Types of studies

We included retrospective or prospective cohort or cross‐sectional studies that assessed the accuracy of an RDT, or compared the accuracy of two or more RDTs, in the same study population (i.e. comparative accuracy studies). We excluded case‐control studies because they are known to overestimate test accuracy (Whiting 2011). Eligible studies included a consecutive series of patients, or a randomly selected series of patients. If the study did not explicitly state that the sampling was consecutive or random, the study was considered unclear but was still included. We excluded studies if they did not present sufficient data to allow us to extract or deduce the number of true positives, false positives, false negatives, and true negatives (i.e. 2 x 2 table data). We also excluded studies published in predatory journals, which is referred to as journals that accept articles for publication for a fee without providing peer‐review or quality checks for plagiarism or ethical approval.

Participants

Studies recruiting people living in P vivax‐endemic areas attending ambulatory healthcare settings with symptoms of uncomplicated malaria were eligible.

We excluded studies if participants:

• had travelled from non‐malarious region to malarious regions, e.g. travellers or displaced populations;

• had been previously treated for their current malaria infection or the test was performed to assess whether treatment was successful, or both;

• had symptoms of severe malaria as defined by the WHO clinical definition (WHO 2014);

• did not have symptoms of malaria as defined by history of fever, headache, or chills/rigors; or

• were recruited through active case finding (for example, door to door surveys).

Index tests

Studies evaluating any immunochromatography‐based RDT specifically designed to detect P vivax malaria. We only included RDTs that met the WHO malaria RDT performance criteria (WHO 2017b).

Target conditions

Studies aimed at detecting P vivax malaria.

Reference standards

Studies that diagnosed P vivax malaria using at least one of the following two reference standards:

• Conventional microscopy of thick blood smears and thin blood smears. Presence of asexual parasites of any density is regarded as a positive smear. Once the diagnosis is established – usually by detecting parasites in the thick smear – the laboratory technician can examine the thin smear to determine the malaria species and the parasitaemia, or the percentage of the patient’s red blood cells that are infected with malaria parasites. The thin and thick smears are able to provide all three of these vital pieces of information. Ideally, blood smears would be examined independently and in duplicate with more than 100 high‐power fields;

• PCR, including quantitative PCR (qPCR), nested PCR (nPCR), and real‐time PCR (rPCR). We also included studies that used loop‐mediated isothermal amplification (LAMP). Most PCR‐based assays for P vivax are only available as laboratory‐developed tests, which means they are rarely used clinically outside of research projects where P vivax malaria is endemic. They are especially useful for diagnosing asymptomatic people as the assays have high sensitivity. Molecular diagnostics theoretically have a lower limit of detection than both RDTs and microscopy depending on the training of microscopists and quality of samples analysed. Significant variation exists between molecular diagnostics developed including type of input material (DNA, RNA, or whole blood), target gene, (number of) species detected, primer/probe composition and concentration, amplification technique (PCR or isothermal), read‐out (gel‐electrophoresis, fluorescence detection, lateral flow), and whether it is qualitative or quantitative. However, no important differences have been found in the accuracy of these tests (Roth 2016).

For studies that used both reference standards, we extracted 2 x 2 data for each reference standard and stratified the analyses by reference standard.

Electronic searches

We searched the following databases up to 30 July 2019 using the search terms and strategy described in Appendix 1: Cochrane Infectious Diseases Group Specialized Register; Central Register of Controlled Trials (CENTRAL), published in the Cochrane Library (issue 7, 2019); MEDLINE (PubMed, from 1966); Embase (OVID, from 1947); Science Citation Index Expanded (SCI‐EXPANDED) and Conference Proceedings Citation Index‐ Science (CPCI‐S), both in the Web of Science, from 1900; LILACS (BIREME).

We also searched the WHO International Clinical Trials Registry Platform (WHO ICTRP; www.who.int/ictrp/en/) and ClinicalTrials.gov (clinicaltrials.gov/ct2/home) for trials in progress, using "vivax malaria", "Plasmodium vivax", and "rapid diagnostic test*" or RDT* as search terms.

Searching other resources

We checked the reference lists of studies identified by the above methods.

Selection of studies

Three review authors (RA, LC and SJ) independently assessed the study eligibility by examining the title and abstract of each article identified by the literature search and excluded obviously irrelevant studies. If a review author considered the abstract to be potentially eligible, we obtained the full‐text article. Three review authors independently assessed each full‐text article against the predefined inclusion and exclusion criteria, as stated in the ‘Criteria for considering studies for this review' section, and resolved any disagreements by discussion. All articles that were excluded after full‐text assessment are listed with reasons for exclusion in the ʽCharacteristics of excluded studies' table. We illustrated the study selection process with a PRISMA flow diagram (Figure 1).

Figure 1

---

---

Study flow diagram.

Data extraction and management

Three review authors (RA, LC and SJ) independently extracted data using a pre designed data extraction form.

We extracted the following data.

• Authors, publication year, and journal.

• Study design.

• Study start date.

• Characteristics study participants (age, gender, co morbidities, and pregnancy).

• Study inclusion/exclusion criteria.

• Study setting.

• Malaria species in study setting.

• Malaria prevalence and endemicity in study setting.

• Reference standard.

• Index test (brand name, target antigen, and batch numbers).

• Additional tests (and their results).

• RDT and reference standard setting.

• Lot testing of RDT used.

• Transport and storage conditions of RDTs.

• Training level of person performing index test.

• Training level of person performing reference standard (and if available the WHO certified training level of the microscopist).

• Number of high power fields observed in microscopy.

• Parasite density of microscopy positive cases or PCR.

• Observers or repeats used.

• Number of indeterminate, missing or unavailable test results.

• Number of true positives, false positives, false negatives, and true negatives.

• Type of molecular amplification assay.

• Volume of blood samples.

• Limit of detection for PCR.

We resolved any discrepancies in data extraction by discussion. We contacted the authors of primary studies when we could not resolve any disagreements.

Assessment of methodological quality

We used the revised tool for the Quality Assessment of Diagnostic Accuracy Studies (QUADAS‐2) to assess the risk of bias and applicability of included studies (Whiting 2011). We tailored the tool to the context of the review as shown in Appendix 2. Three review authors (RA, LC and SJ) independently assessed methodological quality using the tailored QUADAS‐2 tool. We resolved any disagreements through consensus. We used both graphics and text to summarize the results.

Statistical analysis and data synthesis

We stratified all analyses by the type of reference standard used. We plotted estimates of sensitivity and specificity from the included studies in forest plots and in receiver operating characteristic (ROC) space using the software, Review Manager 5 (RevMan 5) (RevMan 2014). We planned to perform meta‐analysis using the bivariate model to estimate summary sensitivities and specificities (summary points) (Chu 2006; Macaskill 2010; Takwoingi 2015b). However, due to sparse data or few studies, we simplified the models to univariate random effects logistic regression models to pool sensitivity and specificity separately (Takwoingi 2015a). We performed meta‐analyses using the 'meqrlogit' command in Stata (STATA 2015). Due to the limited number of included studies we did not perform meta‐analyses to compare the accuracy of different RDT brands as planned. However, we summarized individual study estimates from head‐to‐head comparisons of brands in a table.

Investigations of heterogeneity

We intended to investigate any heterogeneity from the pooled analyses with pre‐specified factors, as stated in our secondary objective. Due to the limited number of studies, we were unable to investigate heterogeneity as planned.

Sensitivity analyses

We did not have sufficient data for sensitivity analyses.

Assessment of the certainty of the evidence

We assessed the certainty of the evidence for comparisons where there were sufficient studies enabling meta‐analyses (i.e. quality of evidence or confidence in effect estimates) using the GRADE approach and GRADEpro Guideline Development Tool software (GRADE 2013; GRADEpro GDT 2015). In the context of a systematic review, the ratings of the certainty of the evidence reflect the extent of our confidence that the estimates of test accuracy are correct. As recommended, we rated the certainty of the evidence as either high (not downgraded), moderate (downgraded by one level), low (downgraded by two levels), or very low (downgraded by more than two levels) for four domains: risk of bias, indirectness, inconsistency, and imprecision. For sensitivity and specificity, the certainty of the evidence initially started as high when there were high‐quality cross‐sectional or cohort studies that enrolled participants with diagnostic uncertainty. If we found a reason for downgrading the certainty of the evidence, we classified the reason as either serious (downgraded by one level) or very serious (downgraded by two levels).

Three review authors (RA, LC and SJ) discussed judgments and reached a consensus. We applied GRADE in the following way.

• Risk of bias: we used QUADAS‐2 to assess risk of bias.

• Indirectness: we considered indirectness from the perspective of test accuracy. We used QUADAS‐2 to assess applicability concerns and looked for important differences between the populations studied (for example, in the transmission intensity as defined by the WHO World Malaria Report or WHO malaria country profiles for the corresponding year), the setting, and the review question.

• Inconsistency: GRADE recommends downgrading for unexplained inconsistency in sensitivity and specificity estimates.

• Imprecision: we considered the width of the confidence intervals (CIs), and asked ourselves, “would we make a different decision if the lower or upper limit of the 95% confidence interval (CI) represented the truth?” In addition, we calculated absolute numbers of true positives, false negatives, false positives, and true negatives, as well as ranges for these values based on the CIs of the pooled estimates of sensitivity and specificity for various prevalences of P vivax malaria; we also made judgements on imprecision using these calculations. We also calculated positive and negative likelihood ratios with their 95% CIs.

Assessment of reporting bias

We did not assess publication bias due to the uncertainty about the determinants of publication bias for diagnostic accuracy studies, and the inadequacy of tests for detecting funnel plot asymmetry (Deeks 2005).