Background
Malaria surveillance has been receiving increasing attention, with the World Health Organization (WHO) recognizing surveillance as a core intervention in its Global Technical Strategy for Malaria 2016–2030 [
1]. Currently, different approaches are used to assess morbidity, mortality, and transmission of malaria and monitor trends over time in endemic countries. Two major sources of malaria data are health management information systems (HMIS) and nationally representative household surveys such as the Demographic and Health Surveys (DHS) and Malaria Indicator Surveys (MIS). However, while case-based surveillance data reported through the HMIS only reflect clinical cases accessing formal health facilities, household surveys are too expensive to provide real-time data for continuous malaria surveillance. Several countries are therefore implementing additional data collection strategies, such as sentinel surveillance sites or school surveys [
2‐
5]. To complement these efforts, pregnant women attending antenatal care (ANC) services have been proposed as a pragmatic sentinel population for the surveillance of community-level malaria prevalence [
6]. Pregnant women data can be routinely collected relying on existing infrastructure and mechanisms, thus, providing a cost-effective and real-time estimate of malaria prevalence.
The relationship between the malaria prevalence in pregnant women and children aged 0–59 months has previously been investigated in a systematic review and meta-analysis which found a strong correlation between both groups [
7]. Yet, only one study recruiting participants from an antenatal clinic was included in the meta-analysis. Additionally, the authors made a pooled analysis of prevalence data obtained from different administrative levels which made it impossible to give any recommendations on what spatial scale ANC prevalence might be used to monitor malaria transmission. Another study by Hellewell et al. found that clinical malaria incidence in children can predict ANC prevalence up to 3 months in the future but not vice versa [
8]. However, data from only five hospitals were analysed and no uniform relationship between prevalence and incidence could be established. Therefore, the validity of large-scale routine ANC prevalence data for monitoring transmission in children has yet to be investigated.
This study contributes to the existing evidence by examining the agreement between nationwide routine ANC test-positivity data and malaria prevalence in children at different administrative levels.
Discussion
This is the first study to validate the use of large-scale routinely collected ANC malaria test-positivity data reported through a national health information system as an innovative and cost-effective approach for malaria surveillance at sub-national level. The analyses performed show that the malaria test-positivity in pregnant women attending ANC is related to other population and model-based malaria endemicity measures. Varying test-positivity differences between pregnant women and the comparison groups could be partially explained by different levels of seasonality, urbanization, and ITN usage. These findings suggest that ANC malaria test-positivity can be useful to assess community-level prevalence and sub-national heterogeneity of malaria endemicity. However, the variability of the test-positivity difference across the ANC test-positivity spectrum was high at all levels of comparison and could not be accounted for by the covariates that were used in these analyses.
Factors that will help to decrease the uncertainty are most likely to vary within the comparison groups and geographically. Travelling patterns, treatment-seeking behaviour or intervention coverage have been shown to vary in space and between different demographic groups with changing impact on their malaria prevalence [
15‐
17]. However, the effect of ITN usage in this study was non-existent or too small to conclude an important influence of ITN coverage on the relationship between the malaria prevalence in children and the test-positivity in pregnant women and it might be rather seen as the consequence of increasing intervention coverage due to a high transmission level. Another factor previously suggested by van Eijk et al., and Hellewell et al. to increase the uncertainty of the relationship is gravidity [
7,
8]. Gravidity is a direct consequence of fertility which differs with the level of urbanization and socioeconomic status which is likely to vary at sub-national levels [
10]. Different levels of fertility determine the proportion of primi- and multigravidae women attending ANC. Van Eijk et al. showed that the malaria prevalence in primigravidae women compared to the prevalence in children provides results with less heterogeneity; thus, gravidity might have to be taken into account when using pregnant women attending ANC as a sentinel population for malaria surveillance [
7]. However, routine ANC data in Tanzania so far does not provide information on gravidity on an aggregated level.
While factors that vary geographically and within the comparison groups can account for the variability found in this study, factors that differ between pregnant women and children are more likely to explain the mean difference at varying transmission intensities. First, the test-positivity in pregnant women might be lower than in children due to their higher capability to clear parasites from their blood after drug administration [
18]. Secondly, infections in pregnant women may be less likely to be detected due to the sequestration of the parasite in the placenta and a higher level of immunity as the product of more cumulative exposure over years and during former pregnancies [
7,
19]. However, earlier studies suggest that the protection against the
Plasmodium falciparum malaria parasite does not merely depend on cumulatively acquired immunity but is also associated with the age of the host independent of previous exposure [
20‐
24]. The reason might be constitutional differences between an adult’s immune system and the immune system of a child which potentially leads to a higher ability of the adult immune system to mount a protective response against the parasite infection [
24]. The interaction between exposure and maturation of the immune system would particularly explain the comparatively high malaria prevalence in school children and the progressively decreasing relative difference between school children and pregnant women with increasing ANC test-positivity in areas of perennial malaria transmission. In settings with seasonal malaria however, the prevalence in school children was more comparable to the test-positivity in pregnant women than in areas with perennial transmission. This finding is likely to be caused by a comparatively low level of immunity in the adult population as malaria immunity has been shown to be short-lived and depending on constant exposure to the parasite, thus, fading during seasons of interrupted transmission [
25,
26].
In contrast, seasonality of malaria transmission did not have a significant effect on the relationship between the malaria prevalence in children aged 6–59 months and the test-positivity in pregnant women attending ANC. However, the overall higher prevalence found by the TDHS-MIS 2015/16 still indicates that pregnant women most likely have a higher level of immunity than young children. The observed malaria prevalence in school children was considerably higher than in children 6–59 months. This finding confirms previous research that found older children to be less likely symptomatic than younger children while serving as a reservoir of malaria infection [
27,
28]. A study by Walldorf et al. showed that in southern Malawi, school children were less likely to use bednets, were less often brought for treatment and often used unreliable treatment sources [
27]. Additional to differences in immunity, these factors may explain a higher prevalence in older children compared to younger children.
Finally, this study found very different patterns for modelled prevalence data compared to survey data. The prevalence predicted by the MAP and the BGM was higher than the test-positivity in pregnant women up to a certain threshold but was consistently lower thereafter. It is difficult to find a biologically meaningful explanation for this finding.
The use of routine ANC test-positivity data in Tanzania is supported by high reporting rates and continuously increasing malaria testing rates. However, testing rates were lower in rural and semi-urban settings which tend to be less accessible than urban areas and might therefore be more prone to commodity stock-outs. Independent of district type, the ANC malaria testing rate was lower in districts with a malaria test-positivity above 5%. It can be hypothesized that in health facilities operating in areas of higher transmission symptomatic patients might be prioritized for malaria testing resulting in a shortage of RDTs at the ANC department. However, further investigations should be undertaken to gain better insights into the reasons for the difference in testing rates between areas of different levels of urbanization and malaria transmission.
This study was not without limitations. The direct comparison of routinely obtained ANC malaria test-positivity and data obtained from surveys and model predictions is complicated due to different methodologies. Selection bias is a potential factor influencing ANC malaria test-positivity in areas where testing rates are low. Additionally, routine data has previously been shown to suffer from data quality issues resulting in a need for thorough data cleaning which may by itself introduce biases [
29]. Lastly, it is not clear how these findings for Tanzania can be transferred to other malaria-endemic countries. So far, ANC test-positivity is not a routine surveillance tool to monitor malaria transmission and no global guidelines exist on the most appropriate application of ANC test-positivity data. With ANC coverage being well above 90% in many sub-Saharan African countries (98% in Tanzania in 2015), and WHO recommending the use of routine information systems for malaria surveillance, this study provides support to the application of this approach on a wider scale despite a number of remaining uncertainties [
6,
10,
30,
31].
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.