key: cord-0665794-ca8dqcph
authors: Medo, Matus; Suster, Martin; Bodova, Katarina; Brazinova, Alexandra; Brejova, Brona; Kollar, Richard; Leksa, Vladimir; Lindbloom, Jana; Nosek, Jozef; Vinar, Tomas
title: Technical comment on The impact of population-wide rapid antigen testing on SARS-CoV-2 prevalence in Slovakia
date: 2021-05-28
journal: nan
DOI: nan
sha: 91dfadc345b242273de653a7d3d09166759f5d97
doc_id: 665794
cord_uid: ca8dqcph

Pavelka et al. (Science, Reports, 7 May 2021) claim that a single round of population-wide antigen testing in Slovakia reduced the observed COVID-19 prevalence by 58%, and that it played a substantial role in curbing the pandemic. We argue that this estimate, which is based on incorrect assumptions, is exaggerated, and that the relief was short-lived with little effect on mitigating the pandemic.

Pavelka et al. 1 analyze the effects of non-pharmaceutical interventions (NPIs) implemented in Slovakia in the fall of 2020 (see Fig.1 for the timeline) as a response to a worsening COVID-19 epidemiological situation 2 . They estimate that one round of population-wide antigen testing coupled with standard NPIs lowered the (observed) prevalence by 58%, and they extrapolate that repeated mass testing decreases the prevalence by the same factor in each round. We propose that: (i) effects of the mass testing were much lower and temporary;

(ii) their repeatability is questionable; and (iii) other externalities should be considered to evaluate long-term effects.

Some problems stem from an incorrect use of terminology. Pavelka et al. define "observed infection prevalence" as the proportion of positive results in mass antigen testing. They use this term interchangeably with the term "prevalence" (proportion of infected individuals in the population 3 ), even in the study title and abstract.

The estimate of a 58% decrease in observed prevalence in counties with multiple rounds of mass testing compares the test positivity on two successive weekends. However, the two population samples are influenced by multiple asymmetrical biases. First, individuals positive in the first round of testing (T1) and their whole households were instructed to quarantine and not to participate in the second round (T2), which caused a significant underestimation of prevalence in T2. This effect is also seen in the microsimulation model of Pavelka et al., where prevalence and test positivity are similar in the pilot round (T0), but they differ markedly in subsequent rounds (Fig 2a) . A similar discrepancy is also shown in Moreover, the authors consider the value of 58% a "robust" estimate (95% confidence interval 57-58%), even though they acknowledge their inability to explain a significant variance among counties, which ranged from 47% to 70% in the counties with at least 50 thousand participants. Pavelka et al. use mass antigen testing both as an intervention and as its efficacy measurement, which is often a source of unforeseen biases 6 . The fact that T2 was only implemented in selected counties provides a quasi-experimental setup to estimate mass testing efficacy. The RT-qPCR incidence (measured independently from the intervention) was reduced by 40% in counties with T2 between the week before T2 and the second week after, compared to a reduction by 22% in counties without T2 (Fig.1a) a 23% reduction attributable to the mass testing. Kahanec et al. 7 corrected for biases stemming from differences in the effectiveness of antigen testing in low-and high-prevalence areas, estimating 25-30% reduction over the same period.

This exaggeration of prevalence decrease is independently confirmed by other data (Fig.1b) :

hospital admissions (10-day lag after infection 8 ) decreased by about 30% from their peak (hardly the "sharp decrease in new admissions" 1 ), and excess deaths (15-20 day lag 9 ) decreased by about 24%, both consistent with a prevalence reduction of 20-30%. In both cases, the temporary two-week decline was followed by a long-term steady increase.

Pavelka et al. use a microsimulation to model prevalence in simulated populations under different parameter settings. They assume 100% test sensitivity, in stark contrast with the well-documented lower sensitivity of antigen tests 10 . Moreover, no simulation scenario takes into account the increase of mobility after T1 (Fig.1c) . Changing these assumptions alone dramatically reduces the cumulative effects of mass testing from 90% to 55% over three rounds (Fig.2b) . The strong effects presented in Fig.3 , estimated from the RT-qPCR incidence. Regrettably, the details of the computation of this number, or of the 70% decrease in prevalence compared to unmitigated growth, are not provided by the authors. However, a nationwide improvement had been observed already before T1 (Fig.1a,b,d) , suggesting a much lower epidemic growth (or even decline) at the time, which makes the basis of this computation unfounded.

Finally, Pavelka et al. conclude that mass testing likely had "a substantial effect in curbing the pandemic in Slovakia". Yet, the effects were short-lived. Mobility increased steadily starting from T1, indicating changes in social behavior (Fig.1c) . The long-term increase of RT-qPCR incidence began during T2, the positivity of routine antigen tests and the average viral load in positive RT-qPCR tests began to increase several days later (Fig.1d) . Hospital admissions started to rise two weeks after T2, followed by a steady increase in excess deaths (Fig.1b) , eventually resulting in overloaded hospitals, a 4-month lockdown from January 2021, and the world's foremost position in reported deaths per capita 11 in February 2021.

The first mass antigen testing in Slovakia was organized under unique circumstances difficult to repeat on the same scale. It was facilitated by an extensive involvement of the army and municipalities, minute-to-minute media coverage, the commitment of numerous medical personnel and volunteers (including international aid), and social effects of participants being tested simultaneously. Spontaneously emerging social pressure likely increased not only testing attendance, but also observance of the quarantine. However, all this led to only a temporary decrease in incidence, much lower than estimated by Pavelka et al. Such a mobilization cannot be conceived as a viable long-term strategy, which has been proven in Slovakia itself, as subsequent population-wide antigen testing in 2021 failed to produce anticipated results.

Based on these arguments, it is questionable whether mass testing in Slovakia should serve as a model for other countries. Changing test sensitivity to 70% instead of 100%, a value considered as realistic by Pavelka et al. Bar 5: Changing test sensitivity to 50%. Bars 6,7: People were motivated to participate in the mass testing by relief from extended measures as demonstrated by the increase in mobility in Figure 1A . By adding gradual growth of Re from 1 before the pilot round to 1.15 one week after, the effect of mass testing is significantly reduced again.

The impact of population-wide rapid antigen testing on SARS-CoV-2 prevalence in Slovakia

On Pilot Massive COVID-19 Testing by Antigen Tests in Europe. Case Study: Slovakia

Epidemiology: An Introduction

Evaluating social and spatial inequalities of large scale rapid lateral flow SARS-CoV-2 antigen testing in COVID-19 management: An observational study of Liverpool

Characteristic spatial scales of SARS-CoV-2 pandemics: lessons from mass rapid antigen testing in Slovakia

Ten common statistical mistakes to watch out for when writing or reviewing a manuscript

The Impact of Mass Antigen Testing for COVID-19 on the Prevalence of the Disease

Transmission, Diagnosis, and Treatment of Coronavirus Disease 2019 (COVID-19): A Review

COVID-19 -Time from symptom onset until death in UK hospitalised patients

Handling and accuracy of four rapid antigen tests for the diagnosis of SARS-CoV-2 compared to RT-qPCR

Slovakia: Coronavirus Pandemic Country Profile -Our World in Data

Estimating epidemiologic dynamics from cross-sectional viral load distributions

The authors would like to thank Michael Z. Lin and Marc Bonten for