Summary of discordant results between rapid diagnosis tests, microscopy, and polymerase chain reaction for detecting Plasmodium mixed infection: a systematic review and meta-analysis

Malaria rapid diagnostic tests (RDTs) are widely used to detect malaria parasites among patients who suspected malaria infections in malaria-endemic areas where microscopy is unavailable. Nevertheless, little is known about the performance of RDTs in detecting Plasmodium mixed infections. The present study aimed to evaluate the discordant results between RDTs and microscopy/polymerase chain reaction (PCR) in detecting Plasmodium mixed infections. The PubMed (MEDLINE), Web of Science, and Scopus databases were systematically reviewed to identify related studies that reported the performance of RDTs in detecting Plasmodium mixed infections. Studies were grouped according to the different RDT types including RDT type 2 (pf-HRP2/pan-aldolase), RDT type 3 (pf-HRP2/pan-pLDH), RDT type 4 (Pf-LDH/pan-pLDH), RDT type 5 (Pf/Pv-pLDH), and RDT type 6 (pf-HRP2/Pv-pLDH) for subgroup analysis. The estimates of the different proportions in each analysis group that were visually summarized in a forest plot showed the odds ratio (OR) and 95% confidence interval (CI). Plots were drawn using RevMan (version 5.3; Cochrane Community). Twenty-eight studies were included in the present study. Overall, the meta-analysis showed that RDTs could detect a significantly higher proportion of Plasmodium mixed infections than microscopy (p = 0.0007, OR = 3.33, 95% CI 1.66–6.68). Subgroup analysis demonstrated that only RDTs targeting Pf-specific histidine-rich protein 2 (HRP2)/pan-specific lactate dehydrogenase (LDH) could detect a significantly higher proportion of Plasmodium mixed infections than microscopy (p = 0.004, OR = 8.46, 95% CI 2.75–26.1). The subgroup analysis between RDTs and PCR methods demonstrated that RDTs targeting Pf-specific HRP2/Pv-specific LDH could detect a significantly lower proportion of Plasmodium mixed infections than PCR methods (p = 0.0005, OR = 0.42, 95% CI 0.26–0.68). This is the first study to summarize the discordant results between RDTs and microscopy/PCR in detecting Plasmodium mixed infections. Malaria RDTs targeting Pf-HRP2/pan-pLDH could detect a higher proportion of Plasmodium mixed infections than microscopy, while RDTs targeting Pf-HRP2/Pv-specific LDH could detect a lower proportion of Plasmodium mixed infections than PCR methods. The results of this study will support the selection and careful interpretations of RDTs for a better diagnosis of Plasmodium mixed-species infections and appropriate treatment of malaria patients in endemic and non-endemic settings.


Methods
Search strategy. Searches of Medline (PubMed), Web of Science, and Scopus were systematically performed using the search terms provided in Supplementary Table S1. The searches were limited to the English language. Searches were carried out and finished on 1 April 2020. All reference lists of all eligible and included studies as well as Google Scholar search was performed to further increase the number of included articles for review.

Definition of malaria RDTs and microscopy.
Types of malaria RDTs were classified according to the study by Bell et al. 7 . They classified malaria RDTs into seven types according to the antigen used in the reagent strip, including type 1 (HRP2 (falciparum-specific), 2 (pf-HRP2/pan-aldolase), 3 (pf-HRP2/pan-pLDH), 4 (Pf-LDH/pan-pLDH), 5 (Pf/Pv -pLDH), 6 (pf-HRP2/Pv-pLDH), and 7 (aldolase). RDTs types 2, 3, 4, 5 and 6 can detect mixed or concurrent infections. Interpretation of Plasmodium mixed-infections by RDT was based on details provided by authors of the included studies. The gold standard for malaria detection is still microscopy where the examinations of thin and thick blood films lead to the demonstration of malaria parasites.
inclusion and exclusion criteria. Cross-sectional studies that reported the number of Plasmodium mixed infections evaluated by any of the five types of RDTs (types 2, 3, 4, 5 and 6) in comparison to microscopy or PCR were included in the present study. Studies reporting the results of RDTs and microscopy from the same patient samples or those reporting the results of RDTs and PCR from the same patient samples were included in the study. The following types of literature were excluded; studies that reported mixed-infections only for RDTs but did not report microscopy or PCR, incomplete data, no RDT results, co-infections with other agents, experimental studies, review articles, case reports and case series, polymorphism/mutation studies, knowledge about malaria/practice assessments, animal/mosquito studies, studies of haematological alterations, guidelines, and clinical drug trials. Studies with no full text and present data in the local language were also excluded.
Data extraction. All studies acquired through the search were stored in EndNote reference manager software (version X9; Clarivate Analytics). The data extractions started with screening the titles and abstracts after Scientific RepoRtS | (2020) 10:12765 | https://doi.org/10.1038/s41598-020-69647-y www.nature.com/scientificreports/ duplicate studies removed. Studies that were not related to the inclusion criteria were excluded. Then, the studies were screened for full-text articles, and those that did not comply with eligibility criteria were excluded with tags indicating the reason for exclusion. The data from full-text articles that passed the inclusion and exclusion criteria were then exclusively examined and extracted by two independent authors (MK and KUK) using an Excel spreadsheet for further analysis. Any inconsistencies relating to included studies and data extraction were resolved by a third or a fourth reviewer (FRM or GDM).

Statistical analysis.
Studies were grouped (subgroup) according to the different RDT types for comparative analysis. The meta-analysis of the proportion of the number of Plasmodium mixed infections per the total number of total malaria positives were performed as follows: (1) the summary estimate of the difference in the proportion (odds ratios, ORs) of RDTs to detect mixed infections compared with microscopy and (2) the summary estimate of the difference in the proportion (ORs) of RDTs to detect mixed infections compared with PCR methods were estimated. The subgroup analysis of RDT types, blood collection methods (finger prick or venipuncture), and types of Plasmodium mixed species confirmed by PCR were analysed in the present study. All analyses were conducted using Review Manager Version 5.3 (Cochrane, UK). The statistical analysis used to calculate the difference between groups was the Mantel-Haenszel test with a random-effects model. The metaanalysis for each study and the overall studies are presented with OR and 95% confidence intervals (CIs) as effect measures and summarized in forest plots. Cochrane's Q test and Higgins's I 2 statistics were performed to assess the heterogeneity of the included studies.
Quality of included studies. The quality of the individual studies included in the present study was assessed by the Quality Assessment of Diagnostic Accuracy Studies (QUADAS) 22 . The tool includes 4 domains including the following: (1) report the review question, (2) develop review-specific guidance, (3) review the published flow diagram, and (4) judge bias and applicability. Each domain was assessed in terms of the patient selection, index test, reference standard, and flow timing. Patient selection was the method of patient selection reported in the included studies. The index test was the RDT method that was conducted and interpreted in the included studies. The reference standards were microscopy or the PCR method that was conducted and interpreted. The flow and timing described any patients who did not receive the index tests or reference standard. Each question was answered with a "yes, " "no, " or "unclear" response. The results of the QUADAS assessment for all included studies were then summarized in the methodological quality graph and summary created by Review Manager.
publication bias. Publication bias is the publication of studies due to the statistical significance of the results 23 , which can lead to overestimated effect sizes and the dissemination of false-positive results 24 . The publication bias was assessed by visual inspection of funnel plot asymmetry (the asymmetrical distribution of the included studies in the graph between the OR and SE (logOR)). The publication bias was also assessed with Egger's test. Both tests aimed to determine small-study effects leading to more or less beneficial summaries of OR estimates 25 .
WHO product testing of malaria RDTs. The WHO product testing of malaria RDTs began in 2008 52 . All companies manufacturing malaria RDTs under the ISO-13485 Quality System Standard were invited to submit up to three tests for evaluation 52 . The results of the WHO product testing of malaria RDTs are demonstrated in Table 2. RDTs from the eight studies 30,34,36,40,41,46,49,51 were not subject to the WHO product testing program as these RDTs were developed and used before 2008, while the results of malaria RDTs from the four studies 28,37,43,47 was not found on the WHO testing product.
Methodological quality of the included studies. The methodology and reporting of the selected studies varied highly ( Fig. 2; Supplementary Fig. 1). All 28 included studies had cross-sectional designs. Most of the included studies (25/28, 89.3%) used a consecutive or random sample of patients. Two studies did not enrol a Scientific RepoRtS | (2020) 10:12765 | https://doi.org/10.1038/s41598-020-69647-y www.nature.com/scientificreports/ consecutive or random sample of patients 21,36 . In another study 19 , the sampling method for participants enrolled was unclear. Microscopic examination was used as the reference standard in 24 studies. PCR was used as a refer-   (Fig. 3). In total, six different RDT types including RDT types 2, 3, 4, 5 and 6 were included in the analysis. One study did not report the type of RDT used in their study 39 . Six studies used more than one RDT type/brand in their studies 21,26,27,35,36,43 . Four studies 28   The summary estimate of ORs between type 6 RDTs and microscopy to detect mixed infections based on the analysis of nine included studies was 1.07 (95% CI 0.74-1.55, p = 0.71, I 2 = 0%). Overall, the significant summary estimate of ORs between all types of RDTs and microscopy to detect mixed infections was found (OR = 3.33, 95% CI 1.66-6.68, p = 0.009, I 2 = 94%). The subgroup analysis of blood collection methods for microscopy was performed using 18 included studies. The results demonstrated that no subgroup difference (p = 0.55) was found among studies using blood from the finger prick method and those using blood from venipuncture. The summary estimate of ORs between all types of RDTs compared to those performing microscopy using blood from the finger prick method to detect mixed infections was significantly different among 11 studies (OR = 4.41, 95% CI 1.72-11.29, p = 0.002). The summary estimate of ORs between RDTs and microscopy using blood from the venipuncture method was significantly different among seven studies (OR = 3.04, 95% CI 1.44-6.43, p = 0.004) (Fig. 4).

Discordance between RDts and pcR.
Overall, 12 studies reported on mixed infections detected by both RDTs and PCR 21,[26][27][28][29]32,33,37,39,[43][44][45] , as shown in Fig. 5. Among 12 studies, three different types of RDTs were reported including RDT types 2, 3, and 6. The summary estimate of ORs between type 2 RDTs and PCR to detect mixed infections was 8.21 (95% CI 4.51-15.0). The summary estimate of ORs between type 3 RDTs and PCR based on the analysis of six included studies to detect mixed infections was 4.05 (95% CI 0.73-7.84, p = 0.07, I 2 = 97%). The summary estimate of ORs between type 6 RDTs and PCR based on the analysis of five included studies to detect mixed infections was 0.42 (95% CI 0.26-0.68, p = 0.0005, I 2 = 0%). Another study with no description on the type of RDT showed that the summary estimate of ORs between RDTs and PCR to detect mixed infections was 0.84 (95% CI 0. 17-4.3). Overall, the summary estimate of ORs between all types of RDTs and PCR to detect mixed infections was 1.17 (95% CI 1.54-0.53, p = 0.42, I 2 = 96%). The subgroup analysis of detection of Plasmodium mixed-species infections between RDT and PCR found that the summary estimate of ORs between RDT and PCR was comparable in P. falciparum mixed infections with P. vivax (OR= 0.81, 95% CI 0.51-1.27, p = 0.36, I 2 = 51%) and in P. falciparum/P. vivax and P. falciparum mixed infections with other Plasmodium spp. (OR: 6.96, 95% CI 1.50-32.4, p = 0.01, I 2 : 99%) (Fig. 6).  there was a symmetrical distribution of the included studies (geometric shapes) in the graph between the OR and SE (logOR) (Fig. 7). The publication bias was further assessed with Egger's test. Egger's test showed no publication bias due to the small-study effects found (p-value = 0.166) (Table S2). Therefore, the summary estimates of ORs in the present meta-analysis were not confounded by publication bias of the included studies.

Discussion
This is the first study to summarize the available data on the discrepancy between RDTs and two gold/reference standards for the detection of malaria mixed infections. The summary ORs of discrepancies of RDT types 2, 3, 4, 5, and 6 in detecting malaria mixed infections compared to microscopy were 4.33, 8.46, 0.99, 0.87, and 1.07, respectively. Even though the overall summary estimate of ORs was significantly observed, subgroup analysis of RDT types demonstrated that only RDT type 3 could detect a significantly higher proportion of Plasmodium mixed infections than the microscopic method. Among the 8 studies conducted in Ethiopia 27,30,31,38,42,45,48,50 , only a study by Ashton et al. 27 28 . The false positive on Pan-pLDH test lines among P. falciparum samples at high parasite densities may be possible to use as the detection limit of the SD Bioline Malaria Ag P.f/ Pan RDT used in the study by Mehlotra et al. because the mean parasitaemia level in samples that were positive for both the PfHRP2 and pan-pLDH test bands was significantly higher than that in those that were positive only for the PfHRP2 band 44 . In addition, the included study by Ashton et al., 2010, demonstrated the false-positive results in Pan-pLDH test lines of P. falciparum (38%) and P. vivax samples which might cause by high parasite densities (> 5,000 parasites/µl) 27 . Therefore, high P. falciparum or P. vivax parasitaemia could lead to incorrect interpretation of RDTs, particularly interpretation of mixed infections. The discordance between RDT types 2 or 3 and microscopy can be explained because RDT type 3 is specific to Pf-HRP2 and pan-pLDH and RDT type 2 is specific to pan-aldolase and thus cannot distinguish between a P. falciparum infection and a mixed infection when both test lines are observed. Other possible causes of discrepancy were false positive results from patients who had received any anti-malarial treatment in the previous four weeks as reported by the authors, parasitized erythrocytes cytoadhered to the microvasculature that were not seen in the peripheral circulation or on blood films although antigen continued to be released yielding RDT positivity 54 , or a low parasite density www.nature.com/scientificreports/ of the mixed infection that was too low to be seen by the microscopists but with sufficient parasite antigen to yield RDT positivity 55 .
The meta-analysis of RDTs and microscopy had no significant discrepancy among RDTs type 2, 4, and 6. In this analysis, the summary results of RDT type 5 performed by Iqbal et al. 46 and RDTs performed by Ranjan and Ghoshal 39 could not be interpreted because there were a small number of studies for subgroup analysis. Overall, the evidence was strong for RDT types 3 and 6 mainly because a large number of studies were available for inclusion. However, the summary estimate of RDT type 3 demonstrated high heterogeneity among the included studies (I 2 = 96%) when compared to those of RDT type 6 (I 2 = 0%). In this study, more than half of the studies (n = 18) relied solely on microscopy as the gold/reference standard for Plasmodium species identification. Therefore, the discordant results between RDTs and microscopy demonstrated in the present study might be due to the imperfect nature of the gold/reference standard because mixed infections with P. falciparum could be missed by microscopy. Because of these results, RDT types 2 and 3 could rectify the diagnosis of P. falciparum in mixed-species infections that might be missed by the microscopy method. These results supported that the selection of the most appropriate RDTs relative to malaria epidemiology and are very crucial to differentiated mixed infections because the identification of Plasmodium mixed-species infections would facilitate appropriate treatment with artemisinin-based combination therapies (ACTs), which could eliminate any mixed infection even if mixed infections were not detected by the gold/reference standard, the microscopy method 56 .
Recently, the sensitivity and specificity for the detection and identification of malarial parasites have been improved using the Nested-PCR method, which amplifies the 18s rRNA gene 57 . It has been proven to be more sensitive and accurate than routine diagnostic microscopy and provides the advantage of a higher proportion of detection in cases of mixed-species infections 57 . In the present study, 12 included studies used PCR as a reference standard for Plasmodium species identifications. The discrepancy between RDT type 3 and PCR (OR = 4.05) appeared to be heavily influenced by the included studies by Berzosa et al. and by Mehlotra et al. in which the individual ORs were extremely high (19.9 and 93.2, respectively). This affirms that when compared with using PCR as the gold/reference standard, the high discrepancy between RDT type 3 targeting Pf-HRP2 and pan-pLDH leads to incorrect interpretation of mixed infections by RDTs, as we discussed earlier in the discrepancy of RDT type 3 and microscopy. The false positive results of RDTs when detecting mixed infections may be associated with decreased age because of the high prevalence of malaria in children, particularly children under 5 years of age, who are likely to develop severe malaria with high parasitaemia 58,59 . The present meta-analysis demonstrated that the significant discordance between RDTs and PCR was found in studies using RDT type 6, which detects the pf-HRP2/Pv-pLDH antigen of malaria parasites. RDT type 6 could detect a lower proportion of Plasmodium mixed infections than the PCR reference method. This finding was similar to three previous studies 21,60,61 . Therefore, the lower proportion of Plasmodium mixed infections detected by RDT type 6 than by PCR demonstrated in the present study might be due to the lower sensitivity and specificity of RDTs than of PCR methods. In practice, PCR methods have a higher sensitivity (approximately 0.0001 parasites/µL) than RDT (approximately 100 parasites/µL) and microscopy (approximately 50-500 parasites/µL) 8 , which allows for the detection of Plasmodium mixed infections at a low parasite density, which are routinely missed in microscopy 62 . The subgroup analysis of Plasmodium mixed-species infections as reported by the 6 included studies demonstrated that a comparable proportion detected P. falciparum mixed infections with P. vivax between RDTs and PCR, while there was a significant difference in the proportion that detected P. falciparum/P. vivax and P. falciparum mixed infections with other Plasmodium species. This subgroup analysis suggested that RDTs had identical results with PCR in detecting P. falciparum and P. vivax mixed infections. In contrast, RDTs had discordant results with PCR in detecting P. falciparum mixed infections with other Plasmodium species. Nevertheless, these results should be further confirmed by full experimental studies.
The present study had limitations. First, RDTs targeting HRP-2 and pan-pLDH or RDTs targeting HRP-2 and pan-aldolase are likely to be positive in P. falciparum mono-infections or mixed-species infections. Regarding this limitation of the RDTs in the included studies, the summary estimates of ORs between RDT types 2 and 3 and microscopy need to be carefully interpreted. Second, the overall evidence of the analysis between RDTs and PCR was weak, mainly because few studies were available for inclusion. Second, the lower sensitivity and specificity of RDTs than those of PCR was due to the limits of detection. The WHO has suggested that the clinical sensitivity of RDTs is highly dependent on conditions including the level of parasite density and the subset of any population, such as young children or pregnant women; thus, the interpretation of RDTs must be carefully interpreted 63 . Third, the sensitivity and specificity of RDTs compared to the gold standard could not be calculated due to data on individual patient were lacking and the data on whether patients who gave positive results for RDT were the same patients who gave positive results for the gold/reference standard or not, as most of the included studies report the number of positive separately between RDTs and microscopy/PCR. Fourth, some eligible studies might have been missed through the search strategy. However, the additional search of reference lists of the included studies and searches of other sources such as Google search and Google Scholar, and performing extensive searching of reference lists and searching other sources with broad search terms, helped to reduce this limitation by further increasing the number of included studies. Fifth, the study aimed to clarify what proportion of Plasmodium mixed-infections could not be confirmed by a positive RDT result, and the proportion of Plasmodium mixed infections were often not the primary target of studies, which led to a low number of studies that were focused on mixed infections. In light of these, although the current data are still suggestive of high discrepancies of RDT type 3 for detecting Plasmodium mixed infections in comparison to microscopy and of RDT type 6 for detecting Plasmodium mixed infections in comparison to PCR methods, they provided a critical advantage on malaria treatment in resource-limited settings in which the results of microscopy could not be obtained. Further studies focused on the diagnosis of Plasmodium mixed-species infections by RDTs are needed to provide a better understanding of the performance of RDTs, guide the development of an improved diagnostic Scientific RepoRtS | (2020) 10:12765 | https://doi.org/10.1038/s41598-020-69647-y www.nature.com/scientificreports/ test for Plasmodium mixed infections, and facilitate the appropriate treatment of patients with ACTs. This will help with the elimination of malaria in endemic and non-endemic areas where laboratory capacity is limited.

conclusion
In conclusion, the present study suggested that malaria RDTs showed some discordant results with microscopy and PCR. The selection interpretation of RDTs can facilitate a better diagnosis of Plasmodium mixed-species infections and appropriate treatment of malaria patients in endemic and non-endemic settings.
consent for publication. All authors have read the manuscript and consent to its publication.

Data availability
The datasets used during the current study are available without restriction and demonstrated in the present manuscript and additional files.