In many malaria endemic regions, children acquire clinical immunity to malaria, and develop anti-parasitic mechanisms during the first few years of life consequent to repetitive exposures 1. Thus, as children grow up, they develop an aptitude to lessen the density of malaria infection and to tolerate malaria parasites in the absence of fever 2. It appears that the capability to tolerate malaria parasites increases with age, and is influenced by the intensity of malaria transmission. For this reason, clinical malaria is age-specific with regional and seasonal variations 34. Additionally, in such areas, observation of asymptomatic malaria parasites carriers in the population is common, and detection of malaria parasites in a blood smear, does not certainly indicate a clinical malaria episode. Although the development of symptoms and/or signs of clinical malaria is complex and multi-factorial 5, in the recent years, emphasis on the relationship between fever risk and parasite density has been largely used as an entry point to define a clinical malaria episode 6.

The concept of "pyrogenic threshold" has been proposed for defining malaria episode in endemic areas 789. The occurrence of fever cases due to other causes in the presence of parasitaemia, may well result in an over-diagnosis of clinical malaria. Thus, estimation of the proportion of fever cases attributable to malaria infection is crucial to establish a more concise definition of clinical malaria. Smith et al 6 proposed a methodology to estimate the attributable proportion of fever using a logistic regression of the fever on a monotonic function of the parasite density 10. In addition, the sensitivity and specificity of specific cut-off values of parasite density can be estimated.

In Mozambique, malaria remains one of the main causes of febrile illness among children. Transmission is perennial with regional variations throughout the country. The incidence of clinical malaria established through weekly active case detection suggests that the risk of clinical malaria is highest between the age of one and three years when children experience an average of more than two episodes per year. Based on a continuous demographic surveillance system and verbal autopsies carried out in Manhiça district, the risk of clinical malaria drops sharply after the age of six 11.

Early diagnosis and prompt treatment of cases, especially in children represent the keystone strategy aiming at reducing the burden of malaria-related morbidity and mortality. However, in most rural areas, diagnosis of clinical malaria is rather presumptive based on fever or history of fever, though a positive blood film is required to confirm the diagnosis in areas where laboratory facilities exist.

This paper reports the results of a national malaria survey, carried out in Mozambique, in which the prevalence and the intensity of malaria infections, the establishment of malaria case definition and its relation to age strata across the country were determined.

A detailed description of the methods used in this survey is presented in a companion paper 12. Briefly, a total of 8,816 children aged below ten years of age, living in households selected from 24 districts chosen randomly in the country, were eligible for the survey. Oral informed consent was obtained and a questionnaire was completed by well-trained team members. Thick and thin smears for parasitological examination and parasite density estimation were obtained during the survey. Axillary temperatures were also recorded using digital thermometers. Overall prevalence of Plasmodium falciparum infections, fever, and malaria parasite infection associated with fever were estimated in each age group for each region and stratum.

Using the method proposed by Smith et al 6, the attributable proportion of fever due to malaria using a logistic regression of the fever on a monotonic function of the parasite density was estimated. The model used was logit(π_i) = α + β(x_i)^τ, and represents a logistic regression model where π_I is the probability that observation i with parasite density xi is a fever case. β(x_i)^t corresponds to model type 3 described by Smith et al and is a monotonic function which is more flexible than the regression on x_i or log(x_i). The models were fitted using maximum likelihood. To constrain the parameter τ to be positive, the maximum likelihood were estimated for the log(τ). Confidence intervals for the estimated attributable fraction, the τ parameter, the sensitivity and specificity were obtained using the bootstrap methodology 13. Analysis were made using Stata Statistical Software (Stata Corporation, College Station, Tx USA).

In this survey, the proportion of fever cases attributable to malaria was 37.8% (95% CI 32.9% – 42.7%), representing the proportion of febrile morbidity that would have been removed if malaria infections were eliminated among children in various settings in the study area. On the other hand, the results of this study highlighted the importance of other fever episodes not attributed to malaria infections in the study area.

In many malaria endemic areas, the burden of disease, a definition based on fever or reported episode of fever within the previous 48 hours in the presence of any level of parasitaemia, would result in an over-diagnosis of malaria cases. This study was conducted during high malaria transmission season, nevertheless, among febrile children approximately one third did not have malaria parasites. This finding has significant implications on the treatment policy, particularly in rural areas where in the absence of laboratorial diagnosis, all febrile cases would be considered as clinical malaria episodes. Moreover, even among febrile parasitized children, the proportion attributable to malaria infection was less than 40%.

The association between malaria infection and body temperature varies significantly among children. Despite, that the definition of clinical malaria have been related to fever episode and presence of parasites in the blood stream, in endemic-malaria areas, manifestations of clinical malaria have a wide spectrum 14 and the parasite density required to trigger fever differs significantly from one individual to another 4. It was clear that all malaria cases had malaria parasites and fever, however, children with fever in the absence of malaria parasites are not necessarily non-malaria cases, since parasitaemia is a fluctuant variable and a child with malaria may have parasitaemia at undetectable levels at a given time.

In the study area, the majority of parasitized children were asymptomatic carriers, and not all fever episodes were associated with malaria parasites, hence very few fever episodes associated with asexual P. falciparum infections were observed. Additionally, the risk of fever among parasitized children was age-dependent, and increased with increasing mean parasite density.

In the study area, the overall proportion of fever cases attributable to the presence of malaria infection, when adjusted for age showed significant variations. Among younger children, it was 43.5% and decreased with increasing age to reach a low of 21.5% among older children. Therefore, the proportion of fever cases attributed to parasitaemia, the sensitivity and specificity of clinical malaria definition was age-specific. Consequently, among children exposed to malaria infection, the outcome or the risk of developing fever as a clinical manifestation of malaria infection, decreased with increasing age. These findings corroborate with results from other studies carried out in highly endemic areas 134.

Regional variations on the proportion of fever attributable to malaria infection among children have been reported from other community-based surveys in endemic-malaria areas 1, highlighting the changing patterns of the relationship between malaria parasites and the host 24. Based on the analysis of the age specific sensitivity and specificity confidence intervals for the proportion of fever attributable to malaria, the sensitivity and specificity for the cut-off-points definition for one parasite/μl and 2,500 parasites/μl were obtained for different age groups. It was crucial to determine age specific sensitivity and specificity, otherwise lack of specificity would result in a biased estimate of case definition and lack of sensitivity would result in a loss of power. However, to obtain accurate estimates of the efficacy of one intervention in clinical studies, diagnostic specificity is more important thus age-specific cut-off and region specific should be considered.

The case definitions derived from this survey, have shown the relationship between parasite density and fever and how this varies with age and region. This may help guide case definition for clinical trials of preventive tools, as well as provide definitions that may improve the precision of burden of disease assessment.

This study confirms that malaria infection remains a major cause of febrile illness during childhood. However, other causes of fever should be considered in case management of febrile illnesses during childhood.

The authors declare that they have no competing interests.

SM made a substantial contribution on conception and design of the study, coordinated and supervised data collection in all regions, interpretation of data, performed all statistical analysis and wrote the manuscript. JA gave a major contribution on data interpretation and statistical analysis and helped to draft and revised the manuscript. AT helped to draft, gave contribution and critically revised the manuscript. PA gave a major contribution on conception and study design and helped to draft and critically revised the manuscript. All authors read and approved the final manuscript.