key: cord-0312169-vhf66cd7 authors: Rufino, J.; Baquero, C.; Frey, D.; Glorioso, C. A.; Ortega, A.; Rescic, N.; Roberts, J. C.; Lillo, R. E.; Menezes, R.; Champati, J. P.; Fernandez Anta, A. title: Using Survey Data to Estimate the Impact of the Omicron Variant on Vaccine Efficacy against COVID-19 Infection date: 2022-01-21 journal: nan DOI: 10.1101/2022.01.21.22269636 sha: 642923fa24e8aa80f87765044e3101cfe546399b doc_id: 312169 cord_uid: vhf66cd7 Data collected in the Global COVID-19 Trends and Impact Surveys (UMD Global CTIS), and data on variants sequencing from GISAID, are used to evaluate the impact of the Omicron variant (in SouthAfrica and other countries) on the prevalence of COVID-19 among unvaccinated and vaccinated population, in general and discriminating by the number of doses. In South Africa, we observe that the prevalence of COVID-19 in December (with strong presence of Omicron) among the unvaccinated population is comparable to the prevalence during the previous wave (in August-September), in which Delta was the variant with the largest presence. However, among vaccinated, the prevalence of COVID-19 in December is much higher than in the previous wave. In fact, a significant reduction of the vaccine efficacy is observed from August-September to December. For instance, the efficacy drops from 0.81 to 0.30 for those vaccinated with 2 doses, and from 0.51 to 0.09 for those vaccinated with one dose. The study is then extended to other countries in which Omicron has been detected, comparing the situation in October (before Omicron) with that of December. While the reduction measured is smaller than in South Africa, we still found, for instance, an average drop in vaccine efficacy from 0.53 to 0.45 among those vaccinated with two doses. Moreover, we found a significant negative (Pearson) correlation of around -0.6 between the measured prevalence of Omicron and the vaccine efficacy. The Omicron variant of SARS-CoV-2 has seen an expressive increase since its initial classification in November 2021 [Oo21] . In South Africa it appears to have out-competed the Delta variant [Hod21] and has rapidly spread into Europe and other regions. Preliminary observations also indicate that it might spread faster and might have higher immune evasiveness than previous variants [KK21] . While vaccination still provides a level of protection against a serious disease [RHRM + 21], recent results [PvSG + 21, NKL + 21, KST + 21, LMD + 21] point towards a reduced level of protection against infection, especially from 15 weeks post the second dose [ASK + 21], and it is likely that the number of breakthrough infections (i.e., infections among vaccinated people) will rise with the spread of Omicron. It is also possible that the rapid spread of Omicron is not only a consequence of high transmissibility but also of immune evasiveness [LMD + 21] . Some of the preliminary models [SLD + 22] showed that high transmissibility in combination with high immune evasiveness could lead to a concerning health system overload [LRSC + 21] . Since the spring of 2020, the University of Maryland in collaboration with Facebook has collected extensive survey data on self-reported symptoms, infection, testing, behavior and, more recently, vaccination status (UMD These methods for classifying cases as positive or negative have two main limitations. First, they do not take into account diagnostic uncertainty, e.g., the same set of symptoms might be associated with some other condition. Second, these criteria are not adaptive to possible changes in the symptoms experienced as conditions change, e.g., as vaccination rates increase or new virus variants emerge. Thus, in this work, we introduce a new machine-learning-based classifier (described in Section 2.2) where the responses of users in the ground-truth set are used to train a model, which is then used to determine the status of users outside that set (users who do not report test information). We use the random-forest technique to design this classifier and the corresponding results are labeled Random Forest in what follows. We refer to the values obtained with each of these five classifiers (namely, Random Forest, UMD CLI, Stringent CLI, Classic CLI, and Broad CLI) as proxy estimates (or proxy for short). We compare each proxy estimate with the estimate of active cases obtained from the official number of cases as described by Alvarez et al. [ÁBC + 21] , where each new case is assumed to remain active for 10 days. These last estimates are called Confirmed. Both Confirmed and the estimates using the various proxies lead to time series with one estimated value per day. Each response to the survey includes a large number of questions (obviously, not all participants answer all questions). For training and inference of the Random Forest classifier, we use only questions with answers holding discrete values. From these we remove questions B7 and B8, which are only used to create the groundtruth set, as well as related questions, such as "B0: As far as you know, have you ever had coronavirus ?" and "B15: Do any of the following reasons describe why you were tested for COVID-19 in the past 14 days?". Finally, we do not use the questions related to vaccination, since we do not want them to influence the classification. The set of questions used can be found in Appendix B. The answers to this set of questions are "dummified" before they are used, i.e., a question with k possible answers is replaced by k binary attributes. The Random Forest model is generated with the randomForest function in R. No hyperparameter tuning is done, and the standard options of the function are used, with the exception of limiting the model to 100 trees to reduce the training time. Observe that the questions in Appendix B include all symptoms, but also have many more questions, including behavioral or demographic aspects. Additionally, the Random Forest classifier can give different weights to different symptoms, while previously proposed symptom based criteria are based on determining only whether a symptom is present or not. Thus, overall the Random Forest classifier is much more versatile than the symptom-based criteria described in the previous section. Additionally, there are other aspects that make the Random Forest classifier(s) more adaptive: • Firstly, we create different models for different countries. It is expected that different countries will have local characteristics, thus training and using the classifier with data from one same country can capture them. • Secondly, we create not one but several models per country: one for each 3-month period. This allows the model to capture and adapt to aspects that change over time, like the level of vaccination, the surge of new variants, or the stringency measures imposed. In order to verify whether the Random Forest classifier provides better proxy estimates than the symptomsbased classifiers, we selected a set of countries and tested the performance of each classifier in the last two quarters of 2021. To this end, we randomly divided the ground-truth set into a training and a testing set, with 70% and 30% of the responses of the ground-truth set in each subset, respectively. Table 1 shows the results for three countries that have detected Omicron in December for the periods of July-September 2021 (2021-Q3) and of October-December 2021 (2021-Q4). The classification performance metrics used are: • Accuracy: Ratio of cases correctly classified over the size of the test set. • Sensitivity / recall: Ratio of cases correctly classified as positive over the number of positive cases. • Specificity: Ratio of cases correctly classified as negative over the number of negative cases. • F-score: Harmonic mean of precision and recall, where the precision is the ratio of cases correctly classified as positive over the number of all cases classified as positive. As can be seen in Table 1 , Random Forest almost always shows the highest performance (marked in bold) among the classification methods used. As another test, we then selected a set with the 20 countries that have the largest number of available responses in the UMD Global CTIS dataset along with South Africa. For each of these countries, the first two 3 . CC-BY 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted January 21, 2022. and that of each of the candidate proxies in the period June 18th, 2021 2 to December 31st, 2021. All time series have one value per day, which is the average of the latest 14 days. We can make two observations from Table 2 . First, Random Forest turns out to be the candidate proxy that exhibits the highest correlation values in most countries. Second, 17 out of the 21 countries exhibit low TPR (≤ 0.1) values in at least one of the first two columns (either official or survey-based TPR), and 11 out of the 21 exhibit low values in both columns, with 7 having values no higher than 0.05 3 . This suggests that such countries tend to keep the case count relatively under control and report data somewhat correctly. We can thus interpret the high correlation between the Random Forest proxy and the Confirmed time series as a sign that this proxy constitutes the most promising option among the five proxies considered. As mentioned, each classifier will be used to determine whether survey responses correspond to positive or negative cases. Hence, the prevalence of COVID-19 estimated by a given classifier is the ratio between the number of positive cases over the total number of responses. Then, we consider four subsets of responses: • Unvaccinated: Participants that respond negatively to the question "V1: Have you had a COVID-19 vaccination?" • Vaccinated: Participants that respond positively to Question V1. • Vaccinated with 1 dose: Participants that respond positively to Question V1 and declare having received 1 dose in Question "V2: How many COVID-19 vaccinations have you received?" is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted January 21, 2022. ; https://doi.org/10.1101/2022.01.21.22269636 doi: medRxiv preprint • Vaccinated with 2 doses: Participants that respond positively to Question V1 and declare having received 2 doses in Question V2. Unfortunately, from the questions in the UMD Global CTIS it is not possible to know whether those with one dose are fully vaccinated, i.e., they have received a one-dose vaccine, or they simply received only the first dose of a two-dose vaccination. Similarly, it is not possible to know whether the participant received a booster shot. For each of these subsets, the prevalence of COVID-19 is computed as the fraction of responses classified as positive among the responses that report a given vaccination status. For each proxy we also estimate the vaccine efficacy (V E ) against illness as in [VRAB21], based on the estimates of prevalence among unvaccinated (P U ) and vaccinated (P V ): The confidence intervals of this metric are obtained using the Katz-log Method [AB15] . The main objective of this work is to evaluate the change in vaccine efficacy due to the Omicron variant. To this end, we evaluate the decrease in vaccine efficacy in South Africa from mid-June 2021 until the end of 2021. Moreover, to ensure that we have sufficient data for our estimates, we concentrate on three time periods in 2021, each lasting about a month, two dominated by the Delta variant: i) June 18 to July 18, 2021, which is the period considered in [VRAB21], and ii) August 9 to September 6, 2021; and one dominated by Omicron: December 1st to 31st, 2021 4 (see Table 3 ). In addition to considering South Africa as a whole, we also study the Gauteng province, which is among the most affected by Omicron in the country. Beyond South Africa, we study the 50 countries for which the UMD Global CTIS has the largest amount of data. We compute for all of them the vaccine efficacy in two periods. • Period 1: The month of October (in which Omicron was still not present). • Period 2: The month of December (in which Omicron was present). A computed efficacy value is only considered if it is non-negative, both prevalences P V and P U are at least 0.01, and the number of samples used to compute them is at least 1000. We only consider further the countries with at least one efficacy value in Period 2. 4 The information on variant presence is obtained from [Our21] , which extracts it from [EBM17] via [Hod21] . 6 . CC-BY 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted January 21, 2022. ; https://doi.org/10.1101/2022.01.21.22269636 doi: medRxiv preprint In the left plot we have the actual ratio (note that the y axis is in logarithmic scale). On the right plot all curves are normalized so the smallest value is 0 and the largest value is 1. We have observed that the information on prevalence of Omicron is available [Our21] with a significant delay. Hence, most countries do not report relevant presence of Omicron until the second half of December 2021. For that reason, we consider the prevalence of Omicron reported in Period 3: from December 15th, 2021 to January 7th, 2022 5 . Furthermore, among the countries mentioned above, in order to have a reasonable estimate of the prevalence of the Omicron variant, we consider only countries whose data is based on sequencing at least 30 virus samples. We say that these are the countries with presence of Omicron and use their estimated Omicron prevalence in Period 3 in some of our results. For all countries with presence of Omicron, we compare the estimated vaccination efficacy using Random Forest among all three vaccination groups and for both periods. For this, we adopt simple statistical methods, such as correlation analysis. Figures 1a and 1b show the prevalence of COVID-19 in South Africa in the period June 18th to December 31st, 2021, with the different proxies. The direct approach of Figure 1a shows a gap from the estimate Confirmed derived from the official number of cases to the other proxies. This gap can be explained by a combination of under-detection in the official number of cases (in South Africa the test-positivity rate is above 15%, as seen in Table 2 ) and the presence of a background of symptoms that never goes to zero. Figure 1b shows that if each curve is independently normalized to the unit scale all proxies closely track the evolution of the official number of cases Confirmed. In Figures 2a, 2b , 2c and 2d we show the COVID-19 prevalence in South Africa among Vaccinated, Unvaccinated, Vaccinated with 1 dose and Vaccinated with 2 doses with the diffferent proxies. We can observe that the UMD CLI and Stringent CLI proxies show a low infection prevalence in the period July-September and the month of December when compared with the Random Forest proxy. This is possibly because UMD CLI and Stringent CLI have a fixed combination of symptoms that did not capture well the new variants Delta and Omicron, while the Random Forest classifier is trained on a 3-month period and can adapt to these changes. On the other hand, Classic CLI and Broad CLI show a high prevalence in the period October-November, when the official data was showing that the number of cases was very low, possibly because of existing symptoms in the population not related to COVID-19. Focusing on the Random Forest proxy, and in Vaccinated (2a) versus Unvaccinated (2b) prevalence, we can observe that although in the unvaccinated population we see a similar magnitude across the two waves (August-September and December) we see that in the Vaccinated group there is a much higher rate of prevalence in the 5 Our World In Data [Our21] stopped sharing the variant data on January 10th, 2022, upon GISAID request. . CC-BY 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted January 21, 2022. ; https://doi.org/10.1101/2022.01.21.22269636 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted January 21, 2022. ; https://doi.org/10.1101/2022.01.21.22269636 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted January 21, 2022. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted January 21, 2022. December wave. This hints at a decrease of vaccine efficacy towards infection with the introduction of Omicron, as we will show next. Figure 3a shows the prevalence in South Africa estimated with Random Forest across the reported vaccination states. Here we confirm the observation that in the December wave there was a disproportionate increase of infections in the vaccinated groups (Vaccinated, Vaccinated with 1 dose and Vaccinated with 2 doses). We also observe that, as expected, subjects vaccinated with two doses show higher protection that those reporting only one dose (with Vaccinated somewhere in between since it combines both groups). As for vaccination efficacy, Figure 3b shows the estimates for South Africa, again with Random Forest. While the data in October-November has lower quality due to the reduced number of cases in that country, we can clearly observe the reduction of vaccine efficacy, towards infection, when contrasting the August-September period to the December period when Omicron dominates. Table 4 quantifies the measurements of estimated efficacy for the three periods of interest and for the five classifiers. We also provide a similar analysis in Table 5 with data restricted to the Gauteng province. Figure 4 shows an area plot, estimated from the UMD Global CTIS data, of the proportion of vaccinated with 1 dose, Vaccinated with 2 doses, and Unvaccinated from June 18th until December 31st, 2021. As can be seen, the ratio of the population vaccinated is low at the beginning of this interval, especially with two doses. Then, we can see a high increase in Vaccinated between July and October. We point out that in each time point of this plot the proportions are provided by a different set of surveys respondents, and it still closely captures the increase of vaccination. From the analysis of the 50 countries with the largest amount of data in the CTIS plus presence of Omicron and a calculated efficacy value, as defined in Section 2.5.2, we obtain a set of 24 countries. In Table 8 (in Appendix C) we show, for reference, the level of vaccination in these countries 6 . The next two tables, Table 9 and 10, present the estimates of virus prevalence in the same countries in the periods of October and December, and also estimates of vaccination efficacy towards infection. Both prevalence estimates and the derived efficacy estimates are obtained by the Random Forest classifier and shown with 95% confidence intervals. When data is insufficient to meet the defined selection criteria (c.f. Section 2.5.2), it is omitted and replaced by "-". Both tables are presented alphabetically by country name and also share a column depicting the most recent data on Omicron prevalence among all virus samples. While Table 9 focuses on the data from individuals that declared their overall vaccination status (using groups Vaccinated, Unvaccinated), Table 10 makes a more detailed characterization by considering the number of doses declared (groups Vaccinated with 1 dose, Vaccinated with 2 doses, Unvaccinated). We also observe that there is less data on individuals with only one dose, since this is a transient state in the vaccination sequence. The full information on sample sizes can be consulted in Appendix C in Tables 11 and 12. Figure 5a shows three pairs of box plots. Each pair allows comparing vaccine efficacy in October and December when considering data from the selected countries. . CC-BY 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted January 21, 2022. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted January 21, 2022. ; https://doi.org/10. 1101 /2022 The next three figures, Figures 5b, 5c and 5d , allow us to see a clear trend when plotting efficacy against the most recent relative level of Omicron presence in each selected country. For each case, we present a smoothed line, in blue, depicting a clear decreasing trend. Table 7 presents estimates for the correlation coefficient (using Pearson correlation) together with the corresponding p-value, which confirms its statistical significance for the usual α = 5%. After its surge in South Africa, the Omicron variant is increasing in prevalence in other countries. Although it is still unclear if this variant is associated to a milder disease [KBPC + 21] several studies have raised concerns over the decrease of vaccine effectiveness against infection [PvSG + 21, NKL + 21, KST + 21, LMD + 21] and this can lead to a wider spread of the virus even in countries with a high vaccination uptake. While we have observed that Omicron reduces the efficacy of vaccines, new studies show that T cells may remain effective with this new variant [AQM22] . Daily participatory symptom surveillance, with widespread deployment in most world countries along the last couple of years, has the potential to offer a new instrument for assessing both global and local trends in health status. While limited in assessing the ground truth, due to the smaller control over the sample design and the need to preserve anonymity, we believe that the vast number of daily survey responses can compensate some of these factors. In this study, we developed a method to adapt and calibrate against the reported SARS-CoV-2 infection status the selection of symptoms, and other covariates from the survey, along different time periods and locations. This was shown to provide a better proxy for assessing the trend in infections and more closely track the official reported cases, in particular in those countries that had a strong surveillance and consistent test positivity rates. Using By January 7th, 2022, there were a limited number of candidate countries exhibiting both a high prevalence of Omicron and a high level of sequencing data supporting it. Nevertheless, we extend the analysis to these countries and show the observed changes in efficacy when comparing the months of October (pre-Omicron) with December (with partial presence of Omicron). Although these results should be confirmed once the level of Omicron becomes more dominant in many countries, we have observed a significant level of correlation of around and beyond −0.6 between vaccine efficacy (with either one or two doses) and the prevalence of Omicron. We must also keep clear that this reduction of efficacy is towards infection, and while it does have impact on transmission it does not imply a reduction of vaccine efficacy in protection against serious disease, hospitalization and death. There are several assumptions that frame our analysis. We assume that UMD Global CTIS answers provide a sample of the population that is interchangeable among the Delta and Omicron dominated periods. Additionally, we did not take into account possible effects from waning immunity and vaccine boost shots, however by considering several different countries we have a mix of different vaccination timings. . CC-BY 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted January 21, 2022. In the UMD Global CTIS the following question is asked: "B1 In the last 24 hours, have you had any of the following?" [The21a]. The following is the list of possible answers (non exclusive): • Fever (B1 1). • Cough (B1 2). • Difficulty breathing (B1 3). • Fatigue (B1 4). • Stuffy or runny nose (B1 5). • Aches or muscle pain (B1 6). • Sore throat (B1 7). • Chest pain (B1 8). 15 . CC-BY 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted January 21, 2022. ; https://doi.org/10.1101 https://doi.org/10. /2022 • Nausea (B1 9). • Loss of smell or taste (B1 10). • Headache (B1 12). • Chills (B1 13). The following is the list of survey questions whose answers are used to create the Random Forest models, and to classify with them the responses: B1 1, B1 2, B1 3, B1 4, B1 5, B1 6, B1 7, B1 8, B1 9, B1 10, B1 11, B1 12, B1 13, B1 14, B1b x1, B1b x2, B1b x3, B1b x4, B1b x5, B1b x6, B1b x7, B1b x8, B1b x9, B1b x10, B1b x11, B1b x12, B1b x13, B1b x14, B3, B5, B6, B9, B10, B11, B12 1, B12 2, B12 3, B12 4, B12 5, B12 6, B13 1, B13 2, B13 3, B13 4, B13 5, B13 6, B13 7, B14 1, B14 2, B14 3, B14 4, B14 5, C0 1, C0 2, C0 3, C0 4, C0 5, C0 6, C1 m, C2, C3, C5, C6, C7, C8, C9, C9a, C12, C13 1, C13 2, C13 3, C13 4, C13 5, C13 6, C14, D1, D2, D3, D4, D5, D6 1, D6 2, D6 3, D7, D8, D9, D10, E2, E3, E4, E7, H1, H2, H3. The questions removed are B0, B7, B8, B15, and all the questions related to vaccination (V-questions). is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint sity of Maryland (UMD) to access their data, specifically UMD project 1587016-3 entitled C-SPEC: Symptom Survey: COVID-19 and CMU project STUDY2020 00000162 entitled ILI Community-Surveillance Study. The data presented in this paper and some of the programs used to process it are openly accessible at https: //github.com/GCGImdea/coronasurveys/tree/master/papers/omicron_efficacy_paper_medRxiv. . CC-BY 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted January 21, 2022. Confidence intervals for ratios of proportions: implications for selection ratios Efstathios Stavrakis, Govind Jeevan, and Christin Glorioso. Estimating Active Cases of COVID-19. medRxiv SARS-CoV-2 T Cell Responses Elicited by COVID-19 Vaccines or Infection Are Expected to Remain Robust against Omicron Global monitoring of the impact of the COVID-19 pandemic through online surveys sampled from the Facebook user base Data, disease and diplomacy: GISAID's innovative contribution to global health Facebook Data for Good. COVID-19 symptom survey -request for data access Covid-19 world symptom survey data api CoVariants: SARS-CoV-2 Mutations and Variants of Interest Partnering with a global platform to inform research and public policy making Short-term projections based on early omicron variant dynamics in england. medRxiv Omicron SARS-CoV-2 variant: a new chapter in the COVID-19 pandemic Analysis: A meta-analysis of early results to predict vaccine efficacy against omicron. medRxiv Joe Hasell, Cameron Appel, Charlie Giattino, and Lucas Rodés-Guirao. A global database of COVID-19 vaccinations Third bnt162b2 vaccination neutralization of sars-cov-2 omicron infection. medRxiv Classification of omicron (b.1.1.529): Sars-cov-2 variant of concern Our World in Data. Data on COVID-19 (coronavirus) by Our World in Data Increased risk of sars-cov-2 reinfection associated with emergence of the omicron variant in south africa Comparing the clinical efficacy of covid-19 vaccines: a systematic review and network meta-analysis Coronavirus Pandemic (COVID-19). Our World in Data Impact of vaccination and non-pharmaceutical interventions on sars-cov-2 dynamics in switzerland Argentina 44509 48807 3077 2778 40276 44590 3704 1884 36115 41783 Belgium 16448 18373 1687 1718 14266 16004 747 463 13327 15269 Brazil 198423 162402 9428 6552 183859 151114 38885 8680 142594 139517 Colombia 34859 33883 5437 2734 28457 30197 9979 7514 18034 22137 Denmark 19591 27284 917 1206 18279 25472 212 217 17781 24684 France 82767 111041 10234 11593 67393 95663 6369 4708 60218 89139 Germany 89348 110359 12601 11868 71980 95530 6655 5490 64611 88548 India 76675 68155 4076 2631 63803 60076 16798 7344 45967 51622 Italy 98712 112754 7023 6095 89120 103305 9066 5108 78852 96124 Mexico 139967 118861 12063 6472 119471 109330 35960 17776 82321 90162 Netherlands 27505 30803 3804 3380 23001 26621 2175 2025 20397 24087 Norway 16746 21862 935 1010 15536 20404 389 304 14980 19724 Poland 30295 38001 5318 6105 23924 30578 2327 2499 21236 27603 Portugal 22758 29352 1299 1368 21017 27340 3470 3172 17180 23631 Romania 45123 24638 11038 4917 32558 19022 4477 2451 27594 16192 Russia 35186 30037 12301 9001 21680 19884 2845 2819 18573 16779 Slovakia 9567 11323 1987 2208 7382 8841 306 487 6989 8215 South Africa 18308 19492 4149 4006 12805 14753 5009 4138 7624 10423 Spain 33455 51568 2035 2625 30652 47444 3814 3574 26453 43223 Sweden 53564 57823 3001 3200 49564 53544 699 443 48380 52348 Switzerland 14863 16755 2906 2617 11585 13742 886 676 10541 12824 Turkey 27159 22854 3238 2307 23033 19844 1473 729 21015 18561 United Kingdom 41812 47072 3080 3174 37421 42421 925 770 36109 41122 Vietnam 48955 39105 8043 1116 37073 36097 17325 3241 19233 32246 India 2899 2231 186 93 1629 1235 623 242 958 939 Italy 558 2610 120 329 394 2158 67 95 322 2035 Mexico 6881 4747 1201 485 5167 4047 2287 1038 2808 2956 Netherlands 487 1441 95 210 367 1179 60 106 299 1046 Norway 147 569 15 39 127 516 10 17 116 495 Poland 1039 2504 298 749 676 1614 90 173 572 1416 Portugal 170 821 17 55 142 742 28 98 112 632 Romania 2579 448 1109 175 1335 239 158 42 1158 186 Russia 1550 775 752 318 727 401 79 70 633 323 Slovakia 276 635 89 216 174 397 14 36 157 360 South Africa 695 2348 249 599 388 1672 214 564 167 1093 Spain 468 2776 65 186 375 2479 80 177 290 2277 Sweden 297 1037 48 103 234 899 8 16 225 878 Switzerland 170 639 61 175 102 445 10 21 90 418 Turkey 1479 1143 288 181 1125 897 136 57 962 818 United Kingdom 1321 2168 141 180 1124 1926 53 59 1060 1851 Vietnam 364 1271 58 35 251 1141 95 76 152 1043 Table 12 : Number of survey responses classified as positive by Random Forest in each period from the countries with presence of Omicron (as defined in Section 2.5.2), for each level of vaccination.