Efforts to reduce the burden of malaria in Africa have intensified through the scaled delivery of effective tools for malaria prevention and disease management, facilitated by increased donor funding and a commitment by national governments and their partners [1]. There have been a number of reports of declining malaria morbidity and mortality, in some cases directly or plausibly linked to the scaling up of control efforts [2–13]. However, many of these reports are from areas with relatively low baseline malaria transmission intensity (South Africa; large parts of Zambia, Ethiopia, Eritrea, Rwanda), island communities (Zanzibar; Bioko, São Tomé and Príncipe), or represent areas with a long history of malaria research and surveillance systems (Kilifi, Kenya and The Gambia). It is striking how little data have been published from areas in Africa characterized by high transmission and less intense rapid scaling of intervention coverage.

In Kenya, following a nationally representative review of pediatric malaria admissions to hospitals between 1999 and 2009 we noted that the declining incidence of malaria admission was not a universal observation across the country [8] despite relatively consistent temporal increases in intervention coverage between sites [9]. Kenyan hospitals where declining malaria admission rates were far less notable, or showed evidence of increasing with time, were located in areas with higher malaria transmission intensity [9]. To extend the empirical observations of possible temporal disease burden changes in Africa across a wider range of transmission conditions and intervention coverage success, we present data assembled from five hospital sites in Uganda between 1999 and 2009 to improve our understanding of the proposed epidemiological transition in Africa.

Across all five hospital sites investigated, malaria admission rates have increased or remained unchanged since the period of prescaled intervention and funding (1999-2003) through a period when Uganda had received significant overseas donor support (2007-2009). Conversely non-malaria pediatric admissions remained relatively constant across all three observation periods 1999-2003, 2004-2006 and 2007-2009 (Table 2). Analysis of monthly time series of malaria admission rates using autoregressive models adjusting for external factors show that there was a significant increase in pediatric admissions from January 1999 to December 2009 (Figure 2; all linear fits P < 0.001) in four of the five sites. The net, proportional increases across all sites in malaria admissions from the beginning of 1999 to the end of 2009 ranged from 47% to over 350% depending on the hospital site. For example in Apac the proportion of malaria admissions was 54% in 1999 and increased to 70% by 2009 while in Jinja the proportion of malaria admissions increased from 49% to 73%.

We have adjusted for a number of factors that might influence a rising temporal hospital presentation of malaria including intrinsic population growth among the pediatric population's served by each hospital, between yearly seasonal variations in rainfall patterns and a proxy for overall hospital use through the adjustment for admissions not diagnosed as malaria. It seems therefore entirely plausible that malaria admissions have risen in four of the five sites included in the analysis. The site located in an area of lower transmission intensity experienced a large 'epidemic' period of malaria between 2005 and 2006 that distorted the linear time series analysis, however even at Kambugu the average malaria admission incidence between 1999-2003 was not significantly different from the period after large funding and implementation activities 2007-2009 (Table 2).

We have not been able to adjust for the proportion of true malaria diagnoses among children admitted to each hospital or changing diagnostic practices that may create a systematic bias in observed trends. This presents a perennial problem for all retrospective hospital reviews where parasitological diagnoses are not available for systematic review or data linkage. Nevertheless similar reviews of admission records, with similar caveats, have been used repeatedly to support claims of declining incidence of malaria admissions [7, 8, 10, 11, 53, 54] and therefore should be viewed with equivalent validity for increasing incidence. Changing practices and support to improve the diagnosis of malaria with time are likely to have resulted in fewer false positives and therefore one might expect this to support an observation of declining malaria diagnosed admissions.

Despite the incomplete nature of centrally reported hospital admission data, the national Health Information Systems (HMIS) of the Ugandan Ministry of Health suggest that the malaria case burden in public health facilities has increased from 3.5 million in 2000 to 10.7 million in 2004 and risen further by 2008 to over 12.2 million cases [46]. Although HMIS data are plagued by gaps and changes in reporting rates between and within years, these parallel data circumstantially support a rising, rather than declining, trend in malaria burden in Uganda. User fees in government health facilities were abolished in 2000, thus potentially creating a systematic bias in observed trends but the increasing trends remained significant after adjusting for non-malaria cases used here as a proxy for changes in hospital usage.

Uganda has been granted almost $267 million in overseas donor assistance since 2002 and this has translated into an average annual external commitment to malaria funding of approximately $2 per person at risk in Uganda [24], within $0.5 of requirements predicted to provide sufficient funding for a complete package of intervention delivery [55]. National coverage with essential interventions remains below international targets set 10 years ago in Abuja, Nigeria [56] and those set by national malaria strategies launched in 2001 and 2005 covering 5-year periods, respectively [18, 21]. By 2009 the proportion of children sleeping under an ITN was only 33% and less than a quarter of children with a febrile illness had access to the nationally recommended first-line treatment. However, there are important differences in intervention coverage between districts that are served by the five hospitals included in this study (Table 1). At Apac and Tororo, two focus districts for PMI support, ITN coverage among children exceeded 40% by 2009 while it was less than 20% in Mubende and Jinja districts. However it is important to consider all facets of malaria control and while ITN and IRS coverage has improved with time, the proportion of febrile children who access efficacious, recommended antimalarial drugs has declined. In 2000, access to CQ was equivalent to the access reported in 2009 to AL in three of the five study areas shown in Table 1. However over 40% of febrile children living in the area around Apac were treated with AL in 2009 while all other districts only managed to report less than 20% use of AL in 2009 [40]. Despite these differences and the combined efforts to provide IRS with PMI support in Apac, increasing intervention coverage has yet to translate into observed reductions in hospitalization. Uganda is still characterized by predominantly high malaria transmission intensity, as judged by the parasite prevalence recorded in children aged less than 5 years. By the end of the surveillance period transmission intensity remained high across Mubende, Jinja, Tororo and notably high at Apac but low in Kambugu (Table 1).

It is important to begin to define why large-scale donor support has not translated into significant intervention coverage. There were concerns about the transparent use of GFATM funding in Uganda that led to suspension of funding in August 2005 and led to the interruption in funding flows and planned activities [57]. Procurement and provision of effective medicines through responsive drug management supply chains has been the subject of international concern around tendering processes [35, 58, 59] and documented shortages of malaria medicines in the periphery [36, 38]. In the 1980s and 1990s many settings witnessed an increase in malaria cases as a result of increasing resistance and declining efficacy of antimalarial drugs. This status quo is perpetuated today where despite the existence of more efficacious drugs these are far from universally available resulting in an overall population effectiveness of effective treatment that could perpetuate a continued rise in disease incidence despite only moderate increases in ITN coverage. The preferred options for delivering ITN remained anchored in client payment systems for many years, albeit subsidized, but not free until late in 2007. Mass ITN distribution campaigns however have not raised coverage to effective levels anticipated to impact upon transmission [60]. The current level of ITN coverage is inadequate to significantly impact on disease incidence given the levels of high transmission intensity that characterize Uganda [60]. Additionally, six different national malaria control program managers have been appointed over the last 10 years and this has inevitably resulted in a loss of institutional memory and consistent stewardship. Despite these shortfalls in malaria control implementation since 2004 the Ugandan national malaria strategy for 2011-2015 has as its vision that 'malaria will no longer be the major cause of illness and death in Uganda and families will have universal access to malaria prevention as well as treatment by 2015' supported by a mission statement that states that the Ministry and its partners will 'reduce the level of malaria infection and consequent malaria death in Uganda by 75% by the year 2015, and to sustain that improved level of control to 2020'.

There is a general sense that all of Africa is witnessing an epidemiological transition that is a direct result of international donor support [61]. This position has been supported by recent reports of a declining malaria incidence across a number of sites in sub-Saharan Africa [2–8, 10–13, 53, 62, 63] and modeled predictions of the malaria-specific mortality effect size of increased intervention coverage [61, 64]. However, as noted in a recent review of the evidence, that still supported the view that scaled intervention has contributed to declining malaria incidence across Africa, there remains a publication bias in favor of 'good news stories' [65] from areas of traditionally low transmission intensity and rapid, concerted scaling of disease prevention and management strategies. The modeling exercises that use the Lives Saved Tool (LiST) [61, 64, 66, 67] make various assumptions about the impact size derived from randomized controlled trials [68] on a fraction of age-under-5 mortality presumed to be malaria from verbal autopsy studies [69]. The LiST model does not account for variations in either intervention effect size or malaria burden by malaria transmission intensity. The predicted impacts on under-5 mortality are also based on single intervention effect sizes, for example the expected numbers of deaths averted defined by the protective efficacy of ITN and the proportion of children sleeping under an ITN. What the models do not accommodate are the simultaneous effects of increasing ITN coverage and declining access to effective medicines when children are febrile. In areas of high parasite transmission these combined effects may cancel, or increase severe complications leading to hospitalization or death.

Given the retrospective nature of hospital admissions to define historical trends there are a number of important caveats. First, inadequacies in routine hospital information systems plague many countries in sub-Saharan Africa and it remains difficult to identify hospitals with complete records covering 10 complete years. We have attempted to locate hospitals in representative malaria ecological settings that did have reasonable recording systems since 1999 and were therefore purposive rather than random in our selection of hospital sites. However, any bias would have favored better performing hospitals more likely to disprove any notion of increasing malaria trends and there was no prior expectation that admissions would increase. Second, despite a rapid increase in donor financing of malaria control in Uganda since 2005 there has not been an equivalent investment in monitoring the impact on transmission through routine parasitological or entomological surveillance. Entomological inoculation rates were high in almost all the sites investigated during the late 1990s [15] and childhood parasite prevalence was high at the end of the surveillance period in 2009 (Table 1) offering only a suggestion, rather than an empirical description that there has been little change in transmission. Third, because of a lack of longitudinal, high resolution data on the coverage of ITN, IRS or use of effective medicines to treat fever any attempt to attribute admission impact to changing intervention coverage through a plausibility framework is weakened by inadequate spatial and temporally resolved data. Finally, intriguing observations, whether they be declining or increasing disease incidence raise a number of important questions which unfortunately we and many others are unable to address retrospectively including the relative impacts of first line drug effectiveness with time, the changing patterns of alternative uses of clinical care providers in the community, the precise changing boundaries of catchment populations to hospitals within an area, patterns of resistance and tolerance to insecticides by dominant vector populations and many other factors that contribute to epidemiological transitions.

The problems facing analysis of available data on malaria trends in Africa are generic and not specific to the study presented here. However there is a concern that a more realistic assessment of the public health impact of scaled malaria control is constrained by the selective nature of reporting only 'good news' stories by donor agencies. Despite the limitations of historical reviews a more systematic country-level meta-analysis of available data is required to provide a more comprehensive understanding of the epidemiological transition in Africa and notably from areas that have the highest transmission intensity. These reviews should begin to explore the wealth of national data on intervention coverage, mortality and financing but must be undertaken subnationally to reflect the diversity of malaria risks common to most countries. Linkage to combined intervention access is non-trivial but with more data in time and space a clearer, less simplistic model of attribution might be possible. Without detailed observational epidemiology claims of success and the platform to decide upon future investments in control and new disease-prevention tools will remain ill informed.

Our analysis of the changes in malaria admissions across Uganda show that the malaria transition witnessed in some parts of Africa is not true of all of Africa. Uganda is a long way from universal coverage of effective interventions, including ITN and access to effective treatments. In areas where transmission intensity is high, disease burdens remain high and most fevers will be associated with malaria infection and ensuring more than 80% coverage of interventions for prevention and prompt treatment is critical. In these areas modest increases in approaches to reduce infection risks or treatment of suspected cases we suggest will not impact on disease burdens to the same extent as proposed by the LiST models [61, 67] in areas of less intense transmission and malaria morbidity and mortality is likely to continue to rise.