This prospective and observational study was conducted from August 2000 to December 2008 in Libreville, the capital city of Gabon, where the climate is equatorial. In this urban area, malaria is predominantly caused by P. falciparum. The population of Libreville was estimated to be 537,540 inhabitants in 2006, with children under five years of age representing 16%.

The Malaria Clinical Research Unit (MCRU) is a branch of the Department of Parasitology of the Medicine Faculty, located in the Centre Hospitalier de Libreville (CHL), the largest public hospital in Gabon. At MCRU, all febrile paediatric outpatients and inpatients, aged less than 11 years old, were prospectively screened for malaria. In this unit, children routinely benefit for free malaria diagnosis based on microscopic examination. Temperature, history of fever, age, sex, and location were collected prospectively.

Throughout the study period, a thick blood film was done for each patient. Blood slides were processed according to the Lambaréné procedure, which is described elsewhere 17. Slides were considered negative if no asexual blood stages of parasites and gametocytes were seen in 100 oil-immersion fields. Quality control of the blood smears was performed: a second microscopist, blinded to the results of the first reading, read the slides. In case of disagreement, slides were controlled by a third microscopist and the mean of the two closest readings of parasitaemia was taken. The case definition of malaria was any febrile child with slide-confirmed parasites.

Monthly rainfall data of Libreville for the last decade were obtained from the meteorological services of the Agency for Air Navigation Safety in Africa and Madagascar (ASECNA).

Demographic and laboratory data were recorded and given to the data manager. Data were entered and cleaned using Epi-info version 3.3.2 (February 9, 2005 CDC Atlanta) and were analysed with STATA version 10 (Stata Corp, College station, USA). Medians with interquartile ranges (IQR) of age and geometric means of parasite densities (GMPD) are presented. Differences between groups were assessed using Pearson's chi-squared or Fisher's exact tests for proportions. For ordinal outcomes, the linear trend test (Cochran-Armitage test for trend) was used. The association between age and risk of malaria by year was assessed using odds ratio and the 95% confidence interval (OR (95%CI)). Linear regression was applied to test the significance of decreases for log-transformed data. To take into account the effect of auto-correlated observations and the season disturbances, an ARMAX model was built including covariates. We postulated a straightforward relationship between proportion of malaria cases and the monthly rainfall average measured by 20-point first-order moving average that was applied to filter seasonal variation and highlight the long-term movements in the data. Deviations in monthly rainfall were examined by comparing monthly values of the parameters (malaria cases, rainfall, time series data) obtained during the study period (9 years).

The significance level of all comparison was set at 0.05. All reported p-values are two-tailed.

During the nine-year study period, slides from 28,092 febrile children were examined at the MCRU. The number of children screened showed two peaks, in 2002 (n = 4867) and in 2008 (n = 4684). There was a decrease in the yearly number of consultations for fever from 2002 to 2005, followed by a significant increase until 2008 (p < 0.01, Cochran-Armitage test for trend) (Table 1).

The median (IQR) age of children consulting for fever was globally of 24(12-48) months. This did not significantly change from one year to another over the nine years (Table 1). Overall, children younger than five years represented 76% of the screened patients; this proportion was nearly the same over the surveillance period (Table 1). More than 55% of the febrile patients screened each year at MCRU were aged 0 to 35 months.

There was a linear decrease in the annual number of malaria cases, from 1,785 in 2002 to 359 in 2007 with a slight increase in 2008 (p < 0.01, Cochran-Armitage test for trend). In 2000, the proportion of febrile illness due to malaria was 45%; this proportion declined to 12% in 2007 and 15% in 2008.

The median (IQR) age of children with a positive blood smear significantly increased from 24(15-48) to 41(21-72) months over the study period (p < 0.01). In 2000, parasite prevalence was highest and the median age of patients with positive slides the lowest (Figure 1). The median age of patients with malaria negative slides did not fluctuate (24(12-48)) over the time. The lowest prevalence of malaria throughout the study period was found among the 0-23 months old patients (Figure 2). The proportion of malaria cases in the febrile children increased with age until 47 months (p < 0.01, Cochran-Armitage test for trend); from four years old, the proportion of malaria cases was similar in the different age groups (Figure 2).

The study population was grouped into those less than five years and the oldest (Table 1). In children younger than five years old, the prevalence of malaria decreased significantly in 2001 (26%), compared to the year 2000 (p < 0.01). A relative increase in malaria cases of 10% was observed between 2001 and 2002, followed by a drop of about two third in 2007 (10%) and 2008 (13%) (p < 0.01, Cochran-Armitage test for trend). In older children (5-10 years), the proportion of malaria cases was significantly higher compared to the youngest; the risk of having malaria was 1.2 to 2.0 in older patients than younger ones (Table 1).

According to the paediatric ward, data were compared between inpatients group (n = 15299) and outpatients group (n = 7502); origin was not known for 5291 children. Between 2000 and 2005, the proportion of malaria cases among children aged less than 24 months was always higher in the inpatients group (25-35%) compared to the outpatients (11-23%) (p < 0.01). From 2006, the same difference was only observed for patients aged 96 to 120 months old ((p < 0.02). During the study period, the median age did not significantly vary among the infected inpatients, ranging from 18 to 24 months old; whereas in the infected outpatients group, it ranged from 21 to 40 months (p < 0.01)

Parasitaemia ranged from 14/μL to 1,611,408/μL. The GMPD did not vary significantly over the study period, and ranged from 6,800 p/μL in 2000 to 7,512 p/μL in 2008. Depending on the age, the GMPD was slightly higher in children less than five years old (6751/μL to 9,779/μL) than in children aged 5-10 years old (3,593/μL to 6,951/μL). No difference was observed when comparing GMPD between inpatient and outpatient groups (results not shown).

Rainfall patterns were similar throughout the study period. June, July and August were the driest months with recorded rainfalls of 0 to 85 mm, except for the years 2000 and 2006 (Figure 3). The rest of the months were generally rainy with the wettest period from October to December (330-400 mm).

Monthly malaria prevalence data showed that there was a slight decrease in the proportion of positive slides from February to May, with a continuous increase between May and October (Figure 3). However, a comparison of the proportions of malaria cases during the rainy (mean prevalence obtained between September and May) and the dry season (mean prevalence obtained between June and August) did not show any difference over the study period. Indeed, analysis of time series auto-correlation on de-seasonalized data showed that there was an absence of correlation between means of monthly rainfall, with an autocorrelation coefficient of 0.48 (95%CI: -1.81-0.86) (standard error 0.68) (p = 0.48). Rainfall had no effect on the intercept of the trends in the proportion of malaria cases observed during the study period (Table 2).