We conducted the meta-analysis following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guideline. We performed a systematic search of the databases PubMed, Cochrane Library, EMBASE, and the Web of Science from their inception to 31 May 2023. The main search terms were “[(COVID-19 OR SARS-CoV-2 OR Coronavirus disease 2019 OR Novel coronavirus) AND (Rapid diagnosis* OR Rapid detection OR Rapid antigen test* OR Antigen assay) AND (false-positive OR false-positivity OR specificity OR accuracy)] AND (cycle threshold OR Ct)] (Table 1). We also searched the previous reviews for relevant studies. We registered this systematic review on PROSPERO (registration number: CRD42022351138). No language restrictions were applied.

We included studies evaluating the specificity or false-positivity of commercially available RADTs for diagnosis of SARS-CoV-2 infection, against RT-PCR as a reference standard. Cohort studies, nested cohort studies, and case-control or cross-sectional studies as well as randomized studies were considered. Studies carried out in various locations, targeted at individuals of any age despite presence of symptoms were pooled in our study. Studies without reporting a Ct value cut-off were excluded.

Two authors (Y-PY and ZJ) independently scanned the titles and abstracts of the search results in Endnote X9 to retrieve relevant records and obtained a full-text review of those eligible articles that met our inclusion criteria. Then, a third author (T-HT) was invited to settle any conflicts or disputes. Figure 1 illustrates the flowchart of the screening process. Based on the previously defined excluded criteria, 51 papers written in the English language were included in the final meta-analysis.

Afterward, data extraction was performed by an author per paper and reviewed by a second. The following data were extracted from included studies using a data-extraction form: first author, nation, sample size, sample condition, index tests, false positives, and Ct cut-off values.

We used STATA Version 17.0 software to conduct the meta-analysis. Subgroup analysis was conducted based on different Ct value cut-offs (<40 or ≥40). The main indicators used were percentage ratios and 95% confidence intervals (CIs) of false positives in the RADTs group relative to those in the RT-PCR group. The I² statistic was used to assess the level of statistical heterogeneity, and an I² value of ≥50% confirmed heterogeneity [15]. We conducted a random-effect model meta-analysis because we expected considerable clinical heterogeneity. In addition to an overall evaluation, we also conducted a sub-group meta-analysis for Ct value cut-off with <40 and those ≥40. We evaluated the quality of the evidence for each outcome using the GRADE (Grading of Recommendations Assessment, Development and Evaluation) protocols, which classified evidence as very low, low, moderate, or high [16].

Although COVID-19 is not a chronic disease, it meets the Wilson criteria for screening due to the following facts: it is an important health problem; its natural history and side effects are well understood; there is a recognizable latent or early symptomatic stage; a test is easy to perform and its acceptably, accurately, reliably, sensitively and specifically easy to interpret; an accepted treatment is recognized for the disease; treatment is more effective if started early; a policy on who should be treated has been in execution; timely diagnosis and treatment are cost-effective, and case finding should be a continuous process [68].

To our knowledge, this is the first systematic review and meta-analysis that explores the relationship between RADTs and Ct value cut-offs. We integrated data from 51 studies to evaluate the false positive rates of RADTs. We found an overall satisfactory false positive rate (0.01, 95% CI: 0.00–0.01) for RADTs compared to RT-PCR. In the stratified analysis, we also found that false positive rates of RADTs did not increase when Ct values of RT-PCR increased (Ct < 40, 0.01, 95% CI: 0.00–0.01; Ct ≥ 40, 95% CI: 0.00–0.01).

Amplification of genomic sequence of RT-PCR is measured in Ct values. Reporting this Ct value or calculating viral load aids in interpretation and clinical decision-making [69]. However, there are controversies on how thresholds for infectivity shall be defined [70]. A previously published meta-analysis [71] showed that

RADTs present high sensitivity and specificity in detecting COVID-19 but not in exploring the relationship between Ct values and sensitivity/specificity. Since it is difficult to standardize Ct values across different systems and research teams, we could not compare the results between the studies included with different Ct values [71]. Many qPCR assays introduced a Ct cut-off of 40 for infectivity, allowing the detection of very few starting RNA molecules [69]. Our study was able to prove further the effectiveness of the association between RADTs and RT-PCR in different groups of Ct-Values.

Furthermore, it is of great necessity to carry out mass testing in communities as a major strategy for transmission control to minimize the spread of the infection because statistics show that those who never develop symptoms or are still in the before-symptom-onset stage can be quite contagious [72]. Therefore, self-testing using RADTs is vital in low-cost, resource-saving infection control [73]. Meanwhile, swab sampling from the oral or anterior nasal (AN) is less invasive and more efficient in improving compliance and testing frequency. Therefore, if such tests are adopted in community screening, especially in densely populated sites like schools and universities, rest homes, clinics, and prisons, they may help to reduce or even eliminate transmission significantly [58]. As it is easy to perform, all individuals can collect samples independently. This would eventually reduce the quarantine time for the potentially infected [69] and allow individuals to return to their regular work as early as possible. But the prerequisite for a reliable and validated SARS-CoV-2 test is compliance with negative and traditional physical preventive measures like face masks and social distancing that must be strictly observed [74]. Besides, such diagnostic capability would also enable policymakers to evaluate and adjust restrictive rules like quarantine duration and business shutdown period to ensure quick recovery of the economy.

The clinicians could also adjust discharge criteria according to Ct values and determine when the patient can discontinue isolation (which will shorten the hospitalization and isolation time of patients further and avoid unnecessary isolation treatment and waste of medical resources). The adjustment of Ct-values has implications in public health screening, enabling contact tracers to focus on persons who are most likely to be infectious. Reduced screening performance induced by the low sensitivity of RADTs makes it essential to guarantee effective repetition in close screening [75, 76].

The COVID-19 pandemic threatened the effective management of hospital risk. Health care providers should respond timely and effectively to rapidly changing regulations and guidelines [77]. During this period, hospitals need to care for COVID-19 patients, but also those with other diseases. The lower false positive rate and costs of RADTs imply that early and suitable detection was effective.

Nevertheless, the consequences of failing to identify and separate those COVID-19 cases affect the quality of care [78, 79]. Cases that were negative on the RADTs but had clinical symptoms were sent for further RT-PCR testing for timely identification of COVID-19 cases.

RT-PCR is the most important indicator for the clinical diagnosis of COVID-19. However, since it requires substantial manpower and time, rapid and accurate laboratory diagnosis technology is very important. The results of this study implied that RADTs could be considered an alternative for the rapid triage of patients. Although manufacturers’ instructions vary, our study provides guidance in the real world: RADTs are easy to conduct, do not require expert knowledge, and the results are obtained within a few minutes, saving time and money. If suspicious patients had similar clinical symptoms but were negative for RCTs, they were further tested with RT-PCR. In addition, RADTs are also suitable for epidemiological analysis, such as group epidemic monitoring and contact tracing. Such integrated strategies could significantly enhance prevention.

RADTs is not only a reasonably inexpensive, simple test with quick results and more accessibility to patients, but is also an important tool that might be more useful in cutting off the chain of transmission by rapidly identifying positive and previous cases, discovering vast numbers of asymptomatic carriers who often migrate from one location to another [80, 81]. In addition, RDTs can provide additional seroepidemiological data and aetiological diagnosis to determine the magnitude of COVID-19 spread within a population [80]. Thus, previous studies indicated that RADTs were recommended for the early detection of patients suspected of having COVID-19 at the peripheral level of the health system and outside hospital settings in low- and middle-income countries [80–82], where there is little access to molecular tests.

In the meta-analysis, heterogeneity exists if the sample estimates for the population risk were of different magnitudes [83]. In this study, we used the random effect model when I² statistics were 89.32%, 91.46%, and 91.02% for Ct values < 40, Ct values ≥ 40, and overall, respectively. Because of the existence of significant heterogeneity in false positive rates, it is important to assess heterogeneity in the meta-analysis. This could be caused by various factors, including population characteristics, study design, sample quality, antigen test manufacturers’ instructions, and Ct cut-off values [7]. We aggregated studies that explicitly reported a Ct cut-off value, but heterogeneity in the results was inevitable.

The strengths of this study are as follows: First, we included all relevant studies from a global database, which are accepted with a relatively high level of evidence. Second, we excluded studies with no control groups to increase comparability and decrease possible heterogeneity. This is because a larger publication bias may exist without control groups. Third, a subgroup meta-analysis was performed to analyze the real association that controlled the independent Ct values. We analyzed the separate effects of the association between RADTs and RT-PCR based on different CT values.

However, there are still several limitations that should be noted when interpreting the findings of this meta-analysis. Firstly, because sources of reagents are very complex and no single diagnosis standard exists, the bias estimated is inevitable. Secondly, ordinary meta-analyses on efficacy render high-quality evidence from randomized controlled trials only. However, it is impossible to randomize people into “RADTs” and “RT-PCR” categories. Thirdly, few studies considered other potential confounding factors. Fourthly, the variable studied is only at Ct values in a rough range with a cut-off of <40 or ≥40, and it has failed to classify CT value cut-offs in more groups, such as below 29, 30–37, 38–40, and above 40, and add any other factors, for example the type of RADTs and RT-PCRs, the day the patient was examined, the type of reagent, the time to complete the examination, etc. Fifthly, when using the GRADE approach to evaluate the quality of the evidence for each outcome, the current evidence from all selected studies was moderate in imprecision. In the future, more comprehensive studies are recommended to improve the quality of evidence. Finally, only English papers were included, it was very difficult to explore the disparity of languages for the relationship between RADTs and RT-PCRs.

In conclusion, the best available evidence supports an association between RADTs and RT-PCR. When Ct-values were analyzed using cut-off <40 or ≥40, this resulted in an estimated false positive rate of only 1%. However, it is limited, and more trials are warranted. Further studies regarding subgroups according to sex and age are essential to clarifying the subgroup effect.