Artemisinin-based combination therapies (ACTs) are the first-line drugs for malaria in most endemic countries, but there are concerns that targeting of ACTs to those in need remains poor. Many patients with malaria do not obtain ACTs, while many others with febrile illness obtain ACTs but do not have malaria parasitaemia (Whitty et al. 2008). Because microscopic testing of malaria has been limited in availability and is often of poor quality (McMorrow et al. 2008; Drakeley & Reyburn 2009; Kahama-Maro et al. 2011), it has been standard practice to diagnose malaria presumptively based on fever, resulting in over-treatment with antimalarials and under-treatment of non-malarial infections (Reyburn et al. 2004; D'Acremont et al. 2009). Such non-malarial febrile illnesses in a minority of cases are caused by bacterial infections but are most often due to self-limiting viral illnesses (Mtove et al. 2011; Leslie et al. 2012; McMorrow et al. 2012). Malaria rapid diagnostic tests (mRDTs) are thought to be an important tool to improve malaria diagnosis and targeting of ACTs (Murray et al. 2008; Drakeley & Reyburn 2009). Parasitological confirmation of malaria prior to treatment is now recommended for patients of all ages by World Health Organization (2010), with mRDTs an important part of the ‘T3: Test, Treat, Track’ initiative to ensure that every suspected malaria case is tested and every confirmed case is treated and tracked in a surveillance system (World Health Organization 2012). Decreases in the proportion of febrile illnesses associated with malaria in many settings (D'Acremont et al. 2010) have led commentators to stress the urgency of expanding mRDT coverage (D'Acremont et al. 2009), although important challenges to effective implementation have been highlighted (McMorrow et al. 2008; English et al. 2009).

Malaria rapid diagnostic test implementation has had a varied impact on clinical decisions. While some studies have demonstrated significant reductions in the proportion of patients obtaining an antimalarial drug (Williams et al. 2008; Msellem et al. 2009; Thiam et al. 2011; Yukich et al. 2012) or reductions in over-treatment of malaria (Kyabayinze et al. 2010; Bastiaens et al. 2011; D'Acremont et al. 2011; Masanja et al. 2012), others have reported under-utilisation of diagnostic tests (Hamer et al. 2007; Nyandigisi et al. 2011) and frequent antimalarial treatment of patients with negative test results (Hamer et al. 2007; Reyburn et al. 2007; Bisoffi et al. 2009; Skarbinski et al. 2009; Nyandigisi et al. 2011). This variability has been documented across studies within the same country. For example, in Tanzania, D'Acremont et al. (2011) demonstrated a reduction in over-treatment with antimalarials, while Reyburn et al. (2007) did not, and McMorrow et al. (2008) reported significant challenges in both microscopy and mRDT use under routine conditions. Another key concern is that increased mRDT use will lead to greater and often inappropriate use of antibiotics as clinicians are faced with unclear treatment decisions for mRDT-negative patients, with potentially adverse consequences for antibiotic resistance (Baiden et al. 2012).

Artemether–lumefantrine (ALu) has been the first-line drug for treatment of malaria in Tanzania since 2004, although roll-out to the public sector did not begin until 2006. Treatment at government hospitals, health centres and dispensaries is intended to be free of charge for children below 5 years old and pregnant women, although this policy is not always adhered to (Njau et al. 2006). In 2006, the National Malaria Control Programme revised policies favouring presumptive diagnosis of malaria to treatment based on parasitological confirmation for patients aged 5 years and above, with treatment based on clinical diagnosis permitted for children under age 5 years (United Republic of Tanzania Ministry of Health & Social Welfare 2006), although implementation remained limited in practice. In 2010, the policy was amended to include parasitological confirmation of suspected malaria cases for all ages (United Republic of Tanzania Ministry of Health & Social Welfare 2011), corresponding with the phased roll-out of mRDTs to all levels of government health facilities from 2009 to 2012.

While mRDT implementation has been studied in many settings, very few studies have reported evaluations of mRDT roll-out under routine operational conditions, where implementation has been purely the responsibility of the government (Bastiaens et al. 2011; Thiam et al. 2011; Masanja et al. 2012; Yukich et al. 2012). Two studies from Senegal and Zambia used routinely collected data to measure the impact on antimalarial consumption but did not assess the appropriateness of case management (Thiam et al. 2011; Yukich et al. 2012), while two studies in Tanzania examined very early results of mRDT roll-out in limited settings [two hospitals (Bastiaens et al. 2011) and one demographic surveillance site (Masanja et al. 2012)]. More commonly, study teams in these studies had participated in implementation to some degree, potentially influencing the findings and limiting their generalisability.

In this article, we report the impact on case management practices of routine government implementation of provision and use of mRDTs in Tanzania, based on large-scale health facility surveys conducted before and after mRDT roll-out. The surveys were conducted in three of mainland Tanzania's 21 regions, reflecting the country's diversity of malaria epidemiology and commodity availability, thereby allowing us to explore the impact of policy implementation under a range of contexts.

We visited 140 health facilities (11 hospitals, 24 health centres and 105 dispensaries) between May and October 2010, and 176 health facilities (13 hospitals, 31 health centres, 132 dispensaries) between April and July 2012.

We have presented results from large-scale health facility surveys in three regions of Tanzania before and after the roll-out of mRDTs in all levels of government health facilities. We have reported on the availability and use of diagnostic testing among febrile patients and case management with ACTs based on facility test results and reference blood smears, and we have also examined the effect of this policy change on prescription of antibiotics. The findings are representative of Mwanza, Mbeya and Mtwara Regions, which encompass considerable diversity in malaria transmission, economic status and culture.

Malaria rapid diagnostic test roll-out in Tanzania led to substantial changes in the provision of malaria case management in public facilities. Post-roll-out nearly 70% of facilities had mRDTs in stock and 60% of health workers had received formal mRDT training. This was associated with a large increase in the proportion of febrile outpatients tested for malaria from 15.5% in 2010 to 54.9% in 2012, and facility mRDT testing was shown to have relatively good sensitivity and specificity. However, there were still major gaps in diagnostic coverage, with almost half of all patients not tested. In Mtwara, the proportion of patients tested (71.5%) and the proportion of health workers with formal mRDT training (79.0%) were higher than in Mwanza and Mbeya, reflecting the more recent roll-out, and therefore, lower potential for training coverage to be eroded by staff turnover or for mRDT stock-outs to arise. The latter was an important factor in suboptimal diagnostic coverage, with the per cent tested increasing to 70% if only facilities with diagnostics available were considered. Other studies have similarly reported less-than-optimal coverage of diagnostic tests (Hamer et al. 2007; Nyandigisi et al. 2011; Masanja et al. 2012). In addition to poor diagnostic availability (Ezeoke et al. 2012), reasons for poor coverage may include inadequate health worker training, negative health worker perceptions of mRDTs, provider workload or patient preferences (Chandler et al. 2008a; Williams et al. 2008; Asiimwe et al. 2012; Baiden et al. 2012). Health centre and dispensary patients paid a flat rate for treatment, so patient willingness-to-pay for tests is not expected to have been a factor in these facilities, but may have contributed in hospital settings where separate fees were sometimes charged for diagnostics.

Provider compliance with test results in terms of prescription has been a major concern in the roll-out of improved services for parasitological confirmation. Reported reasons for low compliance include providers' or patients' lack of confidence in the test result, the presence of symptoms associated with malaria, and the lack of alternative diagnoses identified by the providers or accepted by the patients (Chandler et al. 2008ab, 2010; Asiimwe et al. 2012; Ezeoke et al. 2012). While some studies have reported poor compliance with negative test results (Hamer et al. 2007; Reyburn et al. 2007; Bisoffi et al. 2009; Nyandigisi et al. 2011). our findings correspond with other studies that have demonstrated a substantial reduction in antimalarial treatment after implementation of mRDTs (Williams et al. 2008; Msellem et al. 2009; Kyabayinze et al. 2010; Bastiaens et al. 2011; D'Acremont et al. 2011; Thiam et al. 2011; Masanja et al. 2012; Yukich et al. 2012). Our results show that treatment of patients with a negative facility test in 2012 was less than one in ten, even in facilities with ALu in stock. The proportion of patients negative by reference blood smear who did not receive any antimalarial increased significantly from 57.8% in 2010 to 82.3% in 2012.

The reduction in ACT provision for patients who were not parasitemic led to a significant increase in the composite indicator of patients appropriately treated of 18 percentage points including all facilities and of 29 percentage points considering only those facilities with ALu in stock. However, while over-prescription of ACT was substantially reduced, under-prescription remained a major problem. Only 56.5% of patients with a positive facility test obtained an ACT in 2012, with particularly poor results in Mwanza, where only 18.2% of positive patients obtained ACT. There was weak evidence of a decrease in the proportion of patients positive by reference blood smear who obtained ACT between 2010 and 2012 overall, and a significant fall in Mwanza, although the percentages of patients testing positive overall were low (9.2% in 2010 and 16.5% in 2012) resulting in small subsamples and large confidence intervals. Poor ACT availability was a major factor in the under-treatment of patients testing positive. ACT stock-outs were present in all regions but especially severe in Mwanza, with around 40% of facilities experiencing complete ACT stock-outs at the time of both surveys. Even at facilities where ALu was in stock, nearly 20% of patients overall with a positive facility test result did not obtain ACTs. Fear of future stock-outs prompting drug rationing or the absence of the appropriate weight-specific blister packs may have discouraged health workers from dispensing ACT to some patients testing positive by the facility test (Wasunna et al. 2008). Again at health centres and dispensaries, this is unlikely to be due to patients not wanting to pay for ACTs as flat fees for consultations were generally charged, but willingness-to-pay could have been a factor in hospitals which sometimes charged separately for drugs. These results are in contrast to other studies that found more than 90% compliance with positive test results (Hamer et al. 2007; McMorrow et al. 2008; Williams et al. 2008; Bisoffi et al. 2009; Msellem et al. 2009; Kyabayinze et al. 2010; D'Acremont et al. 2011), perhaps reflecting the greater risk of ACT stock-outs and more limited health worker training coverage and supervision under the routine, operational conditions we evaluated. Another study evaluating early effects of mRDT roll-out in one Tanzanian district, found that in facilities with ACT in stock, 20% of patients testing positive did not receive ACT, similar to results we present here (Masanja et al. 2012).

While the proportion of patients who obtained an ACT significantly decreased from 2010 to 2012, the proportion of patients receiving antibiotics significantly increased to just under half of all patients. An increase in antibiotic prescription after mRDT introduction was also reported by two other studies in mainland Tanzania (Bastiaens et al. 2011; D'Acremont et al. 2011) and one in Zanzibar (Msellem et al. 2009). No specific guidelines were given to health workers on when to prescribe antibiotics as part of mRDT training, but this increase might be expected, given that health workers may be more likely to consider a bacterial diagnosis following negative malaria test results. However, we also observed a significant increase from 31.3% to 47.9% in antibiotic provision among patients not tested between 2010 and 2012. The reason for this is unclear, as antibiotic availability was similarly high during both surveys, and data were not collected on symptoms warranting antibiotic prescription. While data on the aetiology of non-malarial fevers are still limited, a study of children with non-severe febrile illness in northern Tanzania found that while a bacterial pathogen was identified from blood culture in only 0.9% of children, the WHO criteria for pneumonia were met in 48% (Mtove et al. 2011). The latter is considered an indication for antibiotics although pneumonia may also be caused by a virus. In Pakistan, a trial assessed withholding of antibiotics among children with fast breathing and did not find a difference in clinical outcomes, suggesting that bacterial pneumonia may be over-diagnosed (Hazir et al. 2011). To avoid substituting overtreatment with antimalarials with overtreatment with antibiotics, there is an urgent need for better diagnostics of non-malarial fevers and improved training on case management when patients test negative or when diagnostic tests are not available.

This study has several limitations. The sampling probability of health facilities was based on the most recent malaria outpatient data available, which may have suffered from some inaccuracies. Health workers may have been more likely to perform diagnostic tests and less likely to dispense antimalarials to negative patients as a result of the team's presence. To reduce this possible source of bias, the study team emphasised to health workers the importance of following their normal procedures. Community members may also have been drawn to the facility because of the presence of the study team in hopes of obtaining drugs for both sick household members and as a reserve for future periods of drug stock-outs. Research assistants screened all patients arriving at the facility carefully, but it is possible that some patients may not have had a true illness.

The 2012 survey took place slightly earlier in the year than the 2010 survey. The peaks in malaria incidence in Tanzania usually occur just after the short rainy season in November–December and just after the long rainy season in March–May, which may have meant that in 2010, Mtwara was visited after the peak, and in 2012, Mwanza was visited before it fully developed. However, taking into account variable weather patterns, late rains in 2010, and lack of entomological data, it is not clear that this would have affected parasitaemia prevalence.

Finally, care should be taken in interpreting facility-level data in isolation, as improvements seen at government health facilities may not necessarily translate to an increase in coverage at the community level. IMPACT2 household surveys in the same three regions in 2010 and 2012 similarly showed that among patients visiting public health facilities for fever, there was a significant increase in the proportion receiving a diagnostic test (from 28.7% to 46.6%), and a significant decrease in the proportion obtaining ACTs (from 57.4% to 46.1%) (Thomson et al., in draft). However, the household surveys showed no significant change overall in the proportion of patients reporting fever who recalled having a diagnostic test pre- and post-mRDT roll-out (Thomson et al., in draft). This appears to have reflected a reduction in the use of public facilities from 25.3% to 16.8% of reported fevers; as diagnostic coverage was much lower in the private sector, no overall increase in parasitological confirmation at the community level was observed despite the mRDT roll-out in the public sector.

Tanzania has made major strides in scaling up access to diagnostic testing for malaria within public health facilities and reducing overuse of ACTs by patients without parasitaemia. However, this study has also demonstrated the dramatically negative role that ACT and mRDT stock-outs continue to play in malaria case management. Several initiatives are underway in Tanzania to improve public sector commodity supply, including text-based stock-out reporting systems (Barrington et al. 2010), strengthening of zonal distribution capacity and direct delivery to health facilities, but this continues to be a major challenge. Maintaining buffer supplies of drugs and preventing leakage to the private sector may also potentially reduce stock-outs, as well as ensuring that mRDTs are in stock and appropriately utilised.

Priorities for further research include investigating reasons for under-prescription of ACTs for patients testing positive, and high levels of antibiotic prescription, and the potential effect on antibiotic resistance. As countries increasingly implement policies of universal testing prior to treatment for malaria, it will be important to address these concerns in order to improve case management for febrile illnesses and appropriately target ACTs.