Background: In malarious areas, pregnant women are more likely to have detectable malaria than are their non-pregnant peers, and the excess risk of infection varies with gravidity. Pregnant women attending antenatal clinic for their first visit are a potential pragmatic sentinel group to track the intensity of malaria transmission; however, the relation between malaria prevalence in children, a standard measure to estimate malaria endemicity, and pregnant women has never been compared. Methods: We obtained data on malaria prevalence in pregnancy from the Malaria in Pregnancy Library (January, 2015) and data for children (0-59 months) were obtained from recently published work on parasite prevalence in Africa and the Malaria in Pregnancy Library. We used random effects meta-analysis to obtain a pooled prevalence ratio (PPR) of malaria in children versus pregnant women (during pregnancy, not at delivery) and by gravidity, and we used meta-regression to assess factors affecting the prevalence ratio. Findings: We used data from 18 sources that included 57 data points. There was a strong linear relation between the prevalence of malaria infection in pregnant women and children (r=0·87, p<0·0001). Prevalence was higher in children when compared with all gravidae (PPR=1·44, 95% CI 1·29-1·62; I2=80%, 57 studies), and against multigravidae (1·94, 1·68-2·24; I2=80%, 7 studies), and marginally higher against primigravidae (1·16, 1·05-1·29; I2=48%, 8 studies). PPR was higher in areas of higher transmission. Interpretation: Malaria prevalence in pregnant women is strongly correlated with prevalence data in children obtained from household surveys, and could provide a pragmatic adjunct to survey strategies to track trends in malaria transmission in Africa. Funding: The Malaria in Pregnancy Consortium, which is funded through a grant from the Bill & Melinda Gates Foundation to the Liverpool School of Tropical Medicine, UK; US Centers for Disease Control and Prevention; and Wellcome Trust, UK.
We obtained data on the prevalence of malaria infection in pregnant women from the Malaria in Pregnancy Library.11 This library is a comprehensive bibliographic database created by the Malaria in Pregnancy Consortium that is updated every 4 months with a standardised protocol to search more than 40 sources, including PubMed, Web of Knowledge, and Google Scholar.12 We used data up to January, 2015, without language restriction.12 Inclusion criteria were: studies in sub-Saharan Africa, based in either the community or antenatal clinics, that screened pregnant women for malaria parasitaemia by microscopy or rapid diagnostic test, irrespective of the presence of symptoms. We excluded studies that selected only women with a history of fever or malaria, and studies that diagnosed malaria at delivery, so that the data for pregnant women would be comparable with those for women attending antenatal clinic. There was no time limit for inclusion and we did not restrict study selection to those with first antenatal visit data. We undertook a systematic evaluation of studies in pregnant women and extracted data including study location, year of study, study population, inclusion and exclusion criteria used, use of malaria prevention strategies (ITNs, IPTp, or prophylaxis), type of malaria diagnostic test used, and test results. Where sufficient information was available, data were extracted by gravidity group, study site, and malaria season. Where needed, and if possible, we contacted authors of the included studies for additional information. Data on the prevalence of malaria infection in pregnant women were then selected on the basis of the availability of the same prevalence data in children aged 0–59 months collected during the same study period and in the same locality as the data in pregnant women. The contemporaneous prevalence data in children and pregnant women were either extracted from studies reported in the Malaria in Pregnancy Library that also reported data in children, or obtained from surveys that collected data on pregnant women and children simultaneously. We identified these data from the large database of over 28 483 temporally and spatially unique surveys of malaria infection undertaken across Africa since 1980 and described elsewhere,6 and from nationally representative household surveys, such as Demographic and Health Surveys, Multiple Indicator Cluster Surveys, and Malaria Indicator Surveys.13, 14, 15 An overview of the methods used in these surveys has been reported previously.9, 16 The information we extracted from the child records included study population, inclusion and exclusion criteria used, use of ITNs, type of malaria diagnostic test used, and test results. We assessed the quality of studies after considering source population, participant selection, appropriate tests, characteristics reporting, and completeness of outcome data. Quality was classified as low-to-moderate or good. Further details of the methods used to assess quality are included in the appendix. Meta-analyses were conducted using Stata (version 13, StataCorp LP, College Station, TX, USA) using the metan command with input of numerators and denominators for pregnant women and children and the “rr” option to pool the prevalence. We expressed differences between prevalence estimates in pregnant women and children as pooled prevalence ratio (PPR) obtained by meta-analyses using DerSimonian and Laird random-effects models.17 We used random effects models because of the wide heterogeneity in study design and to minimise the effect of study size.18 The extent of heterogeneity was measured using the I2, a measure of the proportion of total variability explained by heterogeneity rather than chance expressed as a percentage,19 with 0–40% representing no or little heterogeneity, 30–60% moderate heterogeneity, 50–90% substantial heterogeneity, and 75–100% considerable heterogeneity.20 To explore determinants of the relation between the prevalence in pregnant women versus children, we examined sources of heterogeneity across studies of the PPR estimates using random-effects meta-regression.21 Regression coefficients were presented as odds ratios (ORs) and their corresponding 95% CIs. We estimated between-study variance (τ2) using the algorithm of residual (restricted) maximum likelihood, and calculated p values, and 95% CIs for coefficients using the modification by Knapp and Hartung.22 For the meta-regression, study-level predictors were considered for inclusion in the initial models if the p value for the univariate association of that variable with the endpoint was <0·2. We considered the effect of the following predictors: gravidity, study period, location of recruitment for pregnant women (community or antenatal clinic), coverage of antimalarial prevention (chemoprophylaxis or IPTp) in pregnant women, type of diagnostic test, malaria transmission intensity, as defined by the average malaria prevalence among children and pregnant women (as a continuous variable and stratified as 40%),23, 24 and ITN coverage. Because there is a high correlation between ITN use in pregnant women and children (appendix), we used data for coverage in children to represent both groups. HIV infection is known to increase the risk of malaria in pregnancy;25 however, unfortunately none of the included studies had a systematic assessment of maternal HIV status. As an approximation of maternal HIV status, we used the information from the prevalence of HIV in women aged 15–49 years in the same study, or data from a Demographic and Health Survey closest to the study date, or data from other sources by country in all people aged 15–49 years (appendix). We did a sensitivity analysis to explore the potential effect of the type of study included (regional survey versus observational study) and of study quality on the primary outcome by comparing the results of (sub)national surveys with local studies, or results from low-to-moderate studies with those from good quality studies. The funding institution had no role in the design and development, data extraction, analysis and interpretation of the data, or preparation, review, or approval of the paper. AMvE had full access to all data and had final responsibility for the decision to submit for publication.