SARS-CoV-2 Diagnostic Tests: Algorithm and Field Evaluation From the Near Patient Testing to the Automated Diagnostic Platform

Introduction: Since the first wave of COVID-19 in Europe, new diagnostic tools using antigen detection and rapid molecular techniques have been developed. Our objective was to elaborate a diagnostic algorithm combining antigen rapid diagnostic tests, automated antigen dosing and rapid molecular tests and to assess its performance under routine conditions. Methods: An analytical performance evaluation of four antigen rapid tests, one automated antigen dosing and one molecular point-of-care test was performed on samples sent to our laboratory for a SARS-CoV-2 reverse transcription PCR. We then established a diagnostic algorithm by approaching median viral loads in target populations and evaluated the limit of detection of each test using the PCR cycle threshold values. A field performance evaluation including a clinical validation and a user-friendliness assessment was then conducted on the antigen rapid tests in point-of-care settings (general practitioners and emergency rooms) for outpatients who were symptomatic for <7 days. Automated antigen dosing was trialed for the screening of asymptomatic inpatients. Results: Our diagnostic algorithm proposed to test recently symptomatic patients using rapid antigen tests, asymptomatic patients using automated tests, and patients requiring immediate admission using molecular point-of-care tests. Accordingly, the conventional reverse transcription PCR was kept as a second line tool. In this setting, antigen rapid tests yielded an overall sensitivity of 83.3% (not significantly different between the four assays) while the use of automated antigen dosing would have spared 93.5% of asymptomatic inpatient screening PCRs. Conclusion: Using tests not considered the “gold standard” for COVID-19 diagnosis on well-defined target populations allowed for the optimization of their intrinsic performances, widening the scale of our testing arsenal while sparing molecular resources for more seriously ill patients.

Introduction: Since the first wave of COVID-19 in Europe, new diagnostic tools using antigen detection and rapid molecular techniques have been developed. Our objective was to elaborate a diagnostic algorithm combining antigen rapid diagnostic tests, automated antigen dosing and rapid molecular tests and to assess its performance under routine conditions.
Methods: An analytical performance evaluation of four antigen rapid tests, one automated antigen dosing and one molecular point-of-care test was performed on samples sent to our laboratory for a SARS-CoV-2 reverse transcription PCR. We then established a diagnostic algorithm by approaching median viral loads in target populations and evaluated the limit of detection of each test using the PCR cycle threshold values. A field performance evaluation including a clinical validation and a user-friendliness assessment was then conducted on the antigen rapid tests in point-of-care settings (general practitioners and emergency rooms) for outpatients who were symptomatic for <7 days. Automated antigen dosing was trialed for the screening of asymptomatic inpatients.
Results: Our diagnostic algorithm proposed to test recently symptomatic patients using rapid antigen tests, asymptomatic patients using automated tests, and patients requiring immediate admission using molecular point-of-care tests. Accordingly, the conventional reverse transcription PCR was kept as a second line tool. In this setting, antigen rapid tests yielded an overall sensitivity of 83.3% (not significantly different between the four assays) while the use of automated antigen dosing would have spared 93.5% of asymptomatic inpatient screening PCRs.

INTRODUCTION
At the time of writing (January 7, 2021), Belgium is emerging from a second wave of COVID-19 epidemic. The World Health Organization (WHO) recommended mass use of reverse transcription real-time PCR (RT-PCR) to detect active SARS-CoV-2 infections (1). However, the unprecedented high volume of samples reaching laboratories led to global scarcities of reagents and delays making prolonged containment measures less acceptable by the population (2). Since then, a new set of diagnostic tools have been developed, such as antigen detection immunoassays or molecular point-of-care tests. These tools could allow diversification of testing strategies and decrease shortages and overflows.
Thanks to their high sensitivity, ranging from 73.9 to 89.5% for high viral load samples [10 5 -10 7 RNA copies/swab (3)], and their overall specificity (4,5), antigen-detection rapid diagnostic tests have been integrated in several countries' testing strategies (6-10) 1,2 . Both Centers for Disease Control and Prevention (CDC) (11) WHO (12) and European Center for Disease Control and Prevention (ECDC) (13) have issued guidelines for their use. However, practical considerations are still lacking (including the best target populations). Meanwhile, several manufacturers have developed molecular point-of-care tests, most of which additionally target influenza and/or RSV (14,15) while others offer wider respiratory syndromic panel (16).
In addition, high throughput antigen-dosing systems based on chemiluminescence enzyme immunoassay represent an interesting alternative (17). This solution, recently deployed in German airports, is a striking example of delocalized laboratory medicine (18).
Following this expansion of available diagnostic tools, a deeper reflection has come to light on the best use of these various testing solutions according to their sensitivity, their turnaround time, the context in which the result will be used (patient vs. population-centered approach), the kinetics of the epidemic and the availability of reagents and consumables (19).
All of the above may partly explain the apparent confusion we are currently witnessing in the deployment of antigen rapid diagnostic tests and/or molecular point-of-care tests in most industrialized countries, either in terms of choosing the most appropriate diagnostic tests or the target population to apply these tests to. We would like to share here the results of evaluations we performed on four antigen rapid diagnostic tests, one automated antigen dosing assay and one molecular point-of-care test for the diagnosis of COVID-19, not only from an analytical "laboratory" point-of-view but also through their field implementation during the second Belgian COVID-19 wave. Using different techniques at different levels in a multistep, integrated, and adaptive diagnostic algorithm helped us to diversify and increase our overall testing capacity.

Population
LHUB-ULB (Laboratoire Hospitalier Universitaire de Bruxelles-Universitair Laboratorium Brussel) is a clinical laboratory serving five university hospitals (containing a capacity of around 3,000 beds) as well as a network of general practitioners in Brussels, Belgium. LHUB-ULB's service area covers 700,000 inhabitants (20). From July to September 2020, patients undergoing a SARS-CoV-2 RT-PCR were retrospectively categorized through a structured algorithm into four categories according to the information provided on the orders: symptomatic outpatients, hospital admissions (symptomatic or not), asymptomatic high-risk contacts, or mandatory screenings. The RT-PCR median C T values from these four groups were compared using the Tukey-Kramer method.

Symptomatic Cases Definition
We used the case definition provided by the Belgian national health institute (Sciensano) for COVID-19 (21). The acute apparition of one major symptom, the presence of two minor symptoms, or the aggravation of chronic respiratory symptoms without any other obvious cause was defined as a possible case (Supplementary Table 1). A confirmed case was a person with a SARS-CoV-2 positive sample.

Antigen Rapid Diagnostic Tests
Four lateral-flow immunoassays were evaluated: Panbio TM COVID-19 Ag Rapid Test Device (Abbott Rapid Diagnostics, Germany), BD Veritor TM SARS-CoV-2 (Becton-Dickinson and Company, USA), COVID-19 Ag Respi-Strip (Coris BioConcept, Belgium) and SARS-CoV-2 Rapid Antigen Test (SD Biosensor, Republic of Korea). Reading was performed by trained operators except for the BD Veritor TM for which an automated reader (BD Veritor TM System) was used.
An analytical performance study was performed using nasopharyngeal swabs. The swabs preserved in universal transport media (UTM) were sent to our laboratory for a SARS-CoV-2 RT-PCR, and then kept refrigerated overnight after the RT-PCR was performed. The four assays were performed at the same time by two trained operators. The amount of UTM engaged was according to the recommendations by each manufacturer for evaluation purposes but not for clinical use.
After the performance study, antigen rapid diagnostic tests were done in point-of-care settings, either a practice within our network of general practitioners, or in the emergency room of the Saint-Pierre university hospital. Each possible COVID-19 outpatient, who was within 7 days of symptoms onset, was offered an antigen rapid diagnostic test and informed that a negative result would require an additional sampling for RT-PCR as recommended at the time (21). Each antigen rapid diagnostic test sampling and test procedure was performed according manufacturer instructions (Supplementary Table 2).
The user-friendliness of each antigen rapid diagnostic test was assessed with a four-part questionnaire adapted from the Scandinavian evaluation of laboratory equipment for point-ofcare testing SKUP/2008/114 evaluation (22).

Molecular Point-of-Care Test
To assess the analytical performance of the Cobas R Liat SARS-CoV-2 & Influenza A/B nucleic acid test (Roche Molecular Systems, USA), nasopharyngeal swabs, which were sent to our laboratory for a SARS-CoV-2 RT-PCR and tested positive, were kept refrigerated overnight before testing. In addition, frozen samples from February 2020 which underwent at that time a Cobas R Liat Influenza A/B & RSV RT-PCR assay were also tested.

Automated Antigen Dosing Assay
Antigen dosing was performed using the Lumipulse R G SARS-CoV-2 Ag (Fujirebio, Japan) assay, expressing the dosage in pg/mL. For biosafety consideration, a viral-deactivation step (56 • C heating for 30 min) was added to the manufacturer's instructions protocol (23).
The analytical performance study was performed on UTM swabs kept refrigerated overnight after a SARS-CoV-2 RT-PCR. All available positive samples were selected. Negative samples were randomly selected to obtain a positive/negative ratio around 2:1.
In the second part of the evaluation, we evaluated the Lumipulse R performance on UTM samples sent to our laboratory for SARS-CoV-2 RT-PCR for patients who required scheduled hospital admission, COVID-19 contacts, or for healthcare workers.

Gold Standard and Statistical Analysis
Analytical performance study of antigen rapid diagnostic tests, molecular point-of-care test and automated antigen dosing were carried out on three different sets of samples.
SARS-CoV-2 RT-PCR was considered as the gold-standard. Except for some antigen rapid diagnostic tests, for which negative results were controlled by various other RT-PCR protocols, samples underwent the RealTime SARS-CoV-2 assay (Abbott Molecular, USA) on our m2000 platform. As detection of both targeted genes (RdRp and N) is performed using the same fluorophore, the C T values of this assay are observed up to 32 cycles, and not comparable with C T values of other RT-PCR assays. Consequently, only the C T values obtained using the RealTime SARS-CoV-2 assay were considered.
Statistical analyses and receiver operating characteristic (ROC) curves were performed using Analyse-it R for Microsoft Excel v3.80.

Antigen Rapid Diagnostic Tests
Ninety-nine UTM samples including 61 positive (C T ranging from 3.86/32 to 30.94/32) were selected. In this frame, the sensitivities of each antigen rapid diagnostic test were ranging from 36.1 to 49.2% ( Table 1). The latest C T detected antigen rapid diagnostic tests was 18.06/32. No false positive result was observed.

Molecular Point-of-Care Test
The agreement of the Cobas R Liat with the m2000 system for SARS-CoV-2 diagnostic was of 90.9% (50/55) for positive samples. C T values correlation between instruments was good (R 2 = 0.931). The Cobas R Liat yielded positive results for all positive samples presenting a C T value below 27.29/32 and yielded positive results for samples with C T of up to 29.11/32. Eighteen of the 19 frozen Influenza A positive samples and 5 of the 6 frozen influenza B positive samples yielded coherent positive results. Agreements for negative samples were of 100% for each parameter.

Automated Antigen Dosing Assay
Two hundred fourteen samples were selected including 136 positive samples. ROC curve analysis yielded an area under the curve (AUC) of 0.893±0.021 (Supplementary Figure 1). The highest Youden Index was at a threshold of 13.75 pg/mL (sensitivity 67.7%, specificity 97.1%). At a threshold set at 1.32 pg/mL [similar to a previous study (17) and to the manufacturer proposed cut-off at 1.34 pg/mL (24)], sensitivity was 78.9% and specificity of 73.9%. To exclude any false positive, the threshold had to be set at 20.27 pg/mL (sensitivity 63.9%). Finally, using a C T < 20/32 as a judgement criterion, the AUC of the ROC curve was 0.984 ± 0.007 (Supplementary Figure 2) with an optimal

Elaboration of the Diagnostic Algorithm
Following these results, we elaborated the algorithm described in Figure 2: whereas the diagnosis of outpatients was mainly based on point-of-care antigen rapid diagnostic tests, the hospital algorithm combined antigen rapid diagnostic tests, molecular point-of-care tests and conventional RT-PCR in an integrative diagnostic strategy. Four clinical situations were further identified: screening of asymptomatic patients, patients requiring immediate admission, symptomatic outpatients with symptoms lasting for less or more than 5 days.  Table 3). The Coris COVID-19 Ag Respistrip had a less satisfactory rating. The main practical issue was its readiness: its "strip-in-a-tube" format was considered by  operators as non-practical and leading to a potential biosafety hazard when the reading is difficult. Notably, SD Biosensor TM and Coris BioConcept did not provide any internal control in their kit. BD Veritor TM was the only kit offering nasal swabbing and automated reading.

Automated Antigen Dosing Assay
Two hundred seventy-nine patients (including 93 asymptomatic patients screened for a scheduled hospitalization) were tested. Their SARS-CoV-2 carriage status was categorized as "unlikely" if dosing below 1.32 pg/mL (n = 219, 78.5%), "possible" if dosing from 1.32 to 20.27 pg/mL (n = 46, 16.5%) and "certain" if dosing higher than 20.27 pg/mL (n = 14, 5.0%). All patients with "certain" results had a positive RT-PCR. Seven patients out of 46 (15.2%) with a "possible" result and five out of 219 (2.3%) with an "unlikely" result were tested positive according to RT-PCR, respectively (Table 4). Thus, the overall sensitivity for asymptomatic patients was of 86.7% (13/15). Hence, using this assay for the pre-admission screening of these 93 patients would have spared 87 RT-PCR (93.5%) for the cost of one missed low-positive (C T = 26.04/32).

DISCUSSION
In most industrialized countries, the large scale use of RT-PCR to detect active SARS-CoV-2 infections has shown limits in its capacity to broadly screen the population while providing timely and therefore meaningful results for optimized prevention and treatment. To fill this gap, SARS-CoV-2 antigen rapid diagnostic tests and molecular point-of-care tests are now considered as an adjunct to the RT-PCRs performed on large automated platforms (25).
Our results provide substantial evidence that no current antigen rapid diagnostic test is sensitive enough to be performed on UTM specimen (i.e., at the laboratory). During the first wave in Europe, we proposed a strategy combining antigen rapid diagnostic tests and RT-PCR, both performed in the laboratory (26). We stopped using antigen rapid diagnostic tests in the laboratory during the declining phase of the epidemic, not because of their low sensitivity [as stated by colleagues (27)], but because the proportion of samples from recently infected patient dropped, impairing these tests' usefulness (28). Regular follow-up of the positivity rate could allow adaptations of antigen rapid diagnostic test strategy as proposed by CDC (11) and ECDC (13). Here, we demonstrate the added-value of antigen rapid diagnostic tests at the point-of-care level for <5-days symptomatic outpatient thanks to their ease-of-use, rapid timeto-result, and low cost.
Our results show slightly lower sensitivity than previously reported (25). Indeed, part of the false negative results observed is likely due to variability in the adherence to protocol regarding sampling, incubation time and DSO. Sensitivity and specificity of such antigen rapid diagnostic tests strongly depend on their good execution and reading which are harder to achieve at the frontline where the expertise of personnel can vary; especially in this time of pandemic when the turn-over is higher than usual. This was confirmed by other recently published studies targeting the same population, with sensitivity ranging from 70.0 to 80.4% (29)(30)(31).
The absence of significant difference between antigen rapid diagnostic tests clinical performances highlights the need to assess their user-friendliness as a main criterion of choice. Our analysis underlined the need to consider very practical aspects such as opening caps while wearing gloves, ensuring biosafety outside a laboratory (see Figure 3) and instructions targeting non-laboratory operators, as recently discussed for low-resource settings (32). Besides, an immediate, in-person communication of a positive result likely allowed a stronger message and a better adhesion regarding quarantine, hygiene and contact-tracing than if done through virtual means, days after the consultation.   The Cobas R Liat yielded stunning performances for a 20min triplex molecular point-of-care test compared to our RT-PCR. However, invalid results were experienced with viscous samples. The addition of a molecular point-of-care test for patients attending the emergency room and needing hospitalization, regardless of the suspicion of COVID-19, allowed a faster management of inpatients avoiding the admission of asymptomatic SARS-CoV-2 carrier in "COVID-free" units, or the admission of SARS-CoV-2-negative patients in COVID-19 units pending their RT-PCR results. Furthermore, influenza and SARS-CoV-2 co-detection allows a better surveillance at a time where the potential co-circulation of the influenza and SARS-CoV-2 is still unknown. The costs of these molecular point-of-care tests stay high and their availability low. Hence, their use should be considered by targeting the best population with regards to the reduction of global costs related to isolation, use of protective equipment and prevention of nosocomial clusters.
In the present study, the Lumipulse R G SARS-CoV-2 Ag showed an overall good analytical performance compared to RT-PCR; and more specifically, to exclude negative and low positive samples using different criteria and cut-off values than the ones proposed by the manufacturer. These cut-offs need to be adapted and chosen regarding the local epidemiology and the objectives of the screening. Our cut-off values diverged from the one proposed in a previous study (17). However, despite the fact we added a viral deactivation by heating, our results yielded a better AUC of the ROC curve. In case of limited access to RT-PCR, such technique can allow testing people who would be otherwise not tested. Its higher throughput and sensitivity than antigen rapid diagnostic tests and its faster timeto-result than RT-PCR make it an interesting intermediary tool. Its low costs and its probable good assessment of infectiousness allow a relevant periodic testing in terms of infection control. Therefore, using antigen dosing could be the best solution to repeatedly test high number of high risk contacts while sparing RT-PCR resources. However, their biosafety must be carefully considered and viral neutralization applied if needed; viscous samples may cause pipetting errors and specific interpretation algorithm should be elaborated.
Our study presents some limitations. We did not consider alternative specimens for SARS-CoV-2 detection such as saliva, the use of serology or broad molecular "syndromic" respiratory panels that could be of use in a larger diagnostic algorithms (33). The emergence of new variants should not impact the value of our algorithm due to the different targets of the assays. However a careful follow-up of their performances over time should be implemented.

CONCLUSION
In conclusion, our study underlines the importance of shifting our attention from a narrow focus on the sole analytical performances of the diagnostic tools available (especially when these are similar) to an integrated approach taking into account (i) practical consideration such as time-to-result, field easeof-use, availability of reagents (ii) target populations (iii) intended use of produced results, and (iv) kinetic of the epidemic. Hence, we elaborated here a diagnostic algorithm based on these considerations to optimize the use of the newly extended arsenal of SARS-CoV-2 direct diagnostic tools, from the decentralized setting to the automated lab, to ensure clinical microbiologists enough ammunition for a reliable and meaningful COVID-19 diagnostic.

DATA AVAILABILITY STATEMENT
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

ETHICS STATEMENT
Ethical review and approval was not required for the study on human participants in accordance with the local legislation and institutional requirements. Written informed consent from the participants' legal guardian/next of kin was not required to participate in this study in accordance with the national legislation and the institutional requirements. Written informed consent was obtained from the individual(s) for the publication of any potentially identifiable images or data included in this article.

AUTHOR CONTRIBUTIONS
NY, CDe, MD, F-ZB, FP, MW, and CDu did the investigations. NY and MH contributed to literature review and the writing of the initial draft. NY, CM, FP, HD, ND, M-LD, and SM contributed to manuscript revision, data compilation, and figure presentation. All authors provided critical review and commentary. NY, ND, FC, MB, and MH contributed to study design, manuscript preparation, literature review, revision, and project administration. All authors contributed to the article and approved the submitted version.