Introduction
With increasing population growth, urbanization, and industrialization have collectively contributed to widespread circulation of dengue virus (DENV), which is primarily transmitted by Aedes species mosquitoes [Reference Senn1, Reference Khan2]. An estimated 390 million people are infected annually, making DENV a major health concern due to recurrent outbreaks of dengue fever (DF) and dengue haemorrhagic fever (DENF). DENV is a single-stranded, positive-sense RNA-enveloped virus with an approximately 11 kb genome that encodes three structural proteins (envelop, capsid, and membrane) and seven non-structural proteins (NS1-NS5) [Reference Khan2, Reference Lee3].
While most DENV infections are asymptomatic, approximately 5–10% of cases can progress to severe forms such as dengue hemorrahgic fever (DHF) or dengue shock syndrome (DSS). These life-threatening complications are primarily driven by vascular fragility, resulting from endothelial dysfunction and a cytokine storm initiated by the host immune system. NS1 protein plays a critical role in this pathology by directly damaging endothelial cells through complement system activation and induction of pro-inflammatory cytokines, including tumour necrosis factor alpha (TNF-α), interleukin-6 (IL-6), and interferon gamma (IFN-γ). These cytokines, released by monocytes, dendritic cells, and T cells contribute to excessive complement activation, leading to vascular leakage, pleural effusion, and plasma extravasation. The resulting loss of endothelial integrity can lead to hypotension, haemoconcentration, and, in severe cases, hypovolemic shock [Reference Khan2, Reference Srikiatkhachorn and Kelley4].
Early diagnosis of DENV is essential for proper management and public health response to the disease [Reference Shrivastava5]. Dengue is primarily confirmed via laboratory testing using various techniques, including viral isolation, molecular assays, and serological methods [Reference Guzmán and Kourí6]. While virus isolation is regarded as the gold standard, it is laborious, time-consuming, and requires biosafety level 3 facilities. Reverse transcription-polymerase chain reaction (RT-PCR) and enzyme-linked immunosorbent assay (ELISA) are more commonly used to detect viral RNA or specific antibodies [Reference Kumarasamy7]. Nevertheless, these methods still require trained personnel and specialized machines, which are often lacking in laboratories in resource-limited settings.
This diagnostic gap has led to the increasing use of rapid diagnostic tests (RDTs), often called lateral-flow-based point-of-care tests (Figure 2). These assays detect DENV antigens or antibodies (e.g. NS1, IgM, or IgG), and are particularly suitable for low-resource settings because of their affordability, user friendly, and minimal storage requirements [Reference Blessmann8]. RDTs can differentiate between primary and secondary dengue infections. In primary infections, IgM is the first antibody to appear, followed by IgG. In contrast, secondary infections are characterized by an early and pronounced IgG response due to immunological memory. NS1 antigen, which is produced early in infection and independent of the host’s immune response, can be detected in both primary and secondary infections, making it a valuable early marker [Reference Yow9].
Aside from RDTs, the tourniquet test (TT), sometimes called capillary fragility test, is a low-cost, rapid physical examination method for the diagnosis and classification of DF. Historically, the World Health Organization (WHO) discouraged its use for diagnosing DHF and DSS. However, revised WHO guidelines now include TT as a diagnostic criterion for DF, dengue with warning signs, and severe dengue, recognizing its practical utility in resource-limited settings [10, Reference Ismail11].
DENV infection increases capillary permeability, a phenomenon exploited by the TT. As part of the procedure, a blood pressure cuff is inflated on the upper arm to a level between systolic and diastolic pressures. After five minutes, the number of petechiae (small, non-raised, purplish-red skin dots caused by capillary haemorrhages) within a defined one-square-inch area is counted. A result of more than 20 petechiae per square inch is considered positive [Reference Halsey12], although some guidelines also consider 10 petechiae or more to indicate a positive result [Reference Asidik13].
Clinical features such as fever and leukopaenia can only identify ‘probable dengue’ in endemic areas. However, when combined with a positive TT (indicative of haemorrhagic manifestations), these features significantly improve diagnostic specificity in distinguishing DF from other febrile illnesses, particularly in adults [Reference Sawasdivorn14]. For paediatric patients, the 1997 WHO clinical case definition for probable DF remains applicable: fever accompanied by a positive TT and leukopaenia during non-epidemic periods or fever with either a positive TT or leukopaenia along with any other clinical symptom during an epidemic [Reference Antunes15].
Given the simplicity and affordability of the TT and the growing availability of RDTs in resource-constrained countries, this study aims to compare their diagnostic accuracy for DF. As infectious diseases continue to emerge and re-emerge at unprecedented rates, particularly in low-resource settings, the availability of simple, cost-effective diagnostic tools like TT and RDTs could prove life saving.
Method
Search strategy
A systematic search was carried out using online bibliographic databases: PubMed, Science Direct, and Scopus. The population, intervention, comparator, and outcome (PICO) question format was used for the search terms. People infected with DENV were included in the population. RDTs (to identify DENV NS1 and antibodies) and a TT (to diagnose dengue) were used as the intervention, while a validated laboratory-based PCR served as comparison. The primary outcome was DENV infection, which was assessed in studies by (1) NS1, IgM, and IgG detection for RDTs, (2) counting the number of petechiae after inflating a blood pressure cuff on a person’s upper arm for TT. A petechiae count of 10 or more indicates a positive test, indicating sensitivity, whereas a count of less than 10 indicates a negative test, indicating specificity.
Search terms (free-text terms and keywords) related to dengue diagnosis were utilized in the right combination with relevant controlled vocabulary (Medical Subject Heading-MeSH), along with Boolean operators to achieve the most comprehensive search results. Each search strategy was refined to improve article relevance and comprehensiveness. The final search was performed using the following terms: ‘tourniquet test’, ‘capillary fragility test’, ‘PCR’, ‘polymerase chain reaction’, ‘rapid diagnostic test’, ‘dengue’, ‘break-bone fever’, ‘diagnosis’, and ‘ELISA’ (see Supplementary Table 1). We looked for additional references in the reference lists of all the included studies that were included (i.e. snowballing) (see Figure 1 below). Only articles written in English were analyzed. Studies were considered if they satisfied the criteria described below: (a) diagnosed DF from any country of the world; (b) contained data from all types of observational research (such as cross-sectional, case report, case-control, cohort studies, and case series) that assessed the TT’s and/or RDT’s diagnostic accuracy for dengue infection; (c) studies examining individuals who initially presented with fever and were later tested for dengue using the index test, TT, or RDTs (testing for the presence of viral antibodies (IgM and/or IgG) or NS1 antigen) and PCR (reference standard), (d) studies using serum samples, plasma, or whole blood (fresh or frozen) from patients clinically suspected of dengue infection exposure.
Figure 1. Flowchart detailing the study selection procedure. STARD = Standards for the Reporting of Diagnostic Accuracy Studies; N = number of papers.
Studies that failed to meet the research topic or inclusion criteria (not connected to the diagnosis of DF, no report on the accuracy, sensitivity, or specificity of the type of test, and/or no data to calculate it) were excluded. Studies having an undetermined methodology, challenge studies, experimental research, etc., were disregarded. Conference abstracts, brief papers with incomplete datasets or presentations, review papers, pieces containing commentary or opinion, protocols, and inaccessible articles were disqualified.
Study selection
All the cohort studies that were found in the databases were evaluated separately by two reviewers (ZB and MBB). Potential studies were divided into groups for full-text reading. Any differences were settled by (NF) and (MUI), and the justifications for including and omitting trials were noted.
Extraction of data and quality assessment
Data were extracted from the full texts of selected articles. The review of full text was done by ZB and MBB. Using this form, we were able to gather data on the study’s design, the types of participants, the index and reference tests, and the total number of participants. For each research study comparing the two tests, a 2x2 table was made. The quality of the data extraction table was developed following the Standards for Reporting of Diagnostic Accuracy Studies (STARD) [Reference Cohen16] and Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) [Reference Whiting17] guidelines. The STARD each has 5 items consisting of background and objective, methods, results, discussion, and registration, and the QUADAS tools also have 4 items each consisting of patient selection, index test, reference standard, flow, and timing sections of the result. Risk of bias and applicability items are included for each of the domains mentioned above. Items were given a score of positive (low-bias risk), negative (high-bias risk), or unclear (insufficient information). The findings section included a description of each evaluation. Furthermore, the preferred reporting items for systematic reviews and meta-analyses (preferred reporting items for systematic reviews and meta-analyses (PRISMA)) were followed [Reference Page18].
Data synthesis and analysis
A 2x2 contingency table was created for each investigation. We calculated likelihood ratios (LRs), predictive values, sensitivity, and specificity. The value of 1 was added to cells in the 2x2 table of TT in the primary research that had 0 in them so that computations could be performed; however, this only occurred in one trial. Although we had intended to exclude primary papers that reported two cells with 0, that did not happen. Each study’s sensitivity and specificity were combined, and the mean output and standard deviation (SD) were estimated.
Result
Study selection
Out of the 675 studies initially identified, eight were excluded as duplicates (Figure 1). The remaining 667 articles were screened based on their titles and abstracts. Of these, 613 were not included for reasons listed below: out of scope (n = 576), systematic/meta-analysis/narrative reviews (n = 3), full text unavailable (n = 4), conference abstracts (n = 1), non-English language (n = 1), not relevant to the research question (n = 16), use of inappropriate diagnostic tests (n = 7), unsuitable study design (n = 5), lack of sufficient statistical data to determine true/false positives (n = 8), and absence of PCR as the reference standard (n = 31).
Figure 2. A diagram illustrating the working principle of Rapid Diagnostic Tests (RDTs).
Ultimately, 23 original research articles met the inclusion criteria and provided information on the sensitivity and specificity of dengue RDTs and the TT compared to PCR-based diagnosis. Of these, 18 studies evaluated RDTs and 5 focused on the TT. Four were prospective cohort studies, while the remaining 19 were retrospective cohort studies (Table 1). The number of participants per study ranged from 67 to 30,760.
Table 1. Shows the Different Studies Evaluating the Performance of Dengue RDTs and TT Using PCR as a Reference Standard.
Abbreviations: CI: Confident interval, PPV: Positive Predictive Value, NPV: Negative Predictive Value, PLR: Positive likelihood ratio, NLR: Negative likelihood ratio, RDTs: Rapid Diagnostic Tests, NS1: Non-Structural Protein 1, PCR: Polymerase Chain Reaction, EIA: Enzyme Immunoassay, DENV-1: Dengue Virus Serotype 1, LAMP: Loop-Mediated Isothermal Amplification, SD: Standard, RT-PCR: Reverse Transcription Polymerase Chain Reaction, qRT-PCR: Quantitative Reverse Transcription Polymerase Chain Reaction IgG: Immunoglobin G, IgM: Immunoglobin M, DENV: Dengue Virus, mAbs: Monoclonal Antibodies, RDT: Rapid Diagnostic Test, PAD: Paper-based Analytical Device, USA: United States of America, DENV-2: Dengue Virus Serotype 2, TT: Tourniquet Test, ZIKV: Zika virus,
Study characteristics
Twenty-three (23) eligible studies were included (Table 1). These studies, conducted between 2001 and 2022, spanned 21 countries: Australia, Thailand, Vietnam, United States of America (USA), Netherlands, China, Canada, England, India, Japan, Germany, Switzerland, Republic of Korea, New Caledonia, Malaysia, Italy, Brazil, Puerto Rico, Peru, Taiwan, and French Guiana. The regional distribution of these studies is also detailed in Table 1.
Of the 21 countries, studies involving RDTs were conducted in 16 countries: Switzerland, New Caledonia, Germany, Japan, India, England, China, the Netherlands, Canada, the USA, Vietnam, Thailand, Australia, the Republic of Korea, Italy, and French Guiana. TT was studied in 5 countries: Taiwan, Thailand, Brazil, Peru, and Puerto Rico.
Peru (n = 13,548) and Brazil (n = 30,760) had the largest study populations, using TT and RDTs, respectively. Most of these studies focused on individuals with suspected acute DENV infection. The most frequently used RDTs were the Standard™ Dengue Duo test, which demonstrated the highest sensitivity among the assays, followed by the Panbio® Dengue RDT. In all studies, RT-PCR was used as the reference standard. Notably, one study involving RDTs reported issues of serological cross-reactivity.
Risk of bias and study quality assessment
Each selected study was assessed using STARD [Reference Cohen16] and QUADAS-2 [Reference Whiting17] guidelines. The STARD tool assesses five domains, including background and objectives, methods, results, discussion, and registration, while QUADAS-2 evaluates four key areas of diagnostic study quality: patient selection, index test, reference standard, and flow and timing.
Due to the use of patient data from databases, potential bias from multiple assessors, or the inclusion of patients with pre-existing conditions, a high risk of bias in the patient selection domain was observed in five studies, while the other 18 studies were evaluated of a low risk of selection bias. Regarding PCR as a reference standard, all studies demonstrated a low risk of bias. For the domain of index, unclear risk of bias was observed in four studies due to insufficient reporting or methodological concerns.
Study quality assessments are illustrated in Figure 3. Only one study employed random sampling [Reference Prabowo28]. Based on the cumulative data, 100%, 56%, and 15% of participants underwent PCR, RDT, and TT, respectively.
Figure 3. Quality Assessment of Diagnostic Accuracy Studies II.
Discussion
A summary of all investigations, including the types of samples used and patient characteristics, is provided in Table 1. Across all studies, a total of 7,513 samples were evaluated using RDTs, with serum, plasma, and whole blood being the primary sample types. Only one study specifically assessed the performance of dengue RDTs for detecting remote prior infection [Reference Echegaray30], providing limited insight into the IgG component of RDTs in individuals with a history of DENV exposure.
Among the 18 RDT studies, four [Reference Shih22, Reference Blacksell31, Reference Kyaw32, Reference Hunsperger35] evaluated the IgM component, while fourteen [Reference Lee3, Reference Foley19, Reference Alidjinou20, Reference Shih22–Reference Vivek25, Reference Tsai27–Reference Fry29, Reference Blacksell31–Reference Kyaw32, Reference Liu34–Reference Hunsperger35] focused on NS1 antigen detection. Six studies [Reference Lee3, Reference Shih22, Reference Tricou24, Reference Vivek25, Reference Fry29, Reference Echegaray30] assessed combined NS1/IgM detection, two [Reference Echegaray30, Reference Kyaw32] evaluated IgM/IgG, and four [Reference Lee3, Reference Shih22, Reference Tricou24, Reference Tsai27] assessed all three markers (IgG, IgM, NS1). This distribution reflects the focus on acute primary infections, in which IgG levels are often absent or minimal depending on the illness stage.
The mean sensitivity of RDTs detecting IgM alone was 61.4% (SD = 28.6), with a specificity of 80.3% (SD = 16.4). For NS1 alone, the mean sensitivity was 79.2% (SD = 13.7), and specificity was 94.6% (SD = 5.1). Combined detection of IgM and NS1 increased sensitivity to 80.4% (SD = 14.5) and specificity to 89.4% (SD = 11.3). Notably, combining all three biomarkers (IgG, IgM, and NS1) yielded the highest diagnostic performance, with a mean sensitivity of 90.98% (SD = 8.4) and specificity of 90.8% (SD = 14.6).
Table 1 provides detailed estimates for each RDT’s likelihood ratios, predictive values, and 95% confidence intervals. The majority of included studies were carried out in dengue-prone areas; they often lacked information on co-infection with other flaviviruses or prior treatment history. This limited the ability to account for potential serologic cross-reactivity, which may lead to false positives, especially in areas where viruses like Zika co-circulate. Only one study thoroughly assessed flaviviral cross-reactivity.
Most RDT studies used whole blood, suitable for point-of-care testing. One innovative study [Reference Vongsouvath33] demonstrated that nucleic acid amplification and DENV typing could be conducted directly from used RDTs, making field-based surveillance more feasible. Results showed strong concordance between NS1 RDT-derived samples and serum-based RT-PCR results, with agreement rates of 82.8% (Vietnam) and 91.4% (Malaysia) using blood, and 91.9% and 93.9%, respectively, using serum. There was 100% concordance in identifying the infecting serotype between the two methods.
Five RDT studies [Reference Alidjinou20, Reference Tricou24, Reference Vivek25, Reference Kyaw32, Reference Liu34] successfully differentiated among all four dengue serotypes, with DENV-1 being the most prevalent. One study compared RDT performance across whole blood, serum, and plasma, as summarized in Table 1. However, due to inconsistent reporting on vaccination status, age, co-infections, and time since symptom onset, subgroup analyses were not possible. Overall, the mean sensitivity and specificity of RDTs were 76.2% (SD = 13.8) and 91.5% (SD = 10.3), respectively (Figure 3).
Furthermore, five studies [Reference Halsey12, Reference Sawasdivorn14, Reference Gregory36–Reference Furlan38] evaluated the TT, including both DF and DHF cases. The mean sensitivity of TT was 48.6% (SD = 24.9), and specificity was 79.5% (SD = 14.9) (Figure 4), with 95% CI ranges varying. 0 to 0.9 was the range observed for positive predictive values (PPV) and 0 to 2.9 for negative predictive values (NPV). These findings suggest that TT is better at identifying true negatives than true positives. The positive likelihood ratio ranged from 0 to 2.9, suggesting limited ability to increase post-test probability of dengue in TT-positive individuals. Similarly, the negative likelihood ratio (0 to 0.99) showed that a negative TT result had limited discriminatory power. TT sensitivity was lowest on day 0 of illness onset and highest on day 7 and beyond, as reported by Halsey et al. [Reference Halsey12]. Specificity remained stable across illness days. Sensitivity was higher in younger patients and females, whereas specificity was slightly higher in males. Repeat testing moderately improved sensitivity (to 60%) but reduced specificity (to 56%).
Figure 4. Graphical representation of mean sensitivity and specificity for both the tourniquet test (TT) and rapid diagnostic tests (RDTs).
Using PCR as the diagnostic standard, TT demonstrated less diagnostic accuracy compared to RDTs. The TT’s mean sensitivity (48.6%) and specificity (79.5%) were significantly lower than RDTs (76.2% and 91.5%, respectively). However, two TT studies [Reference Sawasdivorn14, Reference Ho37] had a high risk of bias, potentially skewing sensitivity downward. A limitation of this analysis is that most TT studies used ELISA, not PCR, as the reference standard. Thus, fewer TT studies met the inclusion criteria. Additionally, the inconsistent reporting across primary studies prevented evaluation of TT’s effectiveness in specific subgroups. It also raises concerns about whether TT offers any real advantage over clinical evaluation alone. Further research using PCR as the gold standard is warranted to fully understand the diagnostic utility of TT. In particular, assessing TT performance by day of illness, gender, age, and dengue subtype could provide valuable insights for clinical application. For RDTs, larger studies are needed to explore the impact of pre-existing immunity and dengue subtype on diagnostic accuracy.
Finally, manufacturers of dengue RDTs should be encouraged to develop and validate new tests with improved sensitivity and specificity, incorporating novel antigen or antibody targets. Such improvements would significantly enhance dengue diagnosis, particularly in low-income countries where rapid, reliable, and affordable tools are urgently needed.
Conclusion
In resource-constrained settings where advanced laboratory diagnostics are unavailable, it is essential to distinguish between the utility of the TT and RDT for dengue detection. Given the TT’s consistently low sensitivity, it should only be used alongside other diagnostic methods for dengue. If employed, particularly in severely resource-limited, dengue-endemic regions, it must be interpreted with caution and only in conjunction with clinical findings. In contrast, RDTs demonstrate superior diagnostic performance compared to TT, particularly when benchmarked against PCR. Therefore, RDTs should be prioritized over TT in the diagnostic approach to DF in low-resource environments.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/S0950268825100460.
Data availability statement
Will be made available on request.
Acknowledgment
We would like to appreciate all those who have contributed to this work.
Authors contribution
ZEB: conceptualization, original drafting, reviewing and editing, analyses. NL: supervision, reviewing and editing. MBB: conceptualization, reviewing and editing, supervision. MUI: conceptualization, original drafting, reviewing and editing, resources.
Competing interests
The authors declare none.
Open access
