Following the recent achievements in global malaria control [1], there is increased emphasis on monitoring these achievements and on refining the epidemiological landscape in order to determine intervention needs and guide implementation [2]. Household surveys, including Malaria Indicator Surveys (MISs) [3], Demographic Health Surveys [4] and Multiple Indicator Cluster Surveys [5] are commonly used to achieve these surveillance and monitoring goals, but they are expensive, time-consuming and technically complicated to undertake. A complementary, inexpensive framework for malaria surveillance may be provided by school malaria surveys [6], which were an important component of early, particularly colonial, malaria reconnaissance, and more recently have contributed towards a nationwide assessment of malaria in Kenya [7].

Building on the Kenyan experience, this paper presents results from the first, large-scale school survey of malaria in Ethiopia. Malaria transmission in Ethiopia is temporally and spatially dynamic [8], with transmission unstable, seasonal, and linked to environmental variables such as altitude and rainfall [9]. In recent years, there has been a marked scale-up of the distribution of long-lasting insecticidal nets (LLINs) and indoor residual spraying (IRS) in Ethiopia [10]. To track this progress and to capture the inherent heterogeneities of malaria transmission in the country, various community-based malaria surveys have been carried out at regional state and national levels [11–13]. The aim of the present work was to generate data for Oromia Regional State to assist in targeting malaria control interventions across this heterogeneous transmission setting.

This first application of school-based surveys to inform targeting of malaria interventions in Ethiopia has revealed a comparable prevalence of Plasmodium to that found in the 2007 national MIS survey [12], but lower than that reported in a 2006 survey conducted by the Carter Center (prevalence 4.1%) [11]. The study presented here found P. vivax prevalence to be comparable to that of P. falciparum, and highlighted the marked spatial heterogeneity in infection observed in Oromia and, indeed, Ethiopia. Furthermore, the results are consistent with other cross-sectional study findings [11], with malaria cases found above 2,000 m, i.e. the current NMCP boundary for classification of an area as malarious and determining inclusion in control activities such as IRS and LLIN distribution [10].

Risk factors for P. falciparum infection identified by crude univariate analysis include history of fever and anaemia. Only fever history was, however, associated with P. vivax infection. These findings indicate that fever in the previous month is predictive of malaria due to both species, but there remained a high proportion of identified infections that were asymptomatic. Cross-sectional surveys from a range of transmission settings have shown that a strong association exists between P. falciparum and reported fever [31]. In low transmission settings, it is often assumed that lack of acquired immunity in the population will cause all Plasmodium infections to elicit clinical symptoms, but the current findings dispute this. Since parasite density was not calculated in this study, it is not possible to determine if asymptomatic infections were due to very low parasite density. Asymptomatic infections will contribute to ongoing transmission in a community, but are unlikely to be detected or treated in a context where only individuals feeling unwell access diagnostic services. If Ethiopia is to achieve focal malaria elimination in areas of current low, unstable transmission, alternative strategies must be used to identify and treat asymptomatic infections and halt transmission. In São Tomé and Príncipe, for example, mass screening by means of cross-sectional country-wide surveys, wherein all residents were screened with a RDT and RDT-positive individuals were treated with an ACT, has contributed to recent dramatic reductions in malaria transmission [32].

Absence of age-dependency for infection in the current findings is consistent with lack of acquired immunity among individuals living in low malaria transmission settings [33], and findings from other surveys in Ethiopia [11]. Prevalence of anaemia was found to be higher in the present instance than in the 2005 national school health survey [34], but lower than in the 2005 Demographic Health Survey [4]. Increased odds of anaemia in males, however, was a common finding in the 2005 national school survey, and has been reported from other countries [35]. Males were also less likely to sleep under a LLIN, which may result in more frequent exposure to Plasmodium infection and resultant anaemia. Overall, these findings indicate that iron supplementation should be considered as a possible school health strategy, targeting boys and girls.

While documented scale-up in LLIN distribution and coverage in Ethiopia has been very successful [36], the present study shows that use of LLINs remains less than optimal among school-age children. This is consistent with other studies indicating that children of school-age are often the least likely to have access to mosquito nets owned by the household [37], as well as other data from Ethiopia indicating that net use does not directly correspond with net ownership [38]. While possession of LLINs in a household will exhibit some indirect protective effect for individuals not sleeping under the net, Ethiopia's policy of universal coverage with LLINs in malaria risk areas [39] must be fully implemented in order to fully contribute to transmission control. There is also a need for additional behaviour-change activities linked to LLIN distribution campaigns and the routine health extension programme, to ensure consistent use of LLINs [40, 41].

Somewhat surprisingly, LLIN use was associated with increased odds of malaria in crude univariate analysis. However, multivariate models did not find such association. Previous cross-sectional studies have found that net use is protective against malaria among school-aged children [42, 43], and other surveys found that a protective effect against malaria was linked with the number of nets per household [11]. IRS, as reported by children to have been conducted in their house, was found to be associated with increased risk of both P. falciparum and P. vivax infection in multivariate models. The lack of protective effect of IRS in these findings may be a result of near-universal resistance to DDT (1,1,1-dichloro-2,2-bis(p-chlorophenyl)ethylene) in Ethiopian anopheline mosquitoes (Reithinger et al., unpublished). The most likely explanation for this association between IRS and increased malaria risk is that the NMCP targets IRS to locations of known malaria endemicity; therefore living in a location where IRS is conducted is predictive of being in a malarious area.

The current surveys found a greater proportion of Plasmodium infections due to P. vivax than previously described in Ethiopia [11, 12], with P. falciparum and P. vivax in equal proportion overall but P. vivax dominating in the highland epidemic ecological stratum (90%). The variation in species distribution may be a result of increased use of artemisinin-based combination therapy and P. falciparum- detecting RDTs at peripheral health facilities, impacting on transmission of P. falciparum and changing the epidemiology of this parasite in Ethiopia. Alternatively, the findings may simply be due to the highly variable and unstable transmission setting, where increased P. vivax cases may be a result of focal epidemics in highland areas at the time of the survey, or a result of the tendency for P. vivax to cause long-term chronic infections and show less seasonality in transmission than P. falciparum[9]. The recent adoption of multi-species RDTs at health posts across Ethiopia will greatly improve the diagnosis and treatment of P. vivax infections. These infections are known to cause morbidity, including anaemia, malnutrition and respiratory distress [44, 45], but are likely to have been under-diagnosed in the past due to use of P. falciparum-detecting RDTs. Challenges remaining in control of P. vivax include examination of drug-efficacy and potential adjustment of national policy, in light of identified foci of chloroquine resistance [46–49], as well as strategies for diagnosing and clearing asymptomatic P. vivax infections. Furthermore, similar to other settings, it is likely that as prevalence of P. falciparum in Ethiopia is reduced by effective control interventions, the burden of malaria attributable to P. vivax will increase [50, 51].

It is the commonly held belief that in low transmission settings, a high proportion of children with malaria would be symptomatic and therefore absent from school. The present findings, however, indicate that although self-reported fever during the previous month is predictive of Plasmodium infection, only a minority of parasitaemic individuals identified in schools reported any fever on the day of the survey. Similar high proportions of asymptomatic Plasmodium infections have been found in other low transmission settings [52]. Attempts were made to ensure that all eligible children enrolled at each school were included in random sampling, but we expect a proportion of enrolled students were absent on the survey day. Provided that the underestimate of true parasite prevalence estimated from school surveys is consistent, this methodology can still be applied to collect valid epidemiological data from schools. Further investigation of the contribution of malaria to school absenteeism should be conducted to evaluate the population representativeness of parasite rates from school-based surveys.

The poor sensitivity of microscopy to detect low-density Plasmodium infections [53] may have affected the outcome of this study. The difficulties in correctly identifying low-density infection may have contributed to the discrepancy in microscopy results between standard and expert examination of blood films. Furthermore, these discrepancies indicate a need to implement a rigorous quality assurance system within the routine laboratory diagnostics system for malaria in Ethiopia, or alternatively, to expand the use of RDTs beyond community-level health care. Molecular techniques, such as polymerase chain reaction (PCR), have a lower detection threshold for Plasmodium than microscopy [52, 54], and may be a more sensitive diagnostic tool in a population where low-density infections are expected. Unfortunately, PCR remains suitable only in a research context, and not as a routine diagnostic tool for malaria.

The current study was unable to create a valid model, based on environmental covariates, to predict malaria endemicity across Oromia, because strong environmental predictors for location of transmission foci were lacking. Risk mapping using similar strategies had previously been successful in Afghanistan, with a comparable prevalence of infection (0.49%) [55]. This inability to develop a risk map based on environmental correlates only indicates that there are additional factors contributing to transmission that were not captured in modelled data. It may also be a result of the spatial and temporal variability of transmission, not adequately captured by a cross-sectional survey approach and microscopy diagnosis. Although it was not possible to determine the exact altitude at which infection was acquired, data indicate that most children live close to the school (median 30 minutes' walk). Therefore, it was assumed that children's homes, the site where infection is likely to have occurred, is at a similar altitude to the school and based on this assumption it was possible to identify two clusters of infection in Oromia, using a method that has successfully described hotspots of malaria at small spatial scale in Kenya and Sudan [56, 57].

Identification of all areas where malaria transmission is ongoing may be possible using an alternative diagnostic method where IgG antibodies to Plasmodium are detected using an enzyme-linked immunosorbent assay, reflecting exposure to infection over a longer time period. This method has been used successfully in other low and unstable transmission settings [53, 58, 59]. Alternatively, routinely reported malaria case data from health facilities has been used to model malaria transmission [60], but these data are subject to bias including incomplete recording and reporting, inconsistent quality of diagnostic services and variable access to health facilities across populations and localities. It may be possible to marry parasitological survey data and routine facility data to capture a reliable estimation of malaria transmission levels, and use these combined data to develop a risk map. This approach requires further investigation to ensure comparability between locations, and representativeness of the underlying population.

While the current cross-sectional surveys have provided data regarding the Plasmodium parasite rates among children attending school, there is a need to conduct a rigorous comparison to indicators determined from standard community surveys, such as MISs. This will determine if findings from school-based surveys are representative of all school-aged children in a community or, indeed, the whole community; if representative, school-based surveys could become an alternative survey method to the more costly and labour-intensive community surveys. While it is not expected that there are differences in risk of infection by age in Ethiopia, there is a need to further explore what proportion of school-absenteeism is due to malaria, as well as whether there are differences in malaria risk between enrolled and non-enrolled children. These findings will define the potential role of schools in malaria surveillance, monitoring and control in Ethiopia and other low transmission settings. Envisaged roles of schools in malaria surveillance could be to provide data on coverage of major interventions and parasite prevalence during routine school surveys, and to alert service providers of epidemics using information on school-absenteeism and from active case finding [6].

Results of cross-sectional school surveys in Oromia demonstrated marked spatial heterogeneity in malaria. Although several foci of infection were identified, large areas appear to be non-endemic for malaria. While these findings allow malaria control interventions to be targeted to identified endemic areas, this likely does not reflect the true extent of malaria in Oromia. Research is ongoing to further validate the use of school surveys in identifying transmission foci, as well as to investigate other potential uses of schools in malaria surveillance including monitoring and evaluation of control programme implementation.