This refers to the heated discussion and lockdown-related antagonism expressed by the public regarding the undesired positive correlation, which was presumed inverse, between testing capacity and COVID-19 cases. The supposed endgame of test ramp-up was to contain the virus, but countries have found themselves in the opposite situation. This may be due to more cases now being detected as a result of increased testing or because testing is not comprehensive and early enough to outweigh viral shedding. This may culminate into the United Kingdom’s “operation moonshot” controversy if the testing rate continues to be less than the infectivity rate [19].

Leakage refers to infectious individuals who are not detected. This includes those with either unreported symptoms or not presenting for testing despite being able to, those sent home due to an unavailability of tests, testing conducted on samples of compromised quality, unreported and untraced contacts, false negatives, and noncompliance to isolation and quarantine rules [20-22].

This refers to the conflicting interest of whether to test before symptom onset or upon reported symptoms. The Test and Trace program has been designed not to test people at the very early stages of infection for fear of missing out on the very cases it is meant to detect. The same is true when people are tested late [23,24]. A hidden “giant” within this paradox and a major contributor to transmissions is asymptomatic and presymptomatic infectivity. Research suggests that the serial interval of COVID-19 is shorter than the incubation period, indicating a possible infectivity multiplier effect before the onset of symptoms [25,26]. This is further compounded by the currently unknown operational false negatives [15].

This refers to testing that is not only being selective but is also not being delivered at the point of care. As a result, a major group of the public is eliminated. This has led to the lack of a comprehensive understanding of disease behavior.

The problem includes delays in testing those reporting symptoms, test-to-results delays, and time lapses in contact tracing. These system delays have led to increasing infections in the face of delivering the highest number of tests in Europe [27]. A disease that is as deadly as the present one does not tolerate turnaround time and mitigation program mistakes, the biggest of which has been the neglect of asymptomatic infectivity.

A literature search was performed on September 9, 2020, and constantly refreshed through December 22, 2020. The search involved all articles in English published in 2020, including gray literature. Search terms in Google Scholar included “[UK] [effectiveness of mass testing] [COVID-19] [SARS-CoV-2] [contact OR tracing] [contact tracing] [effectiveness of test and trace] –Animals –Influenza –HIV –Cancer.” The search was restricted to the year 2020.

An advanced search was performed in ScienceDirect for “[test and trace] OR [contact tracing] AND [COVID-19] AND [SARS-CoV-2] AND [asymptomatic] AND [symptomatic] OR [screening for SARS-CoV-2] OR [mass testing for SARS-CoV-2]” with article titles terms “[UK] AND [test and trace] OR [contact tracing] OR [community screening for SARS-CoV-2] OR [mass testing for SARS-CoV-2].” The search was restricted to the year 2020.

A search in PubMed included “((((((((mass testing for COVID-19 and “contact tracing”) OR (mass testing for SARS-CoV-2 and “contact tracing”)) OR (“test and trace”)) OR (“mass testing” and “symptom-based testing”)) NOT (Animals)) NOT (HIV)) NOT (Influenza)) NOT (Ebola)) NOT (Cancer).”

Finally, a search for “mass testing for COVID-19” AND “contact tracing for COVID-19” OR “mass testing for SARS-CoV-2” AND “contact tracing for SARS-CoV-2” was performed in Mendeley.

The preferential publication of studies was counteracted by ensuring that our search included gray literature. Missing data effect verification was performed by searching for gray literature that sought to compare the effectiveness of the intervention to the control [30].

We performed a detailed screening of the extracted data for individual studies. Extracted data included the study date, author, setting, study design, study objective, type of intervention, outcome, type of participants, strategies used, assumptions, data analysis, results, study limitations, and bias.

Following our study objective, studies for synthesis were grouped according to study outcomes. This was done to help capture the studies whose interventions were geared toward evaluating effects on outcomes of interest [31]. This also facilitated the synthesis of results according to the research questions.

Review findings were synthesized thematically. The quality of studies was critically appraised using the most recent tools based on study design, following the Public Health Ontario MetQAT (Meta Quality Appraisal Tool) 1.0 [32,33]. The methodology and risk of bias of modeling studies were assessed using the Relevance and Credibility Assessment of Modeling Studies tool proposed by Caro and colleagues [34]. Cohort studies were assessed using the Critical Appraisal Skills Program (CASP) tool [35]. The Specialist Unit for Review Evidence (SURE) tool was used to assess cross-sectional studies [36]. Studies were grouped into 6 main categories according to study outcomes, as outlined in the eligibility criteria, for easy analysis and synthesis. The quality of evidence generated by different studies was assessed using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) tool [37].

The direction of effect was used as the standardized metric because there was a lack of precision, which was specific to the effect of the intervention and control in the results presented by different studies. This did not permit the calculation of summary statistics [38]. In light of the above, vote counting was the best match in synthesizing the results. A sign test was used to indicate whether there was evidence of an effect or not. Equivocal effects between the intervention and control were considered to be distributed around the null hypothesis of no effect. This study made use of Synthesis Without Meta-Analysis (SWiM) reporting guidelines to report review results [39].

Tabular and graphical methods were deployed in presenting the results of this study. For the primary objective, the GRADE summary of findings table was used to present the certainty of evidence and a bar chart to present the effect direction of studies. For the secondary objective, forest plots were used to present the proportion of asymptomatic cases of SARS-CoV-2, using an Excel model proposed by Neyeloff et al [40].

Concerning the primary question, the results of studies that evaluated the effectiveness of the intervention and control within the United Kingdom, with low risk of bias, were prioritized since this was in line with the review objective. Real-time studies were also prioritized as these are more likely to resemble reality.

The heterogeneity of studies was assessed following the GRADE risk assessment factors [41]. The lack of a pooled effect size for modeling studies did not warrant us to perform a test for methodological diversity for the primary objective [42]. Regarding the secondary objective, however, variability was assessed by directly observing the confidence intervals on the plotted graphs.

The Relevance and Credibility Assessment for Modeling Studies tool was used to evaluate the methodology and risk of bias of modeling studies [34]. A total of 12 (34%) modeling studies [44-54,56] were included and assessed for risk of bias. Of the 12 studies, 5 (42%) were judged to be at low risk of bias, 4 (33%) to be at moderate risk of bias, and 3 (25%) to be at high risk of bias. The main concerns regarding the risk of bias included inappropriate population and setting: no real-world data set leading to either an unreported or inadequately reported validation process of models. There were issues with either the model validation process or the use of a real-world data set across 7 of the 12 studies (58%) that were rated to be either at moderate or at high risk of bias. Above all, the models were based on a series of assumptions, most of which may not work in real life. A summary of the risk of bias assessment of modeling studies is presented in Table 2. A more detailed risk of bias assessment of models can be found in Table S1 of Multimedia Appendix 3.

The single cohort study [57] included in the review was rated to be at moderate risk of bias, principally due to unsuitable population and setting. This study was implemented in Italy. The study’s risk of bias was assessed using the CASP checklist for cohort studies [35]. In this study, contact tracing was limited to control. There could have been issues surrounding participant selection due to unreported eligibility criteria. In addition, no details were provided about loss to follow-up and how this was managed. Table S1 of Multimedia Appendix 4 provides a detailed risk of bias assessment for this study.

The risk of bias assessment of cross-sectional studies was conducted using the SURE tool [36]. A total of 22 cross-sectional studies were assessed: 5 (23%) were judged to be at low risk of bias, 1 (4%) at moderate risk of bias, and 16 (73%) to be at high risk of bias. The authors of 10 (45%) studies failed to clearly state their study design. The study population and setting were unrepresentative in up to 82% (n=18) of the studies. Contact tracing as part of the intervention was lacking in 27% (n=6) of studies. The authors in 15 of the 22 studies (68%) did not justify their sample size. The fair selection of participants was not clear in 73% (n=16) of studies due to unreported eligibility criteria. Statistical methods used in study analysis were unreported in 45% (n=10) of studies, while the reporting of statistical analysis was judged to be inadequate in 18% (n=4) of studies. Nine studies (41%) did not provide technical details regarding sample collection and management. Additionally, only 50% (n=11) of studies provided technical details about testing. Unreported blinding was observed in 95% (n=21) of studies. Seven studies (32%) did not report limitations, leading to possible study bias. Lack of participant characteristics was also observed in 32% (n=7) of studies. Bias due to conflicting interests was judged to be possible in 18% (n=4) of studies since the authors’ conflicts of interest were not declared. Table 3 displays a summary of the risk of bias rating for cross-sectional studies. A detailed examination of how cross-sectional studies were assessed is found in Table S1 of Multimedia Appendix 5.

Vote counting was deployed as the method to synthesize results, in line with the direction of effect that was used. Studies were prioritized based on their degree of bias in the reported evidence. The GRADE diagram for assessing the quality of evidence was used to grade the evidence presented by the different studies [109].

Of the 12 studies categorized under this outcome, 4 (33%) were at high risk of bias, 3 (25%) were at moderate risk of bias, and 5 (42%) were rated as low. A total of 9 (75%) studies [44,46,47,49,51-55] were voted in favor of the intervention (95% binomial exact [BE] CI 42.8%-94.5%, P=.15). Three of the 12 (25%) studies [45,48,50] showed an unfavorable direction of effect and were voted in favor of the control (95% BE CI 5.5%-57.1%, P=.15). The body of evidence presented by the 11 modeling studies [44-54] for this outcome was downgraded by three levels to “very low.” First, studies were downgraded one level because they were neither randomized controlled trials nor real-time studies. An additional two levels of downgrading were due to serious study bias, interstudy variation, imprecision, and indirectness. The evidence from the lone cross-sectional study by Hagan et al [55] was downgraded by three levels to “very low” as well. It was downgraded by one level because the study was not a randomized controlled trial and was further downgraded by two levels due to methodological issues, imprecision, and indirectness.

The single study found for this outcome [56] was voted in favor of the intervention. This study was judged to be at high risk of bias. The quality of evidence was downgraded by one level given that it was not a randomized controlled trial. Being a model based on assumptions, coupled with study limitations, imprecision, and indirectness, the evidence was further downgraded by two levels. The evidence was classified as very low.

We found no study addressing this outcome. There have been mixed views regarding the safety of mass testing and contact tracing. Some argue that rapid mass testing will lead to false positives and negatives, thereby causing misinformation [79,110]. Others see both rapid mass testing and contact tracing as safety nets against virus spread [111-114]. Both nasopharyngeal and oropharyngeal swaps appear to be slightly invasive. There also exists a body of evidence regarding safety and security concerns from the public on contact tracing [115-117].

Again, no study was found regarding this outcome. Altmann and colleagues [111] found a high level of acceptance for app-based contact tracing. Their investigation was done across different countries including the United Kingdom [111]. It was also reported that there is a higher preference for government contact tracing applications than those managed by private companies [22].

There was no study evaluating this outcome. It remains, however, clear that the test and trace system is not equitable [18]. Testing that is delivered near the patient and at a walkable distance increases equity [118,119].

A total of 13 studies were retained to assess the primary objective. Statistical synthesis for the primary objective was based on the binomial probability test and BE CIs performed in Stata 14.2 (StataCorp LLC). Of the 13 studies, 10 (76.9%) favored the intervention (95% BE CI 46.2%-95.0%, P=.09), with just 3 (23%) studies voted in favor of the control (95% BE CI 5%-54%, P=.09). The above indicates that the intervention is a better strategy than the control in the suppression of SARS-CoV-2 transmissions. The probability that the above estimate is true if the conventional Test and Trace program was truly better than mass testing and contact tracing is just 9%. The 76.9% (10/13) favorable direction of effect is a clear enough majority vote to indicate that mass test and trace is truly more beneficial.

Assuming that the true probability of both mass testing with contact tracing and test and trace being equivocal is .50 under the null hypothesis (H₀: mass test and trace=test and trace), this study observed 10 out of 13 votes (76.9%), which is well above the expected binomial probability mean of 6.5 (SD 1.803) votes. Of the 10 studies, 4 (40%) in favor of the intervention were judged to be at high risk of bias, 3 (30%) at moderate risk of bias, and 3 (30%) at low risk of bias. A total of 23% (n=3) of the retained studies had representative samples and settings. Two of 3 studies (67%) implemented in the United Kingdom [52,54] voted in favor of the intervention were judged to be at low risk of bias. The effect direction plot of different studies, together with the associated risk of bias, is shown in Figure 4.

The results of 6 studies [44,47,52-54,56] were judged to be at low to moderate risk of bias. These studies were prioritized in concluding that the mass testing and contact tracing strategy was more effective in the suppression of community transmission of SARS-CoV-2 and the control of COVID-19 than conventional test and trace. The studies by Emery et al [44], Hill et al [52], and Alsing et al [54] were judged to be at low risk of bias. Two of these (ie, [52,54]) were both representative of the population and evaluated mass testing and contact tracing as a hybrid strategy, in line with the primary objective. Emery et al [44] failed to consider contact tracing but compared the effectiveness of testing based on symptoms and testing irrespective of symptoms. We concluded that the direction of effect will not be different if contact tracing were to be integrated since contact tracing is contingent on testing.

The generated GRADE evidence profile was used to present the synthesis findings regarding the primary objective (Table 4). Table S1 of Multimedia Appendix 6 provides details of how the evidence for different outcomes under the primary objective was graded.

A total of 21 cross-sectional studies and 1 cohort study [57-78] were retained under the secondary objective. There was limited precision in effect estimates with just 27% (6/22) of studies providing data on CIs for the proportion of asymptomatic carriers. Of the 22 studies, 7 (32%) were judged to be at low to moderate risk of bias. A graphical presentation of the asymptomatic proportion from the 22 studies (34 reports) can be seen in Figure 5. The sampled population ranged from 76 to 9,899,828 subjects, with a median sample of 395.5 subjects. The number of detected positive SARS-CoV-2 cases and asymptomatic carriers ranged from 0 to 1321 and from 0 to 300, respectively. Likewise, the mean number of positive cases and asymptomatic carriers were 120.9 (SD 280) and 49.3 (SD 71.1), respectively.

The proportion of asymptomatic cases among those testing positive ranged from 28% (483/1723, 95% CI 25.9%-30.2%) in the community (testing of residents) to 90.3% (28/31, 95% CI 74.2%-98.0%) among care home staff. The overall proportion was found to be 40.7% (1084/2661, 95% CI 38.9%-42.6%) (Figure 6). Two studies [64,65] with sample sizes of 121 and 217 subjects, respectively, detected neither cases nor found any asymptomatic carriers and were excluded in the evaluation of asymptomatic carriers among persons who tested positive.

The prevalence of asymptomatic SARS-CoV-2 cases was highest among homeless shelter residents (186/618, 30.1%; 95% CI 26.5%-33.9%), followed by care home residents (163/781, 21%; 95% CI 18%-24%), and lowest among hospital patients (0/217, 0%; 95% CI 0.0%-1.4%). The overall prevalence for all studies was 0.01% (1084/9,942,878; 95% CI 0.0%-0.0%). Excluding screening in the general population in the studies by Cao et al [76], Gudbjartsson et al [67], and Lavezzo et al [69], overall asymptomatic SARS-CoV-2 prevalence for all other settings was found to be 3.8% (601/15,616, 95% CI 3.5%-4.2%). Figure 7 shows the outcome prevalence in various specific sample populations.

The prevalence among asymptomatic populations from 6 studies [59,62,66,68,74,75] was 3.4% (189/5500, 95% CI 3%-4%). The prevalence in a mixed population from 17 studies [57,58,60-65,67,69-73,76-78] averaged 0.009% (895/9,937,378, 95% CI 0.0%-0.0%) (Figure 5).

Four studies [59-62] evaluated the outcome within the United Kingdom. Treibel et al [59] and Brown et al [61] evaluated the outcome among hospital staff, Graham et al [62] evaluated it in care homes, and Abeysuriya et al [60] among pregnant women at term. The overall asymptomatic SARS-CoV-2 proportion among detected cases in the United Kingdom was found to be 56.6% (120/212; 95% CI 49.6%-63.4%). The proportion of asymptomatic cases among those tested positive ranged from 44.2% (57/129; 95% CI 35.4%-53.2%) in care homes to 85.7% (6/7; 95% CI 42.1%-100%) in pregnancy. Figure 8 shows the relationship of asymptomatic proportion among detected cases and in the sampled population in different settings within the United Kingdom.

The overall prevalence of asymptomatic cases within the United Kingdom was found to be 3.8% (120/3194; 95% CI 3.1%-4.5%) with rates ranging from 2.2% (57/2631; 95% CI 1.6%-2.8%) among hospital staff to 14.9% (57/383; 95% CI 11.5%-18.8%) in care homes. Figure 8 demonstrates a higher overall rate among detected cases in the United Kingdom (120/212, 56.6%) compared to that of all studies (z=4.52, P<.001). We found in this review that asymptomatic cases were 1.4 times (56.6%/40.7%) more likely to be detected among positive cases in the United Kingdom than all studies put together. The overall SARS-CoV-2 prevalence rate in the United Kingdom (120/3194, 3.8%) was similar to that of all studies put together (601/15,616, 3.8%), excluding studies undertaken at the population level.

All unreported and unsuitable CIs were generated in Stata 14.2 (BE) and exported to Excel. The rule of three was applied to the studies by Jameson et al [64] and Callaghan et al [65] due to zero-outcome events in their sampled populations.