Types of studies

Studies sampling a consecutive series of patients, or a randomly selected series of patients were eligible. Where the report did not explicitly state that sampling was consecutive, but we judged that consecutive sampling was most probable, we included the report. We excluded studies if they did not present sufficient data to allow us to extract absolute numbers of true positives, false positives, false negatives and true negatives. Due to resource constraints, we also excluded studies if the report did not present enough information to allow full assessment of eligibility or if the study was reported only in a non‐English language.

Participants

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

We excluded studies if participants:

1. were non‐immune people returning from endemic countries or were mainly recent migrant or displaced populations from non‐endemic or very low endemicity areas;

2. had been treated for malaria and the test was performed to assess treatment outcome;

3. had symptoms of severe malaria;

4. did not have symptoms of malaria;

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

In studies where only a subgroup of participants was eligible for inclusion in the review, we included the study provided that we could extract relevant data specific to that subgroup. If studies included some patients with severe malaria, and we could not extract data specific to a subgroup of participants with uncomplicated malaria, we included the study if 90% or more of the participants had uncomplicated malaria.

Index tests

Studies evaluating any immunochromatography‐based RDTs specifically designed to detect non‐falciparum or P. vivax malaria. We included commercial tests that are no longer available because they may use the same antibodies and very similar technology to tests that are currently available or may become available in the future. Older and more recently available versions of the same test, for example, OptiMAL and OptiMAL‐IT were included separately. We also included prototype tests which are not longer available but which correspond to one of the commercial tests.

Comparator tests

We included studies regardless of whether they made comparisons with other RDT tests or not.

Target conditions

Studies aimed to detect non‐falciparum or P. vivax malaria. Where no distinction was made by species, but over 98% of malaria infections were identified by the reference standard as non‐falciparum or P. vivax, the study was eligible for inclusion.

Reference standards

Studies were required to diagnose non‐falciparum or P. vivax malaria using at least one of the following two reference standards:

1. Conventional microscopy of thick blood smears, thin blood smears or both. Presence of asexual parasites of any density was regarded as a positive smear;

2. PCR.

The reference standard was required to be performed using blood samples drawn at the same time as those for the index tests. Where studies used more than one reference standard, we presented data relating to comparisons with each reference standard.

Electronic searches

To identify all relevant studies, we used the search terms and strategy outlined in Appendix 1 to search the following databases: Cochrane Infectious Diseases Group Specialized Register; MEDLINE; EMBASE; MEDION; Science Citation Index; Web of Knowledge; African Index Medicus; LILACS; and IndMED. We based the search on the following MeSH, full text and keyword terms: Malaria, Plasmodium, reagent kits, diagnosis, diagnostics, RDT, dipstick, MRDD, OptiMal, Binax Now, Parasight, Immumochromatography, antigen detection, antigen test, Combo card. We did not limit the search by language or publication status (although we later excluded non‐English language studies due to resource constraints). We restricted the searches to human studies. We updated the search on 31 December 2013.

Searching other resources

We searched the reference lists of included studies for relevant publications. Due to resource constraints, we did not search any other resources.

Selection of studies

We initially used a single selection procedure to identify studies for inclusion in either of the two Cochrane Reviews in this series. The inclusion criteria differed between the reviews only in the target condition and parasite species. Therefore, of the study characteristics examined, we assessed parasite species last , for example a study listed as excluded due to not presenting sufficient data may also have not been a study of non‐falciparum or P. vivax malaria. One author (KA) initially assessed the titles identified by the search, excluding those obviously irrelevant to the diagnosis of malaria using RDTs. We retained titles where we had any doubt regarding inclusion.

Based on abstract examination, we excluded irrelevant letters, review articles and articles and then excluded other irrelevant notes. Using a pro forma, two review authors (KA and NM) independently assessed the eligibility of the remaining potentially relevant articles based on full text publications. We have listed the excluded studies in the Characteristics of excluded studies table. We resolved any discrepancies by discussion. Where we could not reach agreement, we consulted a third author (PG or PO). Where it remained unclear whether a study was eligible for inclusion because of a lack of detail or poor reporting, we excluded it. Similiarly, we excluded non‐English language reports for logistical reasons.

We named studies according to the surname of the first study author and the year of publication. The study naming used in this review uniquely identifies multiple study cohorts within each study report (for example as 'Bell 2001a' and 'Bell 2001b'), each of which use different reference standards or present data separately for more than one population with different characteristics. More than one RDT may be evaluated in each study cohort, thus the number of test evaluations exceeds the number of study cohorts, which exceeds the number of study reports.

Data extraction and management

Two review authors (KA and NM) independently extracted data and resolved any discrepancies by discussion. In cases of studies where only a subgroup of participants met the review inclusion criteria, we extracted and presented data only for that particular subgroup. Where two versions of one reference standard were used, for example local clinic and expert standard microscopy, or field versus laboratory RDT testing, we only included the one most likely to yield the highest quality results.

For each study, we systematically extracted data on the characteristics of the study, as shown in Appendix 2. We also extracted data relating to the sensitivity of the RDT at different levels of parasitaemia (asexual parasites per µL of blood) as presented by the study authors. For each comparison of index test with reference test, we extracted data on the number of true positives, true negatives, false positives and false negatives in the form of a two by two table. RDT results are dichotomous; microscopy results were deemed positive at any level of asexual parasitaemia; and PCR results used the cut‐off points presented by the study authors. Gametocyte‐only parasitaemia was considered negative; where a study was unclear on how they had classed gametocyte‐only parasitaemia, they were assumed to have used the same classification as ourselves and we included the data in the study.

We extracted data for each study (Smidt 2008), using current manufacturers' instructions in interpreting the RDT results. P. falciparum malaria only was considered as negative for parasitaemia. The target condition was defined slightly differently depending on the type of the test, as follows:

• Types 2, 3 and 4 ‐ Non‐falciparum malaria in the absence of falciparum malaria

RDT Types 2, 3 and 4 are designed to detect non‐falciparum species (mainly P. vivax in most situations) when they occur without concurrent P. falciparum infection. They have two test lines, one specific for P. falciparum and one pan‐malarial line to detect all malaria species. Non‐falciparum malaria is identified by a positive pan‐malarial line and negative P. falciparum line; mixed infections will produce positive results for both the P. falciparum and pan‐malaria lines and are indistinguishable from P. falciparum alone.

Mixed infections detected by microscopy were considered true negative if RDT indicated P. falciparum; true positive if RDT indicated non‐falciparum in the absence of P. falciparum; and false negative if RDT indicated no malaria. This method corresponded to the method most often described by the authors of the included studies, first described by Tjitra 1999.

• Tests using Pf HRP2 and Pv pLDH ‐ P. vivax (whether alone or part of mixed infection)

These types of tests are designed to identify P. vivax parasitaemia specifically, as they have a test line specific to P. vivax. Some also include other test lines, specific to other types of malaria parasite. Test results were considered positive for P. vivax whether or not they also indicated the presence of P. falciparum.

Where study authors interpreted test results or presented data differently, we used all the information presented in the paper to extract data consistent with our own methods; if we were unable to do this, we did not include the data in the analyses.

Assessment of methodological quality

Three researchers (KA, NM and SJ) assessed the quality of each individual study using the checklist adapted from the QUADAS tool (Whiting 2003). We answered each question on the checklist with a yes or no response, or noted unclear if study authors reported insufficient information to enable a judgement, and we documented the reasons for the judgement made. We have summarized the criteria we used in Appendix 3.

Statistical analysis and data synthesis

The comparisons made in this review can be considered in a hierarchy. We classified the data on each test type in the primary studies according to commercial brands. In order to provide a coherent description of the studies contributing to each analysis, we structured the results first by grouping studies according to their commercial brand, then grouping brands to form test types. The analytical strategy thus compared the test accuracy of commercial brands within each test type before making comparisons between test types. Comparative analyses first included all studies with relevant data, and were then restricted to studies that made direct comparisons between tests with the same participants, where such studies existed.

For each test type, we plotted estimates of the observed sensitivities and specificities in forest plots and in receiver‐operating characteristic (ROC) space. These plots illustrate variation in accuracy between studies. Where adequate data were available, we performed meta‐analyses using the bivariate model (Reitsma 2005) to produce summary sensitivities and specificities. Using a random‐effects approach, the model jointly synthesises sensitivity and specificity by allowing for correlation between them across studies. We made comparisons between tests by adding a covariate for brand or test type to the bivariate model to investigate association with sensitivity or specificity, or both. Also, we investigated the effect of test type on the variances of the random effects of logit sensitivity and logit specificity and we included separate variance terms where required. We assessed the significance of the difference in test performance by a likelihood ratio test comparing models with and without covariate terms for sensitivity and specificity. Where inadequate studies were available to estimate all parameters, we simplified the bivariate model to two univariate random‐effects logistic regression models by assuming no correlation between sensitivity and specificity. We fitted the models using the xtmelogit command in StataCorp 2011.

Where more than one commercial brand of the same test type was evaluated on the same patients against the same reference standard, we selected one brand at random from the analysis by test type in order to avoid bias due to inclusion of the same participants more than once in the analysis. We included both brands in any analyses comparing commercial brands.

Investigations of heterogeneity

We inspected forest plots and summary ROC plots to visually assess heterogeneity between study specific estimates of sensitivity and specificity. We planned to investigate the effect of age group, continent where the study took place, and adequacy of the reference standard on summary estimates of sensitivity and specificity by adding each factor as a covariate to the bivariate model.

We did not attempt to assess reporting bias because little is known about how this should be done for diagnostic test accuracy (DTA) reviews.