, availability, price and market share.

Why it is important to do this review ACT subsidy programmes are expensive. The initial costs of the AMFm pilot that ran from 2010 to 2012 were estimated at US$ 343 million: US$ 216 million for the subsidy and US$ 127 million for the supportive interventions (Global Fund). It is therefore important to ensure that such programmes lead to intended outcomes: lower ACT prices; increased ACT availability and usage; crowding-out of older, less effective malaria drugs; and an end to the marketing of artemisinin monotherapy (such marketing might increase the development of Plasmodium falciparum resistance, rendering ACTs ineffective; Perkins 2008). It is also important to make sure these subsidies do not have unintended consequences (such as inappropriate use of ACTs among patients with non-malarial fevers leading to drug wastage and delay in care-seeking for appropriate treatment). Objectives To assess the effect of programmes that include ACT price subsidies for private retailers on ACT use, availability, price and market share. Methods Criteria for considering studies for this review Types of studies Studies that assessed the effect of retail sector ACT price subsidies using the following designs were eligible for inclusion (Appendix 1). Randomised trials. Non-randomised trials. Controlled before-after studies. Interrupted-time-series studies (with a clearly defined point in time when the subsidy occurred, and at least three data points before and three after the subsidy intervention).

Methods Criteria for considering studies for this review Types of studies Studies that assessed the effect of retail sector ACT price subsidies using the following designs were eligible for inclusion (Appendix 1). Randomised trials. Non-randomised trials. Controlled before-after studies. Interrupted-time-series studies (with a clearly defined point in time when the subsidy occurred, and at least three data points before and three after the subsidy intervention). Types of participants Studies involving the following groups of patients and units or channels of delivery of subsidised ACTs were eligible for inclusion. Consumers of retail sector subsidised ACTs (patients seeking treatment for fever or malaria; both children and adults). Private retailers of subsidised ACTs (pharmacies, franchised clinics, drug shops, general stores). Types of interventions Intervention Studies assessing the effect of retail sector ACT price subsidies were eligible for inclusion. Retail sector ACT subsidy programmes (both for-profit and not-for-profit retail sectors). Retail sector ACT subsidy programmes with supportive interventions (e.g. retail outlet provider training, community awareness and mass media campaigns). Comparisons Alternative ACT financing mechanisms aiming to achieve similar goals as retail sector ACT price subsidies (such as the generic Global Fund financing mechanism aiming to expand ACT availability in public health care facilities). Public sector interventions to increase ACT availability funded by the United States President's Malaria Initiative (PMI) (http://www.pmi.gov/).

Comparisons Alternative ACT financing mechanisms aiming to achieve similar goals as retail sector ACT price subsidies (such as the generic Global Fund financing mechanism aiming to expand ACT availability in public health care facilities). Public sector interventions to increase ACT availability funded by the United States President's Malaria Initiative (PMI) (http://www.pmi.gov/). Usual ACT delivery mechanisms (non-subsidised ACT interventions). Types of outcome measures Primary outcomes ACT use (defined as the percentage of patients with fever/confirmed malaria who received an ACT on the day that the fever started or on the following day; Global Fund). Secondary outcomes ACT availability (proportion of all facilities stocking ACTs among outlets with any antimalarials in stock). ACT price (cost to patients of a full child or adult course of ACTs). ACT market share (total volume of ACTs sold or distributed as a proportion of the total volume of all antimalarials sold or distributed via outlets). Use of older antimalarial drugs (AQ, chloroquine, artemisinin monotherapy, SP). Adverse effects (such as the number of people receiving ACTs who do not have malaria). All studies with eligible designs, participants and interventions were considered for inclusion irrespective of whether only the above outcome measures were reported. Search methods for identification of studies Electronic searches We aimed to identify eligible published and unpublished studies. We searched the following databases and grey literature sources in February, 2015.

All studies with eligible designs, participants and interventions were considered for inclusion irrespective of whether only the above outcome measures were reported. Search methods for identification of studies Electronic searches We aimed to identify eligible published and unpublished studies. We searched the following databases and grey literature sources in February, 2015. The Cochrane Central Register of Controlled Trials (CENTRAL; 2015, Issue 1, The Cochrane Library) including The Cochrane Effective Practice and Organisation of Care (EPOC) Group Specialised Register. MEDLINE In-Process & Other Non-Indexed Citations, MEDLINE Daily, MEDLINE and OLDMEDLINE, 1946 to Present (OvidSP). Embase 1980 to 2015 Week 07 (OvidSP). CINAHL 1980 to present (EbscoHost). Regional Indexes (Global Health Library, WHO). LILACS (Global Health Library, WHO). Science Citation Index and Social Sciences Citation Index (ISI Web of Science). EconLit 1969 to current (ProQuest). Global Health 1973 to 2015 Week 07 (OvidSP). Health Management (ProQuest; searched 27/11/2013 because we no longer have access to this database). The searches were done without any language, date or publication status restrictions. See Appendix 2 for all search strategies. Searching other resources Grey literature The Grey Literature Report from The New York Academy of Medicine Library (http://www.greylit.org/; searched 04/11/2013).

Health Management (ProQuest; searched 27/11/2013 because we no longer have access to this database). The searches were done without any language, date or publication status restrictions. See Appendix 2 for all search strategies. Searching other resources Grey literature The Grey Literature Report from The New York Academy of Medicine Library (http://www.greylit.org/; searched 04/11/2013). Websites of the following institutions: the Global Fund, the Roll Back Malaria partnership, Malaria Consortium, Medicines for Malaria Venture, UNITAID, the Clinton Health Access Initiative, PMI, World Bank (Booster Program for Malaria Control in Africa), WHO Global Malaria Program, the United Nations Children's Fund (UNICEF), Drugs for Neglected Diseases Initiative, Centre de Recherche pour le Développement Humain, ACTwatch, Bill & Melinda Gates Foundation, UK Department for International Development, Management Sciences for Health Sustainable Drug Sellers Initiative, Oxfam International and Center for Health Market Innovations (searched 04/11/2013). Trial registries WHO International Clinical Trials Registry Platform (ICTRP; http://www.who.int/ictrp/en/; searched 04/11/2013). ClinicalTrials.gov, US National Institutes of Health (http://clinicaltrials.gov/; searched 04/11/2013). Additional resources We also: hand searched relevant conference proceedings (e.g. Multilateral Initiative on Malaria; searched 04/11/2013); hand searched reference lists of relevant articles (technical reports, reviews; searched 04/11/2013); contacted authors of relevant articles regarding any further published or unpublished work; and

Additional resources We also: hand searched relevant conference proceedings (e.g. Multilateral Initiative on Malaria; searched 04/11/2013); hand searched reference lists of relevant articles (technical reports, reviews; searched 04/11/2013); contacted authors of relevant articles regarding any further published or unpublished work; and conducted cited reference searches for all included studies in ISI Web of Knowledge (searched 19/02/2015). Data collection and analysis Selection of studies Two review authors (NO, PG) independently screened titles, abstracts and full texts of identified articles and applied the pre-specified study eligibility criteria to select studies. GY reviewed all articles selected for inclusion. We resolved any disagreements by discussion. We documented the number of articles screened, assessed for eligibility and selected for inclusion in a PRISMA flow diagram (Figure 3; Moher 2009). Studies initially considered eligible but eventually excluded together with the reasons for exclusions are presented in the Characteristics of excluded studies. Figure 3 Results of the literature search and studies selected

Data collection and analysis Selection of studies Two review authors (NO, PG) independently screened titles, abstracts and full texts of identified articles and applied the pre-specified study eligibility criteria to select studies. GY reviewed all articles selected for inclusion. We resolved any disagreements by discussion. We documented the number of articles screened, assessed for eligibility and selected for inclusion in a PRISMA flow diagram (Figure 3; Moher 2009). Studies initially considered eligible but eventually excluded together with the reasons for exclusions are presented in the Characteristics of excluded studies. Figure 3 Results of the literature search and studies selected Data extraction and management Two review authors (NO, PG) independently extracted outcome data at baseline and endline. We also extracted data on study settings (coverage), participants (patient age groups), retail outlets, interventions (level of price subsidy and duration of subsidy programme), ACT supply (distribution) mechanisms and nature of supportive interventions (Characteristics of included studies; Table 2). Data were entered into a pilot-tested data extraction form. We resolved any disagreements by discussion. Table 2 Characteristics of included studies Study Design Country Age group Interventiona Co-intervention Outlet Coverage Data collection Price subsidy Duration Cohen 2015 Cluster-RCT Kenya All age groups Subsidised AL None Drug shops 3 rural districts Household surveys 80% to 92% 4 months

Data extraction and management Two review authors (NO, PG) independently extracted outcome data at baseline and endline. We also extracted data on study settings (coverage), participants (patient age groups), retail outlets, interventions (level of price subsidy and duration of subsidy programme), ACT supply (distribution) mechanisms and nature of supportive interventions (Characteristics of included studies; Table 2). Data were entered into a pilot-tested data extraction form. We resolved any disagreements by discussion. Table 2 Characteristics of included studies Study Design Country Age group Interventiona Co-intervention Outlet Coverage Data collection Price subsidy Duration Cohen 2015 Cluster-RCT Kenya All age groups Subsidised AL None Drug shops 3 rural districts Household surveys 80% to 92% 4 months Kangwana 2011 Cluster-RCT Kenya Children under age 5 Subsidised AL Training of retail outlet staff Job aids Community awareness activities Specialised and general drug storesb 3 rural districts (9 intervention and 9 control sublocations) Household surveys 95% 1 year Kangwana 2013 Cluster-RCT Kenya Children under age 5 Subsidised AL Training of retail outlet staff Job aids Community awareness activities Specialised and general drug storesb 3 rural districts (9 intervention and 9 control sublocations) Provider surveys Mystery shopper surveys 95% 1 year

Kangwana 2011 Cluster-RCT Kenya Children under age 5 Subsidised AL Training of retail outlet staff Job aids Community awareness activities Specialised and general drug storesb 3 rural districts (9 intervention and 9 control sublocations) Household surveys 95% 1 year Kangwana 2013 Cluster-RCT Kenya Children under age 5 Subsidised AL Training of retail outlet staff Job aids Community awareness activities Specialised and general drug storesb 3 rural districts (9 intervention and 9 control sublocations) Provider surveys Mystery shopper surveys 95% 1 year Sabot 2009 Non-randomised cluster trial Tanzania All age groups Subsidised AL Behaviour change communication (local radio advertisements, wall paintings, themed cultural shows) Small drug shops (duka la dawa baridi) 3 rural districts (2 intervention and 1 control) Outlet exit interviews Mystery shoppers Outlet audits Public facility audits 90% 1 year Talisuna 2012 c Non-randomised cluster trial Uganda All age groups Subsidised AL Training of drug shop attendants Branding (‘ACT with a leaf’) Communication activities Drug shops Private clinics Pharmacies 5 districts (4 intervention and 1 control) Outlet exit interviews Outlet audits 95% 20 months ACT = artemisinin-based combination therapy AL = artemether-lumefantrine NGO = non-governmental organisation RCT = randomised controlled trial

Talisuna 2012 c Non-randomised cluster trial Uganda All age groups Subsidised AL Training of drug shop attendants Branding (‘ACT with a leaf’) Communication activities Drug shops Private clinics Pharmacies 5 districts (4 intervention and 1 control) Outlet exit interviews Outlet audits 95% 20 months ACT = artemisinin-based combination therapy AL = artemether-lumefantrine NGO = non-governmental organisation RCT = randomised controlled trial aAll studies: no ACT subsidy interventions were implemented in the control sites; malaria diagnosis was predominantly presumptive based on the presence of fever; exemption was granted for ACTs to be provided over the counter in the intervention sites; AL was repackaged in weight specific packs and marked with recommended retail prices to inform consumers the maximum amount they should pay bSpecialised drug stores (registered or unregistered pharmacies) and general stores (which sold medicines alongside general household goods) cThere was better availability of ACT in the public sector in the control district because of improvements in the procurement and distribution system, and supply by one NGO

aAll studies: no ACT subsidy interventions were implemented in the control sites; malaria diagnosis was predominantly presumptive based on the presence of fever; exemption was granted for ACTs to be provided over the counter in the intervention sites; AL was repackaged in weight specific packs and marked with recommended retail prices to inform consumers the maximum amount they should pay bSpecialised drug stores (registered or unregistered pharmacies) and general stores (which sold medicines alongside general household goods) cThere was better availability of ACT in the public sector in the control district because of improvements in the procurement and distribution system, and supply by one NGO Assessment of risk of bias in included studies Two review authors (NO, PG) independently assessed the risk of bias in the included studies using the Cochrane EPOC ‘Risk of bias’ tool (EPOC 2014). Quality domains assessed included: allocation sequence generation, allocation concealment, similarity of baseline characteristics and outcome measurements, blinding (personnel and outcome assessors), handling of incomplete outcome data, protection against contamination, completeness of follow-up and reporting of outcomes. We also assessed the following additional cluster-specific sources of bias: recruitment bias, loss of clusters, incorrect analysis and comparability with individually randomised trials (Higgins 2011). Findings were classified into three categories: low (low risk of bias for key quality domains, i.e. allocation sequence generation and concealment), high (high risk of bias for one or more of the key domains) and unclear (unclear risk of bias for one or more key domains). We resolved any disagreements by discussion.

). Findings were classified into three categories: low (low risk of bias for key quality domains, i.e. allocation sequence generation and concealment), high (high risk of bias for one or more of the key domains) and unclear (unclear risk of bias for one or more key domains). We resolved any disagreements by discussion. We did not exclude studies on the basis of their risk of bias ratings; rather, we used these findings to help us better understand weaknesses in the identified evidence. We took into account the risk of bias ratings when synthesising and interpreting results. We report on the results of risk of bias assessment in the ‘Risk of bias’ tables and graphs (Figure 4; Figure 5). Figure 4 Review authors' judgements about each risk of bias item for each included study Figure 5 Review authors' judgements about each risk of bias item presented as percentages across all included studies Measures of treatment effect We have reported measures of subsidy effects as reported in the primary studies. We have presented absolute estimates of effects (percentage point differences, range and median effect sizes) with 95% confidence intervals (CIs) where estimable. For example, for the outcome of ACT use, we reported percentage changes in control and intervention sites from baselines and the absolute percentage point difference (cluster-adjusted) between the changes (i.e. ‘difference-in-difference’ estimates; Table 3). Table 3 Percentage of children with fever who received ACT on the same day or following day of fever onset

Measures of treatment effect We have reported measures of subsidy effects as reported in the primary studies. We have presented absolute estimates of effects (percentage point differences, range and median effect sizes) with 95% confidence intervals (CIs) where estimable. For example, for the outcome of ACT use, we reported percentage changes in control and intervention sites from baselines and the absolute percentage point difference (cluster-adjusted) between the changes (i.e. ‘difference-in-difference’ estimates; Table 3). Table 3 Percentage of children with fever who received ACT on the same day or following day of fever onset Study Design Age group Control Intervention Absolute difference (95% CI) Relative effect (95% CI) Baseline Follow-up Baseline Follow-up Kangwana 2011 Cluster-RCT Children under age 5 5.3% 19.9% 4.7% 44.9% 25.0% (14.1% to 35.9%)a NR Talisuna 2012 Non-randomised cluster trial Children under age 5 NR 2% NR 18% 16.0%b OR 10.0 (4.96 to 18.86)c ACT = artemisinin-based combination therapy CI = confidence interval NR = not reportedOR = odds ratio RCT = randomised controlled trial aThere was no correlation between socio-economic status and use of AL (p=0.875) or Tibamal, subsidised AL (p=0.745) bEstimated assuming similar baseline values in control and intervention groups cAll age groups: patients in the intervention districts had a six-fold increase in ACT use relative to the control district (95% CI 4.22 to 8.44). Use of ACT was higher in the highest socio-economic status stratum compared to the lowest stratum (OR 2.4, 95% CI 1.72 to 3.35, p<0.001); estimated from available data

bEstimated assuming similar baseline values in control and intervention groups cAll age groups: patients in the intervention districts had a six-fold increase in ACT use relative to the control district (95% CI 4.22 to 8.44). Use of ACT was higher in the highest socio-economic status stratum compared to the lowest stratum (OR 2.4, 95% CI 1.72 to 3.35, p<0.001); estimated from available data Unit of analysis issues We assessed whether appropriate analysis was conducted to adjust for clustering in estimating precision of effects in cluster trials (Higgins 2011). Dealing with missing data We contacted authors of primary studies where relevant data were missing or where we required further clarification on the reported data. Where data were not available from the authors, we reported the data as missing; we did not impute or extrapolate values. Assessment of heterogeneity We did not assess inconsistency between the results of individual studies using statistical methods; differences in studies precluded meta-analysis. We documented factors that could modify subsidy effects (such as programme coverage, level of price subsidy, ACT supply mechanism and nature of supportive interventions) in accordance with the established guidance for evaluating complex interventions (Shepperd 2009; Table 2).

erences in studies precluded meta-analysis. We documented factors that could modify subsidy effects (such as programme coverage, level of price subsidy, ACT supply mechanism and nature of supportive interventions) in accordance with the established guidance for evaluating complex interventions (Shepperd 2009; Table 2). Assessment of reporting biases Reporting biases arise when the dissemination of research findings is influenced by the nature and direction of results (Sterne 2011). Reporting biases comprise publication bias, time lag bias, multiple (duplicate) publication bias, location bias, citation bias, language bias and outcome reporting bias. We assessed potential selective reporting of outcomes as one component of risk of bias assessment (we focused on the completeness of reporting of pre-specified outcomes). We did not create funnel plots as planned because of insufficient data (only four studies were included in the review) (Sterne 2011). Data synthesis The included studies utilised varied study designs (randomised cluster and non-randomised cluster trials), enrolled diverse populations (children and adults) and used different effect measures. We therefore did not combine results using statistical methods. We have described individual study results in the ‘Effects of interventions’ section.

ilised varied study designs (randomised cluster and non-randomised cluster trials), enrolled diverse populations (children and adults) and used different effect measures. We therefore did not combine results using statistical methods. We have described individual study results in the ‘Effects of interventions’ section. ‘Summary of findings’ table and assessing the certainty of evidence We assessed the overall confidence in estimate of effect (certainty of evidence) for each outcome using GRADE (Guyatt 2008). This system classifies the certainty of evidence (defined as ‘the extent to which one can be confident that an estimate of effect or association is correct’) into four categories: very low, low, moderate or high. Data from randomised trials start at high quality while data from observational studies start at low quality. Quality of evidence from randomised trials can be downgraded in consideration of five factors: risk of bias or study limitations, directness, consistency of results, precision and publication bias. Similarly, quality of evidence from observational studies can be upgraded in consideration of three factors: magnitude of effect estimate, dose-response gradient and influence of residual plausible confounding.

ctors: risk of bias or study limitations, directness, consistency of results, precision and publication bias. Similarly, quality of evidence from observational studies can be upgraded in consideration of three factors: magnitude of effect estimate, dose-response gradient and influence of residual plausible confounding. Two review authors (NO, PG) independently assessed the certainty of evidence; we resolved disagreements by discussion. We did not assess the certainty of evidence for outcomes where there was insufficient data to permit reliable certainty rating. We did not exclude studies on the basis of GRADE ratings; we took into account the certainty of evidence when interpreting results. We used GRADEpro software (GRADEpro 2015) to create ‘Summary of findings’ (SoF) tables for two comparisons: (1) ‘retail sector ACT subsidies combined with supportive interventions versus no subsidies’ (Summary of findings for the main comparison); and (2) ‘ACT subsidy vouchers versus no subsidies’ (Summary of findings 2). We included all the pre-specified outcomes in SoF table 1 and ACT access (defined as ‘the share of illness episodes treated with ACTs’) and targeting (defined as ‘the share of ACT-takers who are malaria positive’) in SoF table 2. We prioritised findings from randomised trials (data from non-randomised trials were incorporated as footnotes).

e-specified outcomes in SoF table 1 and ACT access (defined as ‘the share of illness episodes treated with ACTs’) and targeting (defined as ‘the share of ACT-takers who are malaria positive’) in SoF table 2. We prioritised findings from randomised trials (data from non-randomised trials were incorporated as footnotes). Subgroup analysis and investigation of heterogeneity We intended to perform subgroup analysis to investigate potential variation in subsidy effects by study design, nature of supportive interventions, socioeconomic status and scale of coverage of subsidies (sub-national versus national programmes). However, we did not perform any of the planned analyses because there was insufficient data to permit reliable analysis. Sensitivity analysis We intended to conduct sensitivity analysis to assess the impact of study quality on results (for example, whether effect estimates are robust to changes in study risk of bias). We did not perform this analysis as no meta-analysis was done. Results Description of studies Results of the search We identified a total of 1705 articles from both the electronic and supplementary searches. We excluded 1684 articles following a review of the titles and abstracts. We retrieved the full texts of 21 articles for detailed eligibility assessment. We excluded 16 of the articles because of ineligible study designs. No ongoing studies were identified. Overall, four studies (five articles) fulfilled the review inclusion criteria (Cohen 2015; Kangwana 2011; Kangwana 2013; Talisuna 2012; Sabot 2009; Figure 3).

texts of 21 articles for detailed eligibility assessment. We excluded 16 of the articles because of ineligible study designs. No ongoing studies were identified. Overall, four studies (five articles) fulfilled the review inclusion criteria (Cohen 2015; Kangwana 2011; Kangwana 2013; Talisuna 2012; Sabot 2009; Figure 3). Included studies We included four studies (five articles; Characteristics of included studies; Table 2) Two studies (three articles) were cluster-randomised trials (Cohen 2015; Kangwana 2011; Kangwana 2013). The remaining studies were non-randomised cluster trials (Sabot 2009; Talisuna 2012). The studies were conducted in rural districts in Kenya (Cohen 2015; Kangwana 2011; Kangwana 2013), Uganda (Talisuna 2012) and Tanzania (Sabot 2009). One trial enrolled children under five years of age (Kangwana 2011; Kangwana 2013); both adults and children were studied in the other trials. One trial (Kangwana 2011; Kangwana 2013) reported adequate power (80%) for primary outcomes. Study power was not reported in the other three studies.

and Tanzania (Sabot 2009). One trial enrolled children under five years of age (Kangwana 2011; Kangwana 2013); both adults and children were studied in the other trials. One trial (Kangwana 2011; Kangwana 2013) reported adequate power (80%) for primary outcomes. Study power was not reported in the other three studies. ACT price subsidies were accompanied with supportive interventions (including retail outlet provider training and community awareness campaigns lasting less than a year) in all except one study (Cohen 2015). The subsidy level was 95% in two studies (Kangwana 2011; Kangwana 2013; Talisuna 2012), 80% to 92% in one study (Cohen 2015) and 90% in the remaining study (Sabot 2009). Post-intervention data collection periods were varied: 4 months (Cohen 2015), 12 months (Kangwana 2011; Kangwana 2013; Sabot 2009) and 20 months (Talisuna 2012). Retail outlets comprised specialised and general drug stores in all the studies. Private clinics were included in one study (Talisuna 2012). ACT supply and distribution chains were varied: a third party procured and delivered ACTs directly to trained outlets from which shopkeepers purchased the treatment at a wholesale price in two trials (Kangwana 2011; Kangwana 2013; Talisuna 2012). In one trial the project managers procured the ACTs directly from the manufacturing drug company (Novartis) and sold them to a wholesaler; drug shops purchased the ACTs from the wholesaler (Sabot 2009). Households were provided with vouchers to purchase subsidised ACTs from retail outlets in one trial (Cohen 2015).

2). In one trial the project managers procured the ACTs directly from the manufacturing drug company (Novartis) and sold them to a wholesaler; drug shops purchased the ACTs from the wholesaler (Sabot 2009). Households were provided with vouchers to purchase subsidised ACTs from retail outlets in one trial (Cohen 2015). Randomised cluster trials Kangwana 2011 evaluated the impact of retail sector ACT (artemether-lumefantrine; AL) subsidies in febrile children aged 3 to 59 months in Kenya. Nine areas were randomly allocated to the intervention arm (ACT subsidy plus supportive interventions, retail provider training and communication awareness activities) and nine to the control arm, with a buffer zone of two areas between selected areas. Subsidised AL was provided to retail outlets from November 2008. Cross-sectional household surveys were conducted before (July to August 2008) and after (July to August 2009) the delivery of the intervention. The primary outcome was the proportion of children reporting fever in the previous two weeks who started treatment with AL on the same day as fever onset or the following day. Secondary outcomes were adequacy of AL doses obtained and consumed and the price paid per pack. Data were collected on 2749 children in the target age group at baseline and 2662 at one year follow-up.

ting fever in the previous two weeks who started treatment with AL on the same day as fever onset or the following day. Secondary outcomes were adequacy of AL doses obtained and consumed and the price paid per pack. Data were collected on 2749 children in the target age group at baseline and 2662 at one year follow-up. Kangwana 2013 evaluated the impact of retail sector ACT subsidies using provider and mystery-shopper survey data collected as part of the randomised trial described above (Kangwana 2011). Data were collected at baseline (July to August 2008) and follow-up (July to August 2009) using provider and mystery shopper cross-sectional surveys. The mystery shopper survey assessed patient-provider interactions and aimed to provide data on actual rather than self-reported provider practice. Outcomes assessed included retail sector ACT availability, provider knowledge and provider dispensing practices. A total of 468 outlets were assessed at baseline and 639 at follow-up in the provider survey. 499 outlets were assessed at baseline and 653 at follow-up in the mystery shopper survey.

rted provider practice. Outcomes assessed included retail sector ACT availability, provider knowledge and provider dispensing practices. A total of 468 outlets were assessed at baseline and 639 at follow-up in the provider survey. 499 outlets were assessed at baseline and 653 at follow-up in the mystery shopper survey. Cohen 2015 studied the impact of ACT vouchers in three rural districts in Kenya. Four drug shops (in four market centres) were selected and all households in the catchment area (within a 4 km radius) of these shops were sampled. Households were randomly assigned to one of three groups: (1) ‘No subsidy’ group (received vouchers to purchase full price ACTs at the pre-AMFm retail price of Kenya shillings 500 (approximately US$ 6.25 in 2009)); (2) ‘ACT subsidy only’ group (received vouchers for ACT subsidies of between 80% and 92%); (3) ‘ACT plus rapid diagnostic test (RDT) subsidy’ group (received vouchers for both subsidised ACTs and RDTs. Two vouchers for ACTs (AL) and two vouchers for RDTs (where applicable) were distributed to each household following a baseline survey; 2789 (95%) out of the 2928 households sampled consented to the baseline survey. The trial was conducted between May and December 2009 (the endline survey was administered about four months after the vouchers had been distributed). We only extracted data on ACT access and targeting in the ‘ACT subsidy’ only and ‘No subsidy’ group. ACT access was defined as ‘the share of illness episodes treated with ACTs’. ACT targeting was defined as ‘the share of ACT-takers who are malaria positive’.

d about four months after the vouchers had been distributed). We only extracted data on ACT access and targeting in the ‘ACT subsidy’ only and ‘No subsidy’ group. ACT access was defined as ‘the share of illness episodes treated with ACTs’. ACT targeting was defined as ‘the share of ACT-takers who are malaria positive’. Non-randomised cluster trials Sabot 2009 assessed the impact of AMFm in three rural districts in Tanzania. Since two of the selected districts were adjacent, randomisation was limited so that one of the adjacent districts served as the control. The selected districts were randomly assigned to receive ACT subsidy, ACT subsidy plus suggested retail price or no ACT subsidy (control). The intervention was implemented in 2007. Data were collected at baseline and during intervention using interviews with drug shop customers, retail audits, mystery shoppers and audits of public and non-governmental facilities. Most consumers interviewed in all districts were from the two least poor socioeconomic status quintiles (59% and 68%, respectively). A range of behaviour change communication (local radio advertisements and wall paintings) highlighting the importance of using ACTs and their availability in private shops was conducted by Population Services International. Outcomes assessed were ACT uptake, availability and price. A total of 216 drug shops (duka la dawa baridi) were studied. The report focuses on data collected between August 2007 and August 2008.

ting the importance of using ACTs and their availability in private shops was conducted by Population Services International. Outcomes assessed were ACT uptake, availability and price. A total of 216 drug shops (duka la dawa baridi) were studied. The report focuses on data collected between August 2007 and August 2008. Talisuna 2012 evaluated the impact of retail sector ACT subsidies in Uganda. The pilot was implemented in 2008 and involved four intervention districts (purposefully selected to receive branded subsidised ACTs) and one control district. Supportive interventions included communication and training activities to improve awareness of the availability of subsidised ACTs and correct dispensing and use of ACTs. Outcomes assessed were ACT uptake, availability, price, purchase within 24 hours of symptom onset and market share. Reported data comprise 1162 interviews at baseline (September 2008) and 5181 interviews at endline (May 2010) from 783 outlets. Excluded studies We excluded 16 studies from the analysis because the study designs did not meet our inclusion criteria (uncontrolled pilots, pre-post surveys or qualitative assessments; Characteristics of excluded studies). Some of the studies excluded were large country evaluations without control sites where the observed effects could be true effects or could be secular and not due to the subsidy intervention (Tougher 2012; Table 4). Details of some of the excluded studies are described below. Table 4 National ACT subsidy programmesa

Excluded studies We excluded 16 studies from the analysis because the study designs did not meet our inclusion criteria (uncontrolled pilots, pre-post surveys or qualitative assessments; Characteristics of excluded studies). Some of the studies excluded were large country evaluations without control sites where the observed effects could be true effects or could be secular and not due to the subsidy intervention (Tougher 2012; Table 4). Details of some of the excluded studies are described below. Table 4 National ACT subsidy programmesa Country Lead organisation Launch year Age group Outlets Coverage Outcome: ACT availability Outcome: ACT price Cambodia PSI 2002 All age groups Pharmacies Drug shops 17 of 20 malaria-endemic provinces At 1 year: very low in private facilities (22% stocked adult ACTs, 6% stocked child ACTs) At 1 year: mean consumer price for adult ACTs (US$ 1.07) 70% higher than RRP (US$ 0.63) Cameroon Government 2007 All age groups Public and private health facilities National At 1 year: low availability of subsidised ACTs at public and private facilities, monotherapies widely available At 1 year: adherence to RRP strong in only one province (Yaoundé Centre)

Country Lead organisation Launch year Age group Outlets Coverage Outcome: ACT availability Outcome: ACT price Cambodia PSI 2002 All age groups Pharmacies Drug shops 17 of 20 malaria-endemic provinces At 1 year: very low in private facilities (22% stocked adult ACTs, 6% stocked child ACTs) At 1 year: mean consumer price for adult ACTs (US$ 1.07) 70% higher than RRP (US$ 0.63) Cameroon Government 2007 All age groups Public and private health facilities National At 1 year: low availability of subsidised ACTs at public and private facilities, monotherapies widely available At 1 year: adherence to RRP strong in only one province (Yaoundé Centre) DRC PSI 2006 Children < age 5 Pharmacies Limited to some districts At 2 years: ACTs available in 20.2% (public facilities), 25.8% (part 1 pharmacies), 20% (drug shops), 8.6% (other private outlets); 66.4% of facilities stocked non-ACTs, 47.8% stocked AMT At 2 years: median price of ACTs: US$ 2.75 (public health facilities), US$ 2.29–4.58 (private facilities), US$ 3.89 (all facilities selling ACTs); ACT price 60% higher than price of the most common antimalarial in outlets selling ACTs Madagascar PSI 2003 Children < age 5 Pharmacies Private providers Community agents National At 5 years: 85.6% (public facilities), 47.5% (part 1 pharmacies), 20% (drug shops), 0.1–16.5% (other private outlets); 34.4% of facilities stocked non-ACT, 0.5% stocked AMT At 5 years: median price of ACTs in facilities selling ACTs was US$ 4.04 (ACTs free in public facilities); ACTs 11.3 times more expensive than the most common antimalarial in outlets selling ACTs

rt 1 pharmacies), 20% (drug shops), 0.1–16.5% (other private outlets); 34.4% of facilities stocked non-ACT, 0.5% stocked AMT At 5 years: median price of ACTs in facilities selling ACTs was US$ 4.04 (ACTs free in public facilities); ACTs 11.3 times more expensive than the most common antimalarial in outlets selling ACTs Rwanda PSI 2007 Children < age 5 Pharmacies National At 18 months: high ACT availability in private pharmacies (80–90% stocked child ACTs compared with 10% at baseline); monotherapies effectively banned Data unavailable Senegal Government 2006 All age groups Pharmacies National At 1 year: proportion of all facilities (public and private) stocking ACTs was 44.8% (adult dose), 58.2% (child), 46.3% (infant); monotherapies widely available At 1 year: strong adherence to RRP in private outlets (observed mean retail price = US$ 1.34; RRP = US$ 1.31) ACTs = artemisinin-based combination therapies AMFm = Affordable Medicines Facility-malaria AMT = artemisinin monotherapy PSI = Population Services International RRP = recommended retail price aThese programmes were rolled out before the 2010–11 AMFm pilot. Two countries - Cambodia and Madagascar - have also been included in the AMFm pilot phase and the results from the pilot are due to be reported in 2012 Source: Yamey 2012

AMT = artemisinin monotherapy PSI = Population Services International RRP = recommended retail price aThese programmes were rolled out before the 2010–11 AMFm pilot. Two countries - Cambodia and Madagascar - have also been included in the AMFm pilot phase and the results from the pilot are due to be reported in 2012 Source: Yamey 2012 Tougher 2012 assessed the effect of AMFm in seven countries (Ghana, Kenya, Madagascar, Niger, Uganda, Nigeria and Tanzania (including Zanzibar)). Nationally representative baseline and endpoint surveys of public and private sector outlets that stock antimalarial treatments were conducted in each of the seven countries (eight national-level pilots). Clusters were selected using probability proportional to size sampling; independent samples were drawn at baseline and endpoint (a full census of outlets was done in Zanzibar because of the small population size). Outcomes assessed were ACT price, availability and market share. These outcomes were assessed against pre-specified success benchmarks after one year of AMFm implementation (Table 1). Data on the implementation process and contextual factors (e.g. supportive interventions, mechanisms of distribution of co-paid ACTs) were collected through key informant interviews and document reviews. Although this study provides important evidence on AMFm effectiveness at a national scale we did not include it because it used a before-after design with no comparator sites. The lack of control sites limits the degree to which observed effects can be attributed to AMFm (findings may have been influenced by secular trends in measured outcomes and concurrent malaria interventions).

eness at a national scale we did not include it because it used a before-after design with no comparator sites. The lack of control sites limits the degree to which observed effects can be attributed to AMFm (findings may have been influenced by secular trends in measured outcomes and concurrent malaria interventions). We identified six national programmes to scale-up subsidised ACTs in Cameroon, Senegal, Cambodia, DRC, Madagascar and Rwanda (Table 4). Although the programmes report results indicative of the kind of effect ACT subsidies can have under ‘real world’ conditions, we did not include them because they lacked comparison groups. In addition baseline data were not available for five of the national programmes (such data were only available for Rwanda's national programme). Risk of bias in included studies We have presented findings on risk of bias assessment using ‘Risk of bias’ tables and graphs (Characteristics of included studies; Figure 4; Figure 5). The overall risk of bias was low in two trials (Cohen 2015; Kangwana 2011; Kangwana 2013); it was high in the other two (Sabot 2009; Talisuna 2012).

in included studies We have presented findings on risk of bias assessment using ‘Risk of bias’ tables and graphs (Characteristics of included studies; Figure 4; Figure 5). The overall risk of bias was low in two trials (Cohen 2015; Kangwana 2011; Kangwana 2013); it was high in the other two (Sabot 2009; Talisuna 2012). Random sequence generation and allocation concealment were judged to be adequate (indicating low risk of selection bias) in two trials (Cohen 2015; Kangwana 2011; Kangwana 2013); in the other two they were inadequate (high risk of selection bias). Baseline outcome measures and characteristics between study groups were comparable in all studies. Blinding of personnel (retail outlet providers) and data collectors (interviewers) was not possible in three studies given the public awareness campaigns around subsidised ACTs (Kangwana 2011; Kangwana 2013; Sabot 2009; Talisuna 2012). Performance and detection biases due to lack of blinding were considered low in all studies. The risk of contamination (potential leakage of subsidy intervention into control sites) was low in all except one trial (Sabot 2009), where it was high. None of the trials were at risk of attrition bias or selective outcome reporting. No additional source of bias was present for the cluster-specific domains except for possible recruitment bias in one trial (Sabot 2009). Effects of interventions See: Summary of findings for the main comparison ACT subsidies combined with supportive interventions versus no subsidies; Summary of findings 2 ACT price vouchers versus no subsidies

Random sequence generation and allocation concealment were judged to be adequate (indicating low risk of selection bias) in two trials (Cohen 2015; Kangwana 2011; Kangwana 2013); in the other two they were inadequate (high risk of selection bias). Baseline outcome measures and characteristics between study groups were comparable in all studies. Blinding of personnel (retail outlet providers) and data collectors (interviewers) was not possible in three studies given the public awareness campaigns around subsidised ACTs (Kangwana 2011; Kangwana 2013; Sabot 2009; Talisuna 2012). Performance and detection biases due to lack of blinding were considered low in all studies. The risk of contamination (potential leakage of subsidy intervention into control sites) was low in all except one trial (Sabot 2009), where it was high. None of the trials were at risk of attrition bias or selective outcome reporting. No additional source of bias was present for the cluster-specific domains except for possible recruitment bias in one trial (Sabot 2009). Effects of interventions See: Summary of findings for the main comparison ACT subsidies combined with supportive interventions versus no subsidies; Summary of findings 2 ACT price vouchers versus no subsidies We have presented effect estimates and certainty of evidence for each outcome in GRADE tables (Summary of findings for the main comparison; Summary of findings 2; Appendix 3; Appendix 4). We did not report the certainty of evidence for outcomes where there was insufficient data for GRADE assessment.

Effects of interventions See: Summary of findings for the main comparison ACT subsidies combined with supportive interventions versus no subsidies; Summary of findings 2 ACT price vouchers versus no subsidies We have presented effect estimates and certainty of evidence for each outcome in GRADE tables (Summary of findings for the main comparison; Summary of findings 2; Appendix 3; Appendix 4). We did not report the certainty of evidence for outcomes where there was insufficient data for GRADE assessment. Comparison 1: ACT subsidies combined with supportive interventions versus no subsidies Three studies were included in this comparison (Kangwana 2011; Kangwana 2013; Sabot 2009; Talisuna 2012). Primary outcome Two studies reported data on ACT use (Kangwana 2011; Talisuna 2012; Table 3). In the first study (Kangwana 2011), ACT subsidy programmes increased ACT usage in children under five years of age by 25 percentage points (95% CI 14.1 to 35.9 percentage points; high certainty evidence). This suggests that in practice, among febrile children under five years of age with an ACT usage rate of 5% without a subsidy, subsidy programmes would increase usage by between 19% and 41%. In the second study (Talisuna 2012), ACT subsidy programmes resulted in a ten-fold increase (95% CI 5.0 to 18.9) in ACT usage in children under five years of age and a six-fold increase (95% CI 4.2 to 8.4) in ACT usage in all age groups (very low certainty evidence).

grammes would increase usage by between 19% and 41%. In the second study (Talisuna 2012), ACT subsidy programmes resulted in a ten-fold increase (95% CI 5.0 to 18.9) in ACT usage in children under five years of age and a six-fold increase (95% CI 4.2 to 8.4) in ACT usage in all age groups (very low certainty evidence). Secondary outcomes Two studies reported data on ACT availability (Kangwana 2011; Kangwana 2013; Sabot 2009; Table 5). In the first study (Kangwana 2011; Kangwana 2013), ACT subsidy programmes increased the percentage of retail outlets stocking ACTs for children under five years of age by 31.9 percentage points (95% CI 26.3 to 37.5 percentage points; high certainty evidence). In the second study (Sabot 2009), ACT subsidy programmes increased the percentage of retail outlets stocking ACTs for patients of any age by 72.2 percentage points (95% CI 65.0 to 79.3 percentage points; very low quality evidence). Table 5 Percentage of retail outlets stocking ACTs Study Design Age group Control Intervention Difference (95% CI) Baseline Follow-up Baseline Follow-up Kangwana 2011 Cluster-RCT Children under age 5 0.5% 5.5% 2.4% 37.6% 31.9% (26.3% to 37.5%) Kangwana 2013 Cluster-RCT Children under age 5 0.5% 5.2% 1.5% 36.8% 31.7% (22.0% to 41.3%) Sabot 2009 Non-randomised cluster trial All age groups 1.0% 0% 0% 72.2% 72.2% (65.0% to 79.3%)a ACT = artemisinin-based combination therapy CI = confidence intervalRCT = randomised controlled trial aDrug shops in population centres were more likely to stock ACTs than those in more remote areas (P < 0.001)

Kangwana 2013 Cluster-RCT Children under age 5 0.5% 5.2% 1.5% 36.8% 31.7% (22.0% to 41.3%) Sabot 2009 Non-randomised cluster trial All age groups 1.0% 0% 0% 72.2% 72.2% (65.0% to 79.3%)a ACT = artemisinin-based combination therapy CI = confidence intervalRCT = randomised controlled trial aDrug shops in population centres were more likely to stock ACTs than those in more remote areas (P < 0.001) Three studies reported data on ACT price outcomes (Kangwana 2011; Kangwana 2013; Sabot 2009; Talisuna 2012; Table 6). In the first study (Kangwana 2011), ACT subsidy programmes decreased the median price for ACT prescribed for children under five years of age by US$ 0.84 (median cost per ACT course without subsidy: US$ 1.08 versus with subsidy: US$ 0.24; high certainty evidence). In the second study (Talisuna 2012), "the maximum recommended retail price was within 10% of the recommended ACT price". In addition, the recommended retail price for an adult course of treatment (US$ 0.47) was not adhered to (the median price at the endline survey was US$ 1.96). In the third study (Sabot 2009), the mean price paid for paediatric ACTs (US$ 0.19) was less than for both SP (US$ 0.51, P = 0.001) and AQ (US$ 0.86, P < 0.001). The mean price for ACTs for any age (US$ 0.58) did not differ from the price of SP (US$ 0.67), but was higher than the price for AQ (US$ 0.48, P < 0.001). Table 6 Change in ACT price to patients

Three studies reported data on ACT price outcomes (Kangwana 2011; Kangwana 2013; Sabot 2009; Talisuna 2012; Table 6). In the first study (Kangwana 2011), ACT subsidy programmes decreased the median price for ACT prescribed for children under five years of age by US$ 0.84 (median cost per ACT course without subsidy: US$ 1.08 versus with subsidy: US$ 0.24; high certainty evidence). In the second study (Talisuna 2012), "the maximum recommended retail price was within 10% of the recommended ACT price". In addition, the recommended retail price for an adult course of treatment (US$ 0.47) was not adhered to (the median price at the endline survey was US$ 1.96). In the third study (Sabot 2009), the mean price paid for paediatric ACTs (US$ 0.19) was less than for both SP (US$ 0.51, P = 0.001) and AQ (US$ 0.86, P < 0.001). The mean price for ACTs for any age (US$ 0.58) did not differ from the price of SP (US$ 0.67), but was higher than the price for AQ (US$ 0.48, P < 0.001). Table 6 Change in ACT price to patients Study Design Age group Outcome definition Group Baseline costsa Follow up costa Data collection method Kangwana 2011 Cluster-RCT Under age 5 Median cost per ACT course Control Intervention US$ 1.08 (US$ 0.18) US$ 0.9 (US$ 0.15) US$ 0.84 (US$ 0.14) US$ 0.24 (US$ 0.04) US$ 0.84 per ACT course (IQR not estimable)b,c Kangwana 2013 Cluster-RCT Under age 5 Cost per ACT course (2 doses) Control Intervention NR 2 doses each US$ 2.46 and US$ 2.22 NR US$ 0.25 Not estimable

Study Design Age group Outcome definition Group Baseline costsa Follow up costa Data collection method Kangwana 2011 Cluster-RCT Under age 5 Median cost per ACT course Control Intervention US$ 1.08 (US$ 0.18) US$ 0.9 (US$ 0.15) US$ 0.84 (US$ 0.14) US$ 0.24 (US$ 0.04) US$ 0.84 per ACT course (IQR not estimable)b,c Kangwana 2013 Cluster-RCT Under age 5 Cost per ACT course (2 doses) Control Intervention NR 2 doses each US$ 2.46 and US$ 2.22 NR US$ 0.25 Not estimable Sabot 2009 Non-randomised cluster trial Under age 5d Mean price for ACT Mean price for SP Mean price for AQ Intervention NR US$ 0.19 US$ 0.51 US$ 0.86 Not estimable Talisuna 2012 Non-randomised cluster trial Under age 5 'Retail price for ACT' Control Intervention NR “Maximum recommended retail price was within 10% of the recommended ACT price)”e Not estimable ACT = artemisinin-based combination therapy AQ = amodiaquine IQR = interquartile range NR = not reported RCT = randomised controlled trial SP = sulphadoxine-pyrimethamine aACT treatment course: six tablets (for children aged 3–35 months) and 12 tablets (for children aged 36–59 months) – cost estimates based on ACT course for children aged 3-35 months bDifference between baseline cost per ACT course (control group: US$ 1.08) and follow-up cost per ACT course (intervention group: US$ 0.24) cAt follow-up, 95.3% of caregivers in the intervention arm who bought subsidised AL said they purchased it at the recommended retail price of US$ 0.25. Of the eight not paying this price, three paid less than US$ 0.25 and five paid between US$ 0.31 and US$ 1.23

bDifference between baseline cost per ACT course (control group: US$ 1.08) and follow-up cost per ACT course (intervention group: US$ 0.24) cAt follow-up, 95.3% of caregivers in the intervention arm who bought subsidised AL said they purchased it at the recommended retail price of US$ 0.25. Of the eight not paying this price, three paid less than US$ 0.25 and five paid between US$ 0.31 and US$ 1.23 dAll age groups: the mean price for ACTs (US$ 0.58) did not differ from the price of SP (US$ 0.67), but was higher than for AQ (US$ 0.48, P < 0.001) eThe recommended retail price for an adult course of treatment – US$ 0.47 – was not adhered to (the median price at the endline survey was US$ 1.96) Three studies assessed ACT market share outcomes (Kangwana 2011; Kangwana 2013; Sabot 2009; Talisuna 2012; Table 7). In the first study (Kangwana 2011; Kangwana 2013), ACT subsidy programmes increased market share of ACTs among children under five years of age by between 23.6 and 63.0 percentage points (high certainty evidence). In the second study (Talisuna 2012), the market share for ACTs for patients of any age in the intervention group was 43% at baseline and 69% at follow-up (control data not reported). In the third study (Sabot 2009), the market share for ACTs for children under five years of age increased by 8.9% (-0.5% to 18.2%), and 35.3% (29.8% to 40.7%) for patients aged at least 16 years. Table 7 ACT purchases, sales or market share Study Design Age group Outcome definition Control Intervention Difference (95% CI)

Three studies assessed ACT market share outcomes (Kangwana 2011; Kangwana 2013; Sabot 2009; Talisuna 2012; Table 7). In the first study (Kangwana 2011; Kangwana 2013), ACT subsidy programmes increased market share of ACTs among children under five years of age by between 23.6 and 63.0 percentage points (high certainty evidence). In the second study (Talisuna 2012), the market share for ACTs for patients of any age in the intervention group was 43% at baseline and 69% at follow-up (control data not reported). In the third study (Sabot 2009), the market share for ACTs for children under five years of age increased by 8.9% (-0.5% to 18.2%), and 35.3% (29.8% to 40.7%) for patients aged at least 16 years. Table 7 ACT purchases, sales or market share Study Design Age group Outcome definition Control Intervention Difference (95% CI) Baseline Follow-up Baseline Follow-up Kangwana 2011 Cluster-RCT Children < age 5 Total volume of AL dispensed at general stores 0% 0% 0% 63.0% 63%a Total volume of AL dispensed in specialised drug stores 1.0% 11.0% 0% 65.0% 55%a Kangwana 2013 Cluster-RCT Children < age 5 Total volume of AL sold to mystery shoppers 0.5% 1.8% 0% 25.4% 23.6% (18.7% to 28.6%) Sabot 2009 Non-randomised cluster trial Adults ≥ age 16 Total volume of ACTs purchased 0% 0% 1.0% 35.0% 35.3 % (29.8% to 40.7%)b,c Children < age 5 Total volume of ACTs purchased 0% 6.0% 0% 53.0% 8.9% (-0.5% to 18.2%)d Talisuna 2012 Non-randomised cluster trial All age groups Total volume of ACTs purchased as a proportion of the total volume of all anti-malarials purchased via outlets NR NR 43% 69% NR

Sabot 2009 Non-randomised cluster trial Adults ≥ age 16 Total volume of ACTs purchased 0% 0% 1.0% 35.0% 35.3 % (29.8% to 40.7%)b,c Children < age 5 Total volume of ACTs purchased 0% 6.0% 0% 53.0% 8.9% (-0.5% to 18.2%)d Talisuna 2012 Non-randomised cluster trial All age groups Total volume of ACTs purchased as a proportion of the total volume of all anti-malarials purchased via outlets NR NR 43% 69% NR Total volume of ACTs purchased 1.8% 5.6% 0.8% 26.2% 21.6%a,e ACT = artemisinin-based combination therapy AL = artemether-lumefantrine AQ = amodiaquine CI = confidence interval NR = not reported RCT = randomised controlled trial SP = sulphadoxine-pyrimethamine a95% CI not estimable from the reported data bThere was no correlation between the socio-economic status of the consumer and the likelihood of buying ACTs cPurchases of SP and AQ in the intervention districts declined from 68.0% to 51.0% and 26.0% to 11.0% respectively. Purchases of SP in the control district increased from 62.0% to 83.0% while for AQ declined from 33% to 16.0% dPurchases of SP in the intervention districts decreased from 7.0% to 4.0% and remained the same at 9.0% in the control district. Purchases of AQ in the intervention districts declined from 91.0% to 36.0%, and from 91.0% to 36.0% in the control district eThe market shares for chloroquine and quinine were 5% and 24% respectively at the end of the pilot; Children less than five years had subsidised ACTs purchased for them more often than those aged above 5 years

dPurchases of SP in the intervention districts decreased from 7.0% to 4.0% and remained the same at 9.0% in the control district. Purchases of AQ in the intervention districts declined from 91.0% to 36.0%, and from 91.0% to 36.0% in the control district eThe market shares for chloroquine and quinine were 5% and 24% respectively at the end of the pilot; Children less than five years had subsidised ACTs purchased for them more often than those aged above 5 years Two studies reported data on use of older antimalarials (Kangwana 2011; Talisuna 2012; Table 8). In the first study (Kangwana 2011), ACT subsidy programmes decreased use of AQ and SP among children under five years of age by 10.4 percentage points (95% CI 3.9 to 16.9 percentage points; high certainty evidence). In the second study (Talisuna 2012), at follow-up, use of quinine was 44% in the control group and 37% in the intervention group (odds ratio 0.76, 95% CI 0.54 to 1.08; all age groups; no baseline data were reported). Table 8 Use of older antimalarials Study Design Age group Outcome definition Control Intervention Difference (95% CI) Relative effect (95% CI) Baseline Follow-up Baseline Follow-up Kangwana 2011 Cluster-RCT Children < age 5 Use of antimalarial monotherapya 29.8% 22.8% 39.0% 12.4% -10.4% (-3.9% to -16.9%) NR Talisuna 2012 Non-randomised cluster trial All age groups Use of quinine NR 44% NR 37% NR OR 0.76 (0.53 to 1.08) CI = confidence interval NR = not reported OR = odds ratio RCT = randomised controlled trial aamodiaquine, suphadoxine-pyrimethamine and quinine

Baseline Follow-up Baseline Follow-up Kangwana 2011 Cluster-RCT Children < age 5 Use of antimalarial monotherapya 29.8% 22.8% 39.0% 12.4% -10.4% (-3.9% to -16.9%) NR Talisuna 2012 Non-randomised cluster trial All age groups Use of quinine NR 44% NR 37% NR OR 0.76 (0.53 to 1.08) CI = confidence interval NR = not reported OR = odds ratio RCT = randomised controlled trial aamodiaquine, suphadoxine-pyrimethamine and quinine None of the three studies measured adverse effects of ACT subsidy programmes. Comparison 2: ACT vouchers versus no subsidies One study assessed the effect of ACT vouchers to households on ACT access and targeting, among other outcomes (Cohen 2015). Compared to an access rate of 19% in the control group, subsidies of 80% or more increased the likelihood that a malaria-like illness was treated with an ACT by 16 to 23 percentage points (representing an 85% to 118% increase). However, subsidies were associated with a high rate of over-treatment of malaria: only 56% of patients taking ACTs from the drug shops tested positive for malaria under the 92% subsidy. The two lower subsidy levels were associated with much higher malaria positivity rates: "drug shop ACT-takers were 18 to 19 percentage points more likely to be malaria-positive under the 88 and 80 percent subsidies than under the 92 percent subsidy" (high certainty evidence). ADDITIONAL SUMMARY OF FINDINGS [Explanation]] Effects of ACT price vouchers on ACT accessibility and targeting

Comparison 2: ACT vouchers versus no subsidies One study assessed the effect of ACT vouchers to households on ACT access and targeting, among other outcomes (Cohen 2015). Compared to an access rate of 19% in the control group, subsidies of 80% or more increased the likelihood that a malaria-like illness was treated with an ACT by 16 to 23 percentage points (representing an 85% to 118% increase). However, subsidies were associated with a high rate of over-treatment of malaria: only 56% of patients taking ACTs from the drug shops tested positive for malaria under the 92% subsidy. The two lower subsidy levels were associated with much higher malaria positivity rates: "drug shop ACT-takers were 18 to 19 percentage points more likely to be malaria-positive under the 88 and 80 percent subsidies than under the 92 percent subsidy" (high certainty evidence). ADDITIONAL SUMMARY OF FINDINGS [Explanation]] Effects of ACT price vouchers on ACT accessibility and targeting Patient or population: Patients seeking treatment for suspected uncomplicated malaria Settings: Three rural malaria endemic districts in Western Kenya Intervention: ACT subsidy (ACT vouchers to households; 80% to 92% subsidy) Comparison: No subsidy (households received vouchers to purchase unsubsidised ACTs at the pre-AMFm retail price) Outcomes Effect Number of participants(studies) GRADE certainty of the evidence Comments

Patient or population: Patients seeking treatment for suspected uncomplicated malaria Settings: Three rural malaria endemic districts in Western Kenya Intervention: ACT subsidy (ACT vouchers to households; 80% to 92% subsidy) Comparison: No subsidy (households received vouchers to purchase unsubsidised ACTs at the pre-AMFm retail price) Outcomes Effect Number of participants(studies) GRADE certainty of the evidence Comments ACT access (percentage of illness episodes treated with ACTs; all age groups) Follow-up: 4 months Compared to an access rate of 19% in the control group, subsidies of 80% or more increased the likelihood that a malaria-like illness is treated with an ACT by 16 to 23 percentage points, that is, 85% to 118% increase. 2,789 households (1 study) ⊕⊕⊕⊕ High Cohen 2015 ACT targeting (percentage of ACT takers who are malaria positive; all age groups) Follow-up: 4 months Subsidies were associated with a high rate of over-treatment of malaria (only 56% of patients taking ACTs from the drug shop tested positive for malaria under the 92% subsidy) 2,789 households (1 study) ⊕⊕⊕⊕ High Cohen 2015 ACT: artemisinin-based combination therapy; AMFm: Affordable Medicines Facility–malaria

ACT targeting (percentage of ACT takers who are malaria positive; all age groups) Follow-up: 4 months Subsidies were associated with a high rate of over-treatment of malaria (only 56% of patients taking ACTs from the drug shop tested positive for malaria under the 92% subsidy) 2,789 households (1 study) ⊕⊕⊕⊕ High Cohen 2015 ACT: artemisinin-based combination therapy; AMFm: Affordable Medicines Facility–malaria About the certainty of the evidence (GRADE)* High: This research provides a very good indication of the likely effect. The likelihood that the effect will be substantially different† is low Moderate: This research provides a good indication of the likely effect. The likelihood that the effect will be substantially different† is moderate Low: This research provides some indication of the likely effect. However, the likelihood that it will be substantially different† is high Very low: This research does not provide a reliable indication of the likely effect. The likelihood that the effect will be substantially different† is very high *This is sometimes referred to as ‘quality of evidence’ or ‘confidence in the estimate’ †Substantially different = a large enough difference that it might affect a decision

research does not provide a reliable indication of the likely effect. The likelihood that the effect will be substantially different† is very high *This is sometimes referred to as ‘quality of evidence’ or ‘confidence in the estimate’ †Substantially different = a large enough difference that it might affect a decision Discussion Summary of main results This systematic review examined evidence from studies that evaluated the effect of subsidy programmes aimed at improving accessibility and use of ACT for treatment of malaria. Four trials (five publications) that included substantive subsidies for private retailers were included in the review. All the studies were carried out in three adjacent countries in East Africa (Kenya, Uganda and Tanzania), and had accompanying interventions, including retail outlet provider training, community awareness and mass media campaigns.

s) that included substantive subsidies for private retailers were included in the review. All the studies were carried out in three adjacent countries in East Africa (Kenya, Uganda and Tanzania), and had accompanying interventions, including retail outlet provider training, community awareness and mass media campaigns. The findings indicate that programmes that include substantive price subsidies (90% or more) for private antimalarial drug retailers improve use of ACTs (by 25 percentage points) among children under five years of age with suspected malaria. In practice this suggests that, among febrile children with an ACT usage rate of 5% without a subsidy, subsidy programmes would increase usage by between 19% and 41%. The findings also indicate that subsidy programmes improve ACT stocking in retail outlets (by 32 percentage points) and lower ACT prices (by US$ 0.84 per dose) for children under five years of age with suspected malaria. The impact on ACT stocking for patients of any age is unknown because the certainty of evidence was very low. Subsidy programmes also improve ACT market share (by between 24 to 63 percentage points) among patients of any age and reduce the use of older antimalarials (by 10 percentage points) among children under five years of age.

n ACT stocking for patients of any age is unknown because the certainty of evidence was very low. Subsidy programmes also improve ACT market share (by between 24 to 63 percentage points) among patients of any age and reduce the use of older antimalarials (by 10 percentage points) among children under five years of age. The findings also show that retail-sector ACT subsidies using vouchers lead to substantial increases in ACT access among people seeking treatment for suspected malaria. However, these subsidies also increase inappropriate use of ACTs (that is, increase the proportion of people who receive ACTs but do not in fact have malaria).

ings also show that retail-sector ACT subsidies using vouchers lead to substantial increases in ACT access among people seeking treatment for suspected malaria. However, these subsidies also increase inappropriate use of ACTs (that is, increase the proportion of people who receive ACTs but do not in fact have malaria). Overall completeness and applicability of evidence The data that were used in this review are from studies where malaria was mostly diagnosed based on the presence of fever in people seeking treatment for suspected malaria. The number of patients with confirmed (parasite-diagnosed) malaria was unclear in most of the studies; we therefore do not know whether subsidies resulted in over-treatment of malaria, except in the study of ACT vouchers where this was measured. None of the identified studies assessed mortality or other clinical outcomes. Furthermore, the studies did not examine patient adherence to subsidised ACTs. Thus, whether the observed improvement in ACT usage would translate into real health benefits remains uncertain. It is also unknown whether the impact of subsidies would vary by scale of coverage (sub-national vs. national subsidy programmes), ACT supply (distribution) mechanisms to retail outlets or socioeconomic status (there were insufficient data to explore effects across these subgroups).

l health benefits remains uncertain. It is also unknown whether the impact of subsidies would vary by scale of coverage (sub-national vs. national subsidy programmes), ACT supply (distribution) mechanisms to retail outlets or socioeconomic status (there were insufficient data to explore effects across these subgroups). All the included studies were conducted in rural communities in low-income countries where malaria remains prevalent. Review findings are mostly generalisable to similar settings. However, the small scale nature of the subsidy programmes included in this review, differences in the ACT supply mechanisms and retail sector distribution chains across settings may limit generalisability in some areas. Zambia, for example, has a more expansive public sector distribution chain compared to the private sector (Patouillard 2010).

nature of the subsidy programmes included in this review, differences in the ACT supply mechanisms and retail sector distribution chains across settings may limit generalisability in some areas. Zambia, for example, has a more expansive public sector distribution chain compared to the private sector (Patouillard 2010). Decisions to incorporate retail sector ACT subsidies into national malaria control programmes need to involve consideration of individual country contexts (which could include local malaria epidemiology, public sector to private sector antimalarial market ratio, diagnostic and distributional capacity of retail outlets, access to rapid malaria diagnostics and treatment seeking behaviours). In addition malaria subsidy policies need to balance the benefits of retail sector ACT subsidies and potential unintended adverse effects (for example, delaying the formal treatment-seeking that is needed for correct diagnosis and treatment of malaria and non-malarial fevers; under-treatment of malaria (under-dosing); failing to diagnose and treat co-morbid non-malarial fevers such as pneumonia; and over-treatment of malaria resulting from inappropriate use of subsidised ACTs in individuals with non-malarial fevers). Such use of ACTs may increase the likelihood of emergence of artemisinin resistance.

tment of malaria (under-dosing); failing to diagnose and treat co-morbid non-malarial fevers such as pneumonia; and over-treatment of malaria resulting from inappropriate use of subsidised ACTs in individuals with non-malarial fevers). Such use of ACTs may increase the likelihood of emergence of artemisinin resistance. Quality of the evidence The randomised trials provided high quality evidence on subsidy effects (Cohen 2015; Kangwana 2011; Kangwana 2013). The certainty of evidence for all outcomes reported in the three trials was judged to be high (i.e. the research provides a very good indication of the likely effect). The findings from the other two studies were susceptible to bias (Sabot 2009; Talisuna 2012). We downgraded the certainty of evidence (for ACT stocking for patients of any age) in one study because of high likelihood of selection bias (non-randomised evaluation) and possible confounding by study site (only one control site included; results were likely to be influenced by site-specific factors; Sabot 2009).

2). We downgraded the certainty of evidence (for ACT stocking for patients of any age) in one study because of high likelihood of selection bias (non-randomised evaluation) and possible confounding by study site (only one control site included; results were likely to be influenced by site-specific factors; Sabot 2009). We excluded several studies because they used ineligible designs prone to bias. The ideal designs to assess the effects of large scale public health interventions such as ACT subsidy programmes are cluster-randomised trials with comparable control sites. However, such experimental designs are rarely feasible in practice. For example, cluster randomisation of regions in the included studies was limited by the need to use existing pharmaceutical retail distribution channels. Consequently, implementation of the subsidy interventions could not be restricted to certain areas as predicted by randomisation processes. Furthermore, identification of comparable control groups remains a challenge given inherent differences in contexts such as health systems arrangements in malaria-endemic settings. This challenge was addressed in one included study (Kangwana 2013) through documentation of the context and processes of subsidy implementation (in line with the recommendations for the evaluation of complex interventions; Craig 2008; Shepperd 2009).

exts such as health systems arrangements in malaria-endemic settings. This challenge was addressed in one included study (Kangwana 2013) through documentation of the context and processes of subsidy implementation (in line with the recommendations for the evaluation of complex interventions; Craig 2008; Shepperd 2009). Potential biases in the review process We excluded many potentially eligible studies because of ineligible study designs. It is possible that some of these studies provide useful information that might complement findings from the four included studies. For example, positive effects observed in the included studies were replicated in one excluded study (Tougher 2012). Consistent findings from different study designs across varying malaria transmission and cultural contexts increase our confidence that observed improvements in ACT use, availability, prices and market share can be attributed to the studied subsidy programmes.

udies were replicated in one excluded study (Tougher 2012). Consistent findings from different study designs across varying malaria transmission and cultural contexts increase our confidence that observed improvements in ACT use, availability, prices and market share can be attributed to the studied subsidy programmes. We intended to include only randomised trials, non-randomised trials, controlled before-after studies with at least two intervention and two control sites and interrupted-time-series studies (Appendix 1). These criteria were necessary to minimise possible confounding of subsidy effects by site-specific factors (such as ACT supply chains, regulatory policies and retail provider behaviours). However, we made a post-hoc decision to include two non-randomised cluster trials (Sabot 2009; Talisuna 2012) which compared intervention sites to only one control site. We therefore cannot rule out the possible influence of site-specific factors on observed subsidy effects (hence the decision to downgrade certainty of evidence in Sabot 2009). In addition, government regulatory interventions to phase out monotherapy (AQ/SP) may have contributed to the observed decline in the use of these drugs in Kenya and Uganda (Kangwana 2011; Talisuna 2012). We also did not assess effects on two outcomes as planned in our protocol: ‘availability of alternative antimalarial drugs in all facilities, private and public (including informal outlets)’ and ‘prices of alternative antimalarial drugs (full adult or child courses)’; we used six outcomes included in the ‘Summary of findings’ tables.

assess effects on two outcomes as planned in our protocol: ‘availability of alternative antimalarial drugs in all facilities, private and public (including informal outlets)’ and ‘prices of alternative antimalarial drugs (full adult or child courses)’; we used six outcomes included in the ‘Summary of findings’ tables. Agreements and disagreements with other studies or reviews We identified one related review and one study of the impact of ACT subsidy programmes (Morris 2014; Tougher 2012). Morris 2014 assessed the impact of retail sector ACT subsidies on ACT use. The review included 40 studies, comprising 10 experimental subsidies in eight countries, non-AMFm programmatic subsidies in nine countries and AMFm subsidies in eight pilots. Reported findings were derived from four experimental subsidies, three programmatic subsidies and five of the eight AMFm pilot subsidies. ACT subsidies substantially increased use of ACTs among patients with suspected malaria: each US$1 decrease in price was linked to a 24 percentage point increase in the fraction of suspected malaria cases purchasing ACTs. There were no differences in ACT use among the poorest and richest groups, rural versus urban populations or children versus adults. The authors concluded that ACT price reductions can increase ACT use for suspected malaria, even within poorer, more remote populations that may be most at risk of malaria mortality.

CTs. There were no differences in ACT use among the poorest and richest groups, rural versus urban populations or children versus adults. The authors concluded that ACT price reductions can increase ACT use for suspected malaria, even within poorer, more remote populations that may be most at risk of malaria mortality. Tougher 2012 assessed the effect of the AMFm in seven countries (Ghana, Kenya, Madagascar, Niger, Uganda, Nigeria and Tanzania (including Zanzibar)). The study used a before-after design with no comparator sites (see Excluded studies for details on study methods). The AMFm resulted in a large increase in quality-assured ACT (QAACT) availability (by 25.8 to 51.9 percentage points) in all pilots except Niger and Madagascar, and a large increase in ACT market share (by 15.9 to 40.3 percentage points), driven mainly by changes in the private for-profit sector. Median price for QAACTs per adult equivalent dose decreased substantially in the private for-profit sector in six pilots; the decrease ranged from US$1.28 to $4.82. The market share of oral artemisinin monotherapies decreased in Nigeria and Zanzibar, the two pilots where it was more than 5% at baseline. The authors concluded that subsidies combined with supporting interventions can be effective in rapidly improving availability, price and market share of QAACTs, particularly in the private for-profit sector.

emisinin monotherapies decreased in Nigeria and Zanzibar, the two pilots where it was more than 5% at baseline. The authors concluded that subsidies combined with supporting interventions can be effective in rapidly improving availability, price and market share of QAACTs, particularly in the private for-profit sector. We did not conduct a quality assessment (risk of bias) for the evidence presented in the related review and study; these findings should therefore be interpreted with caution. However, despite differences in study designs, the conclusions in both studies are consistent with the findings of our review: ACT subsidies combined with supportive interventions increase ACT usage, availability and market share and lower ACT prices for people seeking treatment for suspected malaria. Authors' conclusions Implications for practice The findings of this review suggest that programmes that include substantive price subsidies (90% or more) for ACTs for private retailers combined with provider training and marketing improve use and availability of ACTs and lower ACT prices for children under five years of age with suspected malaria. This research has also shown that subsidy programmes improve market share of ACTs (volume of ACTs distributed as a proportion of total volume of all antimalarials distributed via outlets) and reduce use of older, less effective antimalarials for children. We could not draw any conclusion on the impact on ACT stocking for patients of any age because the certainty of evidence was very low.

are of ACTs (volume of ACTs distributed as a proportion of total volume of all antimalarials distributed via outlets) and reduce use of older, less effective antimalarials for children. We could not draw any conclusion on the impact on ACT stocking for patients of any age because the certainty of evidence was very low. Decisions to incorporate retail sector ACT subsidies into national malaria control programmes need to involve consideration of individual country contexts and weigh the benefits of subsidies against potential unintended consequences (such as over-treatment of malaria resulting from inappropriate use of ACTs among patients with non-malarial fevers). Efforts to scale-up retail sector ACT subsidy programmes in malaria-endemic settings (for example, via licensed community based pharmacies) should be complemented with policies to strengthen health systems (for example, enhanced malaria diagnostics using subsidised rapid diagnostic tests to improve ACT targeting; improved antimalarial drug supply in the public sector; in-service malaria case management training; and routine monitoring and surveillance for safety and impact).

ted with policies to strengthen health systems (for example, enhanced malaria diagnostics using subsidised rapid diagnostic tests to improve ACT targeting; improved antimalarial drug supply in the public sector; in-service malaria case management training; and routine monitoring and surveillance for safety and impact). Implications for research The number of patients with confirmed (blood-diagnosed) malaria was unclear in most of the included studies. Thus, future studies should investigate options to better target subsidised ACTs to patients who actually have malaria (for example, effectiveness of retail sector ACT subsidies combined with subsidies for rapid diagnostic tests). Optimal targeting of subsidised ACTs would increase the likelihood that non-malarial illness such as pneumonia (the symptoms of which are often similar to those of malaria) are promptly diagnosed, treated or referred. Such targeting would also reduce the likelihood of emergence of artemisinin resistance. These studies should ideally use cluster-randomised, interrupted-time-series or plausibility designs (Victora 2004). Future studies should also investigate pharmacovigilance and the extent of under- and over-treatment of malaria resulting from inappropriate targeting of retail sector ACT subsidies. The cost-effectiveness and sustainability of subsidy programmes compared to alternative financing mechanisms and other approaches to expand access to subsidised malaria drugs (such as private retail sector-public partnerships and community based strategies) also need to be investigated.

of retail sector ACT subsidies. The cost-effectiveness and sustainability of subsidy programmes compared to alternative financing mechanisms and other approaches to expand access to subsidised malaria drugs (such as private retail sector-public partnerships and community based strategies) also need to be investigated. We would like to thank Marit Johansen and John Eyers for their assistance with the literature searches and Andy Oxman for comments on the review manuscript. This review is supported by a project funded by UKAid from the UK Government Department for International Development. APPENDICES Appendix 1. EPOC review study designs Suggested terms Definition Exclusions Randomised controlled trial (RCT) OR, preferably, randomised trial An experimental study in which people are allocated to different interventions using methods that are random. Studies with only one intervention or control site We recommend only including cluster randomised trials, non-randomised cluster trials, and CBA studies with at least two intervention sites and two control sites. In studies with only one intervention or control site the intervention (or comparison) is completely confounded by study site making it difficult to attribute any observed differences to the intervention rather than to other site-specific variables. Non-randomised controlled trial (NRCT) OR, preferably, non-randomised trial An experimental study in which people are allocated to different interventions using methods that are not random.

APPENDICES Appendix 1. EPOC review study designs Suggested terms Definition Exclusions Randomised controlled trial (RCT) OR, preferably, randomised trial An experimental study in which people are allocated to different interventions using methods that are random. Studies with only one intervention or control site We recommend only including cluster randomised trials, non-randomised cluster trials, and CBA studies with at least two intervention sites and two control sites. In studies with only one intervention or control site the intervention (or comparison) is completely confounded by study site making it difficult to attribute any observed differences to the intervention rather than to other site-specific variables. Non-randomised controlled trial (NRCT) OR, preferably, non-randomised trial An experimental study in which people are allocated to different interventions using methods that are not random. Controlled before-after (CBA) study A study in which observations are made before and after the implementation of an intervention, both in a group that receives the intervention and in a control group that does not. Interrupted-time-series (ITS) study A study that uses observations at multiple time points before and after an intervention (the ‘interruption’). The design attempts to detect whether the intervention has had an effect significantly greater than any underlying trend over time. Studies that do not have a clearly defined point in time when the intervention occurred and at least three data points before and three after the intervention

vention (the ‘interruption’). The design attempts to detect whether the intervention has had an effect significantly greater than any underlying trend over time. Studies that do not have a clearly defined point in time when the intervention occurred and at least three data points before and three after the intervention Repeated measures study (RMS) An ITS study where measurements are made in the same individuals at each time point. Appendix 2. Search strategies CENTRAL, Cochrane Library #1 MeSH descriptor: [Artemisinins] this term only and with qualifier(s): [Economics - EC, Supply & distribution - SD] 16 #2 MeSH descriptor: [Sesquiterpenes] this term only and with qualifier(s): [Economics - EC, Supply & distribution - SD] 5 #3 #1 or #2 18 #4 MeSH descriptor: [Artemisinins] this term only 657 #5 MeSH descriptor: [Sesquiterpenes] this term only 412 #6 (artemisinin or artemisinins or artemisinine or artemisinines or artemisin or artemisins or artemisine or artemisines or arteannuin or "ching hao su" or chinghaosu or ginghaosu or qinghaosu or quinghaosu or quinhaosu or sesquiterpenes or sesquiterpene or artemether or artemetero or artemetherum or artenam or arthemether or paluther or "co artem" or "co artemether" or coartem or coartemether or riamet):ti,ab,kw 1007 #7 #4 or #5 or #6 1007 #8 MeSH descriptor: [Cost Sharing] this term only 25 #9 MeSH descriptor: [Cost Allocation] this term only 16 #10 MeSH descriptor: [Drug Costs] this term only 1689 #11 MeSH descriptor: [Economics, Pharmaceutical] this term only 236 #12 MeSH descriptor: [Commerce] this term only 112

#6 (artemisinin or artemisinins or artemisinine or artemisinines or artemisin or artemisins or artemisine or artemisines or arteannuin or "ching hao su" or chinghaosu or ginghaosu or qinghaosu or quinghaosu or quinhaosu or sesquiterpenes or sesquiterpene or artemether or artemetero or artemetherum or artenam or arthemether or paluther or "co artem" or "co artemether" or coartem or coartemether or riamet):ti,ab,kw 1007 #7 #4 or #5 or #6 1007 #8 MeSH descriptor: [Cost Sharing] this term only 25 #9 MeSH descriptor: [Cost Allocation] this term only 16 #10 MeSH descriptor: [Drug Costs] this term only 1689 #11 MeSH descriptor: [Economics, Pharmaceutical] this term only 236 #12 MeSH descriptor: [Commerce] this term only 112 #13 MeSH descriptor: [Financial Management] this term only 14 #14 MeSH descriptor: [Budgets] this term only 65 #15 MeSH descriptor: [Rate Setting and Review] this term only 0 #16 MeSH descriptor: [Marketing of Health Services] this term only 43 #17 MeSH descriptor: [Social Marketing] this term only 137 #18 MeSH descriptor: [Financial Support] this term only 20 #19 MeSH descriptor: [Financing, Government] this term only 56 #20 MeSH descriptor: [Financing, Organized] this term only 22 #21 MeSH descriptor: [Fees and Charges] this term only 66 #22 MeSH descriptor: [Fees, Pharmaceutical] this term only 25 #23 (subsidy or subsidies or subsidis* or subsidiz*):ti,ab,kw 143 #24 (financing or funding):ti,ab,kw 2733 #25 (co next pay* or copay*):ti,ab,kw 87 #26 (pocket near/3 pay*):ti,ab,kw 13 #27 (voucher or vouchers):ti,ab,kw 272

#21 MeSH descriptor: [Fees and Charges] this term only 66 #22 MeSH descriptor: [Fees, Pharmaceutical] this term only 25 #23 (subsidy or subsidies or subsidis* or subsidiz*):ti,ab,kw 143 #24 (financing or funding):ti,ab,kw 2733 #25 (co next pay* or copay*):ti,ab,kw 87 #26 (pocket near/3 pay*):ti,ab,kw 13 #27 (voucher or vouchers):ti,ab,kw 272 #28 (financial or monetary) near/3 (support or assistance or help or aid or backing):ti,ab,kw 153 #29 (reduce* or lower* or limit* or share or shared or sharing or cut or cutting) near/3 (cost or costs or price or prices or payment* or spending or expenditure):ti,ab,kw 3383 #30 (drug or drugs or medicine* or medicament* or pharmaceutic*) near/3 (econom* or cost or costs or fee or fees or budget* or affordable or marketing):ti,ab,kw 5562 #31 (#8 or #9 or #10 or #11 or #12 or #13 or #14 or #15 or #16 or #17 or #18 or #19 or #20 or #21 or #22 or #23 or #24 or #25 or #26 or #27 or #28 or #29 or #30) 11904 #32 MeSH descriptor: [Private Sector] this term only 41 #33 MeSH descriptor: [Public-Private Sector Partnerships] this term only 8 #34 privat*:ti,ab,kw 2215 #35 retail*:ti,ab,kw 118 #36 (#32 or #33 or #34 or #35) 2324 #37 (#31 or #36) 14025 #38 (#3 or (#7 and #37)) in Trials 41 MEDLINE, OvidSP # Searches Results 1 Artemisinins/ec, sd [Economics, Supply & Distribution] 178 2 Sesquiterpenes/ec, sd [Economics, Supply & Distribution] 42 3 or/1-2 187 4 Artemisinins/ 4435 5 Sesquiterpenes/ 11844

#34 privat*:ti,ab,kw 2215 #35 retail*:ti,ab,kw 118 #36 (#32 or #33 or #34 or #35) 2324 #37 (#31 or #36) 14025 #38 (#3 or (#7 and #37)) in Trials 41 MEDLINE, OvidSP # Searches Results 1 Artemisinins/ec, sd [Economics, Supply & Distribution] 178 2 Sesquiterpenes/ec, sd [Economics, Supply & Distribution] 42 3 or/1-2 187 4 Artemisinins/ 4435 5 Sesquiterpenes/ 11844 6 (artemisinin or artemisinins or artemisinine or artemisinines or artemisin or artemisins or artemisine or artemisines or arteannuin or ching hao su or chinghaosu or ginghaosu or qinghaosu or quinghaosu or quinhaosu or sesquiterpenes or sesquiterpene or artemether or artemetero or artemetherum or artenam or arthemether or paluther or co artem or co artemether or coartem or coartemether or riamet).ti,ab. 9803 7 or/4-6 17929 8 "Cost Sharing"/ 2004 9 "Cost Allocation"/ 1945 10 Drug Costs/ 12407 11 Economics, Pharmaceutical/ 2550 12 Commerce/ 18154 13 Financial Management/ 15422 14 Budgets/ 9877 15 "Rate Setting and Review"/ 2474 16 Marketing of Health Services/ 14065 17 Social Marketing/ 2076 18 Financial Support/ 3039 19 Financing, Government/ 18745 20 Financing, Organized/ 5804 21 "Fees and Charges"/ 8196 22 Fees, Pharmaceutical/ 1141 23 (subsidy or subsidies or subsidis* or subsidiz*).ti,ab. 4892 24 (financing or funding).ti,ab. 39957 25 (co pay* or copay*).ti,ab. 1557 26 (pocket adj3 pay*).ti,ab. 571 27 voucher?.ti,ab. 1052 28 ((financial or monetary) adj3 (support or assistance or help or aid or backing)).ti,ab. 4544 29 ((reduce* or lower* or limit* or share or shared or sharing or cut or cutting) adj3 (cost? or price? or payment? or spending or expenditure)).ti,ab. 38933

25 (co pay* or copay*).ti,ab. 1557 26 (pocket adj3 pay*).ti,ab. 571 27 voucher?.ti,ab. 1052 28 ((financial or monetary) adj3 (support or assistance or help or aid or backing)).ti,ab. 4544 29 ((reduce* or lower* or limit* or share or shared or sharing or cut or cutting) adj3 (cost? or price? or payment? or spending or expenditure)).ti,ab. 38933 30 ((drug or drugs or medicine? or medicament? or pharmaceutic*) adj3 (econom* or cost or costs or fee or fees or budget? or affordable or marketing)).ti,ab. 11651 31 or/8-30 190651 32 Private Sector/ 7513 33 Public-Private Sector Partnerships/ 1171 34 privat*.ti,ab. 63212 35 retail*.ti,ab. 6482 36 or/32-35 73669 37 randomized controlled trial.pt. 384812 38 controlled clinical trial.pt. 88618 39 pragmatic clinical trial.pt. 109 40 multicenter study.pt. 179454 41 non-randomized controlled trials as topic/ 11 42 interrupted time series analysis/ 16 43 controlled before-after studies/ 25 44 (randomis* or randomiz* or randomly or random allocat*).ti,ab. 586424 45 groups.ab. 1414883 46 (trial or multicenter or multi center or multicentre or multi centre).ti. 156461 47 (intervention* or controlled or control or compare or compared or (before adj5 after) or (pre adj5 post) or pretest or pre test or posttest or post test or quasiexperiment* or quasi experiment* or evaluat* or effect or impact or time series or time point? or repeated measur*).ti,ab. 7687512 48 or/37-47 8282037 49 exp Animals/ 17689330 50 Humans/ 13699500 51 49 not (49 and 50) 3989830 52 review.pt. 1937353 53 meta analysis.pt. 53160 54 news.pt. 166715 55 comment.pt. 612267 56 editorial.pt. 369594 57 cochrane database of systematic reviews.jn. 10975

47 (intervention* or controlled or control or compare or compared or (before adj5 after) or (pre adj5 post) or pretest or pre test or posttest or post test or quasiexperiment* or quasi experiment* or evaluat* or effect or impact or time series or time point? or repeated measur*).ti,ab. 7687512 48 or/37-47 8282037 49 exp Animals/ 17689330 50 Humans/ 13699500 51 49 not (49 and 50) 3989830 52 review.pt. 1937353 53 meta analysis.pt. 53160 54 news.pt. 166715 55 comment.pt. 612267 56 editorial.pt. 369594 57 cochrane database of systematic reviews.jn. 10975 58 "comment on".cm. 612267 59 (systematic review or literature review).ti. 57183 60 or/51-59 6788530 61 48 not 60 5826774 62 3 and 61 76 63 7 and (31 or 36) and 61 192 64 62 or 63 [Artemisinin AND subsidy or private AND Methods] 216 Embase, OvidSP # Searches Results 1 artemisinin/ 4920 2 artemisinin derivative/ 2097 3 sesquiterpene/ 5893 4 sesquiterpene derivative/ 2636 5 artemether/ 2243 6 artemether plus benflumetol/ 1792 7 (artemisinin or artemisinins or artemisinine or artemisinines or artemisin or artemisins or artemisine or artemisines or arteannuin or ching hao su or chinghaosu or ginghaosu or qinghaosu or quinghaosu or quinhaosu or sesquiterpenes or sesquiterpene or artemether or artemetero or artemetherum or artenam or arthemether or paluther or co artem or co artemether or coartem or coartemether or riamet).ti,ab. 13715 8 or/1-7 20543 9 ("Health Policy, Economics and management" or "36").ec. 467908 10 drug cost/ 60104 11 commercial phenomena/ 27879 12 drug marketing/ 26589 13 social marketing/ 2703 14 financial management/ 102311 15 budget/ 20306 16 fee/ 14380 17 medical fee/ 11634 18 pharmacoeconomics/ 6029

7 (artemisinin or artemisinins or artemisinine or artemisinines or artemisin or artemisins or artemisine or artemisines or arteannuin or ching hao su or chinghaosu or ginghaosu or qinghaosu or quinghaosu or quinhaosu or sesquiterpenes or sesquiterpene or artemether or artemetero or artemetherum or artenam or arthemether or paluther or co artem or co artemether or coartem or coartemether or riamet).ti,ab. 13715 8 or/1-7 20543 9 ("Health Policy, Economics and management" or "36").ec. 467908 10 drug cost/ 60104 11 commercial phenomena/ 27879 12 drug marketing/ 26589 13 social marketing/ 2703 14 financial management/ 102311 15 budget/ 20306 16 fee/ 14380 17 medical fee/ 11634 18 pharmacoeconomics/ 6029 19 (subsidy or subsidies or subsidis* or subsidiz*).ti,ab. 5621 20 (financing or funding).ti,ab. 47405 21 (co pay* or copay*).ti,ab. 2279 22 (pocket adj3 pay*).ti,ab. 709 23 voucher?.ti,ab. 1226 24 ((financial or monetary) adj3 (support or assistance or help or aid or backing)).ti,ab. 5864 25 ((reduce* or lower* or limit* or share or shared or sharing or cut or cutting) adj3 (cost? or price? or payment? or spending or expenditure)).ti,ab. 50681 26 ((drug or drugs or medicine? or medicament? or pharmaceutic*) adj3 (econom* or cost or costs or fee or fees or budget? or affordable or marketing)).ti,ab. 18354 27 or/9-26 708923 28 public-private partnership/ 2922 29 privat*.ti,ab. 76514 30 retail*.ti,ab. 7520 31 or/28-30 85785 32 Randomized Controlled Trial/ 359286 33 Controlled Clinical Trial/ 390010 34 Quasi Experimental Study/ 2249 35 Pretest Posttest Control Group Design/ 220 36 Time Series Analysis/ 14920 37 Experimental Design/ 10689

26 ((drug or drugs or medicine? or medicament? or pharmaceutic*) adj3 (econom* or cost or costs or fee or fees or budget? or affordable or marketing)).ti,ab. 18354 27 or/9-26 708923 28 public-private partnership/ 2922 29 privat*.ti,ab. 76514 30 retail*.ti,ab. 7520 31 or/28-30 85785 32 Randomized Controlled Trial/ 359286 33 Controlled Clinical Trial/ 390010 34 Quasi Experimental Study/ 2249 35 Pretest Posttest Control Group Design/ 220 36 Time Series Analysis/ 14920 37 Experimental Design/ 10689 38 Multicenter Study/ 114984 39 (randomis* or randomiz* or randomly or random allocat*).ti,ab. 761973 40 groups.ab. 1772331 41 (trial or multicentre or multicenter or multi centre or multi center).ti. 202335 42 (intervention* or controlled or control or compare or compared or (before adj5 after) or (pre adj5 post) or pretest or pre test or posttest or post test or quasiexperiment* or quasi experiment* or evaluat* or effect or impact or time series or time point? or repeated measur*).ti,ab. 9136496 43 or/32-42 9809305 44 (systematic review or literature review).ti. 67987 45 "cochrane database of systematic reviews".jn. 3776 46 Nonhuman/ 4447094 47 or/44-46 4517005 48 43 not 47 7715761 49 8 and (27 or 31) and 48 551 50 limit 49 to embase 513 CINAHL, EbscoHost Top of Form

42 (intervention* or controlled or control or compare or compared or (before adj5 after) or (pre adj5 post) or pretest or pre test or posttest or post test or quasiexperiment* or quasi experiment* or evaluat* or effect or impact or time series or time point? or repeated measur*).ti,ab. 9136496 43 or/32-42 9809305 44 (systematic review or literature review).ti. 67987 45 "cochrane database of systematic reviews".jn. 3776 46 Nonhuman/ 4447094 47 or/44-46 4517005 48 43 not 47 7715761 49 8 and (27 or 31) and 48 551 50 limit 49 to embase 513 CINAHL, EbscoHost Top of Form # Query Results S1 TX (artemisinin or artemisinins or artemisinine or artemisinines or artemisin or artemisins or artemisine or artemisines or arteannuin or "ching hao su" or chinghaosu or ginghaosu or qinghaosu or quinghaosu or quinhaosu or sesquiterpenes or sesquiterpene or artemether or artemetero or artemetherum or artenam or arthemether or paluther or "co artem" or "co artemether" or coartem or coartemether or riamet) Limiters - Exclude MEDLINE records 51 Regional Indexes, Global Health Library, WHO (artemisinin or artemisinins or artemisinine or artemisinines or artemisin or artemisins or artemisine or artemisines or arteannuin or chinghaosu or ginghaosu or qinghaosu or quinghaosu or quinhaosu or sesquiterpenes or sesquiterpene or artemether or artemetero or artemetherum or artenam or arthemether or paluther or coartem or coartemether or riamet) AND (subsidy or subsidies or subsidis* or subsidiz* or commerce or finance* or funding or affordable or marketing or privat* or retail*) LILACS, Global Health Library, WHO

(artemisinin or artemisinins or artemisinine or artemisinines or artemisin or artemisins or artemisine or artemisines or arteannuin or chinghaosu or ginghaosu or qinghaosu or quinghaosu or quinhaosu or sesquiterpenes or sesquiterpene or artemether or artemetero or artemetherum or artenam or arthemether or paluther or coartem or coartemether or riamet) AND (subsidy or subsidies or subsidis* or subsidiz* or commerce or finance* or funding or affordable or marketing or privat* or retail*) LILACS, Global Health Library, WHO (artemisinin or artemisinins or artemisinine or artemisinines or artemisin or artemisins or artemisine or artemisines or arteannuin or chinghaosu or ginghaosu or qinghaosu or quinghaosu or quinhaosu or sesquiterpenes or sesquiterpene or artemether or artemetero or artemetherum or artenam or arthemether or paluther or coartem or coartemether or riamet) Science Citation Index and Social Sciences Citation Index (ISI Web of Science) Topic search (artemisinin or artemisinins or artemisinine or artemisinines or artemisin or artemisins or artemisine or artemisines or arteannuin or "ching hao su" or chinghaosu or ginghaosu or qinghaosu or quinghaosu or quinhaosu or sesquiterpenes or sesquiterpene or artemether or artemetero or artemetherum or artenam or arthemether or paluther or "co artem" or "co artemether" or coartem or coartemether or riamet) (Topic) AND (subsidy or subsidies or subsidise or subsidised or subsidize or subsidized or subsidising or subsidizing or financing or funding) (Topic) AND

(artemisinin or artemisinins or artemisinine or artemisinines or artemisin or artemisins or artemisine or artemisines or arteannuin or "ching hao su" or chinghaosu or ginghaosu or qinghaosu or quinghaosu or quinhaosu or sesquiterpenes or sesquiterpene or artemether or artemetero or artemetherum or artenam or arthemether or paluther or "co artem" or "co artemether" or coartem or coartemether or riamet) (Topic) AND (subsidy or subsidies or subsidise or subsidised or subsidize or subsidized or subsidising or subsidizing or financing or funding) (Topic) AND (randomised or randomized or randomly or "random allocation" or trial or intervention or interventions or controlled or "control group" or "before and after" or "pre and post" or pretest or "pre test" or posttest or "post test" or quasiexperiment or quasiexperimental or "quasi experiment" or "quasi experimental" or evaluate or effect or impact or "time series" or "time point" or "time points" or "repeated measure" or "repeated measures" or "repeated measurement" or "repeated measurements") (Topic) Science Citation Index and Social Sciences Citation Index (ISI Web of Science) Citation search for all 4 included studies Kangwana 2011 Kangwana BP, Kedenge SV, Noor AM, Alegana VA, Nyandigisi AJ, Pandit J, Fegan GW, Todd JE, Brooker S, Snow RW, Goodman CA. The impact of retail-sector delivery of artemether–lumefantrine on malaria treatment of children under five in Kenya: a cluster randomized controlled trial. PLoS Med 2011;8(5):e1000437. Kangwana 2013

Kangwana 2011 Kangwana BP, Kedenge SV, Noor AM, Alegana VA, Nyandigisi AJ, Pandit J, Fegan GW, Todd JE, Brooker S, Snow RW, Goodman CA. The impact of retail-sector delivery of artemether–lumefantrine on malaria treatment of children under five in Kenya: a cluster randomized controlled trial. PLoS Med 2011;8(5):e1000437. Kangwana 2013 Kangwana BP, Kedenge SV, Noor AM, Alegana VA, Nyandigisi AJ, Pandit J, Fegan GW, Todd JE, Snow RW, Goodman CA. The effect of an anti-malarial subsidy programme on the quality of service provision of artemisinin-based combination therapy in Kenya: a cluster-randomized, controlled trial. Malar J 2013;12(81). Sabot 2009 Sabot OJ, Mwita A, Cohen JM, Ipuge Y, Gordon M, Bishop D, Odhiambo M, Ward L, Goodman C. Piloting the global subsidy: the impact of subsidized artemisinin-based combination therapies distributed through private drug shops in rural Tanzania. PLoS One 2009;4(9):e6857. Talisuna 2012 Talisuna AO, Daumerie PG, Balyeku A, Egan T, Piot B, Coghlan R, Lugand M, Bwire G, Rwakimari JB, Ndyomugyenyi R, Kato F, Byangire M, Kagwa P, Sebisubi F, Nahamya D, Bonabana A, Mpanga-Mukasa S, Buyungo P, Lukwago J, Batte A, Nakanwagi G, Tibenderana J, Nayer K, Reddy K, Dokwal N, Rugumambaju S, Kidde S, Banerji J, Jagoe G. Closing the access barrier for effective anti-malarials in the private sector in rural Uganda: consortium for ACT private sector subsidy (CAPSS) pilot study. Malar J 2012;11:356. EconLit, ProQuest

Talisuna AO, Daumerie PG, Balyeku A, Egan T, Piot B, Coghlan R, Lugand M, Bwire G, Rwakimari JB, Ndyomugyenyi R, Kato F, Byangire M, Kagwa P, Sebisubi F, Nahamya D, Bonabana A, Mpanga-Mukasa S, Buyungo P, Lukwago J, Batte A, Nakanwagi G, Tibenderana J, Nayer K, Reddy K, Dokwal N, Rugumambaju S, Kidde S, Banerji J, Jagoe G. Closing the access barrier for effective anti-malarials in the private sector in rural Uganda: consortium for ACT private sector subsidy (CAPSS) pilot study. Malar J 2012;11:356. EconLit, ProQuest ALL(artemisinin or artemisinins or artemisinine or artemisinines or artemisin or artemisins or artemisine or artemisines or arteannuin or "ching hao su" or chinghaosu or ginghaosu or qinghaosu or quinghaosu or quinhaosu or sesquiterpenes or sesquiterpene or artemether or artemetero or artemetherum or artenam or arthemether or paluther or "co artem" or "co artemether" or coartem or coartemether or riamet) Global Health, OvidSP # Searches Results 1 (artemisinin or artemisinins or artemisinine or artemisinines or artemisin or artemisins or artemisine or artemisines or arteannuin or ching hao su or chinghaosu or ginghaosu or qinghaosu or quinghaosu or quinhaosu or sesquiterpenes or sesquiterpene or artemether or artemetero or artemetherum or artenam or arthemether or paluther or co artem or co artemether or coartem or coartemether or riamet).mp. 8409 2 (subsidy or subsidies or subsidise or subsidised or subsidize or subsidized or subsidising or subsidizing or commerce or financing or funding or affordable or marketing or private or retail or retailer or retailers).mp. 43253

# Searches Results 1 (artemisinin or artemisinins or artemisinine or artemisinines or artemisin or artemisins or artemisine or artemisines or arteannuin or ching hao su or chinghaosu or ginghaosu or qinghaosu or quinghaosu or quinhaosu or sesquiterpenes or sesquiterpene or artemether or artemetero or artemetherum or artenam or arthemether or paluther or co artem or co artemether or coartem or coartemether or riamet).mp. 8409 2 (subsidy or subsidies or subsidise or subsidised or subsidize or subsidized or subsidising or subsidizing or commerce or financing or funding or affordable or marketing or private or retail or retailer or retailers).mp. 43253 3 (randomised or randomized or randomly or random allocation or trial or intervention or interventions or controlled or control group or (before adj5 after) or (pre adj5 post) or pretest or pre test or posttest or post test or quasiexperiment or quasiexperimental or quasi experiment or quasi experimental or evaluate or effect or impact or time series or time point or time points or repeated measure or repeated measures or repeated measurement or repeated measurements).mp. 737158 4 1 and 2 and 3 114 Health Management, ProQuest

3 (randomised or randomized or randomly or random allocation or trial or intervention or interventions or controlled or control group or (before adj5 after) or (pre adj5 post) or pretest or pre test or posttest or post test or quasiexperiment or quasiexperimental or quasi experiment or quasi experimental or evaluate or effect or impact or time series or time point or time points or repeated measure or repeated measures or repeated measurement or repeated measurements).mp. 737158 4 1 and 2 and 3 114 Health Management, ProQuest ALL (artemisinin or artemisinins or artemisinine or artemisinines or artemisin or artemisins or artemisine or artemisines or arteannuin or "ching hao su" or chinghaosu or ginghaosu or qinghaosu or quinghaosu or quinhaosu or sesquiterpenes or sesquiterpene or artemether or artemetero or artemetherum or artenam or arthemether or paluther or "co artem" or "co artemether" or coartem or coartemether or riamet) and ALL (subsidy or subsidies or subsidis* or subsidiz* or commerce or financing or funding or affordable or marketing or private or retail or retailer or retailers) and ALL (randomised or randomized or randomly or "random allocation" or trial or intervention or interventions or controlled or "control group" or "before and after" or "pre and post" or pretest or "pre test" or posttest or "post test" or quasiexperiment or quasiexperimental or "quasi experiment" or "quasi experimental" or evaluate or effect or impact or "time series" or "time point" or "time points" or "repeated measure" or "repeated measures" or "repeated measurement" or "repeated measurements")

" or pretest or "pre test" or posttest or "post test" or quasiexperiment or quasiexperimental or "quasi experiment" or "quasi experimental" or evaluate or effect or impact or "time series" or "time point" or "time points" or "repeated measure" or "repeated measures" or "repeated measurement" or "repeated measurements") The Grey Literature Report from The New York Academy of Medicine Library Combination of "artemisinin-based combination therapies, artemether, coartem, private sector, retail sector, retailers, drug shops, pharmacies, public-private sector partnerships, Affordable Medicines Facility – malaria, AMFm, subsidies, co-payment, financing, vouchers." International Clinical Trials Registry Platform (ICTRP), WHO Combination of "artemisinin-based combination therapies, artemether, coartem, private sector, retail sector, retailers, drug shops, pharmacies, public-private sector partnerships, Affordable Medicines Facility – malaria, AMFm, subsidies, co-payment, financing, vouchers." ClinicalTrials.gov, NIH Combination of "artemisinin-based combination therapies, artemether, coartem, private sector, retail sector, retailers, drug shops, pharmacies, public-private sector partnerships, Affordable Medicines Facility – malaria, AMFm, subsidies, co-payment, financing, vouchers."

Combination of "artemisinin-based combination therapies, artemether, coartem, private sector, retail sector, retailers, drug shops, pharmacies, public-private sector partnerships, Affordable Medicines Facility – malaria, AMFm, subsidies, co-payment, financing, vouchers." ClinicalTrials.gov, NIH Combination of "artemisinin-based combination therapies, artemether, coartem, private sector, retail sector, retailers, drug shops, pharmacies, public-private sector partnerships, Affordable Medicines Facility – malaria, AMFm, subsidies, co-payment, financing, vouchers." Appendix 3. GRADE evidence profile: ACT subsidies combined with supportive interventions versus no subsidies Effects of retail sector ACT subsidy programmes on ACT use, availability, price and market share Population: Patients seeking treatment for suspected uncomplicated malaria Settings: East Africa (Kenya, Uganda, Tanzania) Intervention: Retail sector ACT price subsidies plus supportive interventions (retail outlet provider training, community awareness and mass media campaigns) Comparison: Standard practice (no subsidies) Quality assessment Effect GRADE quality of the evidence Importance Number of participants (studies) Design Risk of bias Inconsistency Indirectness Imprecision Other considerations With ACT subsidy No ACT subsidy Absolute difference (95% CI) ACT use (percentage of children under 5 years of age receiving ACT on the same day or following day of fever onset)

Quality assessment Effect GRADE quality of the evidence Importance Number of participants (studies) Design Risk of bias Inconsistency Indirectness Imprecision Other considerations With ACT subsidy No ACT subsidy Absolute difference (95% CI) ACT use (percentage of children under 5 years of age receiving ACT on the same day or following day of fever onset) 2,662 (1 study) Cluster-RCT No serious risk of bias No serious inconsistency No serious indirectness No serious imprecision None 30.3% (19.4% to 41.2%) 5.3% 25% (14.1% to 35.9%) ⊕⊕⊕⊕High CRITICAL ACT availability (percentage of outlets stocking ACTs for children under 5 years of age) 1 study reported in 2 articles Cluster-RCT No serious risk of bias No serious inconsistency No serious indirectness No serious imprecision None 32.4% (22.5% to 41.8%) <0.5% 31.9% (26.3% to 37.5%) ⊕⊕⊕⊕ High CRITICAL ACT availability (percentage of outlets stocking at least one ACT for patients of any age) 1 study Non-randomised cluster trial Seriousa No serious inconsistency No serious indirectness Seriousa None 72.7% (65.5% to 79.8% 0.5% 72.2% (65.0% to 79.3%) ⊕оסоa Very low CRITICAL ACT price (change in ACT price for children under 5 years of age) Median cost per ACT treatment course (6-12 tablets) 1 study Cluster-RCT No serious risk of bias No serious inconsistency No serious indirectness No serious imprecision None US$ 1.08 US$ 0.24 US$ 0.84 ⊕⊕⊕⊕ High CRITICAL ACT market share (volume of ACTs purchased as a proportion of all antimalarials purchased)

ACT price (change in ACT price for children under 5 years of age) Median cost per ACT treatment course (6-12 tablets) 1 study Cluster-RCT No serious risk of bias No serious inconsistency No serious indirectness No serious imprecision None US$ 1.08 US$ 0.24 US$ 0.84 ⊕⊕⊕⊕ High CRITICAL ACT market share (volume of ACTs purchased as a proportion of all antimalarials purchased) 1 study reported in two articles Cluster-RCT No serious risk of bias No serious inconsistency No serious indirectness No serious imprecision None Range 25.4% to 65.0% Range 0% to 11.0% Range 23.6% to 63.0% ⊕⊕⊕⊕ High CRITICAL Use of older antimalarials (amodiaquine, sulphadoxine-pyrimethamine) among children under five years of age 1 study Randomised trials No serious risk of bias No serious inconsistency No serious indirectness No serious imprecision None 24.0% (17.5% to 30.5%) 34.4% -10.4% (-3.9% to -16.9%) ⊕⊕⊕⊕ High CRITICAL Adverse effects (such as the number of people receiving ACTs who do not have malaria) None of the three studies of ACT subsidies combined with supportive interventions assessed adverse effect outcomes. ACT = artemisinin-based combination therapy; RCT = randomised controlled trial aDowngraded from low to very low due to high likelihood of selection bias (non-randomised design) and confounding by study site (only one control site included; results likely to be influenced by site-specific factors) Appendix 4. GRADE evidence profile: ACT vouchers versus no subsidies Effects of ACT price vouchers on ACT accessibility and targeting

aDowngraded from low to very low due to high likelihood of selection bias (non-randomised design) and confounding by study site (only one control site included; results likely to be influenced by site-specific factors) Appendix 4. GRADE evidence profile: ACT vouchers versus no subsidies Effects of ACT price vouchers on ACT accessibility and targeting Population: Patients seeking treatment for suspected uncomplicated malaria Settings: Three rural malaria endemic districts in Western Kenya Intervention: ACT subsidy (ACT vouchers to households; 80 to 92% subsidy) Comparison: No subsidy (households received vouchers to purchase unsubsidised ACTs at the pre-AMFm retail price) Quality assessment Effect GRADE quality of the evidence Importance Number of participants (studies) Design Risk of bias Inconsistency Indirectness Imprecision Other considerations ACT access (percentage of illness episodes treated with ACTs; all age groups; follow-up: 4 months) 2,789 households (1 study) Randomised trial No serious risk of bias No serious inconsistency No serious indirectness No serious imprecision None Compared to an access rate of 19% in the control group, subsidies of 80% or more increased the likelihood that a malaria-like illness is treated with an ACT by 16 to 23 percentage points, that is, an 85% to 118% increase ⊕⊕⊕⊕High IMPORTANT ACT targeting (percentage of ACT takers who are malaria positive; all age groups; follow-up: 4 months)

2,789 households (1 study) Randomised trial No serious risk of bias No serious inconsistency No serious indirectness No serious imprecision None Compared to an access rate of 19% in the control group, subsidies of 80% or more increased the likelihood that a malaria-like illness is treated with an ACT by 16 to 23 percentage points, that is, an 85% to 118% increase ⊕⊕⊕⊕High IMPORTANT ACT targeting (percentage of ACT takers who are malaria positive; all age groups; follow-up: 4 months) 2,789 households (1 study) Randomised trials No serious risk of bias No serious inconsistency No serious indirectness No serious imprecision None Subsidies were associated with a high rate of overtreatment of malaria (only 56% of patients taking ACTs from the drug shop tested positive for malaria under the 92% subsidy) ⊕⊕⊕⊕ High IMPORTANT ACT = artemisinin-based combination therapy; AMFm = Affordable Medicines Facility–malaria Appendix 5. Planned methods not used in the review Unit of analysis issues We will assess whether appropriate analysis was conducted to adjust for clustering in estimating precision of effects in cluster randomised trials and controlled before-after studies. Where clustering has not been accounted for, we will contact study authors, and if possible work with them to re-analyse the results using standard approaches incorporating measures of intra-cluster correlation coefficients (ICCs) (Higgins 2011). If re-analysis is not possible (e.g. due to lack of estimates of ICCs) we will report effect sizes without measures of precision.

tact study authors, and if possible work with them to re-analyse the results using standard approaches incorporating measures of intra-cluster correlation coefficients (ICCs) (Higgins 2011). If re-analysis is not possible (e.g. due to lack of estimates of ICCs) we will report effect sizes without measures of precision. Assessment of heterogeneity For a subset of studies where meta-analysis is considered appropriate (e.g. where study designs and interventions are sufficiently similar), we will also explore heterogeneity using Chi2 tests (Cochran's Q) and the I2 statistic following guidelines described in the Cochrane Handbook for Systematic Reviews of Interventions. Where heterogeneity is detected, we will explore and report plausible explanations for observed differences.

are sufficiently similar), we will also explore heterogeneity using Chi2 tests (Cochran's Q) and the I2 statistic following guidelines described in the Cochrane Handbook for Systematic Reviews of Interventions. Where heterogeneity is detected, we will explore and report plausible explanations for observed differences. Data synthesis We will consider combining results of subsets of studies using meta-analysis (random-effects method) where between-study differences are considered unlikely to explain variability in treatment effects. We will present data used in the synthesis alongside observed results using tables. Where relevant, dichotomous data will be summarised using risk (or rate) ratios (with 95% confidence intervals (CIs)), while continuous data will be summarised using mean differences (with 95% CIs). Where relevant data can be obtained, inappropriately analysed interrupted-time-series' will be re-analysed using time series regression to account for secular trends and potential autocorrelation (in time) of data; the best fit pre-intervention and post-intervention line will be estimated using linear regression or autoregressive integrated moving average techniques (Lagarde 2012; Ramsay 2003).

e-series' will be re-analysed using time series regression to account for secular trends and potential autocorrelation (in time) of data; the best fit pre-intervention and post-intervention line will be estimated using linear regression or autoregressive integrated moving average techniques (Lagarde 2012; Ramsay 2003). Subgroup analysis and investigation of heterogeneity We anticipate the number of eligible studies will be small so we have therefore limited the subgroup analysis. We will explore consistency of intervention effects in stratified analyses to examine whether the intervention indeed is pro-poor: participant socioeconomic status (SES; low versus high SES quintiles; we will consider the poor as those in the lower three SES quintiles or those living on less than $2 (purchasing power parity) per day (World Bank 2014). The following variables will be explored as possible explanations for heterogeneity: Study designs Nature of supportive interventions (e.g. malaria diagnostics) SES Coverage of interventions (sub-national versus national programmes) Sensitivity analysis We will conduct sensitivity analysis narratively or statistically (for a subset of studies where meta-analysis is considered appropriate) to investigate the influence of study quality. CHARACTERISTICS OF STUDIES Characteristics of included studies [ordered by year of study] Sabot 2009 Methods Non-randomised controlled cluster trial Participants Country: Tanzania Setting (coverage): 2 intervention districts, 1 control district Outlets: Small drug shops (duka la dawa baridi) Age group: All age groups

Sensitivity analysis We will conduct sensitivity analysis narratively or statistically (for a subset of studies where meta-analysis is considered appropriate) to investigate the influence of study quality. CHARACTERISTICS OF STUDIES Characteristics of included studies [ordered by year of study] Sabot 2009 Methods Non-randomised controlled cluster trial Participants Country: Tanzania Setting (coverage): 2 intervention districts, 1 control district Outlets: Small drug shops (duka la dawa baridi) Age group: All age groups Interventions Intervention: Subsidised ACT (AL) Comparison: No ACT subsidy (control) Supportive interventions: Behavior change communication (e.g. local radio advertisements, wall paintings, themed cultural shows) emphasising the importance of using ACTs and their availability in private shops Outcomes ACT uptake, availability and price Notes The project managers procured AL from the manufacturer, Novartis, and sold them to a pharmaceutical wholesaler in Dar es Salaam at an average of US$ $0.11 per dose, 88% below the price offered to public buyers. In one of the intervention districts (Kongwa), the suggested retail price intended to inform consumers of the maximum amount they should pay was set at 300, 600, 900, and 1200 Tanzanian shillings (0.25, 0.50. 0.75. and 1 USD respectively) for the four weight packs respectively; no suggested retail price was included on drugs distributed to Maswa in order to test its effect on price outcomes. Risk of bias Bias Authors' judgement Support for judgement

Notes The project managers procured AL from the manufacturer, Novartis, and sold them to a pharmaceutical wholesaler in Dar es Salaam at an average of US$ $0.11 per dose, 88% below the price offered to public buyers. In one of the intervention districts (Kongwa), the suggested retail price intended to inform consumers of the maximum amount they should pay was set at 300, 600, 900, and 1200 Tanzanian shillings (0.25, 0.50. 0.75. and 1 USD respectively) for the four weight packs respectively; no suggested retail price was included on drugs distributed to Maswa in order to test its effect on price outcomes. Risk of bias Bias Authors' judgement Support for judgement Random sequence generation (selection bias) High risk Quote: "The selected districts were randomly assigned to one of the three arms in the study design: subsidy, subsidy plus suggested retail price, and no subsidy (control). As two of the qualified districts were adjacent, randomization was limited so that one of the adjacent districts served as the control" Comment: Non-randomised design Allocation concealment (selection bias) High risk Quote: "The selected districts were randomly assigned to one of the three arms in the study design: subsidy, subsidy plus suggested retail price, and no subsidy (control). As two of the qualified districts were adjacent, randomization was limited so that one of the adjacent districts served as the control" Comment: Non-randomised design Baseline outcome measurements Low risk Comment: No important differences across study groups on pre-specified subsidy outcome measures.

Allocation concealment (selection bias) High risk Quote: "The selected districts were randomly assigned to one of the three arms in the study design: subsidy, subsidy plus suggested retail price, and no subsidy (control). As two of the qualified districts were adjacent, randomization was limited so that one of the adjacent districts served as the control" Comment: Non-randomised design Baseline outcome measurements Low risk Comment: No important differences across study groups on pre-specified subsidy outcome measures. Baseline characteristics Low risk Quote: "The selected districts were among the few roughly comparable across all indicators, with high malaria transmission, large numbers of private drug shops and, importantly, no malaria related trials (e.g. vaccines) underway" Blinding of participants and personnel (performance bias) All outcomes Low risk Comment: Blinding was not possible for study personnel and ACT providers due to the public information campaign around the subsidised drugs in the intervention arm. Lack of blinding was, however, unlikely to influence results as the outcomes assessed were objective. Blinding of outcome assessment (detection bias) All outcomes Low risk Comment: Blinding was not possible for data collectors due to the public information campaign around the subsidised drugs in the intervention arm. Lack of blinding was however unlikely to influence results as outcomes assessed were objective. Contamination High risk Quote: "As two of the qualified districts were adjacent, randomization was limited so that one of the adjacent districts served as the control"

Blinding of outcome assessment (detection bias) All outcomes Low risk Comment: Blinding was not possible for data collectors due to the public information campaign around the subsidised drugs in the intervention arm. Lack of blinding was however unlikely to influence results as outcomes assessed were objective. Contamination High risk Quote: "As two of the qualified districts were adjacent, randomization was limited so that one of the adjacent districts served as the control" Incomplete outcome data (attrition bias) All outcomes Low risk Comment: The number of drug shops closed or refusing to participate were 30 (13%) and 39 (15%) respectively. Selective reporting (reporting bias) Low risk Comment: Data on pre-specified outcomes reported Recruitment bias High risk Quote: "The selected districts were randomly assigned to one of the three arms in the study design: subsidy, subsidy plus suggested retail price, and no subsidy (control). As two of the qualified districts were adjacent, randomization was limited so that one of the adjacent districts served as the control" Loss of clusters Low risk Quote: “The total number of DLDB audited increased from 200 in August 2007 to 216 in August 2008 due to the opening of new shops, with 30 (13%) and 39 (15%) additional shops closed or refusing to participate at these two time periods respectively.”

Recruitment bias High risk Quote: "The selected districts were randomly assigned to one of the three arms in the study design: subsidy, subsidy plus suggested retail price, and no subsidy (control). As two of the qualified districts were adjacent, randomization was limited so that one of the adjacent districts served as the control" Loss of clusters Low risk Quote: “The total number of DLDB audited increased from 200 in August 2007 to 216 in August 2008 due to the opening of new shops, with 30 (13%) and 39 (15%) additional shops closed or refusing to participate at these two time periods respectively.” Incorrect analysis Low risk Quote: “To assess geographical variation in outcomes, the competition level of all DLDB was calculated using the fixed radius approach...The competitive space of each DLDB was defined as 1 kilometer and each shop was assigned to a competition index category between 0 and 5 based on the number of other DLDB within that radius.” “A repeated measures multivariate regression model was used to compare differences in purchase price while controlling for potentially confounding factors and adjusting for clustering of multiple purchases in the same shops.” Comparability with individually randomised trials Low risk Comment: Included clusters comparable Kangwana 2011 Methods Cluster-randomised controlled trial Participants Country: Kenya Setting (coverage): 3 districts (9 sublocations allocated to intervention, 9 sublocations allocated to control) Outlets: Retail outlets (specialised drug shops and general shops) Age group: Children under 5

Comparability with individually randomised trials Low risk Comment: Included clusters comparable Kangwana 2011 Methods Cluster-randomised controlled trial Participants Country: Kenya Setting (coverage): 3 districts (9 sublocations allocated to intervention, 9 sublocations allocated to control) Outlets: Retail outlets (specialised drug shops and general shops) Age group: Children under 5 Interventions Intervention: Tibamal (subsidised ACT: AL) plus supportive interventions Comparison: No subsidised ACT (control) Supportive interventions: Training of retail outlet staff, job aids, community awareness activities (e.g. workshops, posters and paintings on shops; these activities were designed to make the community aware of malaria, the availability of Tibamal, and the importance of adherence to the medication). Outcomes Primary outcome: proportion of children reporting fever in the past 2 weeks who started treatment with AL on the day or following day of fever onset. Secondary outcomes: adequacy of AL doses obtained and consumed and the price paid per pack Notes At the time of the study, AL had a retail price of around US$ 6.16 (500 Kenyan shillings) compared with an average of around US$ 0.37 for common, older antimalarials such as SP and AQ. The outlets were instructed to sell the packs at a retail price of US$ 0.25, which was printed on the drug packaging, providing a 150% retailer mark-up (exceeding that of AQ and SP, which generally had retail markups of 50% to 100%). Risk of bias Bias Authors' judgement Support for judgement

Notes At the time of the study, AL had a retail price of around US$ 6.16 (500 Kenyan shillings) compared with an average of around US$ 0.37 for common, older antimalarials such as SP and AQ. The outlets were instructed to sell the packs at a retail price of US$ 0.25, which was printed on the drug packaging, providing a 150% retailer mark-up (exceeding that of AQ and SP, which generally had retail markups of 50% to 100%). Risk of bias Bias Authors' judgement Support for judgement Random sequence generation (selection bias) Low risk Comment: A random list of all eligible sublocations was formulated per district in Microsoft Excel. The first intervention sublocation was selected from the top of the list. In order to reduce the potential for contamination a “buffer zone” was created where all sublocations located within two sublocation boundaries of the selected sublocation were removed from the list. The list was reshuffled randomly and the first sublocation on the new list allocated to the control arm. Allocation concealment (selection bias) Low risk Comment: The same procedure as for random sequence generation (described above) was used; intervention allocation could not have been seen in advance. Baseline outcome measurements Low risk Comment: No important differences across study groups on pre-specified subsidy outcome measures Baseline characteristics Low risk Comment: Baseline characteristics of study groups reported and comparable

Allocation concealment (selection bias) Low risk Comment: The same procedure as for random sequence generation (described above) was used; intervention allocation could not have been seen in advance. Baseline outcome measurements Low risk Comment: No important differences across study groups on pre-specified subsidy outcome measures Baseline characteristics Low risk Comment: Baseline characteristics of study groups reported and comparable Blinding of participants and personnel (performance bias) All outcomes Low risk Comment: Blinding was not possible for shopkeepers and community members due to the public information campaign around the subsidised drugs in the intervention arm. Lack of blinding was, however, unlikely to influence results as the outcomes assessed were objective. Blinding of outcome assessment (detection bias) All outcomes Low risk Comment: Blinding was not possible for data collectors due to the public information campaign around the subsidised drugs in the intervention arm. Lack of blinding was, however, unlikely to influence results as outcomes assessed were objective. Contamination Low risk Quote: "No children in the control arm were reported to have received Tibamal (subsidised ACT) at follow-up. In addition, at follow-up 82% of caregivers in the intervention arm had heard of Tibamal, compared to only 7% in the control arm." Incomplete outcome data (attrition bias) All outcomes Low risk Comment: In the control arm, of 1,679 households interviewed at baseline, 152 were lost to follow-up; in the intervention arm, Of 1,609 households interviewed, 114 were lost to follow-up.

Contamination Low risk Quote: "No children in the control arm were reported to have received Tibamal (subsidised ACT) at follow-up. In addition, at follow-up 82% of caregivers in the intervention arm had heard of Tibamal, compared to only 7% in the control arm." Incomplete outcome data (attrition bias) All outcomes Low risk Comment: In the control arm, of 1,679 households interviewed at baseline, 152 were lost to follow-up; in the intervention arm, Of 1,609 households interviewed, 114 were lost to follow-up. Selective reporting (reporting bias) Low risk Comment: Data on prespecified outcomes were reported Recruitment bias Low risk Quote: "A random list of all eligible sublocations was formulated per district in Microsoft Excel. The first intervention sublocation was selected from the top of the list...The list was reshuffled randomly and the first sublocation on the new list allocated to the control arm." Loss of clusters Low risk Quote: "We completed interviews in 2,319 homesteads at baseline (3,288 households), and 2,204 homesteads at follow-up (3,182 households).All randomised clusters included in the analysis." Incorrect analysis Low risk Quote: "A separate analysis allowing for clustering within homesteads was also conducted." Comparability with individually randomised trials Low risk Comment: Included clusters comparable Talisuna 2012 Methods Non-randomised controlled cluster trial Participants Country: Uganda Setting (coverage): 4 intervention districts, 1 control district Outlets: Drug shops (private drug shops, private clinics, pharmacies) Age group: All age groups

Comparability with individually randomised trials Low risk Comment: Included clusters comparable Talisuna 2012 Methods Non-randomised controlled cluster trial Participants Country: Uganda Setting (coverage): 4 intervention districts, 1 control district Outlets: Drug shops (private drug shops, private clinics, pharmacies) Age group: All age groups Interventions Intervention: Subsidised ACT plus supportive interventions Comparison: No ACT subsidy (control) Supportive interventions: Communication activities to improve awareness of the importance and availability of ACTs, and training activities to ensure correct dispensing and use of subsidised ACTs Outcomes ACT uptake, purchase of ACT within 24 hours of symptom onset and ACT price, availability and market share Notes There was better availability of ACT in the public sector in the control district because of: (1) new interventions initiated targeting the community level distribution of ACTs through the public sector; and (2) improvements in the procurement and distribution system in the public sector, based on a push instead of a pull system for the lower level health units. The maximum recommended retail price for the subsidised ACT ranged from 200 Ugandan shillings to 800 Ugandan shillings (US$ 0.10 to 0.40), depending on the target age/doses. Risk of bias Bias Authors' judgement Support for judgement

Notes There was better availability of ACT in the public sector in the control district because of: (1) new interventions initiated targeting the community level distribution of ACTs through the public sector; and (2) improvements in the procurement and distribution system in the public sector, based on a push instead of a pull system for the lower level health units. The maximum recommended retail price for the subsidised ACT ranged from 200 Ugandan shillings to 800 Ugandan shillings (US$ 0.10 to 0.40), depending on the target age/doses. Risk of bias Bias Authors' judgement Support for judgement Random sequence generation (selection bias) High risk Quote: "Four intervention districts were purposefully selected to receive branded subsidised medicines - ‘ACT with a leaf’, while the fifth district acted as the control." Comment: Non-randomised design Allocation concealment (selection bias) High risk Quote: "Four intervention districts were purposefully selected to receive branded subsidised medicines - ‘ACT with a leaf’, while the fifth district acted as the control." Comment: Non-randomised design Baseline outcome measurements Low risk Comment: No important differences across study groups on pre-specified subsidy outcome measures

Allocation concealment (selection bias) High risk Quote: "Four intervention districts were purposefully selected to receive branded subsidised medicines - ‘ACT with a leaf’, while the fifth district acted as the control." Comment: Non-randomised design Baseline outcome measurements Low risk Comment: No important differences across study groups on pre-specified subsidy outcome measures Baseline characteristics Low risk Quote: "Some discordance was observed between the intervention and control districts at baseline in terms of drug consumption habits. Fortunately, the observed disparities did not include the use of ACT. Generating survey-adjusted outputs was intended to provide a more reasonable range of likely values that accounted for this prior to executing tests of significance" Blinding of participants and personnel (performance bias) All outcomes Low risk Comment: Blinding was not possible for drug shop personnel due to the public information campaign around the subsidised drugs in the intervention arm. Lack of blinding was, however, unlikely to influence outcomes as outcomes assessed were objective. Blinding of outcome assessment (detection bias) All outcomes Low risk Quote: "Possible interviewer bias minimised through a week long training to instil strict processes for conducting interviews, and to minimise deviation from the interview script. It was unlikely that the lack of blinding could influence outcomes as it was obligatory for the interviewer to observe and record the details of the actual medicine purchased."

nimised through a week long training to instil strict processes for conducting interviews, and to minimise deviation from the interview script. It was unlikely that the lack of blinding could influence outcomes as it was obligatory for the interviewer to observe and record the details of the actual medicine purchased." Contamination Low risk Quote: "To limit leakage of the intervention to the control area, the control and intervention areas had two intervening buffer districts (Bukedea and Kumi) or a lake between them." Incomplete outcome data (attrition bias) All outcomes Low risk Quote: "...the rate of refusal was generally small (not exceeding 10%) and any effect due to refusal probably did not impact significantly on the outcome measures." Comment: Of the 5,643 observations collected in the final evaluation survey, 5,181 observations resulting from visits to 783 outlets were included in the analysis. Selective reporting (reporting bias) Low risk Comment: Data on pre-specified outcomes reported Recruitment bias Low risk Quote: "Four intervention districts were purposefully selected to receive branded subsidised medicines - ‘ACT with a leaf’, while the fifth district acted as the control."

Incomplete outcome data (attrition bias) All outcomes Low risk Quote: "...the rate of refusal was generally small (not exceeding 10%) and any effect due to refusal probably did not impact significantly on the outcome measures." Comment: Of the 5,643 observations collected in the final evaluation survey, 5,181 observations resulting from visits to 783 outlets were included in the analysis. Selective reporting (reporting bias) Low risk Comment: Data on pre-specified outcomes reported Recruitment bias Low risk Quote: "Four intervention districts were purposefully selected to receive branded subsidised medicines - ‘ACT with a leaf’, while the fifth district acted as the control." Loss of clusters Low risk Quote: “Following the baseline survey, [Consortium for ACT Private Sector Subsidy] initiated the ACT subsidy for the intervention districts and four subsequent rounds of monitoring of cross-sectional surveys were administered in the same manner as the baseline survey, with the final round occurring during the period 20 April - 11 May, 2010 (Baseline: n=1,162; round 1: n=1044; round 2: n=1794; round 3: n=1976 and final round 4: n=5181). A minimum number of 5 interviews per outlet were respected in the final survey round, resulting in a significantly higher sample size.” Incorrect analysis Low risk Quote: “Based on the data-collection methods, a survey-adjusted logistic regression model was used. The outlets were treated as the population sampling units within five strata - the five pilot districts.”

Loss of clusters Low risk Quote: “Following the baseline survey, [Consortium for ACT Private Sector Subsidy] initiated the ACT subsidy for the intervention districts and four subsequent rounds of monitoring of cross-sectional surveys were administered in the same manner as the baseline survey, with the final round occurring during the period 20 April - 11 May, 2010 (Baseline: n=1,162; round 1: n=1044; round 2: n=1794; round 3: n=1976 and final round 4: n=5181). A minimum number of 5 interviews per outlet were respected in the final survey round, resulting in a significantly higher sample size.” Incorrect analysis Low risk Quote: “Based on the data-collection methods, a survey-adjusted logistic regression model was used. The outlets were treated as the population sampling units within five strata - the five pilot districts.” Comparability with individually randomised trials Low risk Comment: Included clusters comparable Kangwana 2013 Methods Cluster-randomised controlled trial Participants Country: Kenya Setting (coverage): 3 districts (9 sublocations allocated to intervention, 9 sublocations allocated to control) Outlets: Retail outlets (specialised drug shops and general shops) Age group: Children under 5

Comparability with individually randomised trials Low risk Comment: Included clusters comparable Kangwana 2013 Methods Cluster-randomised controlled trial Participants Country: Kenya Setting (coverage): 3 districts (9 sublocations allocated to intervention, 9 sublocations allocated to control) Outlets: Retail outlets (specialised drug shops and general shops) Age group: Children under 5 Interventions Intervention: Tibamal (subsidised ACT: AL) plus supportive interventions Comparison: No subsidised ACT (control) Supportive interventions: Training of retail outlet staff, job aids, community awareness activities (e.g. workshops, posters and paintings on shops; these activities were designed to make the community aware of malaria, the availability of Tibamal, and the importance of adherence to the medication). Outcomes AL uptake (provider behaviour), availability of older antimalarials, AL price, AL stocking, provider knowledge, and provider dispensing practices. Generally, outlets that received subsidised AL plus training and job aids performed better than those receiving one or none of these intervention components.

Interventions Intervention: Tibamal (subsidised ACT: AL) plus supportive interventions Comparison: No subsidised ACT (control) Supportive interventions: Training of retail outlet staff, job aids, community awareness activities (e.g. workshops, posters and paintings on shops; these activities were designed to make the community aware of malaria, the availability of Tibamal, and the importance of adherence to the medication). Outcomes AL uptake (provider behaviour), availability of older antimalarials, AL price, AL stocking, provider knowledge, and provider dispensing practices. Generally, outlets that received subsidised AL plus training and job aids performed better than those receiving one or none of these intervention components. Notes At the time of the study, AL had a retail price of around US$ 6.16 (500 Kenyan shillings) compared with an average of around US$ 0.37 for common, older antimalarials such as SP and AQ. The outlets were instructed to sell the packs at a retail price of US$ 0.25, which was printed on the drug packaging, providing a 150% retailer mark-up (exceeding that of AQ and SP, which generally had retail markups of 50% to 100%). Generally, outlets that received training and job aids performed better than those receiving one or none of these intervention components. Risk of bias Bias Authors' judgement Support for judgement

Notes At the time of the study, AL had a retail price of around US$ 6.16 (500 Kenyan shillings) compared with an average of around US$ 0.37 for common, older antimalarials such as SP and AQ. The outlets were instructed to sell the packs at a retail price of US$ 0.25, which was printed on the drug packaging, providing a 150% retailer mark-up (exceeding that of AQ and SP, which generally had retail markups of 50% to 100%). Generally, outlets that received training and job aids performed better than those receiving one or none of these intervention components. Risk of bias Bias Authors' judgement Support for judgement Random sequence generation (selection bias) Low risk Comment: A random list of all eligible sublocations was formulated per district in Microsoft Excel. The first intervention sublocation was selected from the top of the list. In order to reduce the potential for contamination a “buffer zone” was created where all sublocations located within two sublocation boundaries of the selected sublocation were removed from the list. The list was reshuffled randomly and the first sublocation on the new list allocated to the control arm. Allocation concealment (selection bias) Low risk Comment: The same procedure as for random sequence generation (described above) was used; intervention allocation could not have been seen in advance. Baseline outcome measurements Low risk Comment:No important differences across study groups on pre-specified subsidy outcome measures

Allocation concealment (selection bias) Low risk Comment: The same procedure as for random sequence generation (described above) was used; intervention allocation could not have been seen in advance. Baseline outcome measurements Low risk Comment:No important differences across study groups on pre-specified subsidy outcome measures Baseline characteristics Low risk Quote: "To control for potential confounders the covariates considered were outlet type (specialized drug store or general store), distance of shop to nearest road, whether any staff had clinically related training, and district. All covariates significant at a p-value of <0.2 were retained in the regression model." Blinding of participants and personnel (performance bias) All outcomes Low risk Comment: Blinding was not possible for shopkeepers due to the public information campaign around the subsidied drugs in the intervention arm. Lack of blinding was, however, unlikely to influence results as the outcomes assessed were objective. Blinding of outcome assessment (detection bias) All outcomes Low risk Comment: Blinding was not possible for data collectors due to the public information campaign around the subsidised drugs in the intervention arm. Lack of blinding was, however, unlikely to influence results as outcomes assessed were objective.

Blinding of participants and personnel (performance bias) All outcomes Low risk Comment: Blinding was not possible for shopkeepers due to the public information campaign around the subsidied drugs in the intervention arm. Lack of blinding was, however, unlikely to influence results as the outcomes assessed were objective. Blinding of outcome assessment (detection bias) All outcomes Low risk Comment: Blinding was not possible for data collectors due to the public information campaign around the subsidised drugs in the intervention arm. Lack of blinding was, however, unlikely to influence results as outcomes assessed were objective. Contamination Low risk Quote: "It is also possible that there was some contamination of the control arm outlets, which could have heard some of the communication activities. However, results indicated that such exposure was low, with only 1% of control arm respondents saying that they had attended the Tibamal training, 14% having heard of Tibamal, and no outlets stocking Tibamal." Incomplete outcome data (attrition bias) All outcomes Low risk Quote: "In both the provider and mystery-shopper surveys, at baseline and also at follow-up, less than 10% of outlets were not interviewed either because the respondent refused to be interviewed or the outlet was closed during visits." Selective reporting (reporting bias) Low risk Comment: Data on pre-specified outcomes were reported

Incomplete outcome data (attrition bias) All outcomes Low risk Quote: "In both the provider and mystery-shopper surveys, at baseline and also at follow-up, less than 10% of outlets were not interviewed either because the respondent refused to be interviewed or the outlet was closed during visits." Selective reporting (reporting bias) Low risk Comment: Data on pre-specified outcomes were reported Recruitment bias Low risk Quote: "A random list of all eligible sublocations was formulated per district in Microsoft Excel. The first intervention sublocation was selected from the top of the list...The list was reshuffled randomly and the first sublocation on the new list allocated to the control arm." Loss of clusters Low risk Quote: "We completed interviews in 2,319 homesteads at baseline (3,288 households), and 2,204 homesteads at follow-up (3,182 households).All randomised clusters included in the analysis." Incorrect analysis Low risk Quote: "A separate analysis allowing for clustering within homesteads was also conducted." Comparability with individually randomised trials Low risk Comment: Included clusters comparable Cohen 2015 Methods Cluster-randomised controlled trial Participants Country: Kenya Setting (coverage): Three rural districts (Busia, Mumias and Samia in Western Kenya) Outlets: Retail outlets (drug shops) Age group: All age groups

Incorrect analysis Low risk Quote: "A separate analysis allowing for clustering within homesteads was also conducted." Comparability with individually randomised trials Low risk Comment: Included clusters comparable Cohen 2015 Methods Cluster-randomised controlled trial Participants Country: Kenya Setting (coverage): Three rural districts (Busia, Mumias and Samia in Western Kenya) Outlets: Retail outlets (drug shops) Age group: All age groups Interventions Households were randomly assigned to one of three groups: (1) “No subsidy” group (“received vouchers to purchase unsubsidized ACTs at the market price of KSh (Kenyan shillings) 500 (just under $6.25). This treatment arm was meant to capture the no-subsidy status quo that prevailed in Kenya prior to the AMFm pilot, in which over-the-counter ACTs were expensive and RDTs were not available in drug shops.”) (2) ACT subsidy only group (3) ‘ACT plus RDT’ subsidy group (received vouchers for both subsidised ACTs and RDTs) The ACT used in the study was Coartem (AL). Supportive interventions: None. Outcomes Outcomes of interest to current review: ACT accessibility (“the share of illness episodes treated with ACTs”); ACT targeting (“the share of ACT-takers who are malaria positive”) Notes “Within the two ACT subsidy groups (‘ACT subsidy only’ and ‘ACT+RDT subsidy’), households were randomly assigned to an ACT subsidy level of 92, 88 or 80 percent (corresponding to $0.50, $0.75 and $1.25 for an adult dose, respectively). The 92 percent subsidy level corresponds to the Kenyan government's target retail price of KSh 40 under the AMFm.” Risk of bias

Notes “Within the two ACT subsidy groups (‘ACT subsidy only’ and ‘ACT+RDT subsidy’), households were randomly assigned to an ACT subsidy level of 92, 88 or 80 percent (corresponding to $0.50, $0.75 and $1.25 for an adult dose, respectively). The 92 percent subsidy level corresponds to the Kenyan government's target retail price of KSh 40 under the AMFm.” Risk of bias Bias Authors' judgement Support for judgement Random sequence generation (selection bias) Low risk Quote: “The randomization of households was done using a computerized random number assignment algorithm and was stratified by drug shop, by the household's distance to the drug shop (in quartiles) and by the presence of children in the household.” Allocation concealment (selection bias) Low risk Comment: The same procedure as for random sequence generation (described above) was used; selection bias due to foreknowledge of treatment allocation is considered unlikely.

Random sequence generation (selection bias) Low risk Quote: “The randomization of households was done using a computerized random number assignment algorithm and was stratified by drug shop, by the household's distance to the drug shop (in quartiles) and by the presence of children in the household.” Allocation concealment (selection bias) Low risk Comment: The same procedure as for random sequence generation (described above) was used; selection bias due to foreknowledge of treatment allocation is considered unlikely. Baseline outcome measurements Low risk Quote: “There are no significant differences across treatment groups, other than for the number of acres owned and the age distribution in the household. In particular, our control group has slightly older household heads, with, as a consequence, a significantly higher fraction of adults. Since age is highly correlated with malaria positivity, a lack of balance across treatment groups in the age composition of households could confound estimates of treatment assignment on uptake and targeting, even though the magnitude of the age differences is not large. Therefore, unless otherwise noted, we control for the age of the household head in all of our results. ”

a lack of balance across treatment groups in the age composition of households could confound estimates of treatment assignment on uptake and targeting, even though the magnitude of the age differences is not large. Therefore, unless otherwise noted, we control for the age of the household head in all of our results. ” Baseline characteristics Low risk Quote: “There are no significant differences across treatment groups, other than for the number of acres owned and the age distribution in the household. In particular, our control group has slightly older household heads, with, as a consequence, a significantly higher fraction of adults. Since age is highly correlated with malaria positivity, a lack of balance across treatment groups in the age composition of households could confound estimates of treatment assignment on uptake and targeting, even though the magnitude of the age differences is not large. Therefore, unless otherwise noted, we control for the age of the household head in all of our results.” Blinding of participants and personnel (performance bias) All outcomes Low risk Comment: Lack of blinding considered unlikely to impact on outcomes Blinding of outcome assessment (detection bias) All outcomes Low risk Comment: Lack of blinding considered unlikely to impact on outcomes Contamination Low risk Comment: Households were provided with vouchers for subsidised and non-subsidised ACTs; cross-group contamination is considered unlikely.

Blinding of participants and personnel (performance bias) All outcomes Low risk Comment: Lack of blinding considered unlikely to impact on outcomes Blinding of outcome assessment (detection bias) All outcomes Low risk Comment: Lack of blinding considered unlikely to impact on outcomes Contamination Low risk Comment: Households were provided with vouchers for subsidised and non-subsidised ACTs; cross-group contamination is considered unlikely. Incomplete outcome data (attrition bias) All outcomes Low risk Quote: “Only five percent of households surveyed at baseline were not reached at endline, and attrition was balanced across treatment arms.” Selective reporting (reporting bias) Low risk Comment: Data on pre-specified outcomes were reported. Recruitment bias Low risk Quote: “We selected four drug shops, in four rural market centers and sampled all households in the catchment area (within a 4km radius) of each of these shops.” “The randomization of households was done using a computerized random number assignment algorithm and was stratified by drug shop, by the household's distance to the drug shop (in quartiles) and by the presence of children in the household.” Loss of clusters Low risk Quote: “Only five percent of households surveyed at baseline were not reached at endline, and attrition was balanced across treatment arms.”

Recruitment bias Low risk Quote: “We selected four drug shops, in four rural market centers and sampled all households in the catchment area (within a 4km radius) of each of these shops.” “The randomization of households was done using a computerized random number assignment algorithm and was stratified by drug shop, by the household's distance to the drug shop (in quartiles) and by the presence of children in the household.” Loss of clusters Low risk Quote: “Only five percent of households surveyed at baseline were not reached at endline, and attrition was balanced across treatment arms.” Incorrect analysis Low risk Comment: Multivariable regression analysis allowing for clustering used (e.g. “If more than one household member got sick simultaneously, we include all concurrent first episodes, and therefore cluster the standard errors in all illness episode regressions at the household level.” Comparability with individually randomised trials Low risk Comment: Included clusters comparable ACT = artemisinin-based combination therapyAL = artemether-lumefantrineAQ = amodiaquineRDT = rapid diagnostic testSP = sulphadoxine-pyrimethamine Characteristics of excluded studies [ordered by year of study] Study Reason for exclusion Laxminarayan 2006 Not an RCT, NRCT, CBA or ITS: mathematical modelling study comparing impact of introduction of ACT subsidy with scenarios in which artemisinin monotherapy and partner drug monotherapy are used in a small proportion of patients in the absence of ACTs MENTOR 2010 Not an RCT, NRCT, CBA or ITS: uncontrolled pilot of ACT implementation in two municipalities

Characteristics of excluded studies [ordered by year of study] Study Reason for exclusion Laxminarayan 2006 Not an RCT, NRCT, CBA or ITS: mathematical modelling study comparing impact of introduction of ACT subsidy with scenarios in which artemisinin monotherapy and partner drug monotherapy are used in a small proportion of patients in the absence of ACTs MENTOR 2010 Not an RCT, NRCT, CBA or ITS: uncontrolled pilot of ACT implementation in two municipalities Alba 2010 Not an RCT, NRCT, CBA or ITS: annual census of drug shops, retail audits of public, mission and private outlets complemented with demographic surveillance system data Cohen 2010 Not an RCT, NRCT, CBA or ITS: pre/post survey examining equity and spatial distribution of outcomes in the delivery of subsidised private sector ACTs Yeung 2011 Not an RCT, NRCT, CBA or ITS: documentation of programmatic experience of implementation of subsidised ACT in the private sector Smith 2011 Not an RCT, NRCT, CBA or ITS: census of public and private facilities, chemists, pharmacies, other malaria medicine retailers Rutta 2011 Not an RCT, NRCT, CBA or ITS: pre/post program evaluation (approximately one year after the introduction of subsidised AL in accredited drug dispensing outlet) Tougher 2012 Not an RCT, NRCT, CBA or ITS: before-and-after design with no comparator sites Yadav 2012 Not an RCT, NRCT, CBA or ITS: periodic retail audits of accredited drug dispensing outlets to examine availability and price of subsidised ACT during the first year of the AMFm

Rutta 2011 Not an RCT, NRCT, CBA or ITS: pre/post program evaluation (approximately one year after the introduction of subsidised AL in accredited drug dispensing outlet) Tougher 2012 Not an RCT, NRCT, CBA or ITS: before-and-after design with no comparator sites Yadav 2012 Not an RCT, NRCT, CBA or ITS: periodic retail audits of accredited drug dispensing outlets to examine availability and price of subsidised ACT during the first year of the AMFm Yamey 2012 Not an RCT, NRCT, CBA or ITS: ‘mixed-methods’ design, triangulating data from a literature review with information from interviews with experts Davis 2013 Not an RCT, NRCT, CBA or ITS: interview study of ACT availability and use in the private sector of five AMFm phase 1 countries O Meara 2013 Not an RCT, NRCT, CBA or ITS: study examines factors associated with retailers' likelihood of stocking subsidised AL and the association between price and sales for AL, quinine and sulphadoxine-pyrimethamine Kedenge 2013 Not an RCT, NRCT, CBA or ITS: qualitative study (focus group discussions) to understand the impact of subsidising ACTs in the retail sector Malm 2013 Not an RCT, NRCT, CBA or ITS: document review of policies, guidelines, reports, meeting minutes, and Internet search of literature on implementation of AMFm Fink 2013 Not an RCT, NRCT, CBA or ITS: periodic household and retail outlet surveys

Kedenge 2013 Not an RCT, NRCT, CBA or ITS: qualitative study (focus group discussions) to understand the impact of subsidising ACTs in the retail sector Malm 2013 Not an RCT, NRCT, CBA or ITS: document review of policies, guidelines, reports, meeting minutes, and Internet search of literature on implementation of AMFm Fink 2013 Not an RCT, NRCT, CBA or ITS: periodic household and retail outlet surveys Tougher 2014 Not an RCT, NRCT, CBA or ITS: before-and-after design with no comparator sites ACT = artemisinin-based combination therapyAL = artemether-lumefantrineAMFm = Affordable Medicines Facility-malariaCBA = controlled before-afterITS = interrupted time seriesNRCT = non-randomised controlled trialRCT = randomised controlled trial DATA AND ANALYSES This review has no analyses. CONTRIBUTIONS OF AUTHORS NO, GY and PG designed the review methodology and undertook study identification and selection. NO extracted data. NO and PG assessed study quality and analysed data. NO prepared the first draft of the review. All authors participated in the interpretation of results and writing of the final manuscript.

IONS OF AUTHORS NO, GY and PG designed the review methodology and undertook study identification and selection. NO extracted data. NO and PG assessed study quality and analysed data. NO prepared the first draft of the review. All authors participated in the interpretation of results and writing of the final manuscript. DECLARATIONS OF INTEREST GY led the San Francisco ‘hub’ of the Evidence-to-Policy initiative (E2Pi; http://www.e2pi.org), an independent, non-profit policy think tank based at the University of California, San Francisco during the writing of this review. E2Pi is funded by a core grant from the Bill & Melinda Gates Foundation, which was one of the funders of the pilot phase (Phase 1) of the AMFm. E2Pi was also contracted and funded by the Global Fund to estimate benchmarks of success for the AMFm Phase 1 (the final paper is published at http://www.theglobalfund.org/documents/amfm/E2PI_EstimatingBenchmarksInAMFm_Report_en/). GY was on a fixed salary at UCSF. PG is a member of the WHO Malaria Treatment Guidelines Panel that recommended a global shift to ACTs in 2006. PG is corresponding author of an analysis published in the Lancet in 2004 that provided some of the evidence underpinning the WHO decision. NO has no known conflicts of interest. SOURCES OF SUPPORT Internal sources KEMRI-Wellcome Trust Research Programme, Kenya. External sources Effective Health Care Research Consortium, UK. Cochrane Effective and Practice and Organisation of Care Group, Norway. Evidence-to-Policy initiative (E2Pi), Global Health Group, University of California, USA.

NO has no known conflicts of interest. SOURCES OF SUPPORT Internal sources KEMRI-Wellcome Trust Research Programme, Kenya. External sources Effective Health Care Research Consortium, UK. Cochrane Effective and Practice and Organisation of Care Group, Norway. Evidence-to-Policy initiative (E2Pi), Global Health Group, University of California, USA. UKAid (Department for International Development), UK. DIFFERENCES BETWEEN PROTOCOL AND REVIEW We intended to include only randomised trials, non-randomised trials (with at least two intervention and two control sites), controlled before-after studies (with at least two intervention and two control sites) and interrupted-time-series studies (Appendix 1). We, however, made a post-hoc decision to include two non-randomised cluster trials (Sabot 2009; Talisuna 2012) that compared intervention sites to only one control site (we downgraded the certainty of evidence in Sabot 2009 and acknowledged possible confounding associated with these designs). We added adverse effects (unintended consequences of ACT subsidies) as a secondary outcome. We did not consider two pre-specified secondary outcomes: availability of alternative antimalarial drugs in all facilities, private and public (including informal outlets); and prices of alternative antimalarial drugs (full adult or child courses; we prioritised direct outcomes presented in the summary-of-findings table). A number of methods planned in the protocol were not implemented in the review. These methods could be relevant for future updates of this review and are summarised in Appendix 5.