key: cord-0810505-duu01dob
authors: Ricon-Becker, I.; Tarrasch, R.; Blinder, P.; Ben-Eliyahu, S.
title: A seven-day cycle in COVID-19 infection and mortality rates: Are inter-generational social interactions on the weekends killing susceptible people?
date: 2020-05-08
journal: nan
DOI: 10.1101/2020.05.03.20089508
sha: b47a0bdd04d1a5b9887c91a2e144a4217e2bfecc
doc_id: 810505
cord_uid: duu01dob

We observed a significant seven-day cycle in (i) reported COVID-19 new casesa (in 7/12 countries), and in (ii) reported COVID-19 deathsb (in 7/12 countries), based on data from the 12 developed North-American and European countries that reported more than 1,000 deaths by April 28th (Spaina, Italya and Switzerlanda, Belgiumb, Canadab and the USAb, Germanya,b, the Netherlandsa,b, Swedena,b and the UKa,b). Daily reported numbers (March 29th to April 28th) are based on Our World in Data database (derived from the European Center for Disease Control and Prevention, ECDC). In all seven countries, numbers of new cases peaked on Thursday-Friday, five days after the weekend, corresponding with a reported ~5-day lag between contact with an infected person and the manifestation of clinical symptoms. Death tolls peaked on Wednesday-Thursday in all seven countries, ~12-14 days following a weekend, corresponding with the reported median of 14-day hospitalization before death. One may suspect that the weekend restricts the availability of testing or reporting of new cases, that are eventually reported during the following week. However, to completely account for both observations, this artifact should occur uniformly in all the above-mentioned countries and consistently along the entire month studied. Moreover, deaths are defined events and seems less likely to be recorded or reported inaccurately, at least for the vast majority of cases. Thus, we hypothesize that an increase in inter-generational social interactions occurs during the weekend, which facilitates transfer of COVID-19 from younger people to older vulnerable individuals. These and additional infected older people account for most recognized COVID-19 cases and deaths, which may occur at regular time intervals, specifically in this vulnerable population. Additional explanations may include weekly rhythms in immune functions, hospital care quality, or other various health-related behaviors. Our hypotheses should be re-tested and refined based on databases that accurately report events times when these will become available.

explanations may include weekly rhythms in immune functions, hospital care quality, or other various health-related behaviors. Our hypotheses should be re-tested and refined based on databases that accurately report events' times when these will become available.

On April 28 th , 2020, there were over 3 million reported cases of SARS-COVID-2 (COVID-19) worldwide, and over 200,000 deaths 1 . In many countries strict social distancing protocols and guidelines have been put in place to limit the spread of the disease. Yet, there seems to be a world-wide tendency to negate many of these restrictions, at least within certain subpopulations. In a few countries, there are no formal restrictions on social interaction. The price of not imposing such restrictions, or of defying them, is not easy to quantify and remains ambiguous. Also, it is unknown whether temporary or partially loosening of restrictions, during a weekend or a holiday, has significant impact on the number of infected people or death.

To address these questions, we scanned Our World in Data database 1 (derived from European Center for Disease Control and prevention data) for (i) daily new cases and (ii) for daily new deaths. We observe rhythmicity in both indices. To test if the observed patterns are not accidental, and whether they follow a 7-day cyclic rhythm, we studied all 12 developed North American and European countries that reported more than 1,000 deaths by April 28 th . In order to validate the results, data from the US were also collected from an additional source (Worldometer 2 ), and were analyzed in parallel.

Analyses were performed separately for each country and for daily numbers of deaths and daily numbers of new cases. In order to account for random variability, raw data was smoothed using a moving average with a window size of three days. Then, in order to account for global trends in each time series, the differences between every two consecutive time points (days in this case) were calculated and used for further analyses. Next, autocorrelations were calculated using an independence model method for calculating standard errors (implemented in SPSS, Version 25) using lags ranging between 1 and 16 days. An autocorrelation is the correlation of a signal with a delayed copy of itself as a function of a given lag. The results represent the similarity between observations as a function of the time lag between them. As such, a significant autocorrelation indicates the presence of recurring pattern such that values in the series can be predicted based on its preceding values. For example, if an autocorrelation with a lag of 7 days is positive and significant, it means that the values of the series tend to repeat themselves every seven days. We hypothesized that autocorrelations with a lag of 7 days will be significantly positive. We considered here significant autocorrelations coefficients that were outside the . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 8, 2020. . 95% confidence interval. Finally, for the purpose of comparison of the detrended time series with a fixed 7-day periodic sinusoidal we fitted the following model ( ) = 1 * (sin (2 + ⁄ 2 + 2 ) ⁄ ) + 3 where b1 represents the amplitude, b2 phase shift and b3 an offset and c was held constant to match the desired period of 7 days. The fit was computed with a simplex search method as part of the Matlab Optimization Toolbox (Matworks Inc).

The original database had 2 missing data points for the number of deaths (one for Spain and one for Switzerland). Missing values were replaced by the average between the adjacent days.

We first analyzed the number of deaths cases reported daily and found a significant periodic cycle of seven days in seven out of the twelve countries in the dataset. These countries were: Belgium, Canada, Germany, the Netherlands, Sweden, the USA and the UK. Fig. 1 presents a sample of the results obtained in a subset of countries (Germany, Sweden, Spain and the USA) selected to demonstrate the variability we observed in the current analysis. Qualitatively, a periodic pattern can be observed in the original time series for Germany, Sweden and Spain (left top panel); this pattern becomes more pronounced in the detrended data (black line, left bottom panels). The autocorrelation analysis (right panels) unequivocally demonstrated the presence of such periodicity in the time series from Germany and Sweden (the other countries with significant periodicity can be seen in Supp Fig. 1&2 ). As stated in the method section, autocorrelations falling beyond the 95% confidence interval are significant at the .05 level. For the US data, significant autocorrelations were obtained with lags of 6 and 8. Since the average of those cycles is 7, the US can be considered to have cyclic death period ranging between 6 and 8 days, with an average of 7 days. Furthermore, the analyses performed on the number of deaths on an additional database 2 for the US, showed a significant autocorrelations with the 7 days lag (see Supp. Fig. 5 ). Taken together, we can therefore consider the presence of seven-day period also for the US. Similar patterns emerge from the analysis of the records of new cases reported daily (Fig 2 [ where 7-day lag autocorrelations were significant for Germany, Sweden and Spain] and Supp Fig 3&4 [ 7-day lag autocorrelations significant for Italy, the Netherlands, Switzerland and UK]). Here, seven out of twelve countries showed significant autocorrelations with a 7-day lag. These countries were: Germany, Italy, Netherlands, Spain, Sweden, the UK, and Switzerland. It should be noted that once such a component is present in the time series, one should expect to see also significant components at multiples of the same fundamental period (i.e 14 days, 21 etc) as can be observed also in analyses presented here albeit only for 14 days as our data is limited to one month. In toto, . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 8, 2020. . https://doi.org/10.1101/2020.05.03.20089508 doi: medRxiv preprint our analysis of the data provided by the European Center for Disease Control identified a salient periodic component, centered around a seven-day cycle. The lower left pane presents differences between every two consecutive days (black line) with a fixed 7-day periodic sinusoidal superimposed (in red). The right pane presents autocorrelations for lags ranging between 1 and 16 days, with 95% confidence intervals. Significant autocorrelations fall outside the 95% interval. 

A 7-day cyclic pattern is evident in 7 out of the 12 countries studied, regarding COVID-19 new cases, and in 7 out of the 12 countries regarding number of deaths, with peak incidence of new cases occurring mostly on Fridays, and peak incidence of death occurring mostly on Thursdays and Fridays, uniformly in all the countries identified.

Several characteristics of the COVID-19 pandemic and disease course are known and hold relevance to these observations and to our speculated interpretations. First, the disease is commonly transmitted through respiratory droplets during un-protect social interactions 3 . Second, older people account for most recognized cases of the disease, and for the great majority of fatalities 4 and asymptomatic people carry and transfer the virus 5 . A trend towards increased social mixing during the weekends has been reported (including in countries present in our analyses 6 ). Third, epidemiological studies indicated that the median duration between contacting a sick/carrying individual and exhibiting clinical symptoms of COVID -19 infection is ~5 days 5 and that the median duration of hospitalization in people that do not survive the virus is ~14 days 7 .

Because the restrictions imposed in many countries or adopted voluntarily include limiting workplace hours for those in high-risk populations, susceptible/older people may become infected at higher rates during weekend-days compared to weekdays, as a result of increased social interactions with younger relatives or friends, as indeed indicated previously 6 . Under this assumption, and in accordance with the above known parameters, it follows that these vulnerable individuals will (i) exhibit clinical signs of COVID -19 infection at higher rates ~5 days after the weekend, on Thursday-Friday, and (ii) may have . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 8, 2020. . a greater risk of dying ~14 days following symptoms onset, on Thursday-Friday, as indeed reflected in the current observations. Higher infection rate during weekend days may stem from lower social distancing, or higher frequency of interactions between young and old individuals during the weekend. Furthermore, as most new cases of COVID -19 occur in people older than 70, and the great majority of deaths are within this age group, we believe that this vulnerable population is infected by younger relatives or friends unaware of carrying the virus. If this postulation indeed accounts for the observed 7-day pattern, many lives may be saved by greater adherence to social distancing, especially during the weekend. In a few countries (e.g., Sweden), the levels of oscillation were more than 50% of the average levels, suggesting large potential for saving lives. In the US, the low range of daily deaths in the period assessed is about 1700, while the high range is about 2500. We believe that social distancing may lower such difference. This potential should be weighed against the expected mental and medical costs of loneliness and isolation.

One cannot ignore alternative explanations that may account for the evident seven-day cycle in the numbers of reported new cases and reported COVID-19 mortality. A leading plausible explanation could be delayed reporting of these events. That is, while new cases and deaths are distributed evenly along all weekdays, reports of these events are delayed during the weekend, erroneously reported and ascribed to 2-4 days later, while events occurring on weekdays are reported without delay. This scenario may be relevant to reports of new cases that need to be confirmed or that depend on the decision to check and/or hospitalize a person. However, it is less likely to account for a significant and welldefined event such as the death of a patient.

Additional explanations may include weekly rhythms in hospital care quality 8 . Our hypothesis should be re-tested and refined based on databases that accurately report events' times, when these will become available. We hope that our analysis will trigger local health authorities to ascertain or improve the accuracy of the reported dates, and, if proven to be accurate, act accordingly to prevent more actively the spread of COVID19 during weekends.

Supp. Fig 1 & 2 Fig.1 . The upper left pane depicts raw data (gray points) with smoothed data overimposed (red line). The lower left pane presents differences between every two consecutive days (black line) with a fixed 7-day periodic sinusoidal superimposed (in red). The right pane presents autocorrelations for lags ranging between 1 and 16 days, with 95% confidence intervals. Significant autocorrelations fall outside the 95% interval. . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 8, 2020. . https://doi.org/10.1101/2020.05.03.20089508 doi: medRxiv preprint Figure 1 . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 8, 2020. . https://doi.org/10.1101/2020.05.03.20089508 doi: medRxiv preprint Figure 2 . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 8, 2020. . https://doi.org/10.1101/2020.05.03.20089508 doi: medRxiv preprint Supp. Fig. 1 . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 8, 2020. . https://doi.org/10.1101/2020.05.03.20089508 doi: medRxiv preprint Supp. Fig. 2 . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 8, 2020. . https://doi.org/10.1101/2020.05.03.20089508 doi: medRxiv preprint Supp. Fig. 3 . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 8, 2020. . https://doi.org/10.1101/2020.05.03.20089508 doi: medRxiv preprint Supp. Fig. 4 . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 8, 2020. . https://doi.org/10.1101/2020.05.03.20089508 doi: medRxiv preprint Supp. Fig. 5 . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 8, 2020. . https://doi.org/10.1101/2020.05.03.20089508 doi: medRxiv preprint

Coronavirus Pandemic (COVID-19) -Statistics and Research -Our World in Data

Modes of transmission of virus causing COVID-19: implications for IPC precaution recommendations

Updated understanding of the outbreak of 2019 novel coronavirus (2019-nCoV) in Wuhan

Asymptomatic Transmission, the Achilles' Heel of Current Strategies to Control Covid-19

SOCRATES: An online tool leveraging a social contact data sharing initiative to assess mitigation strategies for COVID-19

The epidemiology and pathogenesis of coronavirus disease (COVID-19) outbreak

Making sense of the evidence for the "weekend effect