Introducing rapid diagnostic tests for malaria to drug shops in Uganda: a cluster-randomized controlled trial

Abstract Objective To evaluate the impact – on diagnosis and treatment of malaria – of introducing rapid diagnostic tests to drug shops in eastern Uganda. Methods Overall, 2193 households in 79 study villages with at least one licensed drug shop were enrolled and monitored for 12 months. After 3 months of monitoring, drug shop vendors in 67 villages randomly selected for the intervention were offered training in the use of malaria rapid diagnostic tests and – if trained – offered access to such tests at a subsidized price. The remaining 12 study villages served as controls. A difference-in-differences regression model was used to estimate the impact of the intervention. Findings Vendors from 92 drug shops successfully completed training and 50 actively stocked and performed the rapid tests. Over 9 months, trained vendors did an average of 146 tests per shop. Households reported 22 697 episodes of febrile illness. The availability of rapid tests at local drug shops significantly increased the probability of any febrile illness being tested for malaria by 23.15% (P = 0.015) and being treated with an antimalarial drug by 8.84% (P = 0.056). The probability that artemisinin combination therapy was bought increased by a statistically insignificant 5.48% (P = 0.574). Conclusion In our study area, testing for malaria was increased by training drug shop vendors in the use of rapid tests and providing them access to such tests at a subsidized price. Additional interventions may be needed to achieve a higher coverage of testing and a higher rate of appropriate responses to test results.


Introduction
In areas where malaria is endemic, the appropriate management of febrile illness and the effective use of resources for malaria control rely on the availability and use of diagnostic tests. 1 In the absence of diagnostic tests, antimalarial drugs are often taken for illnesses that have similar symptoms to those of malaria. [2][3][4][5] Failure to diagnose malaria can lead to poor case management, a waste of scarce health resources and increased risk of antimalarial resistance. 1,6 The non-treatment or delayed treatment of malaria contribute substantially to malaria-attributable child mortality. 7,8 In Uganda, only a minority of febrile illnesses are treated with artemisinin combination therapy -i.e. the recommended first-line treatment for malaria -and many of such episodes go untreated. 9 Similar observations have been made in Kenya, the United Republic of Tanzania and other African countries. [10][11][12][13] The World Health Organization (WHO) recommends parasitological confirmation of malaria before antimalarial drug use. 14 Although the current Global malaria action plan of the Roll Back Malaria Initiative calls for universal access to malaria testing, 15 such access remains a distant goal in most countries with endemic malaria. A study in six African countries found that only 4-31% of children with febrile illnesses were tested for malaria. 9 In many countries, patients and caregivers rely heavily on a loosely regulated private sector for malaria treatment. 16,17 In consequence, the engagement of the private sector has become an increasingly common strategy in malaria control programmes -as reflected, for example, in the pilot Affordable Medicines Facility-malaria (AMFm). 18 The development of inexpensive and simple rapid diagnostic tests for malaria has opened the possibility of widespread access to malaria diagnosis. These antigen detection tests have been shown to be as effective as routine microscopy in malaria diagnosis 19 and can be safely performed by individuals with only basic training. 20 Although research from Cambodia, 21 Somalia 22 and Uganda 23 has shown that the distribution of rapid diagnostic tests by the private sector is feasible, we know very little of the impact of this approach on population-level rates of malaria diagnosis and purchase of antimalarial drugs. We therefore conducted a trial in eastern Uganda to investigate the impact -on malaria diagnosis and the purchase of antimalarial drugs -of training the vendors from licensed drug shops to test patients with a rapid diagnostic test for malaria. The trained vendors were also encouraged to buy the test, at a subsidized price, from local wholesale providers. The study took place in Uganda's eastern region, where the annual transmission rates for malaria exceed 100 infective bites per person 24 and presumptive symptom-based treatment remains commonespecially when, as commonly occurs, treatment is sought outside the higher level public-health facilities. [25][26][27] Malaria is responsible for 30-50% of outpatient visits and 9-14% of inpatient deaths in Uganda. 28 Objective To evaluate the impact -on diagnosis and treatment of malaria -of introducing rapid diagnostic tests to drug shops in eastern Uganda. Methods Overall, 2193 households in 79 study villages with at least one licensed drug shop were enrolled and monitored for 12 months. After 3 months of monitoring, drug shop vendors in 67 villages randomly selected for the intervention were offered training in the use of malaria rapid diagnostic tests and -if trained -offered access to such tests at a subsidized price. The remaining 12 study villages served as controls. A difference-in-differences regression model was used to estimate the impact of the intervention. Findings Vendors from 92 drug shops successfully completed training and 50 actively stocked and performed the rapid tests. Over 9 months, trained vendors did an average of 146 tests per shop. Households reported 22 697 episodes of febrile illness. The availability of rapid tests at local drug shops significantly increased the probability of any febrile illness being tested for malaria by 23.15% (P = 0.015) and being treated with an antimalarial drug by 8.84% (P = 0.056). The probability that artemisinin combination therapy was bought increased by a statistically insignificant 5.48% (P = 0.574). Conclusion In our study area, testing for malaria was increased by training drug shop vendors in the use of rapid tests and providing them access to such tests at a subsidized price. Additional interventions may be needed to achieve a higher coverage of testing and a higher rate of appropriate responses to test results.

Methods
The study was designed as a clusterrandomized controlled trial, with extensive monitoring of the health behaviour of households before and after a rapid diagnostic test was made available in local drug shops. Since the study was designed to explore both cross-sectional and pre-post differences, 67 (85%) of the study villages were randomly selected to receive the intervention while the remaining 12 (15%) were used as a control group. The study was implemented between March 2011 and April 2012. Training in the use of the diagnostic test occurred between 21 June and 6 July, 2011. A simple random number draw -generated by Stata/SE version 11.0 (StataCorp. LP, College Station, United States of America) -was used for the selection of study villages and households and the assignment of villages to the intervention or control arm.

Villages
The sampling frame for the study included all of the villages in the districts of Budaka, Bukedea, Kibuku, Kumi, Ngora and Pallisa that had at least one drug shop licensed and registered with the ministry of health. Although the district health office initially identified 92 such villages, only 79 still had at least one licensed drug shop at the time the main study was launched. These 79 villages were included in the cluster randomization ( Fig. 1).

Households
The households in all 92 target villages were listed before the launch of the study. Subsequently, 25 households in each target village were randomly selected, visited for a baseline survey to record basic demographic characteristics and health behaviours, and re-visited every month for 9 months (Fig. 1) to monitor their health problems and treatment seeking. During the tenth and final follow-up survey -i.e. the so-called endline survey -all consenting adults and -if their caregivers gave consent -the children in the study households were tested for malaria using the rapid diagnostic test.

Drug shops
Each licensed drug shop listed by the ministry of health in the 79 villages and included in the randomization was visited for a baseline survey in April 2011 and an endline survey in March 2012.

Intervention
Vendors from 108 drug shops in the 67 intervention villages were invited for 2 days of training in the use of a rapid diagnostic test for malaria -i.e. the CareStart Malaria HRP2 (Pf) test (Access Bio, Somerset, USA). The training, described previously, 23 was facilitated by trainers approved by the national ministry of health. These trainers reviewed the signs and symptoms of uncomplicated malaria and severe illness and instructed the trainees on how to perform the diagnostic test and on the recommended first-line treatment for malaria. The first day of training was classroom-based while the second day involved practical experience in a health facility. Trainees were not given specific instructions on when to recommend testing and -apart from the promotion of artemisinin combination therapy -were not given treatment algorithms. No retail price for the test was suggested. Our main objective was to observe whether and how the trainees chose to integrate the test into their normal practice. Upon successful completion of the training, each trainee was given a free box of 40 test kits, gloves, an instruction leaflet, results slips and a sharps disposal box. Any new staff employed by the drug shops in the intervention villages after the initial training were invited to a similar training course that was run in October 2011.
Trained drug shop vendors were visited monthly to track their stocking and usage of the rapid diagnostic test and compliance with the recommended protocols for testing. Test kit storage, administration and post-use disposal were monitored using a 17-point checklist. Information on the shop's stock of the test kits and price of the test for patients was recorded. At the same time, any questions the vendors had about the test's administration were answered. Every 3 months, four unused test kits were collected from each shop holding the kits and sent for lot testing at the Foundation for Innovative Diagnostics Laboratory at the Pasteur Institute of Cambodia, in Phnom Penh. Every kit investigated in this way passed lot testing.

Test kits
According to WHO, the test investigated has a panel detection score of 98.7%, a false-negative rate of < 1% and a total false-positive rate of 2.4%. 29 Because this test, like other tests based on the detection of histidine-rich protein II, can remain positive for up to 5 weeks after a cured infection, a higher false-positive rate may occur in settings where malaria

villages
Randomization is highly endemic. 30 Trainees were told that any patient who tested positive who had also taken antimalarial drugs in the previous 4 weeks should be referred to a facility where they could be checked for malarial infection by microscopy. The test kits were purchased and imported, at a cost of 0.70 United States dollar (US$) per kit, by the study team. They were then sold to a prominent pharmaceutical wholesaler in Kampala at a subsidized price of US$ 0.12 per kit. The wholesaler's regional pharmacy in the city of Mbale subsequently sold the kits -exclusively to our trainees -at a price of US$ 0.19 per kit. 23

Data entry and analysis
Data were entered using the CSPro 4.0 package (United States Census Bureau, Suitland, USA) and mainly analysed, using Stata version 11.0, in a multivariate linear probability difference-in-differences model. A P-value of 0.05 or less was considered statistically significant. The primary outcome was whether a household member with febrile illness was tested for malaria. Secondary outcomes included the medication taken to treat febrile illness -if any -and where treatment -if any -was sought. The main independent variable was whether the illness occurred in a village in our intervention or control arm. As vendors from 15% of the drug shops targeted for training did not complete their training, intention-to-treat effects were estimated.
To control for seasonal effects and for village characteristics, a full set of village and monthly fixed effects were included in the fully adjusted model. Estimated robust standard errors were clustered at the village level. 31,32 Although adjustments were also made to remove the potential effects of a behaviour change campaign that was rolled out in the later stages of our intervention, the roll-out of the campaign was orthogonal to the main treatment and did not affect our main results.
Our study was powered to detect an increase in the fraction of fever cases seeking health care at drug shops using the rapid diagnostic kit from 10% to 20%. Assuming an incidence of one episode of febrile illness per person-year, a mean household size of five individuals, 40% of households seeking treatment at a drug shop and an intra-class correlation of 0.05, the study was powered to detect the targeted 10% difference in the 9-month intervention period with a probability of 0.92.

Ethics
Ethical approval for this study was given by the Harvard School of Public Health (protocol # P19371-106) and the Uganda National Council for Science and Technology (protocol # HS805).

Trial registry
The trial was registered as clinical trial NCT01652365 at clinicaltrials.gov.

Drug shops
Out of the 108 registered drug shops, 92 shops completed training. Fourteen vendors declined to be trained, one shop was relocated outside our study area and the vendor from another shop was considered to have failed the training. Each of the 92 shops had at least one staff member who successfully completed training (Fig. 1).
Over the monitoring period from July 2011 to March 2012, shops run by successful trainees bought a mean of 146 (median: 40) diagnostic kits from the local wholesaler (Table 1). Overall, 13 440 test kits were bought and the shops investigated 10 412 patients using the test kits that they had been given or bought. However, 37 such shops (40%) did not purchase any of the test kits and most of the others bought only small numbers of the kits (Table 2). Together, just three shops accounted for 32% (3346) of all of the rapid diagnostic tests performed on patients.
The mean price paid by a patient assessed with the rapid diagnostic kit was US$ 0.43 -representing a 125% markup on the kit's wholesale price. According to the data collected in our household surveys, the median prices paid for artemisinin combination therapy for individuals younger and older than 5 years were US$ 0.57 and US$ 0.76, respectively. The corresponding costs of quinine -which accounted for 72% of purchases of antimalarial drugs other than artemisinin combination therapies -were US$ 0.57 and US$ 0.38, respectively. 12 Fig. 2 illustrates the households investigated and illnesses captured over the study period. In total, 25 758 episodes of illness were reported by study households across the 10 survey rounds. Respondents reported the presence of fever in 22 697 (88.1%) of these episodes and we focused on these episodes of febrile illness in our analysis. For 4364 of the reported episodes of illness -including 3908 episodes of febrile illness -treatment was sought at one of the study drug shops.

Households
The study households in the intervention and control villages appeared similar in terms of their baseline demographic characteristics ( Table 3). The estimated prevalence of malaria at the time of the endline survey was 43% in   Table 4 presents the characteristics of the episodes of febrile illness recorded pre-intervention, in monthly morbidity surveys. Most individuals (4448) with febrile illness sought treatment in the public sector, at a private-sector drug shop or pharmacy, or at a private-sector clinic or hospital. However, 21.5% (1218/5666) of such individuals sought no care. Roughly half (1039/2322) of the reported visits to drug shops for the treatment of febrile illness were to a shop that was involved in our study.
During the pre-intervention period, malaria testing and antimalarial drug use among cases of febrile illness were significantly less likely in the intervention villages than in the control villages -25.9% (1258/4861) versus 38% (306/805; P = 0.002) and 54.3% (2641/4861) versus 68.2% (549/805; P = 0.003), respectively. This difference appears to be a reflection of a relatively larger fraction of patients from the control villages seeking treatment at a public facility. In the pre-intervention period, the percentage of febrile patients seeking care in the private sector who were tested for malaria was similar across the two study arms (40% [54/135] versus 33.6% [235/700]; P = 0.565). Table 5 shows the population-level crude estimates and the corresponding -and, generally very similar -adjusted estimates of the intervention's impact on testing, medication choice and treatment seeking. Among all cases of febrile illness and among cases of febrile illness in children younger than 5 years, according to the crude model, the in-

Research
Malaria tests in drug shops, Uganda Jessica Cohen et al.
tervention significantly increased the probabilities of being tested for malaria, by 6.0 (P = 0.015) and 7.6 percentage points (P = 0.015), respectively. According to the same model, the intervention increased the likelihood of taking any antimalarial drug or artemisinin combination therapy by 4.8 and 1.6 percentage points and reduced antibiotic usage by 0.3 of a percentage point -but none of these differences reached statistical significance. Appendix A (available at: https://cdn1.sph.harvard.edu/wpcontent/uploads/sites/358/2012/08/ AppendixA.pdf ) demonstrates the robustness of these main results, which remained similar if we (i) used patientreported episodes of suspected malaria rather than all fever episodes, (ii) confined our analysis to the episodes of illnesses that occurred before the roll-out of the behaviour change campaign, or (iii) used logistic regression instead of linear probability models. Fig. 3 shows the fractions of patients visiting the drug shops in intervention and control villages who were tested for malaria before and after roll-out of the intervention. Prior to the first training course about the rapid diagnostic test, the fraction of patients tested for malaria in a study drug shop was similar in the intervention and control villages, 8.9% (70/786) and 10.6% (12/113), respectively. After the first training course, the fraction of patients tested for malaria in a study drug shop in the control arm remained almost unchanged (33/334) but the corresponding value in the intervention arm almost doubled (390/3011; P < 0.001). Nearly 90% (3112/3458) of patients investigated using the rapid diagnostic test gave a positive result. The reliability of the test results is discussed in Appendix A. Fig. 4 summarizes our data on the patients who, during the intervention period, visited and were tested at the study drug shops in the intervention villages. Although over 80% (285/342) of such patients who tested positive for malaria purchased an antimalarial drug of some kind, nearly 45% (21/48) of those who tested negative did the same. Test-positive patients were more than twice as likely to purchase artemisinin combination therapy as test-negative patients (40.9% [140/342] versus 16.7% [8/48]; P < 0.001). However, less than half of the test-positive patients -and 31.5% (590/1871) of the untested pa-

Medication purchased
ACT: artemisinin combination therapy. Notes: Medication is shown separately for the patients who were tested for malaria with a rapid diagnostic kit and found positive, the patients who were similarly tested and found negative and the patients who were not tested.
Malaria tests in drug shops, Uganda Jessica Cohen et al.
tients -purchased artemisinin combination therapy. We found no differences in antibiotic purchases according to testing status or test result.

Discussion
In countries where malaria is highly endemic, many episodes of febrile illnesses are treated at drug shops. Our results indicate that, by offering training and access to subsidized rapid diagnostic tests to private drug shops, it is possible to increase malaria testing rates significantly. Three of the most commonly voiced concerns regarding use of rapid tests for malaria diagnosis by the private sector are poor adherence to protocols, the potential crowding out of public-sector treatment and increased antibiotic purchases by patients who have a negative result. We found no evidence to support any of these concerns. In addition, as we previously reported, our monthly monitoring of the drug shops indicated generally high levels of compliance with recommended treatment, storage, waste management and test-kit administration protocols. 23 From a policy perspective, a major challenge for any health initiative in the private sector is the achievement of adequate uptake and coverage. In our study setting, the potential impact of the intervention was limited by the need to work only with drug shops that were licensed by the Ugandan Ministry of Health. Inclusion of unlicensed outlets would have substantially increased the reach of our programme but may also have complicated monitoring and quality control. Even among the licensed shops with trained staff, however, uptake was limited. Only about 60% (55/92) of such shops chose to stock the test kits and fewer than 20% (390/2261) of the febrile patients who visited a drug shop that had a trainee were actually tested for malaria. While no detailed information on the reasons for the relatively weak uptake of the rapid tests was collected, higher rates of uptake could probably be achieved by combining behavioural change efforts with stronger financial incentives. The price subsidies for the test kits could be increased, drug shops could be paid incentives to perform tests and provide appropriate treatment to the test-positive patients, and the level of any subsidy could be correlated with the quality of the testing provided by each shop. Greater impact could potentially also be achieved if the test-related training and subsidy could be combined with training on the appropriate treatment of non-malarial illnesses -although this would require policy-makers to support the movement of non-medical professionals further into formal case management.
A second main challenge, from a policy perspective, is the remarkably low uptake of artemisinin combination therapies that we observed. Only 40.9% of the patients who tested positive for malaria actually purchased such therapies. This modest amount of uptake does not appear to have been driven by lack of access, given that 84.2% (96/114) of our study drug shops reported having artemisinin combination therapies in stock at the time of our endline survey. 12 It is possible that patients perceived the prices of such therapies to be too high -compared with those of other antimalarial treatments -even though, in Uganda at the time of our study, the prices of such therapies had been substantially lowered by subsidies from the Affordable Medicines Facility-malaria programme. 12,18 Furthermore, use of rapid tests for malaria diagnosis in the private sector appears to be a feasible and potentially effective way to increase testing rates and improve overall case management. As discussed in greater detail elsewhere, 33 the subsidizing of such tests for use in the private sector is likely to yield highest returns in settings where malaria prevalence is low and treatment seeking in the private sector is common. The accurate diagnosis of malaria could eliminate the wasteful use of antimalarial drugs for non-malarial illness and improve the management of malaria and other febrile illnesses -particularly if the private sector is properly incentivized and equipped to treat the true causes of non-malarial illness appropriately. ■  Домохозяйства сообщали о 22 697 случаях лихорадочного заболевания. Наличие экспресс-тестов в местных аптеках значительно увеличило вероятность проверки любого лихорадочного заболевания на малярию на 23,15% (P = 0,015) и его лечения противомалярийными препаратами на 8,84% (P = 0,056). Вероятность приобретения комбинированных препаратов на основе артемизинина увеличилась на статистически незначительный показатель 5,48% (P = 0,574). Вывод В нашей области исследования частота тестирования на малярию была увеличена путем обучения фармацевтов аптек использованию экспресс-тестов и предоставления им доступа к подобным тестам по субсидированной цене. Для достижения более ширового охвата тестированием и более высокого процента принятия соответствующих мер по результатам тестов могут потребоваться дополнительные мероприятия.
Malaria tests in drug shops, Uganda Jessica Cohen et al.