Background
Almost half of the world’s population is at risk of infection with
Plasmodium vivax [
1]. Most cases originate in South East Asia and the Western Pacific and a significant number occur in South America and the Horn of Africa [
1,
2].
P. vivax is associated with substantial morbidity and severe and fatal disease in endemic countries [
3‐
6].
P vivax infection is characterized by comparatively low parasitaemia compared to
Plasmodium
falciparum, and in co-endemic regions vivax parasitaemia is often conservatively documented as
P. falciparum reducing the reported prevalence [
3]. Many nations evaluating their prospects for malaria elimination are endemic for
P. vivax and as successful control programmes reduce the risk of
P.
falciparum the relative proportion of
P.
vivax infection rises. Accurate data on
P.
vivax prevalence and transmission patterns is important for the progress of elimination campaigns and focusing malaria control efforts [
4].
Light microscopy examination of blood films is the main method for detecting peripheral parasitaemia and for differentiation of
Plasmodium species [
5]. The World Health Organization (WHO) guidelines for malaria diagnosis and treatment recommend that malaria treatment be given only after a positive parasitological test result from either microscopy or a rapid diagnostic test (RDT) [
6]. The advantages of light microscopy make it an ideal diagnosis tool in resource poor settings however its accuracy depends upon the technician’s skill level and can be adversely affected by operational constraints or technical problems. Low-density infections often remain undetected by microscopy [
7,
8]. As
P.
vivax is characterized by low level parasitaemia microscopy may not be the most appropriate tool for accurate diagnosis. Molecular methods, e.g. polymerase chain reaction (PCR), depend on DNA amplification approaches and have higher sensitivity than microscopy [
9]. Despite the greater sensitivity of PCR it is not widely used due to the lack of a standardized methodology, high costs, and the need for highly-trained staff. PCR is increasingly used in epidemiological studies, but rarely used in routine clinical diagnosis.
Most
P.
vivax cases occur in low transmission settings where asymptomatic carriage occurs relatively frequently [
10]. With the reliance on symptoms driving the presentation of infected patients and microscopy used as the main diagnosis tool, asymptomatic and sub-microscopic infections are likely to remain undetected and untreated in vivax endemic populations. Sub-microscopic infections of
P. falciparum have been shown to infect mosquitoes and transmit malaria [
11]. Sub-microscopic
P. vivax infections may similarly contribute to the infectious reservoir and maintain transmission. A review on asymptomatic malaria in Brazil highlighted the importance of accurate diagnosis and detection of all
Plasmodium cases for the malaria control programmes to be effective [
7]. A prevalence ratio reported from a study in Peru indicates that 78 % of infections would go undetected if microscopy alone was used in surveillance programmes [
8]. In a study from Iran microscopy did not detect any vivax infection in the 900 samples, whereas PCR detected ten positive samples out of 871 [
12].
In a review exploring the proportion of sub-microscopic
P. falciparum infections, Okell et al. documented reduced sensitivity of microscopy which detected on average 50 % of falciparum infections detected by PCR [
13]. Cheng et al. recently reviewed 25 publications (31 surveys) from 1996 to 2010 reporting a mean of 69.5 % sub-microscopic
P.
vivax infections [
14]. The current review extends the work of Cheng et al. with a meta-analysis and the inclusion of additional studies; since 2010, a further 11 prevalence studies reported microscopy and PCR
P. vivax prevalence and an additional seven were identified from 1996 to 2010. The aim was to re-quantify the extent of sub-microscopic
P. vivax infections detected by PCR and identify whether this ratio varies with transmission intensity, location of study or laboratory methods.
Methods
This systematic review used a predefined protocol and followed the PRISMA guidelines [
15]. PubMed, EMBASE and the Cochrane Library were searched using MeSH search terms and Boolean operators: “malaria” AND “PCR” AND “vivax” up to the 15 September 2014. The search was restricted to English language publications with no time limits. Results of each search were exported and duplicates removed. This high-yield search strategy was used to ensure capture of all relevant articles. Titles and abstracts of all articles were initially scanned to identify prevalence studies. Case reports, case series, efficacy studies, entomological surveys, non-human studies, immunological studies, genetic sequencing studies and other non-relevant articles were excluded. Full texts of articles identified as potentially relevant in this initial screen were assessed against the full eligibility criteria using a standardized form. References of relevant articles were scanned to identify additional studies.
Selection criteria
Eligible studies reported the prevalence of P. vivax by microscopy and PCR in the same population living in a malaria endemic country. If only a subset of participants were tested by microscopy or PCR, the study was included providing the subset was selected randomly. Exclusion criteria included: studies of imported malaria, studies where a large-scale intervention was implemented before the measurement of prevalence (e.g. insecticide-treated bed nets or mass drug administration), studies in pregnant women, studies that select participants based on parasitaemia levels, malaria symptoms or malaria diagnosis, studies of a population not representative of a defined endemic area (e.g. studies in refugees or migrant workers), studies with less than 20 blood samples tested by either method, and studies where no P. vivax was detected by PCR or microscopy. When the presence of malaria symptoms was not specifically stated it was assumed that participants seeking treatment or care at health facilities were symptomatic and these studies were excluded. In cohort studies, data were extracted from the baseline observation only.
From each eligible study the following information was extracted: month and year of sample collection, location, age criteria for inclusion, age range of included participants, prevalence of P. vivax malaria by microscopy, prevalence of P. vivax malaria by PCR, number of false-positives (i.e. number of samples microscopically positive and PCR negative) and number of false-negatives (i.e. number of samples microscopically negative and PCR positive) and PCR and microscopy methodology. Authors were not contacted for further information and no studies were excluded on the basis of quality.
Statistical analysis
The parasite prevalence by microscopy and PCR were calculated for each study, once for
P.
vivax monoinfections and again for all
P.
vivax infections (i.e. infections where
P.
vivax was detected either alone or with any other
Plasmodium species). Prevalence of infection detected by microscopy was compared to the prevalence of infection detected by PCR to produce a microscopy:PCR prevalence ratio and the sub-microscopic prevalence was calculated as: PCR prevalence minus the microscopy prevalence. PCR was considered the reference standard and the numbers of false-positives and false-negatives were used to calculate sensitivity and specificity estimates for microscopy when this information was available. Log transformed microscopy: PCR prevalence ratios and the sample size for each study were used to derive inverse-variance weighted fixed effects and random effects meta-analysis combined estimates [
16]. The Kruskal–Wallis method was used for nonparametric comparisons, and Student’s
t test for parametric comparisons. For categorical variables percentages and corresponding 95 % confidence intervals (95 % CI) were calculated using Wilson’s method. Proportions were examined using χ
2 with Yates correction or by Fisher’s exact test. The Chi squared heterogeneity statistic was used to assess the between-study heterogeneity. Forest plots and combined random-effects estimates were produced for sub groups (e.g. microscopy method) to determine if heterogeneity was accounted for by any methodological differences. All analyses were performed using Stata software (11.0; StataCorp).
Discussion
Data from 40 published studies show that PCR detects 67 % more P.
vivax infections than microscopy in surveys of endemic populations. This result is similar to findings by Cheng et al. who reported on 25 studies in which P.
vivax was not detected by microscopy in 69.5 % of estimates. The additional 18 studies included in the current analysis generally reported higher prevalence by both microscopy and PCR; the mean PCR prevalence was 18.6 % (range 0.18–73.2 %) compared to a range of PCR prevalence estimates of 0.2–59.5 % reported by Cheng.
Both reviews are also consistent with a literature review of
P. falciparum PCR and microscopy prevalence reported by Okell and colleagues in which microscopy detected 51 % of all parasitaemias detected by PCR [
13]. There are several characteristics of
P.
vivax infections that make diagnosis by microscopy more difficult than diagnosis of
P. falciparum including the fact that
P.
vivax and
P. falciparum are co-endemic in most vivax endemic areas and the detection of lower levels of
P. vivax parasitaemia is difficult in the presence of higher levels of
P. falciparum parasitaemia [
3]. PCR has greater sensitivity to identify mixed infections [
64], and this was apparent in the discrepancy between microscopy and PCR being greatest when considering all
P. vivax infections (mono and mixed infections). Similar results are apparent in cross sectional surveys where PCR is as much as 30-fold more efficient at detecting mixed infections than microscopy alone [
17]. The difference between prevalence ratios when detecting monoinfection and prevalence ratios when detecting all
P.
vivax infections, showed a similar trend although this did not reach statistically significance likely due to the limited number of prevalence pairs.
Only ten studies reported sensitivity and specificity measurements and these studies specifically aimed to measure these parameters for either PCR or microscopy. The wide range of microscopy sensitivities mirrored published sensitivities although microscopy sensitivity <50 % is less frequently reported in the literature than observed here. The sensitivity and specificity of microscopy is highly dependent on slide quality and the skill of microscopists [
65]. Unfortunately these factors could not be considered in this analysis; both are subjective and rarely reported. The low prevalence of vivax malaria could explain the variability in sensitivity estimates as the number of positive results for both PCR and microscopy used in sensitivity calculations was small.
The studies included in this review are representative of the geographical distribution of
P. vivax, with most but not all
P.
vivax endemic countries represented. The inclusion of only English publications introduced a bias towards studies from English speaking countries and a large number of potentially eligible publications from Brazil were not available in English [
66]. The analysis focused on blood samples representative of endemic populations with a high proportion of included studies adopting comprehensive sampling strategies in an attempt to determine covariates in endemic populations; however the detection of sub-microscopic infections in high-risk groups or other populations excluded from this review also have public health relevance that warrants investigation.
With only 54 paired prevalence estimates included in this meta-analysis, the power to detect key covariates was limited. Studies where patients were selected on the basis of malaria symptoms were excluded and this one criterion significantly reduced the number of studies available. Five studies were excluded as no P. vivax was detected by either technique, in addition, prevalence ratio could not be calculated for nine prevalence pairs when no P. vivax was detected by microscopy, the least sensitive detection method. Therefore, the extent to which microscopy failed to detect infections of P. vivax in low transmission areas is likely underestimated. Sample size is an important consideration in the design of studies where few infections are detected. The studies in this review that did not microscopically detect any P. vivax infections but did detect vivax infections with PCR were of varying samples sizes and two were large studies (N = 3316, N = 1527). This indicates that when only using microscopy, increasing the sample size alone will not allow detection of all existing P. vivax infections.
The detection limit of both techniques is dependent on the volume of blood examined. In theory, less blood is examined using standard microscopy techniques (~5 µL) than when a PCR assay is used [
60,
67]. This meta-analysis did not consider the exact volume of blood examined since the amount of blood used for DNA extraction for PCR was reported in only 18/40 studies and the amount of blood taken from participants reported in only 5/40 studies. Furthermore, nearly all studies that did report the amount of blood used for DNA extraction reported using 200 µL. The type of blood sample and DNA extraction method used for PCR analysis was examined and while the PCR prevalence was higher in studies that used whole blood versus studies that used filter paper this likely reflected regional variation and since the majority of studies using whole blood were from Asia and the majority of studies using filter paper being from the Asia Pacific or South America.
Future studies should examine the prevalence ratios of all malaria species in the same population to see if the sensitivity and specificity of microscopy and PCR varies according to the species examined. Although PCR is unlikely to be routinely used for screening in low-income countries, quantification of sub-microscopic infections under different transmission settings could be used to accurately estimate the true prevalence of
P.
vivax infections. Mathematical models could predict the extent of sub microscopic infections given a number of key parameters on the population characteristics and the transmission setting. Bayesian approaches have already been used to estimate disease prevalence in the absence of a gold standard diagnostic test or when the gold standard has imperfect sensitivity and specificity [
68]. The current review has identified gaps in the information required for mathematical models designed to accurately estimate prevalence of
P. vivax infection. Health planning and decision making by malaria control and elimination programmes in
P.
vivax endemic areas will require reliable estimates of parasite prevalence. The results of this review highlight the benefits of investing in PCR techniques to inform malaria control programmes. In areas focused on elimination it is vital that all reservoirs of
P.
vivax malaria are detected especially since rates of asymptomatic carriage can be substantial with a known ability to transmit [
69].
Authors’ contributions
CM, MA and CD conceived of the study and developed the study design. CM searched the literature, extracted and analysed the data and prepared the manuscript. RP, MA and CD advised on the systematic review and meta-analysis. All authors read and approved the final manuscript.