Age-specific malaria seroprevalence rates: a cross-sectional analysis of malaria transmission in the Ouest and Sud-Est departments of Haiti

The samples analysed in this study were collected from four sites located in the Ouest and Sud-Est department of Haiti in the communes of Gressier and Jacmel, between February and May 2013. A map of Haiti including the enrollment locations, study communes, and departments is presented in Figure 1. Enrollment was based on convenience sampling from both clinical and non-clinical settings, as part of a larger study on host protective genetic factors [17]. Study sites included a rural community, two schools, and a clinic located in the Ouest and Sud-Est departments of Haiti, with participation open to all individuals attending each site. Healthy children were enrolled from the Christianville School in Gressier and from the Hossana Baptist School in Jacmel. Patients and healthy family members attending Portail Leogane Clinic in Jacmel were enrolled on a voluntary basis. Individuals from Chabin were enrolled from community members seeking general health services at a mobile clinic. Participants at each location were given opportunities to ask the enrolling physicians questions during information sessions prior to consent. After obtaining consent from participants or their guardian, local clinicians collected approximately 3 ml of blood by venipuncture into serum separation tubes from participants, which were centrifuged immediately at 6,000 RPM for two minutes after collection was complete. All serum samples were stored at -80°C in the University of Florida field laboratory in Gressier, until shipment to the Emerging Pathogens Institute, in Gainesville, FL, for analysis and storage. Serum samples were collected from a total of 823 participants between the ages of two and 80. Malaria infected participants (5/823), determined via rapid diagnostic test (RDT) were excluded from analysis to reduce the effect of positive individuals seeking treatment for current infections. Participants missing age data (3/823) were also excluded, resulting in a final sample size of 815 (483 females and 332 males). Of the 815 samples analysed, all were screened for previous exposure using the antigen AMA-1, but only 759/815 samples were screened using the MSP-1, due to limited amounts of serum from some participants.

Ethical approval to conduct this research was obtained from the Haitian-based Ethical Review Committee, the University of Florida Institutional Review Board, and the Office of Research Protections, United States Army Medical Research and Materials Command.

Serum samples were screened for antibodies against AMA-1 and MSP-1₁₉ using an indirect enzyme-linked immunosorbent assay (ELISA). Serum samples of subjects were diluted in 5% non-fat skim milk in phosphate buffered saline (NFSM-PBS). ELISA plates were coated in duplicate with the respective antigen diluted in 5% NFSM-PBS to a final concentration of 1 ml/ml for AMA-1 and 0.5 ml/ml for MSP-1, before overnight incubation at 4°C. The next day, antigen was removed and plates were washed five times with 0.05% tween-20 in PBS-K and then blocked for one hour with 5% NFSM-PBS to reduce non-selective binding. Following additional wash, diluted serum samples as well as positive and negative control sera were plated in duplicate and incubated for two hours at 4°C. Horseradish peroxidase conjugated rabbit anti-human IgG secondary antibody was diluted 1:1,000 in 5% NFSM-PBS and added to the plate. After one hour, plates were washed seven times and treated with 3,3’, 5,5’-tetramethylbenzidine (TMB) substrate solution in the dark for 20 minutes to allow sufficient colour development and stopped with 2 M sulfuric acid.

After measurement of the samples in duplicate, the average absorbance at 450 nm was used to determine the thresholds for the classification of a sample as seropositive or seronegative. Given the low incidence and relatively large sample size for each antigen (n = 815 or 759), the seronegative population responses for AMA-1 and MSP-1 should follow a normal distribution function with sample mean μ ^ and sample standard deviation σ ^. This behaviour is presented in Figure 2, where the suspected seronegative populations approximated normal distributions with μ ^ = 0.191 , 0.140 and σ ^ = 0.0427 , 0.04 for AMA-1 and MSP-1, respectively. Thresholds for positive AMA-1 and MSP-1 responses were classified by the addition of three, four, and five sample standard deviations to the sample mean, such that ≥ 99.7% of seronegative population members would not be classified as seropositive. The resulting thresholds using three, four, and five, sample deviations for AMA-1 were 0.319, 0.362, 0.405 and 0.260, 0.30, 0.340 for MSP-1. Responses above three standard deviations of the mean suspected average were considered seropositive, given the minimal impact using more conservative thresholds had on the estimated prevalence (see Discussion).

The observed cross-sectional seroprevalence was used to estimate the seroconversion rate using a method similar to those previously described [11, 18]. Briefly, an age specific seroconversion model was fit to the prevalence of AMA-1, MSP-1, and AMA-1 or MSP-1 seropositive population members using all participants (aged 2 to 80) and separately for participants less than 20 years of age to estimate the rate of seroconversion (λ). Participants were separated into “age classes” to depict aggregated seroprevalence, with wider age ranges used in older groups, due to the lower number of participants over the age of 20. Since the AMA-1 and MSP-1 responses are long-lasting and the estimation of reversion rates has been suggested to be unreliable with cross-sectional data, a reversion rate (ρ) of zero was used for the final analysis [13], however non zero reversion rates were also explored. Without seroreversion, the probability of infection (prevalence) at age x was modeled using the equation P(x) = [1 ‒ exp(-λ * x)]. When seroreversion was included in the model, the probability at age x was modeled using the equation P(x) = λ/(λ + ρ) [1 ‒ exp(-(λ + ρ)x)]. The functions were optimized (using R) to give estimated seroconversion/reversion rates λ ^ and ρ ^, as well as their standard errors for the calculation of 95% confidence intervals for λ ^ and ρ ^. Odds ratios for the probability of having a previous exposure, as determined by a positive ELISA response, were calculated using a simple logistic regression by age category.

Characteristics of the resulting study population can be found in Table 1.

Using the average absorbance from each serum sample and a threshold of three standard deviations from the sample mean to indicate the presence of AMA-1 or MSP-1 antibodies, 172 of 815 (21.1%) had the presence of AMA-1 antibodies, 179 of 759 (23.6%) had the presence of MSP-1 antibodies, and 247 of 815 (30.3%) had the presence of either AMA-1 or MSP-1 antibodies. Using thresholds from three to five standard deviations gave similar estimates of seroprevalence that ranged from 16.3% to 21.1% for AMA-1, 17.4% to 23.6% for MSP-1 and 24.0% to 30.3% for either AMA-1 or MSP-1. The seroprevalence by age group (ranging from 2 to 80 years) is presented in Table 2 and Figure 3.As expected, the prevalence of seropositive participants increases with participant age. The prevalence of previous infections determine by AMA-1 response was not significantly different (p value > 0.05) between age classes under 20 years of age. However compared to participants younger than 20 years of age, the likelihood of previous infection in those over 20 years of age was 6.0 times higher (95% CI OR 4.03, 8.99). For MSP-1 responses, differences in prevalence compared to the youngest age class were significant (p value = 0.023) beginning at the third age class (9 to 13 years of age), with those over 20 years of age having 3.7 times the likelihood of previous infection (O.R. – 3.7; 95% CI = 2.48, 5.53). Of participants who were tested with both AMA-1 and MSP-1 (n = 759), 104 were classified as seropositive using both antigens (13.7%), 75 were MSP-1 positive and AMA-1 negative (9.9%), 57 were AMA-1 positive and MSP-1 negative (7.5%), and 523 were negative for both AMA-1 and MSP-1 (68.9%). The distribution of positive ELISA responses for AMA-1 or MSP-1 antigens appears graphically in Figure 4, with regions I, II, III, and IV representing the average absorbance value for the response to the AMA-1 and MSP-1 antigens for samples with AMA-1(+)/MSP-1(+) results, AMA-1(-)/MSP-1 (+) results, AMA-1(+)/MSP-1(-) results, and AMA-1 (-)/MSP-1(-) results, respectively.

The prevalence and the estimated probability of previous infection as determined by AMA-1 and MSP-1 antigen response are presented in Figure 5 for the entire study population (2 to 80 years) and those less than 20 years of age. The estimated SCR for AMA-1 (top panel) from the entire study population was 0.016 (SCR – 0.016; 95% CI = 0.013, 0.018) and 0.014 (SCR – 0.014; 95% CI = 0.012, 0.017) when fit to data from participants 20 years or younger. For MSP-1 (middle panel), the estimated SCR from the entire study population was 0.018 (SCR – 0.018; 95% CI = 0.016, 0.021) and 0.019 (SCR – 0.019; 95% CI = 0.015, 0.022) for participants 20 years or younger. Using AMA-1 or MSP-1 to define seropositive participants (bottom panel), the estimated SCR from the entire study population was 0.025 (SCR – 0.025; 95% CI = 0.022, 0.028) and 0.026 (SCR – 0.026; 95% CI = 0.022, 0.302) for participants 20 years or younger. When a non-zero seroreversion rate (SRR) was used to model the seroprevalence data from all participants aged 2 to 80 years of age, the following conversion and reversion rates were derived: 0.014 (95% CI λ ^ 0.011, 0.017) and -0.006 (SRR– 0.006; 95% CI = -0.016, 0.003) for AMA-1 responses, 0.0204 (SCR – 0.204; 95% CI = 0.016, 0.025) and 0.008 (SRR – 0.008; 95% CI = -0.005, 0.022) for MSP-1 responses, and 0.0273 (SCR – 0.0273; 95% CI = 0.022, 0.032) and 0.007 (SRR – 0.007; 95% CI = -0.003, 0.018) for either an AMA-1 or MSP-1 response.

One of the primary limitations of this study was that serum samples were collected using a convenience sample, which limited our ability to infer how this sample population represents Haiti as a whole. Findings may have been skewed by potentially enrolling participants from clinics (n = 203), however, this potential sampling bias was adequately addressed by excluding all malaria RDT positive individuals (5/815) from final analysis. This study also only screened for previous exposure to P. falciparum, although the likelihood of finding other species or mixed infections remains low, given recent reports, and the presence of host protective factors [5, 20]. As with other ELISA protocols, setting a threshold for the classification of a sample as seropositive is subject to interpretation. To validate the method used in this study, thresholds using absorbance values of four and five standard deviations above the suspected seronegative population mean were also evaluated and fell within the calculated confidence intervals for seroprevalence and SCR using only three standard deviations.