Government restriction efficiency on curbing COVID-19 pandemic transmission in Western Europe

In this article we assess the effectiveness of the restrictions, implemented at government level, to curb the first wave of the COVID-19 outbreak (January-May 2020) in seven European countries. The analysis put in evidence a strong correlation (correlation coefficient greater than 0.85) between one of the statistical parameters of the distribution representing the temporal evolution of the weekly number of patients admitted into the Intensive Cure Unit (ICU), i.e., the skewness, and the Stringency Index, an aggregate synthetic variable that reflects the level of implemented restrictions by a specific country on a scale between 0 (no restrictions) and 100 (full lockdown). Then, to assess if the skewness is consistent in effectively reflecting the applied restrictions, we computed the skewness for non-Covid flu outbreaks during four years from 2014-2015 to 2018-2019 and Covid-19 outbreak, where no restrictions were applied in Italy. The results highlight a significant difference between the values of the skewness for the normal flu with respect to the COVID-19 outbreak. This large difference put in evidence that the implemented restrictions modify the skewness of the ICU hospitalized patient number distribution. The skewness then can be used as a valid indicator to assess if the restriction had any effect on pandemic transmission and can be used as a support for decision makers.


Introduction
The World Health Organization (WHO) confirmed pandemic (March 2020) the new 2 Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) infection that 3 originated in Wuhan, China, in December 2019 (first reported cases), then spreading 4 to Italy and successively worldwide. Some recent studies highlight how the 5 meteorological conditions and air-pollution influenced the pandemic transmission in 6 several parts of the world [1][2][3][4][5][6][7] other than the social and economical factors [8,9] or 7 how fast asymptomatic carriers are found [10]. However, to curb the outbreak 8 transmission, following the Chinese example, the European governments implemented 9 different and progressive restrictions, as imposing face masks indoor and/or outdoor, 10 closing schools, restaurants, theaters, canceling sport events, concerts and any social 11 activity gathering public, up to ban people from traveling and leaving their homes 12 except for unavoidable and urgent needs, e.g., grocery, medical visits [11]. The applied 13 restrictions were not uniformly applied in Europe. They also produced a considerable 14 negative impact on the economy [12]. In this article, we assess, from a statistical point 15 of view, the effectiveness of the applied restrictions in seven European western countries as Italy, France, Sweden, Finland, Germany, Belgium and The Netherlands. 17 Other studies tried to assess the restriction effectiveness [11], but using indexes or 18 variables strongly dependent on testing policy or using models [13]. In this analysis, 19 for the first time, we assess a possible correlation between the different statistical 20 parameters characterizing the distribution of the weekly number of patients 21 hospitalized into the Intensive Care Unit (ICU) in the seven European countries and 22 the averaged Stringency Index (SI; [15]), a synthetic index that represents a metric to 23 quantify the severity of the applied restriction. It is represented by a single number 24 obtained aggregating the different restriction indexes and it ranges from 0 (no 25 restriction) to 100 (full lockdown). Once assessing which statistical parameter of the 26 ICU distribution shows the stronger correlation with the SI, we test its significance 27 level, i.e. if the statistical parameter is consistent and then it is able to effectively be a 28 proxy of the implemented restrictions. To validate the latter speculation, we check if 29 there is a significant difference of the considered statistical parameter values for 30 COVID-19 and the distribution of ICU hospitalized patients during several 31 pre-pandemic flu outbreaks where no restrictions were applied in Italy (data not 32 available for the other countries). 33 We chose, among all the possible variables, the hospitalized ICU patient number 34 because this variable is much more reliable with respect to other ones, e.g., the weekly 35 number of people tested positive [14]. The hospitalized ICU patient number is then a 36 proxy to assess the effective infection rate.

38
The objective of this study is to assess the restriction effectiveness from a statistical  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 March 1, 2022. ; https://doi.org/10.1101/2022.02.28.22271636 doi: medRxiv preprint index, we simply averaged the values along the first pandemic wave (January-May 53 2020).

54
By the way, the SI by itself cannot assess the real effects of the restrictions on 55 pandemic transmission and, thus, cannot quantify the benefits in reducing the severe 56 infection cases. For example, a certain country can implement the maximum level of 57 restrictions (SI=100, full lockdown) without having the capacity to enforce them. 58 Under those circumstances, the real SI value might have an offset because the citizens 59 are partially following the imposed rules. This situation leads to unexpected results 60 about the mitigation. Instead, an instance of a direct measure of the effects of the 61 restrictions can be given by the daily number of infected people. But again, this latter 62 variable is often an inaccurate metric due to several reasons, such as the impossibility 63 to track many asymptomatic cases or, in particular for the first wave, the reduced 64 ability in testing the population, thus often leading to a report relied upon an 65 underestimation of daily cases. Hence, following the indications in [14], we chose the 66 hospitalized ICU patient number thanks to its robustness with respect to other 67 variables.

68
The distribution of the number of ICU hospitalized patients over time can be 69 considered to assess the effectiveness of the imposed restrictions that are quantified by 70 the SI. Indeed, it can represent an empirical probability distribution function 71 measuring the probability that a person can enter the ICU because of a disease caused 72 by the SARS-CoV-2 virus. It is worth to be remarked that the distributions for all the 73 countries are temporally aligned, thus having that the first value into the sample space 74 coincides with the day when the first cases have been reported in each country, as 75 shown in Fig 1. A further issue to be addressed is how to compare the SI (a number) 76 with a probability distribution function. To this aim, we need to sum up it. Hence, we 77 computed, for each country, some of the most common synthetic indexes, i.e., the 78 moments of the distribution, to characterize its shape. Our guess is indeed that if we the sample standard deviation, σ Y , is defined as follows [16]: where µ Y is the mean of X, i.e.: The higher the variance (or, equivalently, the standard deviation), the greater the 88 variability of the distribution around its mean.

89
The skewness instead characterizes the degree of asymmetry of a distribution 90 around its mean. While the standard deviation is non-dimensional quantity, that is, 91 have the same units as the measured quantities, the skewness is a pure number that 92 characterizes only the shape of the distribution. It is defined for the random variable 93 Y as follows [16]: February 23, 2022 3/8 . CC-BY 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 March 1, 2022. ; where s Y is the skewness on the N scalar observations, {y i } N i=1 , of Y . For a unimodal 95 distribution, negative skew indicates that the tail is on the left side of the distribution, 96 and positive skew the vice versa. A zero value means that the tails on both sides of 97 the mean balance out overall. The normal distribution has a skewness equal to 0 (that 98 is symmetric).

99
The kurtosis is also a non-dimensional quantity. It measures the relative 100 peakedness or flatness of a distribution. The index is defined for the random variable 101 Y as follows [16]: where k Y is the kurtosis on the N scalar observations, with a kurtosis greater than 3 is termed leptokurtic, less than 3 is termed platykurtic, 104 and equal to 3 is called mesokurtic (i.e., the kurtosis of the normal distribution).

105
Leptokurtic distribution has heavier (fatter) tails than the normal distribution. is defined as follows: where σ X and σ Y are the two standard deviations for X and Y calculated as in (1), 117 and σ X,Y is the sample covariance: with µ X and µ Y computed as in (2). ρ ·,· is a normalized measurement of the 119 covariance, such that the result always has a value between −1 and 1. It is worth to be 120 noted that the measure can only reflect a linear correlation of variables ignoring other 121 types of relationships. A value of 1 implies that exist a linear equation describing the 122 relationship between X and Y , with all data points lying on a line. The correlation 123 sign is determined by the regression slope: a value of +1 implies that all data points lie 124 on a line for which Y increases as X increases (i.e., X and Y are perfectly positively 125 correlated), and vice versa for −1 (i.e., X and Y are perfectly negatively correlated).

126
A value of 0 implies that there is no linear dependency between the two variables.

144
Before computing the statistical analysis, the empirical distribution probability 145 distribution function of the ICU hospitalizations for each country is shown in Fig. 1.   shows a certain degree of correlation, but still weak (R 2 =0.67).

Fig 3. Correlation Kurtosis. Vs. SI The multivariate regression between the Stringency Index and the Kurtosis
The skewness instead, refers to a distortion or asymmetry that deviates from the 155 perfect bell-shaped distribution. 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 March 1, 2022. ; In this case, the multivariate regression of the skewness and the Stringency Index is 157 very high, with an R 2 =0.87. This result implies that the Stringency Index accounts   temporally shifted to make the maximum coincident (Fig. 1) . CC-BY 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 March 1, 2022. ; https://doi.org/10.1101/2022.02.28.22271636 doi: medRxiv preprint correlate (R 2 =0.67). As further step, we compared for Italy the skewness value during 189 COVID-19 with those obtained for normal flu outbreaks prior to 2020. The