Background
Dengue is an emerging infectious disease endemic to more than 100 tropical countries [
1]. An estimate of up to 390 million dengue infections happened in the year 2010 alone, of which only a quarter were detectable [
2]. Dengue threatens a pandemic with the spread of Aedes mosquito, the vector that carries the pathogen dengue virus that comes in four serotypes (DENV1–4), to subtropical regions throughout the world [
3].
In Malaysia, dengue incidence stood as high as 396.4 cases/100,000 population with case fatality rate ranging from 0.20–0.28% in recent years. In 2015 alone, dengue virus infected up to 120,000 people and caused 336 deaths [
4]. Despite an absence of medical treatment to date, early disease recognition and timely intervention with proper fluid management and supportive care can prevent mortality due to dengue infection [
5].
The obstacle to early dengue diagnosis lies in its diverse and unspecific clinical symptoms that resemble other diseases, which leads to delay in health-seeking and misdiagnosis of the disease [
1,
6]. Laboratory tests such as virus isolation (VI), reverse transcription-polymerase chain reaction (RT-PCR), haemagglutination inhibition (HI), and enzyme-linked immunosorbent assay (ELISA) for the detection of dengue non-structural antigen-1 (NS1) or dengue-specific immunoglobulin (IgM/IgG) can confirm diagnosis. But they are resource-intensive and not suitable for low-resource settings [
7].
The growing demand for point-of-care diagnostics gave rise to many dengue rapid diagnostic tests (RDT) that flooded the market in the past two decades [
7,
8]. Recent development in biosensors for the rapid diagnosis of acute dengue infection, if proven accurate, may intensify the competition and make RDT more affordable to more patients who need it [
9]. In this evaluation study, we aimed to estimate the diagnostic accuracy and utility of a newly developed biosensors-based dengue NS1 assay, and one each of commercially available NS1 ELISA and NS1/IgM/IgG RDT.
Methods
Study design
This evaluation study employed a prospective cross-sectional design. Its reporting followed Standards for the Reporting of Diagnostic Accuracy Studies (STARD) guidelines [
10].
Ethical statement
The study adhered to the principles of the revised 2013 Declaration of Helsinki [
11] and obtained ethical approval from the Medical Research and Ethics Committee, Ministry of Health Malaysia (NMRR-17-853-34,393) and University Malaya Medical Center (MRECID.NO: 2017426–5171). Written informed consent was obtained from each participant and minor assent where appropriate.
Population
The study site was Shah Alam Section 7 Health Clinic, a public clinic with the highest number of dengue patients located in the dengue-endemic district of Petaling, Selangor state, Malaysia. Febrile patients aged 9 months and above with symptoms fulfilling World Health Organisation (WHO) 2009 criteria for suspected dengue that sought treatment at this clinic from 13th November 2017 to 30th March 2018 during normal working hour were recruited consecutively [
6]. Patients in need of emergency care or with pre-existing conditions that were prone to complications from blood sampling were excluded. Both capillary and venous blood samples were collected from each patient using EDTA tubes for immediate onsite index tests. Another venous sample in plain tube was also collected together, centrifuged, aliquoted, and stored at − 80 °c for laboratory-based diagnostic tests in a virology laboratory in University of Malaya, Malaysia. No convalescent sample was taken. Socio-demographic background and clinical history were captured using structured questionnaire through face-to-face interview at the same setting. The 500 target sample size calculated using single proportion sample size formula based on a 50% disease prevalence was expected to achieve 5% absolute precision with 95% confidence for both sensitivity and specificity estimates [
12].
Index tests
ViroTrack Dengue Acute (BluSense Diagnostics, Denmark), SD Bioline Dengue Duo and SD Dengue NS1 Ag ELISA (Standard Diagnostics, Korea) were evaluated in this study for comparison. The first two are RDT intended for point-of-care use. They were tested onsite on both capillary and venous blood samples by medically trained research assistants blinded to the clinical pictures of the research participants. Additionally, ViroTrack Dengue Acute was also tested onsite on the same day with serum samples extracted from the plain tubes.
ViroTrack Dengue Acute
ViroTrack Dengue Acute (BluSense Diagnostics, Denmark) is a biosensors-based semi-quantitative immuno-magnetic agglutination assay packed in a polymer centrifugal microfluidic cartridge. Its diagnostic mechanism was detailed out previously [
13]. Briefly here, for each test, a ViroTrack micofluidic loaded with 30 mcl of blood sample was inserted into a portable opto-magnetic reader – the BluBox. The sample was centrifuged, metered, and mixed with magnetic nanoparticles (MNPs) pre-coated with anti-dengue antibodies located within the cartridge. Dengue NS1 antigen, if present, formed sandwich agglutination with these MNPs and were forced to rotate under an oscillating magnetic field, which modulated the intensity of a laser beam passing through them. A photodetector with a Blu-ray optical pickup unit would then measure the phase difference between the modulated light transmission and the applied field, which corresponded to the level of dengue NS1 antigen. This measurement was presented in a relative unit and interpreted by the BluBox according to a pre-set threshold value, where positive was defined as > = 27, negative if < 23, and equivocal (EQ) if 23–26.9 unit. The whole process after the insertion of the microfluidic was automatic and the result was ready in less than 15 min. The results were recorded by one research assistant and verified by three others independently. For analysis, a patient was considered tested positive for ViroTrack Dengue Acute if either capillary or venous sample was positive, EQ if both were EQ, and negative for all other combinations. No repetition was done for EQ results.
SD Bioline Dengue Duo
This is a commercially available rapid immunochromatographic test that comes in a combo of two joint cassettes, one for NS1 and another for IgM/IgG. Only 100 mcl blood sample was needed for the NS1 assay, while serology required 10 mcl followed by assay diluent. Results were interpreted according to manufacturer’s instruction by two independent research assistants 15–20 min after the application of specimen, where appearance of a test line was considered positive in the presence of a control line. Presence of only control line was considered negative [
14]. Discrepancies between first and second interpreters were resolved with the help of a third interpreter. For analysis, a patient was considered tested positive to an assay on SD Bioline Dengue Duo if either capillary or venous sample was found positive, and negative if both were negative.
SD Dengue NS1 Ag ELISA
SD Dengue NS1 Ag ELISA (Standard Diagnostics, Korea)
, a commercially available direct sandwich ELISA, was performed together with reference tests and interpreted according to manufacturer’s instruction [
15]. Test was considered valid if the negative and positive controls absorbance values were within set ranges. Cut-off value was calculated by adding 0.3 to the mean absorbance for negative controls. A sample was considered positive if its absorbance was equal to or larger than the cut-off value, and negative if lower.
Reference standard
The reference tests comprised of iTaq Universal SYBR Green One-Step real-time RT-PCR (Bio-Rad Laboratories, Hercules, CA), Panbio Dengue Early ELISA, and SD Dengue IgM and IgG capture ELISA (Standard Diagnostics, Korea). They were performed according to the manufacturers’ instructions as described in detail previously [
16‐
19]. They were chosen in reference to a previous study [
20]. These tests were conducted from 6th December 2018 to 13th April 2019, up to around 1 month from sample collection, by trained laboratory personnel blinded to the clinical information and results of the point-of-care index tests. A laboratory-confirmed dengue was defined as 1) RT-PCR positive, or 2) Panbio NS1 ELISA positive; while a presumptive dengue tested negative for both the above, but positive for IgM ELISA [
20]. Both laboratory-confirmed and presumptive dengue were included in the analysis as dengue positive; while patients who did not fall into any category above were taken as dengue negative without further laboratory tests. On top of that, a combination of “Recife” method and IgM/IgG ratio from ELISA was used to classify dengue positive patients into primary and secondary dengue, whereby primary was defined as IgG negative with positive on either IgM, NS1 or RT-PCR; while secondary – positive IgG with negative IgM and positive on either RT-PCR or NS1. If both IgM and IgG were present, IgM/IgG ratio > =1.2 was considered as primary dengue, while < 1.2 - as secondary dengue [
6,
16,
21,
22].
Data analysis
Descriptive analysis was used to describe the sociodemographic and clinical characteristics of the participants. The interrater agreement between the first and second interpreters for each assay in SD Bioline Dengue Duo was assessed using Kappa statistics (k). It was also computed for the agreement of test results between capillary and venous samples for this combo, while ViroTrack Dengue Acute also had additional results for capillary-serum and venous-serum. Agreement was interpreted as poor if k was < 0, slight if 0–0.2, fair if 0.2–0.4, moderate if 0.4–0.6, good if 0.6–0.8, and excellent if 0.8–1.0 [
23].
The true positive (TP), false negative (FN), false positive (FP), and true negative (TN) of each index test and various combinations of the components of combo test as compared to the reference standard were used to calculate various diagnostic accuracy parameters and their 95% confidence intervals (95%CI) using standard formulas [
24,
25]:
Sensitivity (SN) = TP/(TP + FN);
Specificity (SP) = TN / (TN + FP);
positive predictive value (PPV) = TP / (TP + FP);
negative predictive value (NPV) = TN / (TN + FN);
positive likelihood ratio (LR+) = SN / (1-SP);
negative likelihood ratio (LR-) = (1-SN) / SP;
area under curve (AUC) = (SN + SP) / 2; and.
diagnostic odds ratio (DOR) = (TP/FN) / (FP/TN).
Subgroup analyses by exposure (serotype, day of illness or DOI, dengue infection status) and outcome (lab-confirmed vs presumptive dengue) were also performed to compare sensitivity estimates. For diagnostic utility, post-test probabilities (95%CI) of dengue for positive and negative test were calculated for each test assay and their combinations. For that, pre-test probability (prevalence) of dengue among participants of this study was first converted to odds: prevalence/(1-prevalence). This pre-test odds were then multiplied with corresponding LR to obtain post-test odds that in turn were converted back to probabilities: odds/(1 + odds) [
24].
Data analysis was performed using STATA version 12 (StataCorp, TX, US). All inconclusive and missing test results, whether of reference standard or index tests, were excluded from the analysis.
Discussion
Most published dengue RDT evaluation studies used serum samples. Some studies also used whole blood specimen and only one used capillary blood [
26‐
29]. The excellent agreement between the results tested on capillary, venous, and serum samples in this study demonstrated that all three of them can be used on RDT, provided that anticoagulant-coated tool is used for the collection of whole blood specimen. The validity of results from capillary blood has practical implication when minimal invasiveness and/or rapidity is required such as in young children or during massive dengue outbreak.
No study was published prior to this for the evaluation of ViroTrack Dengue Acute, while SD Dengue NS1 Ag ELISA and SD Bioline Dengue Duo have been extensively evaluated. The individual components of the SD Bioline RDT had point estimates of sensitivity and specificity for the diagnosis of acute dengue that ranged within 44.4–94.9% and 70.9–100.0% for NS1 [
30‐
32], 10.0–98.0% and 66.0–100.0% for IgM [
18,
32,
33], and 38.8–90.1% and 92.5–100.0% for IgG [
22,
34,
35], respectively. On the other hand, the point estimates of sensitivity and specificity of SD Dengue NS1 Ag ELISA published previously were 55.2–76.8% and 94.6–98.6%, respectively [
15,
23,
36]. The results in this study fell within the above range except for the IgG component of the SD RDT, where both sensitivity and specificity were lower than previously found. This difference may be attributed to the underlying difference in study design, patient population, definition of reference standard, and other study characteristics [
37‐
39].
Since differences in study characteristics would modify the outcomes, it is difficult to directly compare diagnostic accuracy and utility between different studies without proper assessment. In other words, the apparent difference in diagnostic accuracy and utility parameters between two tests evaluated in two different studies may be due to the difference between their study characteristics instead of the actual performance of the tests themselves. However, diagnostic tests evaluated within the same study on the same patient population under the same condition can be directly compared [
37]. In this study, among all the individual assays, SD Dengue NS1 Ag ELISA had the best performance followed by ViroTrack, SD Bioline Dengue Duo NS1, IgM and IgG components. This outcome is expected as both ELISA and ViroTrack employed objective result read-out based on physical properties of light transmission in contrast to the subjective interpretation of SD RDT [
8,
13‐
15,
18,
30].
The sensitivities of all the evaluated assays turned out as expected in the subgroup analyses. The NS1-only assays performed better in detecting dengue infection in the first 5 days versus 6 days and above as NS1 antigen is actively produced and secreted in the first week [
7,
18,
22,
29,
30,
40‐
43]. However, the decline in performance was more obvious for SD NS1 RDT as reducing level of NS1 towards the end of the week might produce a test line too faint to be detected with naked eyes [
8]. On the contrary, serology-only assays performed better after day 5 when IgM and IgG started to be secreted to counter the infectious agent [
5,
7,
17,
18,
22,
29,
30,
40‐
44]. The observation that SD IgM performed better than IgG, especially after day 5, is most likely due to the use of IgM ELISA in the definition of the reference standard.
This study discovered that NS1-only assays had higher sensitivity in detecting primary and laboratory-confirmed dengue, while serology-only assays were better in secondary and presumptive dengue. Again, the reason for this lies predominantly in the definition of reference standard, where primary and laboratory-confirmed dengue were mostly those tested positive on RT-PCR and/or Panbio NS1 ELISA; while secondary and presumptive were more dependent on IgM and IgG capture ELISA for their definitions. This similar trend was observed repeatedly in previous studies with some variations attributable to differences in study characteristics [
18,
23,
29,
31‐
33,
35,
36,
40].
The finding that all NS1-only index tests had lower sensitivity while SD IgM RDT had it higher for the detection of DENV-1 as compared to other serotypes was unexpected. Some previous studies did not demonstrate difference [
27,
45], while others did [
18,
30,
31,
46,
47]. Further analysis (data not shown) showed that there were more patients beyond the 5th day of illness among those with DENV-1, which may partially explain these results. On the other hand, the failure of all assays to detect the sole DENV-4 was probably due to chance, as this patient had secondary dengue (with low IgM level) on 5th day of illness, when the level of free and detectable IgG and NS1 happened to be too low after their union in vivo [
23,
45,
47,
48]. Lastly, the serotype distribution found in this study agreed with previous finding [
49].
Notwithstanding the underperformance of SD Bioline Dengue Duo individual components, the diagnostic accuracy drastically improved when they were combined and interpreted as dengue positive if found positive on either one component. Combination of SD NS1 and IgM improved sensitivity, NPV and LR- (better rule-out test if found negative) without compromising on the other parameters, while the addition of IgG component further improved the rule-out parameters at the expense of rule-in parameters. This trend was in line with previous findings [
15,
23,
27,
29,
30,
32,
34,
35,
40,
41,
50,
51]. The repercussion of this finding is that dengue combo test is always superior to individual assay RDT as combo has the ability to detect dengue infection regardless of the phase of illness [
40,
41].
This recommendation is further backed up by the diagnostic utility from this study. As shown in Table
4, tested positive for SD NS1/IgM RDT was able to double the probability of dengue infection in a patient to be more than 90%, leaving only less than 10 out of 100 wrongly diagnosed non-dengue patients. Although all individual assays except SD IgG RDT were able to more or less match this performance, they could not reproduce the same when it came to negative tests. If a patient was tested negative for any of the individual assays, the best post-test probability of dengue was still 23.0%. In other word, at least 23 out of 100 dengue patients were FN that might have been misdiagnosed. In contrast, SD combo tests were able to reduce FN to just 15 to 18 per 100 dengue patients. While dengue FP may not be a big concern due to its small proportion and relatively less harmful supportive treatment unless in the case of misdiagnosis of other more severe diseases, high number of FN might lead to late diagnosis and delayed administration of required life-saving treatment for dengue patients. As such, in a clinical setting, especially in primary care, it is important for a dengue RDT to act as a screening tool that can detect more cases with minimal FN.
Although this study found that SD Bioline Dengue Duo, a dengue RDT based on immunochromatographic principle, was a more preferred tool when used as a combo test; the new biosensors-based ViroTrack Dengue Acute was not without its own advantages. It requires only 30 mcl of blood specimen compared to 100 mcl required for SD NS1 RDT. This feature is important when blood sampling is difficult and yield is scarce, such as among paediatric patients. Its test procedure requires minimal training, and is as simple and user-friendly as that of SD RDT at least in terms of sample loading, but its test duration is slightly shorter. Still, after loading blood sample, ViroTrack must be inserted into a Blubox, an electronic device that runs on electricity. This may create a bottleneck if more than one sample must be run concurrently, unless commercialised version of Blubox comes with the ability to run multiple tests. Yet, the presence of Blubox increases test sensitivity and reduces number of false positive from delayed interpretation common for immunochromatographic tests [
8]. Furthermore, its ability to connect with local and global network reduces the rate of manual transcription error in reporting [
52], and makes results available immediately upon diagnosis to treating doctors and public health authority for quick intervention [
53]. Nevertheless, ViroTrack must first be available in the form of a combo test and proven to be more accurate, cost-effective, and beneficial in future studies for this new technology to be popularised.
The strength of this study lies in its sound methodology and application. As mentioned above, it is difficult to directly compare diagnostic performance between evaluation studies due to different study characteristics [
37]. In the same way, it is also fundamentally incorrect to directly apply their results into daily practice or for policy making. Good performances reported in case-control or laboratory-based studies may be due to biases instead of the discriminatory power of the evaluated tests [
7,
39]. In contrast, the cross-sectional prospective design in an actual primary care clinical setting seen in this study produced a more realistic set of diagnostic performance parameters that is true to other similar clinical settings, making the application of its results easier and more valid. Besides, it complied with STARD-guidelines for quality assurance [
10]. In addition, diagnostic utility presented here is more intelligible to clinical practitioners and policy makers compared to the usually reported diagnostic accuracy. With simple calculation based on the formulas provided here or using rough estimation, the diagnostic utility of the three index tests evaluated in this study can be estimated for any clinical setting [
24,
54]. However, it should be cautioned that this exercise took into account only dengue diagnostics without consideration of other diseases. Practitioners should also use existing clinical reasoning for differential diagnosis.
In Malaysia, four previously conducted dengue RDT evaluation studies were exclusively laboratory-based and only one employed cross-sectional prospective design [
18,
41,
55,
56]. This study was the first conducted prospectively among consecutively sampled patients in primary care setting. It provided better and more updated insight on the application of dengue RDT in Malaysia. Furthermore, it was the first that evaluated a biosensors-based RDT in this setting and compared it with extensively used RDT and NS1 ELISA for a more comprehensive understanding of their relative performance, which is absent in most other studies that evaluated only either RDT or ELISA. Two limitations of this study are, first, only single sample was collected from each patient, making the reference standard based on serology presumptive rather than conclusive [
6,
20,
39]; and second, the diagnostic utility calculated was based on the pre-test probability of disease based on the WHO 2009 guidelines without taking into account of the haematological result. Nevertheless, these limitations perfectly reflect the actual situation faced by clinicians in the front line, making the study results more realistic and applicable to the real-world circumstances.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.