High value of rapid diagnostic tests to diagnose malaria within children: A systematic review and meta-analysis

Background Children aged under five years accounted for 61% of all malaria deaths worldwide in 2017, and quicker differential diagnosis of malaria fever is vital for them. Rapid diagnostic tests (RDTs) are strips to detect Plasmodium specific antigens promptly and are helpful in resource-limited areas. Thus, our aim is to assess the diagnostic accuracy of RDTs for malaria in children against the gold standard. Methods MEDLINE, Web of Science, EMBASE, Cochrane Library, the China National Knowledge Infrastructure, Wanfang, and Sinomed databases were systematically searched on August 23, 2019. Studies that compared RDTs with microscopy or polymerase chain reaction in malaria diagnoses for children were eligible. Relevant data were extracted. The quality of studies was evaluated using the revised Quality Assessment of Diagnostic Accuracy Studies instrument. Meta-analyses were carried out to calculate the pooled estimates and 95% confidence intervals of sensitivity and specificity. Results 51 articles were included. For diagnostic accuracy, the pooled estimates of the sensitivity and specificity of RDTs were 0.93 (95% confidence interval (CI) = 0.90, 0.95) and 0.93 (95% CI = 0.90, 0.96) respectively. Studies were highly heterogeneous, and subgroup analyses showed that the application of RDTs in high malaria transmission areas had higher sensitivity but lower specificity than those in low-to-moderate areas. Conclusions RDTs have high accuracy for malaria diagnosis in children, and this characteristic is more prominent in high transmission areas. As they also have the advantages of rapid-detection, are easy-to-use, and can be cost-effective, it is recommended that the wider usage of RDTs should be promoted, especially in resource-limited areas. Further research is required to assess their performance in WHO South-East Asia and Americas Region.


VIEWPOINTS PAPERS
standard. The exclusion criteria were as follows: (1) Studies that were case reports, reviews, editorials, letters, comments, and conference abstracts. (2) Studies that did not present enough information to extract or calculate the number of true-positives, false-positives, true-negatives, and false-negatives. The title and abstract of all relevant articles were read by three reviewers independently according to the inclusion and exclusion criteria during the first round of screening. Then the full texts of the eligible studies were rechecked based on the same criteria. Any disagreement between three reviewers was resolved by discussion.

Data extraction and quality assessment
The following data were independently extracted by three reviewers from eligible studies using Microsoft Excel 2016 (Microsoft Inc, Seattle WA, USA): (1) study characteristics: journal, publication year, first author and his/her institution, study period, study setting, and study design. (2) participants' characteristics: the inclusion and exclusion criteria, sample size, the number of malaria cases, the age range and sex distribution of participants, and the parasite density of Plasmodium. (3) RDTs characteristics: commercial brand, and specific Plasmodium species and antigens detected. (4) RDTs performance: the reference standard, and the number of true-positives, false-positives, true-negatives, and false-negatives. Any discrepancy between three reviewers was resolved by discussion. If only a subset of participants met the selection criteria, data were extracted only for the subgroup.
The revised Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) tool was used to assess the methodological quality of the eligible studies [23], as recommended by the Cochrane Collaboration. The tool has four domains: patient selection, index test, reference standard, and flow and timing. Each domain was scored as "high/low risk of bias" and "high/low applicability concern", except for the last one which only contains the risk of bias section. If insufficient data were reported, the corresponding section would be classified as "unclear". Reviewers assessed the quality of studies independently using Review Manager 5.3 and discussed the inconsistencies. The criteria for each section are listed in Appendix S3 in the Online Supplementary Document.

Statistical analysis
We estimated the sensitivity and specificity of each study with 95% confidence intervals (CI) and presented the results in forest plots. Then we used the Midas module in Stata 12.1 (StataCorp LLC, Texas, USA) to calculate the pooled estimates of the sensitivity, specificity, positive likelihood ratio, negative likelihood ratio, and diagnostic odds ratio. Midas is a comprehensive program for undertaking meta-analysis of diagnostic test accuracy in Stata. Its primary data synthesis is based on the bivariate mixed-effects regression framework. A hierarchical summary receiver operating characteristic curve (HSROC) was fitted and funnel plots were presented respectively to show the comprehensive diagnostic value of RDTs and the potential publication bias among eligible studies. Compared with the summary receiver operating characteristic (SROC) model, the HSROC model allows more between and within-study variability [24]. It was adopted as studies included are expected to show considerable heterogeneity in diagnostic accuracy [25]. We also performed the Q test to assess the heterogeneity among the included studies. The extent of heterogeneity was quantified by I 2 measure [26]. If the heterogeneity was significant (I 2 >50%), we used Meta-disc 1.4.0 software (the Unit of Clinical Biostatistics team of the Ramón y Cajal Hospital, Madrid, Spain) to explore whether a threshold effect existed. Furthermore, meta-regression was conducted to investigate the potential sources of heterogeneity. Covariates included local malaria transmission type, study design, sampling method, reference standard, sample size, geographic location, blinding status, RDTs type, and target antigens. In the regression, the accuracy measure was relative diagnostic odds ratio (RDOR). The coefficients of covariates indicated the change in the diagnostic performance of the RDTs under each study per unit increase in the covariates. In other words, P < 0.05 represented that the corresponding covariates were the major sources of heterogeneity. Subgroup analyses based on the sources of heterogeneity were subsequently conducted by Stata (Stata Corp, College Station, TX, USA).
For meta-regression and subgroup analyses, transmission type was divided into four categories: high, low-to-moderate, mixed and unclear. The transmission type was classified as "high" if the authors described it as "hyperendemic", "perennial", "holoendemic" or "high"; "low-to-moderate" when it was "mesoendemic", "sporadic", "moderate" or "low"; "mixed" if it was described as "seasonal" or when multiple sites of different transmission types were included; "unclear" if the transmission type was not mentioned.

Results of the search
A total of 9731 relevant articles were identified. After removing duplicates, 5933 articles were selected. 5861 articles were excluded based on the criteria. The full texts of 72 articles were evaluated and 51 of them were eventually included. Among them, the most common reason for exclusion was the lack of data for a 2x2 table. The detailed process for selection is shown in Figure 1.
Out of 51 studies included in the review, two evaluated the validity of RDTs when they were used to monitor the effects of ACT treatment, and the others assessed the diagnostic capacity of RDTs. Since HRP2 will be cleared slowly from bloodstream if the treatment of P. falciparum is successful and it can contribute to a higher false-positive rate of RDTs [27], our analyses were grouped into two parts according to the usage time of RDTs (before or after ACT treatment). There were 9 studies that adopted multiple types of RDTs and/or reference standards: 6 studies evaluated 2 types of RDTs [28][29][30][31][32][33], 2 studies evaluated 3 types of RDTs [18,34], and 2 studies selected two different types of reference standards [32,35]. Particularly, in one study, two staffs read the RDTs strips respectively, so there were two different results for each type of RDTs [29]. As enrolled children were asked to retest blood samples during the follow-up regularly, the studies that assessed the RDTs capacity for monitoring the effect of ACT treatment all have multiple test evaluations. As a result, we had 82 test evaluations reporting a total of 57 312 test results. Among them, 34.45% of tests (19 746) showed the positive result against the reference standard.
Most of the studies included were conducted in Africa (n = 47), and the rest of them happened in Asia (three in India and one in Pakistan). The detailed characteristics of the included studies were summarized in Table 1.

The methodological quality of the included studies
The overall methodological quality of studies included was relatively high, as the scores of risk of bias and applicability concerns in the four domains were mainly low. Most studies (n = 41) enrolled consecutive or random sample of patients. None of them were case-control studies. 17 studies adopted double-blind method, and 1 study was single-blind, but the rest did not supply sufficient information. 6 studies selected PCR as the reference standard, 43 studies took microscopy as the reference standard, and 2 studies chose both. 31.37% (16/51) of the eligible studies were scored to have a high risk of bias in the flow and timing domain since they did not include all patients in the analysis. 35.29%(18/51) of included studies had a high applicability concern in patient selection domain because of their unrepresentative samples. The results of the methodological quality were shown in Figure 2, Panel A and Panel B.
Sensitivities of tests ranged from 0.00 to 1.00, and specificities from 0.08 to 1.00 (Figure 3).    Figure 4. The area under the curve is close to 100%, indicating that the performance of RDTs was satisfactory. High heterogeneity was observed between studies (Cochrane' s Q = 2182.22, I 2 = 100.00, P < 0.001), thus we explored its source through the threshold effect analysis and meta-regression. The results suggested that there was no threshold effect between studies (P = 0.06), while transmission type, sampling method and study design were the major sources of heterogeneity (P < 0.05). The results of the meta-regression were shown in Table 2.
In addition, the effect of each variable on the accuracy of RDTs was presented by forest plots if the variable was categorical ( Subgroup analyses were conducted and the results were shown in Table 3. In brief, RDTs conducted in high malaria transmission areas had higher sensitivity but lower specificity compared to low-to-moderate areas. The studies with consecutive or random sample of patients presented higher sensitivity than others.   Both sensitivity and specificity estimated by prospective cohort studies appeared to be higher in comparison with cross-sectional studies. A funnel plot was presented in Figure 7. It demonstrated the existence of publication bias (P = 0.04), and it was found that studies with high accuracy results tended to be published.
Many studies discussed the diagnostic value of HRP2 based RDTs vs LDH based RDTs [73,74], and the problem about which type of RDTs is better still exists. Although target antigens were not the major sources of heterogeneity, a subgroup analysis was performed based on it. Results showed that HRP2 based RDTs had higher sensitivity but lower specificity than RDTs that did not contain HRP2 ( Table 3). But there was no statistically significant difference.

RDTs capacity for monitoring the effect of ACT treatment
19 tests evaluated RDTs' capacity of monitoring the effect of ACT treatment [33,60]. All were conducted in Africa. Fourteen tests took microscopy as the reference standard and five took PCR.
Since the days after initial treatment is an important factor affecting the accuracy of RDTs, we analyzed the results based on this framework. Consequently, after categorizing based on the follow-up period, there was no more than 3 tests within each category, and thus we could not perform a meta-analysis (To perform Midas, a minimum of four 2 × 2 tables is required). In the lack of statistical pooling, we presented the findings in a narrative table (Table 4).
In short, the specificity of HRP2 based RDTs increased with the follow-up period. And at early stages after the initial treatment, the specificity of Pf-LDH based RDTs was much higher than HRP2 based RDTs.

DISCUSSION
Our results demonstrated that RDTs had relatively high sensitivity and specificity for malaria diagnosis in children and all the findings were reported based on the PRISMA Checklist (Appendix S4 in the Online Supplementary Document) [75]. Since there is no previous systematic review focused on children,   the results were compared with those of the whole population. Abba' s research found that the sensitivity of RDT varied between 0.915 and 0.995, while its specificity ranged from 0.906 and 0.987 [9], which is comparable to our results. Moreover, in high transmission areas, the sensitivity and specificity were higher among children (0.96 and 0.92, respectively) than the whole population (0.937 and 0.896, respectively) [9]. This may relate to the fact that because adults have greater immune status than children, adult patients with malaria is more likely to have lower parasite density [60,76], and it could be difficult for RDTs to detect the low concentration of antigens among them. This characteristic makes RDTs more suitable for childhood malaria detection in high transmission areas. Another research conducted by Li calculated the accuracy of HRP2 based RDTs [15]. Comparatively, it had lower sensitivity (0.94 vs 0.96, respectively) but higher specificity (0.91 vs 0.86, respectively) in children than in adults.
Besides high diagnostic accuracy, RDTs also have the advantages of rapid detection and are easy-to-use, making it feasible to utilize it at primary health care centers. These advantages can be particularly important for P. falciparum detection, as it can progress rapidly from an uncomplicated febrile illness to potentially deadly disease [77]. Furthermore, compared to microscopy or PCR, the diagnostic cost of RDTs is relatively low, with a low cost of RDT strips and the training fees for laboratory staff. A few studies have been undertaken to evaluate the economic value of RDTs and they demonstrated that in comparison with microscopy, RDTs are more cost-effective if the whole treatment course have been taken into account [78][79][80]. Therefore, as most of the malaria-endemic areas have limited resources, RDTs is of high value to be used there. For instance, in a large proportion of African lower-level health facilities, technical expertise and microscopy were not available for children [81]. Likewise, almost half of the suspected malaria patients seek care in the private sector in Africa [82], which could be even less equipped.
Considering the endemicity of malaria, RDTs performed in high transmission areas had higher sensitivity but lower specificity than those conducted in low-to-moderate areas. This may be because low-density infection represents a significant proportion of malaria infections among children in low-transmission settings [83,84], leading to a higher false-negative rate. Furthermore, for HRP2 based RDTs, the remaining HRP2 antigen will last for several weeks in peripheral blood after a successful treatment, leading to false-positive results [27,57]. This is more common in high transmission areas since the children there may be infected with P. falciparum several times across their lives [28,60].
Though we did not impose any restriction on the country or region, only four studies conducted in Asia were included, and the rest of them were all performed in Africa. However, each endemic area has its own epidemiological characteristics, and the evidence of Africa cannot verify the applicability of RDTs in other VIEWPOINTS PAPERS areas. For instance, in the WHO South-East Asia Region, where the incidence rate was 7.0 per 1000 population at risk in 2017, both P. falciparum and P. vivax were dominant parasites [3,85]. P. knowlesi infection was also widely distributed there [86]. Meanwhile, most countries are confronted with the problem of limited resource. For example, India carries a high proportion of disease burden, however, microscopies were not accessible for suspected children in poor, remote villages [36,44]. Another endemic area is the WHO Americas Region, where the incidence rate was 7.3 per 1000 population at risk in 2017 [3]. Evidence demonstrated that a large proportion of P. falciparum lacked pfhrp2 or pfhrp3 or both genes there [87], which may lead to invalidity of HRP2 based RDTs. Therefore, corresponding research conducted in these areas is urgently needed.
There are two limitations to be considered in this study. First, since the parasite density of patients is a critical factor for the sensitivity of RDTs, we intended to perform a subgroup analysis. However, almost half of the included studies did not report the geometric mean parasite densities of patients, so linear regression could not be performed. Furthermore, it seemed that there was no widely-recognized standard for the classification of Plasmodium parasite density, and most of the studies classified it differently. Also, because none of the studies provided individual-level data, we could not classify the parasite density by ourselves. As a result, we could not add this factor into meta-regression and subgroup analyses, which might introduce bias. Second, our findings may be more transferable to Africa as most of the included studies were conducted there.

CONCLUSIONS
This systematic review shows the high value of RDTs in malaria diagnosis among children. Considering current prevalence of malaria, RDTs should be a suitable diagnostic test for children, especially in resource-limited areas.