Background
Falciparum malaria remains a public health priority and a major cause of morbidity and mortality in tropical areas [
1]. The observed pathophysiology of
Plasmodium falciparum malaria infection is strongly dependent upon endemicity, age and level of immunity [
2]. In malaria endemic areas, where transmission is perennial and stable, parasitic tolerance has been described and explained by a partial immunity acquired over years [
3,
4]. Age is, therefore, a major indicator of the duration of exposure to malaria parasite. However, acquired immunity is not completely protective but can be effective against clinical symptoms and severe form of the disease [
5].
Several interventions have been implemented during the two last decades in the fight against malaria. These interventions include treatment by artemisinin-based combination therapy (ACT), use of long-lasting insecticide-impregnated bed nets for exposed populations and use of rapid diagnostic tests for malaria diagnosis. These interventions led to considerable reduction in the number of clinical episodes and deaths due to malaria [
1]. Despite the significant progress achieved, the disease remains a major problem. The World Health Organization (WHO) reported 212 million clinical malaria cases and 429,000 deaths in 2015. Children under 5 years old are particularly susceptible to malaria illness, infection and death. In 2015, malaria killed an estimated 303,000 under-fives globally, including 292,000 in the African Region [
1].
Monitoring malaria transmission intensity and strengthening malaria control measures are necessary since incidence of severe disease and mortality increase with transmission intensity [
6‐
8]. The ongoing changes in malaria endemicity become a key determinant of the progress achieved. Moreover these changes determine the time required to reach the step where elimination could be foreseen [
9]. Standard measurements of malaria transmission based on entomological inoculation rate (EIR) and parasite prevalence are expensive, time-consuming and lack of precision because of micro-heterogeneity of malaria transmission [
10‐
12]. In addition, both EIR and parasite prevalence are affected by seasonality [
13,
14]. Furthermore, assessing malaria transmission intensity and evaluating the impact of interventions are complicated in areas where transmission has been substantially reduced. Therefore, alternative approaches are required to assess malaria transmission and evaluate intervention programmes.
The use of mathematical models and prevalence of anti-malarial antibodies constitute alternative approaches to evaluate malaria endemicity [
15,
16]. Mathematical models allow the determination of seroconversion rate (SCR) and seroreversion rate (SRR) which are, for a given time interval, the rates with which a seronegative subject becomes seropositive and a seropositive subject turns back to seronegative, respectively. Predictive seroprevalence (prevalence of antibody responses) can be calculated by using maximum likelihood methods. Seroprevalence reflects cumulative exposure and thus is less affected by seasonality because antibody responses can persist for years after infection. Serological markers have been previously used to assess malaria transmission intensity [
15‐
17], to detect recent changes in malaria endemicity [
17,
18], to evaluate effectiveness of malaria eradication efforts [
19,
20], and SCR has been shown to correlate with EIR [
15]. However, these single antigen-based serological approaches have significant limitations related to the sensitivity of the assay and the antigenic polymorphism that may affect detectable antibody responses [
21‐
23].
An alternative method could be the use of a combination of antigens rather than a single one since recent studies have shown that responses to crude parasite extract or multiple antigens can overcome [
17,
24,
25] these limitations.
Almost all reported mathematical approaches used the catalytic model without considering the effect of several factors (covariates). Taking these covariates into account could improve the effectiveness of serological models in assessing the intensity of malaria transmission [
15,
16,
18,
25‐
29].
Incorporation of covariates such as the number of clinical attack or the use of long-lasting insecticide-impregnated bed nets (LLIN) in the model, allows to take into account the heterogeneity of the population and the effect of interventions affecting malaria transmission.
Previous studies have estimated SCR with reversible catalytic model using cross-sectional surveys and high correlation was shown between serological measurements of transmission and EIR [
15,
16,
18]. However, given the fact that longitudinal surveys are often considered in sero-epidemiological studies and the increasing use of cross-sectional serological data to investigate malaria transmission, it would be important to assess consistency of cross-sectional results with SCR estimated from longitudinal survey in order to validate this approach. In this context, Arnold et al. [
30] investigated catalytic model in longitudinal cohort in order to validate this approach. The authors compared SCR using antibody responses against the merozoite surface protein
1–19 antigen measured in a prospective longitudinal cohort of children aged 0–11 years old to SCR calculated in a cross-sectional survey in a different group ranging from 0 to 90 years old. The comparison between the two approaches was done exclusively in children up to 11 years.
One of the objectives of the present study was to explore if the incorporation of some variations such as allowing age-varying to SCR or adding covariates in the reversible catalytic model could improve the precision or accuracy of exposure estimates using serological measurement of antibody responses to whole parasite extract. The availability of longitudinal data on a cohort whose age ranged from 0 to 90 years allowed the analysis of SCR and SRR values in different age group as studies have shown that antibody responses conversion and reversion varied with age and antigen [
31‐
34].
A second objective of this study was to investigate the utility of
P. falciparum crude schizonts extract which is a mixture of antigens, to monitor malaria transmission intensity and to detect temporal changes in malaria epidemiology in Dielmo, a Senegalese rural holoendemic area, using the age-specific reversible catalytic model described in previous studies [
15,
16]. In Dielmo village, various malaria control interventions have been implemented since 2000, leading to significant reduction in malaria transmission and incidence [
35]. Thirdly, the effects of potential covariates such as malaria control interventions and clinical malaria attacks on the SCR and SRR were investigated. Lastly, the use of cross-sectional data versus longitudinal data was also analysed and a comparison was performed between reversible catalytic model and alternative catalytic model in longitudinal data for assessing transmission evolution through SCR or SRR.
Discussion
In malaria pre-elimination context with reduced clinical episodes and transmission intensity due to implemented strategies against the disease, standard assessment tools lack of sensitivity and more sensitive tools are necessary to accurately evaluate both the level of exposure and the impact of malaria control interventions. These tools can also help to detect resurgence in transmission intensity. In recent years, several studies described the estimation of malaria transmission using model based on the seroprevalence of antibody responses against recombinant blood stage antigens, such as MSP1, MSP2, AMA-1, MSP-1_19, GLURP [
15‐
18,
27‐
30,
43,
44] and sporozoite stage antigens of
Plasmodium [
45]. These serological markers could be potentially sensitive, especially in areas of low transmission [
29] as antibody responses, in particular to blood stage antigens, have been shown to persist for several years after transmission has ceased [
46‐
49]. Thus, antibody responses against blood stage antigens could be detected even if transmission is lower. However, the use of specific single antigen might induce a lack of sensitivity and lead to a significant underestimation of the immune response to
P. falciparum infection. The variation in individual immune reactivity as well as the different duration of antibody responses against different antigens [
34,
50] and antigenic polymorphism may affect serological responses, thus the results obtained with specific recombinant antigens [
21‐
23,
51]. As an example, Ondigo et al. showed that antibody half-life may be short, intermediate or long depending on age and antigen and concluded for a need to associate antigen with similar half-life to improve precision and accuracy of malaria exposure estimates [
52]. Recently, Helb et al., in a study aimed at improving tools to reliably measure
P. falciparum exposure, observed that
P. falciparum-specific antibody responses differ by antigen and, in contrast with Ondigo et al., concluded to the need of selecting and combining different antigens with different kinetics for improving estimates of
P. falciparum exposure [
53].
These combinations have been demonstrated to improve the accuracy and the precision of malaria infection exposure. However the technologies used (multiplex, microarrays or cytometry) are quite expensive and the required equipment is not always available (or only in very few, high-level laboratories) in the countries where monitoring of malaria infection is needed. In a country with limited resources, there is probably a space for the use of less sophisticated tools, such as ELISA assays, and based on the results of this study, the use of crude extract could be a good choice. In fact, crude extract containing a plurality of antigens can overcome the constraints associated to the choice of specific antigens and give a better sensitivity, allowing detection of
P. falciparum exposure in areas of low-level transmission [
22,
24].
In this current study, a crude schizont extract of
PfSch07/03 was used in ELISA to examine the dynamic of antibody responses in individuals living in Dielmo (Senegal), an area where malaria transmission has substantially decreased [
35]. An estimation transmission model based on prevalence of antibody responses against
PfSch07/03 was applied to assess the applicability of this model in the study area.
Results showed a significant increase both in level and prevalence of antibody responses with age within surveys. These observations confirmed the general trend that anti-malaria immune responses varied with age in endemic areas [
38]. For all age groups, comparisons of both OD magnitude and seropositivity prevalence showed that antibody level and seroprevalence were generally significantly higher in 2000 and decreased over time to 2012. The decreased of antibody responses correlated with the decrease of transmission level induced by control interventions deployed in Dielmo [
35]. These results confirmed the high sensitivity of serological responses to detect changes in exposure to malaria infection [
15,
16].
Maximum likelihood method from age-specific seroprevalence reversible catalytic model was used to estimate seroconversion rate (SCR). SCR estimates were shown to correlate closely with EIR, the gold standard measurement of malaria transmission. The result was consistent with previous studies that have shown good correlation between EIR and SCR in low transmission areas [
15]. Importantly, heterogeneity of SCR by time-point survey corresponded to that of transmission in the study site. Indeed, the study locality has for long been a holoendemic malaria area in which transmission has varied from 142.5 (in 1990) to 482 (in 2000) infective bites/person/year [
35]. A recent study in this locality has shown that both transmission level and malaria attacks dropped since the introduction of LLINs in 2008 combined with ACT [
35]. The data showed the recent change about the level of transmission in Dielmo. Some cautiousness should be considered on the interpretation of the relationship between SCR and EIR. The variability of these measures should be taken into account in order to avoid misinterpretation. However, estimating the level of malaria transmission with the serological tool is relatively simple and could have an interest in the assessment of control interventions all the more so the antibodies response remain detectable for years.
SCRs obtained by combining all sectional data were statistically different from those obtained with time-point survey. In cross-sectional studies, SCRs gave precise information in the level of transmission and detected previous changes in malaria exposure. This difference could be explained by the heterogeneity in the transmission in Dielmo during these periods [
35] and by antibody responses variability [
38]. In contrast, the study of Arnold et al. [
30] found a similarity between the prospective estimate of the seroconversion rate from the longitudinal data and that estimated from the cross-sectional survey using data that covered the same period (cross-sectional sample in 2012). These contrasting results could be explained partly by the difference of samples in these two studies. Indeed the samples used by Arnold et al. [
30] included only children aged 0–11 years old. Another explanation could be the dramatic reduction in malaria transmission over the 12 years study period, while in the study by Arnold et al. [
30]. the transmission was constant over the study period.
The observation that SCR averaged over the entire period in the longitudinal analysis differed from the weighted average of the stratum-specific SCRs could be the consequence of the large change in SCR values over the study period. In view of the results in the present study. It seemed that the use for longitudinal data of catalytic models with constant SCR and SRR is not suitable in areas with rapid decline in malaria transmission. In these areas models allowed for time-varying SCR and SRR could be more adapted for longitudinal data in order to take into account the changes in SCRs following the reduction of malaria infection exposure.
Investigation of factors which might affect the SCR and SRR showed that both clinical episodes and the use of mosquito nets had significant impact on these parameters. In agreement with previous studies [
18] showing that the incidence of the disease has a significant influence on the estimation of SCR, clinical episodes were positively associated with the SCR of antibody responses to
PfSch07/03. Similarly, the use of mosquito nets reduced the risk of antibodies jumping from seronegative to seropositive state. This is consistent with previous studies that have shown significant decrease in the level of immunity in the locality after the introduction of LLINs [
38].
Results of LRT and Wald test were in favor of the alternative model suggesting that the previously described catalytic model [
15,
16] was restrictive both in cross-sectional and in longitudinal studies in Dielmo. However, this model gave some interesting results, among which, the correlation between corresponding EIRs obtained by serological data and observed EIR (entomological measurement). These results suggested the applicability of the catalytic model to serological data. However, as described by Yman et al. [
54] the use of catalytic model has some weakness such as the conversion of serological data to binary outcomes through a threshold model and the choice of this threshold. Furthermore, dichotomizing serological measures (seronegative
vs seropositive) may lead to a loss of information due to the high range of serological values of the seropositive group. Several authors proposed alternative models using continuous serological data, as the density model [
41,
42] or the antibody acquisition model [
54]. Interestingly Pothin et al. have shown that while seroconversion rate from the catalytic model and exposure rate from the density model measure different quantities, a high correlation has been observed between these two measures [
43]. This consolidates the use of the catalytic model in this study. However, in the antibody acquisition model, Yman et al. found that in settings of moderate and intense transmission, this model based on continuous data gave better precision and accuracy compared to catalytic models [
54], even if the authors agreed that this approach has some disadvantages in particular if the data are not log-normally distributed [
54]. Arnold et al. proposed the use of quantitative antibody levels as a valuable method to measure changes in transmission or difference in exposure for pathogens that elicit a transient antibody response or for monitoring populations with very high- or very low transmission [
55]. Indeed, in previous studies on the population of the present study, in parallel to seroprevalence estimations, quantitative antibody levels were measured in different age groups and compared between two very different periods of malaria transmission [
38]. The results showed that the levels of antibody responses added complementary and valuable information to seroprevalence values.
Comparisons of the different methods with the same data could be interesting to envisage.
Authors’ contributions
AKD and ATB designed the study. FD, BD, MMF performed the laboratory experiments. JF, ND, CS, JFT, RP, AT and ATB coordinated the fieldwork, sampling, archiving and withdrawal. ON, PSP, AKD and ATB performed the modelisation studies. ON, PSP, AKD, MN and ATB wrote the manuscript. All authors contributed to the revision of the manuscript, and have seen and approved the final version. All authors read and approved the final manuscript