Malaria remains the world's most important tropical parasitic disease, and one of the major public health challenges in the poorest countries of the world, particularly in sub-Saharan Africa. With the new move towards malaria eradication[1] and the scaling-up of malaria control interventions, there is a renewed energy and drive to maximize the impact of control tools in each epidemiological context. Where malaria transmission is seasonal, optimal timing of control becomes particularly important.

Knowing the duration, start and end of the malaria transmission season is important in terms of planning control strategies. For example, to maximize impact, Indoor Residual Spraying (IRS) should be carried out prior to the onset of the malaria season. Also, the cost-effectiveness of some malaria control strategies may be influenced by the intensity of seasonality. For example, in a setting where transmission only occurs during four months of the year, giving intermittent preventive treatment (IPT) to pregnant women (IPTp) or infants (IPTi) all year round would be less cost-effective than in a perennial transmission setting.

Most malaria endemic settings have "seasonal peaks" of malaria cases, which are usually described in terms of the duration and timing of the rains during a given study period. However, this may vary from year-to-year, giving a variety of subjective descriptions of seasonality for a single site. For example, in Manhica (Mozambique), malaria has been described as being perennial with "some" [2], "substantial"[3], and "marked"[4] seasonality, all of which are difficult to quantify. The challenge is how to translate this variation into a quantitative profile of malaria seasonality for a given area. This was highlighted by the need to undertake a pooled analysis of epidemiological data obtained from a systematic review, where it was necessary to define seasonality into distinct categories (Carneiro et al, submitted).

To date, several attempts have been made to describe the seasonality of malaria endemic areas such as the Climate and Malaria Resource Room (CMRR) [5] and Mapping Malaria Risk for Africa (MARA) collaboration [6]. The CMRR provides an interactive map that displays the number of months during the year when climatological conditions are considered to be suitable for malaria transmission. Although these maps provide a macro-scale illustration of where and when climatically-suitable conditions for malaria transmission exist, these continental maps do not account for micro-climatic conditions that can potentially have a significant role on actual transmission at a local scale[7]. The MARA collaboration, has developed maps that describe the expected duration and timing of the transmission season by country based on climate suitability models[8, 9]. Although an attempt was made to validate these seasonal maps against parasite prevalence data, the statistical correlation between parasite prevalence and the number of months at malaria risk was found to be poor[6]. Furthermore, several authors have reported that the use of parasite prevalence data is not ideal for describing malaria seasonality [10–12], because at very high transmission levels malaria prevalence does not vary by season [13].

More recently, a different approach to define seasonality was carried out by Mabaso and others who aimed to predict seasonality from environmental covariates. They defined a seasonality concentration index to model the relationship between environmental covariates and seasonality in malaria incidence[14] and EIR data [12]. The authors reported that sites tended to show stronger seasonality of clinical malaria in all-year round transmission settings than in areas with shorter duration of malaria transmission, but no investigation was made on variations between years to look at consistency of findings[14]. To develop robust definitions, several years of data from each place are needed as there are annual variations, both in rainy seasons and in the intensity and timing of peaks in malaria.

Given that the intensity and seasonality of malaria transmission varies widely within and between countries in sub-Saharan Africa, the choice and timing of malaria control interventions might be different across distinct epidemiological contexts. This paper aims to present a simple method for describing the seasonality of malaria disease in a given site, to assist policy makers in deciding when and where malaria interventions should be delivered.

In this paper, a standardized and practical approach for defining seasonality of malaria is presented based on a simple statistical analysis of the monthly distribution of malaria disease from a wide range of epidemiological settings across malaria endemic countries. A proposed definition of 'marked seasonality', where 75% or more of the episodes occurred in six or less months of the year, appeared to discriminate well between studies with 'marked seasonality' of cases and those with a more regular distribution of cases and perhaps one or two seasonal peaks. Furthermore, consistent results were observed when more than one outcome or more than one calendar year was available from the same site. With regards to monthly EIR data, however, a reliable assessment of seasonality could only be made in sites with high transmission (i.e. > 100 ibpppy), where monthly EIR data was greater than 0 for most consecutive months of the year. The 'concentrated period of malaria' in less intense transmission settings is difficult to interpret as it is not clear whether there is actually no transmission during those months of the year or whether monthly EIR measurements in these areas suffer from poor sensitivity (as monthly sporozoite rates tend to be very low or null yielding zero EIR values during times of the year when there may actually be malaria transmission[50]). Furthermore, for a given site, results from EIR data were not fully consistent with those of malaria morbidity outcomes suggesting that the use of monthly EIR data is limited when looking at seasonal variations of malaria transmission.

For example, Kilifi (Kenya) was found to show 'borderline seasonality' when monthly EIR data was analysed, but 'no marked seasonality' in the two sites reporting data on all hospital admissions with malaria. This discrepancy might also be due to the fact that Kilifi is usually reported as having two rainy seasons[17, 51–55]. The proposed definition of 'marked seasonality' might not be appropriate for sites with strongly bimodal seasonal patterns, as the highest proportion of cases may not occur in consecutive months. Difficulty in defining malaria seasonality in these areas has also been reported in previous attempts[12]. Further work is, therefore, needed to provide a quantitative description of malaria seasonality in these settings.

The range of months of the concentrated period of malaria for sites showing 'marked seasonality', were found to be concordant with both the rainy periods reported in the literature and first/last month of the MARA seasonality database indicating that our approach works well for identifying these aspects of seasonality in these study settings (see Additional file 1). Again, the few discrepancies found between the concentrated period of malaria and the rainy season corresponded to those sites reporting two peaks of rain, corroborating the poor fit of our method in identifying the concentrated period of malaria in these settings. A better fit was usually observed between the concentrated period of malaria and the rainy season in sites where the source of information on the rainy season was the paper that reported the monthly malaria data. This is likely to be due to year-to-year variation in the onset of the rainy season, resulting in a better fit when the onset of the rainy season is matched to the period of clinical surveillance. As can be seen in Additional file 1, the concentrated period of malaria tended to start one month after the start of the rainy season. As it usually takes a few weeks from the start of the rainy season for a sufficient increase in mosquito abundance to reflect a substantial increase of malaria cases[26, 56], it is therefore crucial to take this into consideration to maximize impact of malaria control interventions. For example, IRS interventions should be carried out at least one month prior to the expected onset of the identified concentrated period of malaria in a given area.

The proposed method for identifying the start and end of a concentrated period of malaria overcomes some of the limitations of using the low-spatial resolution seasonality MARA maps to identify the start and end of the malaria season at a local scale which usually results in errors particularly in areas where environmental conditions are most variable[57, 58].

A common definition of malaria seasonality is also crucial for the comparability of studies. The proposed method has already been applied to describe the age-pattern of malaria of differing severities across a range of transmission settings on a series of pooled analyses of existing data (Carneiro et al, submitted and Roca-Feltrer et al, manuscript in preparation). The use of a standardized definition of malaria seasonality in these comparative studies has improved our ability to discriminate studies with 'marked seasonality' from those with more regular transmission and seasonal peaks.

With the rapid scaling up of malaria control interventions in sub-Saharan African countries, continued surveillance is needed to monitor changes in transmission intensity levels and in the burden of malaria disease. Several authors have already reported a drop in hospital malaria admissions after analysing several years of surveillance data [50, 59–61]. Although there may be several reasons for the observed declines in malaria morbidity, authors have argued that decreases in transmission intensity and/or increases in malaria intervention coverage are likely to have played an important role in the changing pattern of malaria epidemiology in these settings. But does malaria seasonality change over time as transmission levels decrease in a given setting? In an area of marked seasonality in The Gambia, for example, the same seasonal peaks were observed every year between 2001 and 2007[50] suggesting that the proposed method for defining a concentrated period of malaria may remain useful even when transmission intensity decreases over time. Further work is needed however to assess whether changes in seasonality may occur with declining transmission intensity in areas of perennial transmission.

Although results from these analyses are encouraging for defining malaria seasonality, in practice there is little reliable monthly data on clinical and severe malaria in many parts of sub-Saharan Africa, questioning its utility on a wide scale. Health Management Information Systems (HMIS) -when working well-have been shown to provide consistent seasonal variation in the number of cases seen at outpatient clinics, coinciding with the pattern of malaria transmission [62]. Although HMIS are often unreliable, sentinel sites -with usually improved quality of data collection-can be used to further explore the applicability of this method. This could help to inform policy makers at a local level to optimize the timing and frequency of indoor residual spraying with insecticides or campaigns to promote the use of insecticide-treated bed nets, as is being done with epidemic early warning systems [63, 64]. Application of this approach to routinely collected local data could also be beneficial in targeting seasonally focussed malaria interventions, such as the seasonal administration of intermittent preventive treatment to children (IPTc) [65–67].

Currently, there is no standard way of defining malaria seasonality resulting in a wide range of definitions reported in the literature. Where there is considerable seasonality, optimal timing of control becomes particularly important. Here a simple approach for describing the seasonality of malaria is presented, that discriminates well between studies with 'marked seasonality' (75% or more of malaria episodes occurring in six or less months of the year) and those with less seasonality. This method also helps to identify the start and end of a 'concentrated period of malaria' at a local scale, which could inform policy makers about the optimum timing and frequency of deployment of malaria control interventions. A poor fit, however, was observed in sites with two seasonal peaks. Further work is needed to explore the applicability of this method, using routine health information system data.