One protocol amendment (notified May 29, 2020; granted June 23, 2020) was made to change the control group from a pre-implementation control group to a contemporaneous control group. This change was made in recognition of the fact that most patients tested for COVID-19 before the start of the trial were ambulatory community patients who were tested in hospital as part of the containment phase of the pandemic, and were therefore not comparable to patients presenting with acute respiratory illness who were recruited into the intervention group of the trial.
For the intervention group, eligible participants were those who met the following criteria: age 18 years or older; capacity to give written informed consent (or, where capacity was lacking, consultee assent could be obtained); a provisional decision had been made by the assessing clinical team to admit the patient to hospital; located in either the acute medical unit, emergency department, or other acute areas; could be recruited within 24 h of presentation; and had an acute respiratory illness, or did not have acute respiratory illness but was suspected to have COVID-19 according to the current Public Health England (PHE) case definition. An episode of acute respiratory illness was defined as a provisional diagnosis of acute pulmonary illness—including pneumonia, bronchitis (non-pneumonic lower respiratory tract infection), and influenza-like illness—or an acute exacerbation of a chronic respiratory illness (including exacerbation of chronic obstructive pulmonary disease, asthma, or bronchiectasis). Patients were excluded if they declined nasal or pharyngeal swabbing, or had previously been included in the study and were presenting again within 14 days after the previous enrolment. The protocol originally allowed for recruitment of symptomatic members of hospital staff; however, this provision was abandoned after only a single staff member was enrolled.
The contemporaneous control group consisted of adults aged 18 years or older who presented with acute respiratory illness or suspected COVID-19 to the emergency department or acute medical unit during the study period (March 20 to April 29, 2020). These patients were eligible for inclusion in the intervention group but were not enrolled because of the capacity of the research team—we had insufficient research staff to recruit all patients with suspected COVID-19 during the day and did not have resources to deploy research teams overnight. Patients in this group were not asked to provide consent, and we collected routinely obtained, fully de-identified data (including demographic, clinical, and outcome data) retrospectively from hospital systems after local data protection assessment and approval.
COVID-19-positive status was defined as PCR positivity for SARS-CoV-2 on either assay.
To allow an assessment of diagnostic accuracy in the point-of-care testing group, if results were discordant between point-of-care and laboratory PCR testing, further PCR testing was done with two additional CE-marked SARS-CoV-2 assays (COVID-19 genesig Real-Time PCR assay [Primerdesign, Chandler’s Ford, UK] and VIASURE SARS-CoV-2 Real Time PCR Detection Kit [CerTest Biotec, Zaragoza, Spain) in another regional laboratory, with operators masked to the original results.
Demographic and clinical data were collected at enrolment and outcome data collected retrospectively from case notes and electronic systems. The ALEA and BC platforms were used for data capture and management.
As post-hoc measures, we assessed the proportion of COVID-19-positive patients enrolled into other clinical trials, and time from admission to enrolment in other clinical trials among COVID-19-positive patients.
All outcomes were measured for the duration of hospitalisation or up to 30 days (whichever was shortest), unless otherwise specified.
The sample size of 500 patients in the point-of-care testing group was chosen pragmatically, based on the availability of the QIAstatDx Respiratory SARS-CoV-2 Panel test kits. The control group consisted of all contemporaneously identified patients who presented in the same time period as the intervention and fulfilled the inclusion criteria in the same admission pathways. It was anticipated that the number included in the control group would be similar, based on the time periods for recruitment to point-of-care testing and the proportion of potentially eligible patients who were recruited. These numbers were considered to be sufficient to provide enough power for comparisons between groups and to estimate the diagnostic accuracy with acceptable precision. Although not formalised in the study design, this sample size corresponds to more than 90% power for a hazard ratio (HR) of 1·25 for turnaround time (equivalent to decreasing the median time to results from 24 h to <20 h, or increasing the proportion of patients with results within 24 h from 50% to 58%). Because the prevalence of COVID-19 during the study was highly speculative at the time of study conception, a formal sample size calculation for the evaluation of diagnostic accuracy was not done. However, a sample size of 500 patients in the point-of-care testing group would provide 80% power to give an approximately 90% chance of achieving a 95% CI width no larger than 10%, based on a sensitivity of 90% and a prevalence of 30%.
Statistical analysis was done by a dedicated medical statistician from the University of Southampton Clinical Trials Unit (SE) who was independent from the study team. Analysis was done with GraphPad Prism (version 7.0) and Stata (version 16) software. The use of multiple imputation was planned if missing data were to exceed 5% for the primary outcome or for key secondary outcomes, but was not needed.
Baseline characteristics and outcomes were compared between groups with use of χ2 tests for equality of proportions for binary data, and with independent-samples t tests (for mean values) or Mann–Whitney U tests (for median values) as appropriate for continuous data. Time to results and time to definitive ward arrival had no censoring. For duration of hospitalisation, deaths were right-censored at 30 days. Median differences and corresponding CIs were calculated with the Hodges-Lehmann estimate. Enrolment into other COVID-19 studies was only evaluated in COVID-19-positive patients.
For the assessment of diagnostic accuracy (point-of-care testing group only), measures were calculated on the basis of a composite reference standard of PCR positivity by any assay when confirmed by a second assay. Results are presented as sensitivity, specificity, likelihood ratios, and predictive values. CIs for sensitivity, specificity, and accuracy are exact Clopper–Pearson CIs, and for the likelihood ratios CIs were calculated using the Log method.
CIs for comparison of proportions are based on the Newcombe–Wilson method. CIs for individual proportions are based on the Wilson–Brown method except for measures of diagnostic accuracy as above.
This study was prospectively registered with the ISRCTN on March 18, 2020 (ISRCTN14966673).
Role of the funding source
The funders of the study had no role in the study conception, design, conduct, data analysis, or manuscript preparation. The corresponding author had full access to all data and the final responsibility to submit for publication.
Table 1Baseline characteristics of patients
NEWS2=National Early Warning Score 2.
Table 2Primary and secondary outcome measures
Table 3Diagnostic accuracy measures for QIAstat-Dx Respiratory SARS-CoV-2 Panel and laboratory PCR in the point-of-care testing group (n=469)
Results from each assay were compared against a composite reference standard (PCR assay with confirmation by a second assay), which showed 177 positive cases (prevalence 37·7% [33·3–42·3]) and 292 negative cases.
To our knowledge, this study is the first to assess the clinical impact of molecular point-of-care testing for COVID-19 for acute admissions, and shows that routine use of point-of-care testing can deliver rapid, accurate, and actionable results to clinical and infection control teams. The use of point-of-care testing led to a large reduction in the time to availability of results compared with laboratory PCR, and this reduction was associated with improvements in infection control measures and patient flow, with patients spending around 1 day less in assessment areas and having fewer bed moves before arriving in definitive COVID-19-positive or COVID-19-negative clinical areas. Less time spent in assessment areas means that non-infected patients spend less time unknowingly exposed to infected patients and are therefore less likely to acquire nosocomial infection. In addition, the rapid identification of COVID-19 patients in assessment areas could mean that health-care workers are less likely to be exposed and infected because COVID-19-positive patients would be rapidly moved to COVID-19-positive areas rather than staying in assessment areas for more than 24 h, where personal protective equipment recommendations are less stringent.
The fewer bed moves in the point-of-care testing group equates to a cost and time saving for hospitals because each bed space must be decontaminated after a patient has vacated it, and cleaning staff are also less likely to be exposed to heavily contaminated environments. Some patients who received point-of-care testing received their results while still in the emergency department and were transferred directly to definitive clinical areas, bypassing the assessment cohort wards entirely. If an even quicker turnaround time for results could be achieved, it is possible that all patients could have their results returned while still in the emergency department so that assessment cohort areas would become unnecessary.
The utility of routine point-of-care testing in facilitating early enrolment into clinical trials has not been fully recognised and should be highlighted. Although there were no approved therapeutic agents available during the current study, subsequently both the antiviral agent remdesivir and the corticosteroid dexamethasone have been shown to be efficacious in treating patients with COVID-19-associated pneumonia who require supplementary oxygen or respiratory support.
Routine point-of-care testing will enable the early identification of patients with COVID-19 as they are being admitted to hospital, facilitating rapid directed therapy with these agents in a test-and-treat paradigm maximising therapeutic benefit.
In addition to testing symptomatic acute admissions to hospital, point-of-care testing could also be used for assessing elective hospital admissions, primary care patients, hospital staff, and care home staff and residents, as well as for airport screening, school screening, and even population-level screening. However, because of the insufficient availability of suitable point-of-care testing platforms for all these uses at present, prioritisation is necessary and should initially be given to acute admission to hospitals to prevent nosocomial infections.
The findings of this study highlight the shortcomings inherent to instituting PCR assays based on a single gene target for a novel virus, without the availability of robust quality-assurance systems. Not all point-of-care testing platforms that are currently available have been shown to be sufficiently sensitive for use in secondary care, where the consequences of false-negative result can be very serious.
Point-of-care testing platforms with appropriate levels of accuracy must be selected based on the intended use case. We would also point out that point-of-care testing must be undertaken under a robust overarching governance structure that includes all elements of the testing process, including pre-analytic and post-analytic steps.
The detection of other respiratory viruses by the QIAstat-Dx Respiratory SARS CoV-2 Panel was infrequent during this study, presumably because of reduced circulation of viruses resulting from physical distancing measures, or because of viral interference from SARS-CoV-2. In Europe, COVID-19 incidence is currently low; however, a second wave is expected this winter which could coincide with seasonal epidemics of other viral infections, including influenza and respiratory syncytial virus infections. Therefore, the use of syndromic point-of-care testing for SARS-CoV-2 and other viruses will be vital for hospitals to rapidly differentiate the causes of acute respiratory illnesses and manage patients appropriately.
This study had a number of limitations, the most important of which was its non-randomised nature. The groups differed at baseline in terms of their respiratory symptoms and signs and NEWS2 scores, which can be explained by the higher prevalence of COVID-19 in the point-of-care testing group compared with the control group. Similarly, this higher prevalence can also explain the longer length of stay and higher rate of antibiotic use and ICU admission in the point-of-care testing group. Patients in the point-of-care testing group were recruited during the day by research staff and eligible patients were highlighted initially by clinical staff in the emergency department. It is likely that patients considered to be at high likelihood of COVID-19 were prioritised for point-of-care testing by clinical staff, leading to these differences.
We attempted to control for bias through the use of multivariable analyses for key outcomes. The multivariable analyses were based on a directed acyclic graph representing the research team’s knowledge of variables related to group assignment and time to results or definitive ward arrival, allowing us to identify and control for confounding variables while avoiding spurious associations between group and outcome. However, it is possible that other unrecognised confounders could exist that affect the relationships between group and outcome. We believe the plausibility and magnitude of effect for these outcomes make it highly unlikely that the process of group assignment would significantly alter the conclusions of the study. Although the results of this study are compelling, they are not fully definitive and ideally should be confirmed with a randomised trial. However, the relatively low incidence of COVID-19 in the UK makes conducting such a randomised trial difficult. In addition, there remain uncertainties around the ideal implementation model for point-of-care testing in hospitals. Different models for deployment include nurse-delivered point-of-care testing and laboratory technician-delivered testing, and the most appropriate and cost-effective models will vary between health-care institutions.
Another limitation of this study was that the same swab could not be used for both point-of-care testing and laboratory testing, meaning that a second swab was obtained contemporaneously for laboratory testing, which could have contributed to the differences in diagnostic accuracy in terms of swabbing technique. Our estimates of diagnostic accuracy are also complicated by the use of the PHE RdRp assay as our comparator. Because of the poor sensitivity of the RdRp assay, we cannot be sure that the QIAstatDx Respiratory SARS-CoV-2 Panel did not generate false-negative results that were also not detected by the RdRp assay but would have been detected by a more sensitive assay. In addition, several samples identified as positive by point-of-care testing could not be tested by the RdRp assay because samples were not sent to the laboratory, which could have affected the overall measures of performance. Finally, because this study was done in symptomatic adults presenting to hospital, the effect of point-of-care testing in other patient groups such as children, community-dwelling adults, and those who are asymptomatic or pauci-symptomatic, is currently unknown.
In summary, routine use of point-of-care testing for emergency admissions was associated with a large reduction in time to results and improvements in infection control measures, patient flow, and recruitment into other clinical trials, compared with laboratory PCR testing. The QIAstat-Dx Respiratory SARS-CoV-2 Panel had high diagnostic accuracy for the detection of COVID-19. Resources should urgently be made available to support the implementation of appropriate point-of-care testing platforms in emergency departments and admission units in hospitals in preparation for the next phase of the pandemic.
NJB assisted with the design of the study, screened and recruited patients, and collected data. SP, VVN, CTM, NJN, HW, and LP screened and recruited patients and collected and collated data. SK and NJC did the discrepancy analysis of samples for PCR testing. FB and HP were responsible for data extraction and management. GB collected and processed samples. BV did the independent performance evaluation for the QIAstatDx Respiratory SARS-CoV-2 Panel. SE analysed the data. TWC reviewed the medical literature, conceived and designed the study, oversaw the conduct of the study, participated in the interpretation of data, and drafted and wrote the manuscript. All authors reviewed and contributed to the manuscript during its development.
TWC has received speaker fees, honoraria, travel reimbursement, and equipment and consumables free of charge for the purposes of research outside of this submitted study from BioFire Diagnostics and BioMerieux; consultancy fees from Synairgen Research, Randox Laboratories, and Cidara Therapeutics; is a member of an advisory board for Roche and a member of two independent data monitoring committees for trials sponsored by Roche; and has acted as the UK chief investigator for an investigational medicinal product study sponsored by Janssen. All other authors declare no competing interests.