Rapid immunochromatographic tests for the diagnosis of dengue: a ...

Introduction

Dengue is an acute viral disease caused by a virus transmitted mainly by Aedes aegypti. This arthropod-borne flavivirus has four distinct serotypes: DENV-1, DENV-2, DENV-3, and DENV-4, which constitute an antigen complex of the Flavivirus genus, Flaviviridae family 1.

For more information, please visit Accu-Tell.

Dengue virus is present in more than 100 countries of the Asia-Pacific, Americas, Middle East, and Africa 2,3,4, with 3 billion people (40% of the world population) at risk of infection in tropical and subtropical regions, with 50 to 100 million infections per year 2,4,5. It is an important arthropod-borne viral disease in terms of human morbidity, mortality and economic impact. Many challenges remain concerning disease control and prevention programs based on vector reproduction and elimination, clinical aspects and pathogenesis 5.

The clinical presentation of dengue infection is highly unspecific varying according to the circulating serotype 5. Differential diagnosis of dengue in urban areas of large metropolises in Latin America, where malaria is not endemic, includes influenza 6,7. In Brazil, since 8, also zika and chikungunya are co-circulating 9, making the diagnosis on a clinical basis unreliable. Thus, diagnostic optimization for adequate clinical management to prevent complications caused by dengue requires better, easier and more efficient rapid tests with good accuracy for case management during the earlier state of infection.

Among the rapid tests, those using the immunochromatographic technique (ICT) to detect the presence of nonstructural protein 1 (NS1) play an important role in early diagnosis of dengue fever (up to seven days from the onset of symptoms) 10. Reference standards such as virus isolation, PCR or PRNT have the great disadvantages of being laborious, time consuming, require specific reagents, equipment, trained personnel and are high cost. ELISA IgM/IgG has been important for health surveillance and distinguishes between primary and secondary infectious in cases previously confirmed by RT-PCR or virus isolation but presents cross-reactivity with other members of the Flaviviradidae family 6.

We found five systematic reviews with meta-analysis on the subject 4,6,11,12,13. Alagarasu et al. 11 included only publications on IgA ICT. Another meta-analysis included nine studies on NS1 ICT 4 and the systematic review by Blacksell et al. 6 assessed a single commercial test (Panbio ICT - Abbott Laboratories) in 11 studies, showing wide variability between them. These reviews point out the high specificity of the ICT, but with heterogeneous sensitivities, requiring a critical assessment that includes the various types of ICT and brands available on the market as well as their evaluation in acute and convalescent samples. In fatal cases, NS1 strip showed better sensitivity (78.3%) than ELISA NS1 10.

A recent systematic review 12 on the economic impact of dengue’s ICT favored a relatively obsolete diagnostic strategy based on IgM Panbio for acute cases. However, it identified only two studies, one using primary observational data 14 and the other, a simulation modeling design 15.

In children, when it could be difficult to access blood samples, some studies were carried out in saliva and urine 16,17. Muso et al. 17 suggested that only 19% of the studies detected zika virus in saliva, concluding that it could not replace blood tests. In a recent review, Colonetti et al. 18 included three studies for dengue diagnosis evaluating salivary IgM, which provided sensitivity of 86% and specificity of 93%. Two included studies evaluating salivary IgA showed a pooled sensitivity of 69% and a pooled specificity of 98%. Despite these results and the low methodological quality of the studies included in the meta-analysis, the authors concluded that it is still soon to claim that IgA is better than IgM to diagnose dengue 18.

This study aimed to review the literature on the accuracy of ICT using as the reference test any type of PCR, ELISA, or virus isolation, in suspected dengue cases with up to seven days since the onset of fever for NS1 ICT and with no restriction on the days of fever for IgA, IgM/IgG, or NS1/IgM/IgG ICT.

Methods

This was a systematic literature review of observational diagnostic studies reported in the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement 19. The protocol was previously registered on the site PROSPERO number CRD.

Data sources and search strategy

The research question was: Are point of care immunochromatographic tests accurate for early detection of dengue infection? Does the test performance vary according to age, sex, dengue serotype, reference tests or whether it is a primary or secondary infection, acute or convalescent phases? These questions guided the eligibility criteria expressed in the PICO (Patient, Intervention, Comparison and Outcomes) format:

Population: blood/serum or plasma samples from patients with febrile illness suspected of dengue with up to seven days of fever in the acute phase of the disease and with no time limit in the convalescent phase;

Intervention (index tests): ICTs with detection of IgA, NS1, IgM/IgG, or NS1/IgM/IgG, read within 60 minutes;

Comparator (reference standard): PCR, ELISA NS1 or IgM, virus isolation, or a combination of two or three of these;

Outcome (diagnostic parameters): sensitivity, specificity, likelihood ratios, and positive and negative predictive values in ICTs for dengue, besides the information on time and effect measures, according to the case.

We excluded articles that: use inappropriate reference tests, index test limited to the detection of IgG antibodies or that takes more than 60 minutes to perform, incomplete description or partial examination of sample, small sample size or insufficient data to calculate accuracy parameters.

In case of doubt we directly contacted the authors. We did not limit the search based on study design nor on language of publication.

Two researchers conducted the searches up to October for journal articles or congress proceedings publications since inception in MEDLINE via PubMed, Science Direct, Scopus, Web of Science, Ovid MEDLINE JBrigs, SCIRUS, BIREME and EMBASE, with no restriction on language or study design. We also searched gray literature using Google Scholar. Our search strategy in MEDLINE via PubMed employed the keywords: (“dengue/diagnosis”[MeSH Terms]) AND (diagnostic reagents and test kits [MeSH Terms]), generating the following strategy: “humans”[MeSH Terms] AND (“Dengue” OR “Dengue Virus”) AND (sensitiv*[Title/Abstract] OR specificity[Title/Abstract] OR “sensitivity and specificity”[Mesh Terms] OR “Reference Values”[Mesh] OR diagnosis*[Title/Abstract] OR diagnosis[Mesh] OR diagnosis[Subheading]) AND (((“Serologic Tests” OR Immunoassay OR “Reagent Kits, Diagnostic”) AND (Bedside OR Rapid)) OR “Point-of-Care Systems” OR “NS1” OR “NS-1” OR “Viral nonstructural proteins” OR Immunochromatogra* OR Immunochromatography OR bioeasy OR bioline OR bioline OR panbio OR core OR ag-strip OR strip OR Duo OR biorad OR “Reagent Strips”). We used equivalent strategies in the other databases and employed Zotero Standalone 4.0 for Windows (https://www.zotero.org/) in the search and filing of references.

Study selection

Initially, three pairs of reviewers (V.E.M./C.A.F.A., L.V.B.F./S.R.L.P., and Y.H.M.H./S.R.L.P.) independently selected the study abstracts. We held consensus meetings, and in case of disagreement, a third reviewer external to the pair judged the article’s relevance. In the second stage, pairs of reviewers (V.E.M./C.A.F.A., V.E.M./S.R.L.P., and Y.H.M.H./S.R.L.P.) read the full-text articles, also independently. Disagreements arising in the consensus meetings of the respective pairs were also resolved with a third external reviewer.

Data extraction and assessment of risk of bias

We designed a standardized form to extract the following variables by the pairs of reviewers: study design, commercial test names, test manufacturing countries, type of detection used, reference test used, number of study participants, number of confirmed dengue cases, non-dengue cases, measures of accuracy, virus serotype, and time since onset of fever.

We used the Quality Assessment of Diagnostic Accuracy Studies (QUADAS 2) 20 to assess the quality of the selected articles, risk of bias, and applicability. The tool consists of 14 items distributed across four domains that assess patient selection, index test, reference test, flow and timing.

Data synthesis and analysis

We used the “reference standard” defined in each selected study for comparison with the index test to determine the true-positive (TP), false-positive (FP), false-negative (FN), and true-negative (TN) values. Diagnostic accuracy, sensitivity (Sn), specificity (Sp), positive predictive value (PPV), negative predictive value (NPV), positive (LR+) and negative likelihood ratios (LR-), diagnostic odds ratio (DOR). We estimated positive (PP+) and negative post-test probabilities (PP-) in scenarios of 25%, 50%, and 75% prevalence.

For each ICT (IgA, NS1, IgM/IgG, NS1/IgM/IgG), we performed a meta-analysis for each measure of diagnostic accuracy listed above, with the respective 95% confidence intervals (95%CI). The analyses were performed with the Winpepi (http://www.brixtonhealth.com/pepi4windows.html) and Stata XIV (https://www.stata.com) packages using the MIDAS command (Meta-analytical Integration of Diagnostic Accuracy Studies) performing the bivariate mixed-effects binary regression modeling framework. Meta-analyses were conducted according to the different analytes and/or brands.

We calculated the I2 statistic to detect significant overall and inter-subgroup heterogeneity 21. We considered I2 values greater than 50% as high evidence of heterogeneity in data. In the presence of I2 point estimate higher than 50%, we performed meta-analysis using random effects model 22.

We analyzed study heterogeneity graphically and through the I2 test. We explored possible causes of clinical heterogeneity between studies through subgroup analyses: disease phase (acute or convalescent), by the most extensively assessed brand name, and overall quality of studies according dimensions of QUADAS 2 (low versus high or unclear risk of bias) 20.

Assessment of publication bias used the Deeks graph, where p-value < 0.05 was considered significant 23.

Results

Characteristics of included studies

The initial search identified 3,791 publications. After removing duplicates, we reviewed 3,783 abstracts, and selected 108 articles for reading the full-texts, of which 57 were selected for this review ( Figure 1). The studies assessed multiple ICT brand tests with different analytes: five assessed IgA 24,25,26,27,28, 21 NS1 10,27,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47, 12 IgM/IgG 27,34,48,49,50,51,52,53,54,55,56,57, and 25 NS1/IgM/IgG 29,36,37,40,46,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72,73,74,75,76,77 ( Table 1). The total number exceeds since some studies evaluated more than one ICT brand tests and type of analyte. Those articles evaluating NS1/IgM/IgG estimated not only the accuracy parameters for the three analytes, but also for each analyte separately.

Although planned, stratified analysis was not available in original studies, except for different analytes.

The 57 studies were performed mainly in Asia (33; 57.9%) and the Americas (18; 31.6%), only one in Oceania and mostly (94.1%) published in English.

The included studies analyzed 29 ICTs, using as the reference tests RT-PCR, real-time PCR, semi-nested PCR, NS1 ELISA, IgM ELISA, IgG ELISA, IgM antibody capture enzyme-linked immunosorbent assay (MAC ELISA IgM), IgG antibody capture enzyme-linked immunosorbent assay (GAC ELISA IgG), or virus isolation ( Table 1).

Quality assessment of the studies

According to the assessment of methodological quality conducted with the QUADAS 2 tool, of the 57 included studies, only six 29,30,38,39,69,78 did not show risk of bias, and 25 (43.8%) of the them showed high risk of bias regarding the patient selection process ( Figure 2), mainly due to case-control design. Ten of them showed high risk of bias concerning flow and timing, mainly for excluding patients from analysis or for adopting inappropriate intervals between index and reference tests. Concerning reference standard, 31 studies were unclear and three showed high risk of bias, mainly due to not informing about blinding.

For more information, please visit Dengue Combo Rapid Test Cassette.

However, we did not find any major conflicts that could compromise applicability in relation to patients included, index or reference tests in these studies from those targeted by our review questions.

Rapid immunochromatographic tests with IgA detection

A total of 2,051 samples from patients with suspected dengue virus infection were analyzed (median 342, interquartile range [IQR]: 100-914) in the five studies selected for this part of the review 24,25,27,27,28. One of them showed results in acute and convalescent samples 24. Pooled estimate of the IgA tests showed a sensitivity of 88% and specificity of 90% ( Table 2). It was not possible to assess publication bias for these tests due to the small number of studies included in the analysis. The pooled estimate in the acute phase showed slightly higher sensitivity (92.8 vs. 88) and the same specificity (90%) compared with the analysis which included convalescent samples. The performance of this test for screening was better than NS1 or IgM/IgG due to better sensitivity ( Table 2), but lower than tests with three analytes.

Forest plots ( Figure 3) showed similar results between studies, except for one case-control 26 which included mainly primary infections compared to secondary infections (5:1), despite high statistical heterogeneity (I2 = 93%). IgA ICT tests in scenarios with prevalence of 25% showed the positive post-test probability still moderately high (75%) compared to conclusive (90% and 96%) results in epidemic scenarios ( Table 2). Besides that, the negative post-test probabilities were reasonable up to 12% and 18% even in outbreaks ( Table 2).

Only one study 26 reported the serotypes tested ( Table 1). This study assessed the performance according to serotype (DENV-1 and 2), showing heterogeneous sensitivities (Sn = 52.4% in DENV-1 and 73.9% in DENV-2).

Three studies 24,26,27 included primary and secondary dengue infection cases without stratified analysis.

Rapid immunochromatographic tests with NS1 detection

Tests based exclusively on NS1 evaluated three brands up to : Bio-Rad, Panbio, Alere/Bio_Easy. These totalized 21 studies up to the seventh day of the disease 10,27,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,50, one of which 34 presented the results of two settings, one in Malaysia and the other in Vietnam. The tests were assessed in 6,618 samples from patients with suspected dengue (median 241). Of 21 studies, 18 reported the serotypes tested, totaling 852 samples of DENV-1, 582 DENV-2, 501 DENV-3, and 510 DENV-4 ( Table 1), but did not show stratified performance analysis.

The pooled estimates for all NS1 tests showed sensitivity of 74%, and higher specificity of 99%. The lower sensitivity values were obtained for NS1 Bioeasy in a Brazilian sample of DENV-4 outbreak 30 as well as for the brand Asan 37.

Bio-Rad Dengue Rapid Test was used for NS1 detection in 14 of the 21 studies 10,31,32,33,35,39,40,41,43,44,45,46,47,50 (4,678 samples). Sensitivity ranged from 49.4% 39 to 98.9% 31 and specificity from 91% 35 to 100% in 8 studies 32,33,35,39,41,43,45,46. The pooled estimate for the Bio-Rad Dengue Rapid Test showed sensitivity of 79% and specificity of 100% ( Table 2). The post-test probability after a positive result in NS1-based ICT was above 95% in three different hypothetical scenarios of dengue prevalence of 25%, 50% and 75%.

Several recent studies tested SD Duo Bioline ICT but only showed NS1 results. We opted to describe these on Table 1, but to exclude them from the meta-analyses since there was not blinding of other analyte results in the same cassette.

Assessment of the individual studies did not show publication bias (p-value = 0.09).

Rapid immunochromatographic tests with IgM/IgG detection

Seven studies assessed tests with both IgM/IgG detection 27,48,50,51,52,53,54, using 2,597 samples (median 178). Seven studies identified the dengue serotypes, with a total of 251 DENV-1, 176 DENV-2, 193 DENV-3, and 77 DENV-4. Most studies except one evaluating exclusively IgM/IgG ICT were published up to ( Table 1).

These tests presented the lowest values of pooled estimates of sensitivity (54%), with inadequate values of negative likelihood ratios (NLR > 0.4) ( Table 2). Thus, the post-test probabilities after negative results were inconclusive, particularly for epidemic scenarios of prevalence. In the convalescent phase of the disease, the pooled estimate of accuracy showed, as expected, higher sensitivity (Sn = 62.6%, 95%CI: 36.7-82.9), than in the acute phase, 53.8% (95%CI: 41.4-65.8), and high specificity in all phases of the disease (94% and 94.7%,). Specificity was lower for recent studies 27,54.

Panbio Dengue Duo IgM/IgG was the most widely assessed test, with pooled sensitivity and specificity of 56% and 90% ( Figure 3; Table 2).

We detected no publication bias (p = 0.13).

Rapid immunochromatographic tests with simultaneous NS1/IgM/IgG detection

Ten studies that assessed that type of test included a total of 3,361 patients (median 447) with suspected dengue, with 289 DENV-1, 225 DENV-2, 52 DENV-3 and 39 DENV-4 36,37,40,46,58,59,69,70,77,79 ( Table 1).

The best performance was observed for these tests with pooled positive and negative likelihood ratios, of 19.2 and 0.09, respectively. The post-test probability after negative and positive results in endemic (25%) and epidemic (75%) scenarios of dengue prevalence were below 25% and above 85%, respectively. The pooled estimate of sensitivity was 91% and specificity, 96% ( Table 2). Carter et al. 79 obtained the poorest performance in sensitivity. After excluding it, the pooled results were unchanged, Sn = 92% (87-95%) and Sp = 96% (92-98%).

Some recent studies also reported results for each analyte separately even when testing ICT composed of a cassette with three analytes. We describe these “only results” on Table 1 without including these meta-analyses, since this was only a statistical analysis and not a practical use of a test with a single analyte in a cassette.

We observed no asymmetry in the assessment of publication bias in the studies (p-value = 0.09).

Discussion

This was a systematic review addressing the dengue virus detection methods in commercially available ICTs, obtained through a search of nine large databases, with 57 studies included. One strategy used to increase the tests’ performance was the simultaneous test of the three analytes NS1, IgM, and IgG 40,46,56,58,67,71. In our review, these ICTs showed high pooled estimates, better than those of IgA ICTs. Among the ICTs with serological detection assessed in this review, those with IgA detection stood out as having the best accuracy, with high pooled sensitivity and specificity in the acute phase compared to IgM/IgG ICT.

IgA tests showed the best performance in triage of patients in acute phase of the disease. They were twice as positive among cases with up to seven days of dengue fever when compared to those in the convalescent phase. Still, these studies did not analyze the tests according to phase of disease (acute/convalescent), thus making it impossible to claim that this same performance would be maintained in the initial days of the disease.

The current review showed an excellent pooled specificity (99%-100%) in the acute phase of the disease in ICTs with exclusive detection of NS1, six times more positive among dengue cases when compared to IgA ICTs during the same phase of the disease. These findings corroborate those of Lima et al. 10, who suggested the best performance of NS1 to confirm dengue cases in the acute phase of disease.

The systematic review published by Alagarasu et al. 11 assessed IgA ICTs, including three studies with lower estimate sensitivity of 72% and similar specificity (89%). However, the wide confidence intervals in the measures of accuracy both in our review and in Alagarasu et al. 11 make its use for screening questionable.

In recent years, several authors have questioned the use of IgM/IgG serology to detect dengue and other flaviviruses, due to the tests’ proven cross-reaction with the Zika, yellow fever, and chikungunya viruses, thus limiting their use in scenarios with co-circulation of these viruses 6,80,81.

The systematic review by Zhang et al. 4 showed pooled estimates to these NS1 ICTs similar to our review, with Sn = 71% and Sp = 99%. Both in Zhang et al. 4 and in our review, the performance of NS1 ICT in scenarios with 25%, 50%, and 75% of dengue prevalence pointed to increasing positive post-test probability, ranging from 99 to 100%. When used in screening, these tests should be coupled with a diagnostic algorithm in order to optimize their performance, due to the high number of false-negatives 4.

The accuracy of IgM/IgG ICTs had the worst performance and studies about this ICT were interrupted in . The systematic review by Blacksell et al. 6 assessed the Panbio ICT in the acute phase of the disease and the summary measures were superior to those in our review. Among other factors, these differences can be attributed to the samples’ characteristics related to the convalescent phase or samples with mostly primary infection 6,11.

Only two studies included in this review reported a potential conflict of interest 31,46. Only one 31 reported sensitivity results that differed from the pooled sensitivity in our review.

In addition to the review’s originality, one of its strengths was the scope of the literature search, which included all types of commercially available ICTs for dengue detection, with subgroup analysis according to the ICT detection method in each of the principal commercial ICTs, and when possible, according to the phase of the disease (acute/convalescent).

The review’s limitations include the low methodological quality of the included studies and the lack of data for adequate characterization of the samples (27/34, 79.4%), either by age bracket (21/34, 61.8%) or dengue serotype (16/34, 47.1%), which prevented such subgroup analyses. Another limitation was the high heterogeneity detected in all the types of ICTs that were assessed, possibly due to the differences between the characteristics of the samples included by the studies. These differences were related to the age of the included patients, predominant type of infection (primary or secondary), serotypes assessed, disease phase assessed by the tests (acute/convalescent), and different reference tests (real-time PCR, RT-PCR, in-house ELISA, MAC-ELISA, among others). This heterogeneity may not be explained by the different reference standards since only three studies did not used at least one test with high specificity (100% for RT-PCR or ELISA NS1) 10. Thus, the sensitivity of ICTs does not seem to be penalized by the reference standards. Similarly, the almost perfect specificities of ICTs were not influenced by non-optimal sensitivities (89.5%) of reference tests.

The three systematic reviews that included ICTs pointed to the same limitations described above 4,6,11. Guidelines like the Standards for Reporting Diagnostic Accuracy Studies (STARD) 82 and tools like QUADAS 2 20 have contributed to the standardization of reporting by accuracy studies, as indicated by Blacksell et al. 83. We emphasize that peer-reviewed journals and regulatory agencies should require the use of both these guidelines in order to assist future reviews and the elaboration of recommendations or protocols. Future studies should investigate cost-effectiveness, decision tree or a combination of multiple tests, including ICT in the diagnostic algorithm.

If you want to learn more, please visit our website Chagas Rapid Test.