Abstract
The vast and rapid spread of COVID-19 calls for immediate action from policy-makers, and indeed, many countries have implemented lockdown measures to varying degrees. Here, we utilized nationwide surveys that assess COVID-19 associated symptoms to analyse the effect of the lockdown policy in Israel on the prevalence of clinical symptoms in the population. Daily symptom surveys were distributed online and included questions regarding fever, respiratory symptoms, gastrointestinal symptoms, anosmia and ageusia. A total of 2,071,349 survey responses were analysed. We defined a single measure of symptoms, Symptoms Average (SA), as the mean number of symptoms reported by responders. Data were collected between March 15th to June 3rd, 2020. Notably, on the population level, following severe lockdown measures between March 15 th and April 20th, SA sharply declined by 83.8% (p < 0.05), as did every single symptom, including the most common symptoms reported by our responders, cough and rhinorrhea and\or nasal congestion, which decreased by 74.1% (p < 0.05) and 69.6% (p < 0.05), respectively. Similarly, on the individual level, analysis of repeated responses from the same individuals (N = 208,637) over time also showed a decrease in symptoms during this time period. Moreover, the reduction in symptoms was observed in all cities in Israel, and in several stratifications of demographic characteristics. Different symptoms exhibit different reduction dynamics, suggesting differences in the nature of the symptoms or in the underlying medical conditions. Between May 13th and June 3rd, following several subsequent lockdown relief measures, we observed an increase in individual symptoms and in SA, which increased by 31.42%. Overall, these results demonstrate a profound decrease in a variety of clinical symptoms following the implementation of a lockdown in Israel, and an increase in the prevalence of symptoms following the loosening of lockdown restrictions. As our survey symptoms are not specific to COVID-19 infection, this effect likely represents an overall nationwide reduction in the prevalence of infectious diseases, including COVID-19. This quantification may be of major interest for COVID-19 pandemic, as many countries consider implementation of lockdown strategies.
Introduction
Since its isolation in December 2019, the novel coronavirus COVID-19 has spread to almost every country in the world. In Israel, the first infection of COVID-19 was confirmed on February 21st 2020, and in response, the Israeli Ministry of Health (MOH) employed a series of steps in an attempt to mitigate the spread of the virus in Israel. These steps were gradually aggravated; on March 9th all Israelis returning from abroad were instructed to begin a 14-day home-isolation period upon their arrival; on March 11th gatherings were limited to a maximum of 100 people; on March 12th educational institutions were closed; on March 25th new regulations within a distance of 100 meters; and on March 31th the above regulations were further exacerbated forbidding any gatherings of people from different households. Policy changes in Israel thus far were enforced on the entire Israeli population. In several cities with high infection rates, stricter regulations were implemented. April imposed new challenges for the Israeli population and the government, given the Jewish holiday Passover which is traditionally celebrated with family. To prevent gatherings during the holiday period, between April 8th and 15 th, harsher restrictions were imposed, with a national quarantine declared on the eve of Passover. On April 19th, a few of the restrictions were lifted and more employees were allowed to return to work and limited social events such as prayers and weddings were allowed to take place under severe restrictions on the numbers of people. In the days that followed, further relief regulations were approved, including partial reopening of schools on May 3rd, and opening of restaurants and bars on May 26th.
Policy-makers in Israel, as well as others worldwide, struggle to find the most suitable policy which will slow the rate of infection while maintaining the functionality of the economy. In the absence of a vaccine against COVID-19, control of disease spreading through large-scale physical distancing measures appears to be an effective means of mitigation 1. These may include closures of workplaces, schools, and a general lockdown. While many countries have implemented different policies and various degrees of lockdowns 1-3, the effect of these measures on the prevalence of clinical symptoms in the general population has not been quantified to date. Here, we utilize data obtained from nationwide, daily, one minute surveys, deployed by us from the early stages of COVID-19 spread in Israel 4, to analyse the effect of actions implemented in Israel on the prevalence of different clinical symptoms and at high geographical resolution.
Methods
Data
To obtain real-time information on symptoms across the entire Israeli population, we developed a simple one-minute online questionnaire 4. The survey was posted online (https://coronaisrael.org/) on March 14th, and participants were asked to fill it on a daily basis and separately for each family member, including those who are unable to fill it out themselves (e.g., children and eldery). The survey is filled anonymously to maintain individual privacy and was filled out 2,071,349 times until June 3rd.
The survey includes questions on respondents’ age, gender, geographic location (city and street), isolation status, smoking habits, prior medical conditions, body temperature measurement and self-reporting of symptoms experienced in the 24-hours prior to filling the survey. All prior medical conditions and symptoms included in the questionnaire were meticulously chosen by medical professionals, based on symptoms which were described as prevalent in patients with COVID-19 infection (e.g. cough and fatigue) as well as symptoms which were less prevalent in these patients (e.g. nausea and vomiting) 5,6 to allow better discrimination of individulas with possible COVID-19 infection. Responders were also asked questions regarding their adherence to the MOH guidelines including the number of people they have been in close contact with in the 24 hours prior to answering the survey (separately for adults and children) and the frequency of wearing a face mask outside of home. Newer versions of the survey introduced a unique identifier which allows identification of repeated responses of the same responder, while keeping the anonymity of the responders. The complete questionnaire can be found in Section 1 of the Supplementary Appendix (Figure S1).
The initial symptoms included cough, fatigue, myalgia (muscle pain), shortness of breath, rhinorrhea or nasal congestion, diarrhea and nausea or vomiting. Additional symptoms, including sore throat, headache, chills, confusion and loss of taste and/or smell sensation, were added in later versions. Efforts were devoted to reaching underrepresented populations through several channels, including call centers and media appearances.
Responses from participants who reported age over 120 years old or body temperature less than 35 or more than 43 degrees Celsius were excluded from the analysis (497 and 670 responses, respectively). Another 21,330 responses which contained invalid values in other data fields were excluded.
In order to analyse additional factors that may affect the prevalence of symptoms in the population, we utilized several additional external databases. Data on the number of COVID-19 cases in Israel per data was obtained from the Israeli MOH 7,8. Meteorological information on the weather in Israel during the study period was obtained from the Israel Meteorological Service 9. Information on mobility of individuals in Israel was obtained from the Google COVID Community Mobility Reports10. In order to estimate the prevalence of flu-like symptoms in the Israeli population in previous years, data from the Israeli MOH were obtained 11.
Statistical analysis
To estimate the changes following the Israeli lockdown policies, we analyzed the prevalence of reported symptoms throughout the period of data collection. We analysed the complete study population, and performed several sensitivity analyses by stratifying this population according to several demographic characteristics such as city of residence, population density in city of residence, and education level (see Section 4 in supplementary appendix). We further employed Inverse Probability Weighting (IPW), to examine the sensitivity of our estimates to possible biases in the data (see section 2 in supplementary appendix). Along with the analysis of individual symptoms, we also analyzed the average number of symptoms reported per date, which we term Symptoms Average (SA). When analyzing individual symptoms, we examined each symptom starting from the time it was included in our survey, and additionally an average of all symptoms which were incorporated in the survey at the same starting time (see Section 1 in Supplementary appendix). When comparing prevalence of reported symptoms on different dates we used a two sided t-test to evaluate the significance of change in symptom distribution. To capture the quantitative differences in the rate and timing of symptoms reduction, we fit a sigmoid function to each symptom as well as SA and analyze its parameters (see Section 3 in supplementary appendix).
We further analyzed the subset of responses which include a unique identifier, allowing us to follow repeated responses of the same individuals over time. We examined individuals who reported repeatedly over several weeks of the study, separated them into individuals who initially reported or did not report the presence of a specific symptom, or any symptom, and analyzed changes in their reports over the following weeks.
Results
Overall, 2,071,349 responses were collected during the study period of March 14th and June 3rd, 2020. Of these, 186,979 responses were excluded (Methods) and a total of 1,884,370 responses, 1,797,240 (95.38%) adults and 87,130 (4.62%) children were eventually included in the study. The characteristics of the responders are described in Table 1.
Characteristics of study population
First, we analysed the general effect that the restrictions implemented by the Israeli MOH had on mobility patterns and social distancing in the Israeli population. As expected, while physical distancing policies were implemented, the self-reported number of children and adults with whom a person has been in close contact with (defined as within approximately 2 meters for more than 15 minutes) during the last 24 hours was stable and low. A rise in these numbers was observed once restrictions began to release, on April 20th. First, a rise in the number of self-reported adults, starting on April 20th, when more workplaces were opened and small social events were allowed to take place, followed by an increase in the number of self-reported children which rose starting from May 3rd, when schools partially reopened (Figure 1D, Table 2). A similar trend was observed in data on the mobility of the Israeli population10, demonstrating an increase in both workplace and recreation mobility as the restrictions were gradually lifted (Figure 1F, Table 2).
A: Relative change of the Symptoms Average (SA) based on a 7-day running average. B: Relative change of symptoms reported since March 15 th based on a 7-day running average C: Relative change of symptoms reported since March 29th based on a 7-day running average. D: Relative change in number of contacted people since March 15th based on a 7-day running average. E: Relative change in temperature since March 15th based on a 7-day running average. F: Relative change in mobility since March 15th based on a 7-day running average. G: Number of responses per day. H: Cumulative number of confirmed COVID-19 patients in Israel per day. Dashed vertical lines mark dates with COVID-19 National related events, which are shown in Table 2 (marked in bold).
A: Relative change of the Symptoms Average (SA) based on a 7-day running average. B: Relative change of symptoms reported since March 15th based on a 7-day running average C: Relative change of symptoms reported since March 29th based on a 7-day running average. D: Relative change in number of contacted people since March 15th based on a 7-day running average. E: Relative change in the number of responders who report wearing a mask outside. F: Number of new confirmed COVID-19 cases per day.
COVID-19 related national events by date. Dates marked in bold represent events which are shown as vertical dashed lines in Figure 1.
Next, we calculated the prevalence of the various clinical symptoms reported by the responders throughout the study period, and analysed the changes following policies which were implemented during this time period (Figure 1A-C, Table 2). Our analysis revealed a significant reduction in the prevalence of all symptoms as physical distancing policies became more strict. The largest reduction was observed in cough, which decreased by 74.1% (p < 0.05) from March 15th to April 20th. The most common symptom reported by responders, rhinorrhea and\or nasal congestion, decreased by 69.63% (p < 0.05). We also observed a reduction of 83.83% in SA (p < 0.05).
In order to model the dynamics of different symptoms in time, a sigmoid function was fit for each of the symptoms (see Section 3 in supplementary appendix). In this representation, the slope of the function represents the rate of the decrease in the prevalence of the symptom in time, while the shift represents how many days after March 18th the symptom prevalence begins to decrease. Interestingly, although the prevalence of all symptoms decreased following the lockdown measures, different symptoms presented different patterns. For example, the largest rate of decrease (slope coefficient 0.509) was observed for muscle pain, while shortness of breath had a much smaller decrease rate (slope coefficient 0.1). In addition, while some symptoms, such as muscle pain and rhinorrhea or nasal congestion started declining quickly after the implementation of the lockdown policies, (shift of roughly 2 days) moist cough and sore throat, started declining much later (shift of 16 days). These variations suggest differences in the nature of the different symptoms, or in their underlying medical conditions.
Several sensitivity analyses showed a similar trend in symptoms reduction when stratifying the population by different characteristics. Geospatial analysis revealed a similar trend when separately analyzing the SA in different cities in Israel (a few sample cities are shown in Supplementary Figure S5 in Section 4 of supplementary appendix). The decrease in the SA is also evident in all cities, despite significant population and geographical differences between them. Moreover, this trend is also visible when subgrouping the population by different demographic characteristics such as population density in the city of residence, district of residence, average number of people per household and average employment percentage. Finally, weighting each individual in the study population using IPW, to reduce possible selection bias, also showed a similar decrease in prevalence of symptoms and SA (Supplementary Figure S2 in Section 2 of supplementary appendix).
Following the decrease in the prevalence of individual symptoms and in the SA, in the period following several releases in physical distancing restrictions, we observe an increase in all reported symptoms and in SA. Between May 13th and June 3rd, we saw a rise of 31.42% in SA, and the least reported symptom on May 13th - loss of taste and\or smell, rose by 72.27% until June 3rd. Along with the increase in reported symptoms, we see a rise in the number of people a responder interacts with, and a decrease in the number of responders which reported wearing a mask when leaving their homes.
The average temperature in Israel during the study period showed a slight upward trend, which coincides with seasonal changes in Israel during the study period, with an exception of a heat wave towards the end of May. This temperature trend does not explain both the decrease and the final increase that we observed in the symptoms during the study period. In addition, data obtained from the Israeli MOH on Influenza-like symptoms in Israel from previous years revealed a very low rate of influenza-like illness in the time period which is parallel to the time in which our survey has been distributed. During this time period in previous years, there were less than 1 weekly clinic visits due to influenza-like symptoms per 10,000 people, across all age groups 11, indicating that the symptoms reduction observed is not due to seasonal changes. Finally, although the rate of responses to the survey changes over time (Figure 1), at least 4,500 responses were received every day and a median of 16,000 responses was recorded throughout the whole study period. Additionally, to overcome the daily variability in the number of responders and obtain a more robust estimation, all analyses done were on smoothed data with a 7-day running window. Thus, the decrease in all reported symptoms is probably not a result of lacking reports, but rather a true trend of reduction in symptoms.
Analysis of repeating responses answered by the same individual for several weeks show a similar pattern of symptoms reduction. Less than 60% of symptomatic individuals in the first week of April, were still symptomatic in the second week of April. Moreover, less than 40% of symptomatic individuals in the first week of April were still symptomatic by the last week of May (Figure 3). Examining the responders which did not report any symptom in the first week of April, we see almost all of them did not report any symptom in the following weeks as well. Similar patterns repeat in all individual reported symptoms (Supplementary Figure S6), demonstrating the decrease in symptoms does not only occur in the population-level but also in an individual-level.
Left. Percentage of individuals who reported a symptom each week, out of the population of individuals who reported any symptom in the first week. Right: Percentage of individuals who reported a symptom each week, out of the population of individuals who did not report any symptom in the first week
Discussion
Here, to the best of our knowledge, we quantify for the first time the effect of lockdown policies on the prevalence of self-reported clinical symptoms in the population. We show that implementation of various measures of lockdown across time, which manifested in decreased physical contact between individuals in the population, are accompanied by a dramatic decrease in the prevalence of numerous clinical symptoms on the population level. Furthermore, analysis of repeated responses from the same individuals also demonstrated a decrease in reported symptoms on an individual-level. When these lockdown restrictions were loosened, the prevalence of these clinical symptoms gradually rose.
While physical distancing has been used effectively in past epidemics, decreasing human-to-human transmission and reducing morbidity and mortality 12-14, one of the concerns in implementing this strategy is that viral spread will be renewed upon relaxation of these measures 13,14. We present a data driven approach to estimate the effect of implementing physical distancing restrictions, as well as the effect of relieving them, on symptom prevalence. This may be viewed as a Symptometer, which detects sensitive changes in the prevalence of symptoms in the population. Recently, we established together with other researchers an international Coronavirus Census Collective, the CCC, aimed at utilizing information on daily self-reported symptoms as means to track disease spread and predict outbreak locations 15. Our approach may be further implemented globally to estimate and compare different lockdown strategies taken by different countries.
Previous studies during the current COVID-19 pandemic have estimated the effects of lockdown policies on clinical variables that are more specific to COVID-19 infection, such as the effect of the lockdown on the virus doubling time 1. We believe that our study provides a more global view on the overall reduction in the prevalence of infectious diseases during the lockdown, including, but not limited to, COVID-19 infection. Since the symptoms included in our survey are not specific to COVID-19 infection, the prevalence of other infectious diseases are most likely reflected by them. We observe variations in the rate of decrease of different symptoms which suggests differences in the nature of different symptoms, or differences in the medical conditions they represent. This is supported by the fact that the decrease in the prevalence of symptoms is global, and is not limited to specific symptoms which were described as common in individuals with COVID-19 infection. In addition, while the number of confirmed patients with COVID-19 infection in the Israeli population is increasing (16,506 on May 12th), it still represents a small fraction (0.18%) of the Israeli population 8 and is unlikely to be the sole contributor to the large decrease in symptoms visible in our data. Many infectious diseases other than COVID-19 are transmitted by Aerosol transmission 16 or physical contact 17, which are greatly reduced when implementing physical distance measures, and will be therefore affected by these measures. For example, a previous study found a decrease of 42% in the overall diagnoses of respiratory infections in children during school closure as part of an organized labor dispute during a previous influenza outbreak 18.
One may argue that the reduction in symptoms viewed in our data is due to other factors, which are not related to the lockdown policies, such as the normal seasonal variation of Influenza infection. We believe that this is not the case, since we started distributing our survey in the middle of March, when according to data obtained from the Israeli Ministry of Health, the levels of infection caused by Influenza were already extremely low. Furthermore, data on Influenza like symptoms in Israel from previous years suggest a very low rate of influenza like illness in the time of year in which our survey was distributed (with less than 1 weekly clinic visit due to influenza like symptoms per 10,000 people) 11. In addition, it is unlikely to be the result of changes in the weather during this time period, as the average temperature in Israel during the study period was relatively stable, with the exception of a few days of hotter than average temperatures around May 20. Trends in the data which directly reflect the policies enforced during the study period, such as the number of self-reported contacted people and percentage of responders who report on wearing a mask outside home, also supports our conclusion that the observed reduction in symptoms is related to the lockdown policies.
We observed an increase in the prevalence of all reported symptoms following the loosening of the lockdown restrictions, and gradual return of the economy to its full functionality. We believe that this quantification of the effect of loosening the lockdown policies on symptoms prevalence is of major interest both in general and in particular for COVID-19, as many countries consider which actions to contain the spread of the virus they should take and also in light of the upcoming possible 'second waves’ of infection. Israel is one of the few countries that has opened up relatively early and relatively quick, and the experience gained here can be used by other countries who still lag behind in their opening timeline relative to Israel.
Our study has several limitations. Our surveys were deployed as a voluntary tool, and thus the population of the study, which was defined by the responders to the survey, is prone to selection bias, and may not represent the entire Israeli population across all geographic locations. In an attempt to reduce this bias, the survey was distributed in six different languages to reflect the most common languages spoken in Israel and substantial efforts were made to reach disadvantaged populations by engaging leaders in local religious communities, and promoting the survey through both Hebrew and Arabic-speaking media channels. Furthermore, we perform several sensitivity analyses to evaluate the robustness of our estimates, which all show similar results (Figure S5). Of note, even if selection bias exists, it will most likely be present throughout the entire study period, and is therefore less likely to lead to the relative changes that are seen in the symptoms prevalence during the study period.
Despite the preliminary nature of our results, we believe that it is important to publish them now, as time is crucial. Overall, our study quantifies the effect of lockdown policy on the prevalence of several clinical symptoms, which characterize both COVID-19 and other infectious diseases. Such information is critical at this time, and may inform decision-makers worldwide as they consider actions to prevent the ongoing spread of coronavirus.
Data Availability
Tables of de-identified, aggregated data are available at https://github.com/hrossman/Covid19-Survey.
Data availability statement
Tables of de-identified, aggregated data are available at https://github.com/hrossman/Covid19-Survey.
Code availability statement
Analysis source code is available at https://github.com/hrossman/Covid19-Survey. Source code for the questionnaires is available at https://github.com/hasadna/avid-covider as an open source project, and can be readily adapted to use in other countries.
Ethics Declarations
The study protocol was approved by the institutional review board (IRB) of the Weizmann institute of science. Informed consent was waived by the IRB, as all identifying details of the participants were removed before the computational analysis. Participants were made fully aware of the way in which the data will be stored, handled and shared, which was provided to them and is in accord with the privacy and data-protection policy of the Weizmann Institute of Science (https://weizmann.ac.il/pages/privacy-policy).
Competing Interests Statement
The authors declare no competing interests.
Authors contribution
A.K., A.G., H.R., S.S. & T.M. conceived the project, designed and conducted the analyses, interpreted the results and wrote the manuscript. T.K., A.L., D.K., I.K., S.S., & N.G. designed and conducted the analyses, interpreted the results and wrote the manuscript. O.C. directed the organizational and logistic effort and interpreted the results. A.A.Z & A.I. provided and interpreted data. B.G, Y.D., R.B., V.S. & E.S. conceived and directed the project and analyses, designed the analyses, interpreted the results, wrote the manuscript, and supervised the project.
Acknowledgments
We thank the following for their contributions to our efforts: Y. Landau, T. Eldar, S. Kasem, T. Bria, S. Avraham, B. Kirel, A. Terkeltaub, N. Kalkstein, E. Krupka, M. Loitsker, Y. Sela, T. Bareia, J. Shalev, A. Brik, L. Samama, S. Cohen, J. Bitton, R. Priborkin, G. Beryozkin, G. Dardyk, S. Lanzman, N. Epstein, N. B. Shizaf, E. Daian, M. Bialek, M. Suleiman, R. Maor, Z. Maor, Y. Dinur, J. Klinger, R. Zaida, M. Eden, N. Hirshman, O. Dovev, Y. Grossman and the Public Knowledge Workshop (‘Hasadna’).
Supplementary appendix
COVID-19 daily questionnaire
Inverse Probability Weighting
Modelling symptoms dynamics
Sensitivity Analysis
Individual level analysis of changes in symptoms prevalence
1. COVID-19 daily questionnaire
COVID-19 Daily questionnaire *Questions that the responder is required to answer, & Questions that should be filled only once #Questions that were added in newer versions of the questionnaire
Integration time of different symptoms to our survey:
Symptoms reported from March 15th: shortness of breath, rhinorrhea or nasal congestion, cough, fatigue, nausea and vomiting, muscle pain, fever, diarrhea. We define SA from 15/3 as the average of these symptoms.
Symptoms reported from March 29th: cough, chills, confusion, loss of taste or smell, sore throat. We define SA from 29/3 as the average of these symptoms.
2. Inverse Probability Weighting
As our survey is voluntary, the population of responders may suffer selection bias. To estimate the sensitivity of our results to this bias we employed weighting of the individual data so it will better represent the true target population (Israel’s population), thus reducing selection bias. Weights are calculated by binning individuals into city and age group bins, and then weighting each bin to represent its true proportion in the target population. City was defined as a geographical area as defined by the Israel Central Bureau of Statistics (ICBS) 20. Respondents were associated with cities using the address provided in the questionnaire. Age was divided into 4 groups; 0-17, 18-34, 35-54 and 75-120, using the ICBS information, and each responder was assigned an age bin using the age provided in the survey.
A: Relative change of the Symptoms Average (SA) based on a 7-day running average. B: Relative change of symptoms reported since March 15th based on a 7-day running average C: Relative change of symptoms reported since March 29th based on a 7-day running average. D: Relative change in number of contacted people since March 15th based on a 7-day running average. Dashed vertical lines mark dates with COVID-19 National related events, which are shown in Table 2 (marked in bold).
3. Modelling symptom dynamics
We define the function as follows:
![Embedded Image](https://www.medrxiv.org/sites/default/files/highwire/medrxiv/early/2020/06/17/2020.04.27.20076000/embed/graphic-12.gif)
Where x is the aggregated mean of the symptom over the population, and a,b,c are the coefficients we fit. a determines the slope of the function, the bigger the value the steeper the slope. b centers the function and determines where the slope starts on the x axis, the bigger it is, the later in time the function will decrease. c determines where the slope starts on the y axis, the bigger it is the higher the symptom average was in time zero.
The table presents the fitted coefficients of the sigmoid function. a determines the slope of the function, b centers the function and determines where the slope starts on the x axis. c determines where the slope starts on the y axis
Each panel shows the relative change in symptoms prevalence since its integration time to our survey. Black dots - actual relative value, red line - fitted function.
A: Different symptoms decrease over time. In each subplot the black curve is the 3-day running average and the red curve is the fitted sigmoid function over the symptom.B: Relative decrease of the fitted sigmoid function SA based on a 3-day running average. C: Relative decrease of the fitted sigmoid function on different symptoms which were collected prior to the 18th of April based on a 3-day running average. D: Relative decrease of the fitted sigmoid function on different symptoms which were collected from the 24th of April based on a 3-day running average.
4. Sensitivity Analysis
Stratification of the population by several demographic characteristics was done by associating each response with a residence city, according to the address reported in the survey. Cities’ demographics features such as population density, and average housing density were obtained from the Israel Central Bureau of Statistics.
A: Selected cities. B: Geographic districts. C: Subgroups by residence city’s population density. D: Subgroups by average housing density in residence city. E: Subgroups by age groups.
5. Individual level analysis of changes in symptoms prevalence
Left: Percentage of individuals who reported the symptom each week, out of the population of individuals who reported the symptom in the first week. Right: Percentage of individuals who reported the symptom each week, out of the population of individuals who did not report the symptom in the first week.