key: cord-0796815-zvgg5duf authors: Miyazawa, D.; Kaneko, G. title: Face mask wearing rate predicts country's COVID-19 death rates date: 2020-06-23 journal: nan DOI: 10.1101/2020.06.22.20137745 sha: ad5164385808f9aa12cf366a5ec428f0e586af9a doc_id: 796815 cord_uid: zvgg5duf Identification of biomedical and socioeconomic predictors for the number of deaths by COVID-19 among countries will lead to the development of effective intervention. While previous multiple regression studies have identified several predictors for the number of COVID-19-related deaths, little is known for the association with mask non-wearing rate possibly because the data is available for limited number of countries, which constricts the application of traditional multiple regression approach to screen a large number of potential predictors. In this study, we used the hypothesis-driven regression approach to test the association with limited number of predictors based on the hypothesis that the mask non-wearing rate can predict the number of deaths to a large extent together with age and BMI, other relatively independent risk factors for hospitalized patients of COVID-19. The mask non-wearing rate, percentage of age [≥] 80 (male), and male BMI showed Spearman's correlations up to about 0.8, 0.7, and 0.6, respectively, with the numbers of deaths per million in 22 countries from mid-March to mid-June, 2020. The observed numbers of deaths per million were significantly correlated with those predicted by the lasso regression model including four predictors, age [≥] 80 (male), male BMI, and mask non-wearing rates from mid-March and late April to early May (Pearson's coefficient = 0.919). The multiple linear regression models including the mask non-wearing rates, age, and obesity-related predictors explained up to 75% variation of the number of deaths per million in the 22 countries with little concerns about multicollinearity. Furthermore, linear regressions using the mask non-wearing rate in mid-March as the sole predictor still explained up to 72% of the variation of the numbers of deaths from March to mid-June, emphasizing the importance of the strongest predictor. Although further verification is needed to identify causes of the national differences in COVID-19 mortality rates, these findings highlight the importance of the mask, age, and BMI in predicting the COVID-19-related deaths, providing a useful strategy for future regression analyses that attempt to contribute to the mechanistic understanding of COVID-19. correlations were found in age ≥ 80 groups. In contrast to the mask non-wearing rate, correlations between the number of deaths and age-related predictors were highest in mid-May and decreased thereafter. In general BMI showed lower correlations with the number of deaths per million with a clear sexdependent difference. Male BMI was more closely associated with the number of deaths per million than female BMI. Spearman's correlation . 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. The copyright holder for this preprint this version posted June 23, 2020. . https: //doi.org/10.1101 //doi.org/10. /2020 whereas those for other three parameters showed time-dependent increase and reached the apparent plateau around day 50 (mid-May). We subsequently made scatter plots showing the relationship between the number of deaths per million and predictors that showed high Spearman's correlation (Fig. 3) 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. The copyright holder for this preprint this version posted June 23, 2020. . https: //doi.org/10.1101 //doi.org/10. /2020 predictors including the mask non-wearing rates (mid-March), age ≥ 80 (male), and male BMI. As expected, the number of deaths per million did not show a clear correlation with the mask nonwearing rate (late April to early May). The number of deaths per million showed exponential type increases in most plots, in particular with the mask non-wearing rate (mid-March). We therefore applied the logarithmic transformation to the number of deaths per million. The transformed number of deaths per million on May 13 showed significant linear correlation with the mask non-wearing rate in mid-March (r = 0.865, P = 9.739e-08), mask non-wearing rate in late April to early May (r = 0.530, P = 0.009296), age ≥ 80 (male) (r = 0.658, P = 0.0008717), and BMI (male) (r = 0.682, P = 0.0004762). The relationship between the number of deaths per million (May 13) and age ≥ 80 (female) appeared to be similar in the scatter plot (data not shown), but their linear correlation did not reach the statistical significance with a lower Pearson correlation coefficient compared to that of age ≥ 80 of males (r = 0.416, P = 0.05381). age, and mask non-wearing rate. The number of deaths per million on May 13 was used with (bottom panels) and without (top panels) logarithmic transformation. Next, we used the least absolute shrinkage and selection operator (lasso) regression to find the potential effects of these predictors on the log-transformed number of deaths per million. The lasso regression estimates the regression coefficients with the lasso penalty terms, retaining important . 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 June 23, 2020 . . https://doi.org/10.1101 explanatory variables, while less important variables are removed from the regression (Tibshirani, 2011) . When all parameters in Fig. 2 were included together with age ≥ 80 (female), the lasso regression selected BMI (male) (beta = 0.21610188), age ≥ 80 (male) (0.34906818), the mask nonwearing rate (mid-March) (0.06019861), and the mask non-wearing rate in late April to early May (-0.02008978) with an intercept of -6.65720405. Age ≥ 80 (female) was omitted from the regression. The observed number of deaths per million showed high correlation with the number predicted by the lasso regression model ( Fig. 4 ; Pearson's coefficient = 0.918, P = 1.635e-09). We performed a hypothesis-driven linear regression using three or two of predictors selected by the lasso regression. Because the mask non-wearing rate is available only for 22 countries (Supplementary File 1), we first selected BMI (male), age ≥ 80 (male), and the mask non-wearing rate in mid-March for this regression (Table 1, model 1). This multiple regression model explained 76.0% variation of the number of deaths per million on May 13 (P = 2.075e-06), identifying the significant effects of mask non-wearing rate in mid-March and age ≥ 80 (male). BMI (male) also showed a trend for association, but the association was not statistically significant. Other combinations of three predictors also resulted in significant associations between the number of deaths per million on May 13 and the selected predictors with ~0.7 adjusted R square values (Table 1 ). We found several important features in the multiple regressions. First, the effect of mask nonwearing rate in mid-March was significant in all regressions tested with low P values, and exclusion . 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. The copyright holder for this preprint this version posted June 23, 2020. . https://doi.org/10. 1101 of this predictor markedly decreased the adjusted R square (model 2). Second, BMI (male) showed a relatively low association with the number of deaths on May 13, being significant only when the strongest predictor, mask non-wearing rate in mid-Match, was not included the regression (model 3). Third, the highest adjusted R square of 0.7934 was obtained when we include predictors age ≥ 80 (male) and mask non-wearing rates of mid-March and late April to early May (model 4). All regression coefficients were statistically significant in this model. Regressions including two predictors also underscored the importance of the mask non-wearing rate in mid-March as the predictor for the number of deaths per million on May 13. Namely, the mask non-wearing rate in mid-March was always a significant predictor (models 5-7) with adjusted R square of up to ~0.75. While BMI (male) and age ≥ 80 (male) were also significantly associated with the number of deaths per million on May 13, exclusion of the mask non-wearing rate in mid-March dropped the adjusted R square to ~0.65. 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 June 23, 2020. Because the mask non-wearing rate in mid-March showed the strong association with the number of deaths on May 13, we also built several single linear regression models for predicting the number of deaths from other dates using this predictor (Table 2) . Overall, changes in adjusted R square showed the similar tendency with those of the Spearman's correlation coefficients (Fig. 1) , in which the value gradually increased during March and April, reaching the plateau in mid-May. Overall, the mask nonwearing rate in mid-March explained up to 72% of variations of the number of deaths per million. 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 June 23, 2020. (3.111e-07) * P < 0.05, ** P < 0.01, *** P < 0.001 Lastly, we examined the relationship between the mask non-wearing rate, age ≥ 80 (male), and male BMI (Fig. 5) . The mask non-wearing rate in mid-March showed significant correlations with both age ≥ 80 (male) and male BMI. Multiple linear regressions found both age ≥ 80 (male) and male BMI as significant predictors of the mask non-wearing rate with regression coefficients of 5.224 (P = 0.042559) and 8.377 (P = 0.000682), respectively, with an interception of -160.020 (P = 0.006040). This multiple regression model explained 56.8% variation of the mask non-wearing rate in mid-March . 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. The copyright holder for this preprint this version posted June 23, 2020. . https://doi.org/10.1101/2020.06.22.20137745 doi: medRxiv preprint (P = 0.0001325). Compared to the mask non-wearing rate in mid-March, that in late April to early May showed a weak association with age ≥ 80 (male) and male BMI (Fig. 5) . Accordingly, in the following multiple linear regression analysis, age ≥ 80 (male) and male BMI explained only 29.4% variation of the mask nonwearing rate although the regression was still statistically significant (P = 0.01419), The regression coefficients of men BMI and age ≥ 80 (male) were 8.422 and 3.270, respectively (P = 0.0141 and 0.3776, respectively). Only significant correlations are shown (P < 0.05). The number of deaths per million on a certain date highly depends on when the epidemic began in the country, in particular in the early stages of the pandemic. In this study, we have tentatively determined March 20 as the start date of a global pandemic. It is likely that this assumption is a reasonable approximation for our hypothesis because all employed predictors (mask non-wearing rates, age and BMI) tended to show weak correlations with the number of deaths per million at the onset of the pandemic followed by marked increases as the progress of the pandemic. These features enabled us to select a specific date (May 13), at which most parameters showed the high correlations with the number of deaths per million, and to build regression models that well explain the variation of the response variable with limited number of predictors. On the other hand, it is noteworthy that the agerelated predictors were highly correlated to the number of deaths per million from the very beginning . 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. The copyright holder for this preprint this version posted June 23, 2020. . https://doi.org/10.1101/2020.06.22.20137745 doi: medRxiv preprint of the pandemic and then remained constant or gradually declined. Below we discuss the potential biomedical and socioeconomic mechanisms by which each predictor affects the number of deaths by COVID-19. We have proposed possible mechanisms by which obesity could be a risk factor for the severity of COVID-19 and H1N1 influenza infections (Miyazawa, 2020) . Namely, respiratory failure is the most important pathology that contributes to the severity of both COVID-19 and H1N1 influenza infections. Since obese patients generally show a restrictive breathing pattern and reduced lung volumes, the obesity-hypoventilation syndrome can lead to respiratory failure in COVID-19 patients, being a risk factor especially for patients with severe symptoms. Also, obesity has been reported to be a risk factor for the development of acute respiratory distress syndrome (Zhi et al, 2016) , which is a serious clinical manifestation of COVID-19 (Simonnet et al., 2020) . According to Moriconi et al. (2020), in patients of COVID-19, inflammatory markers were higher in obese group than non-obese group at admission, and obese group showed a worse pulmonary clinical picture, with lower PaO2. Thus, obesity could be a risk factor for already hospitalized COVID-19 patients or even for those with serious symptoms. These potential mechanisms may explain the reason why the association of BMI (male) and the number of deaths increased May to June 2020. Typical COVID-19 patients are hospitalized a few days to a week after infection and in some cases the symptom become severe a few weeks after hospitalization. If the obesity affects the last phase of the transition of the disease, there must be some delay in the increase in correlations between BMI and the number of deaths per million compared to the spread of COVID-19. However, we cannot rule out the possibility that obese people have high chance to be infected because obese adults are known to inhale air average 50% more per day than non-obese adults (Brochu, 2014), which may lead to increase the chance to inhale the virus. In contrast to BMI, age ≥ 80 (male) was correlated with the number of deaths from the onset of the global pandemic. This may reflect the common etiological feature that old people are more susceptible to infection and have higher risk of death after infection compared to young people because of the dysfunction of immunity. The precise reason why older age contributes to the number of COVID-19-related deaths remains to be elucidated, while the immune dysfunction has been proposed as a potential mechanism (Mueller, McNamara, & Sinclair, 2020) . Meanwhile, it is controversial whether the reduced immunity simply contributes to the higher mortality after infection. Excessive immunity could cause the cytokine storm and ARDS in COVID-19 in some cases (Ye, Wang, & Mao, 2020) . Furthermore, two patients with X-. 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. The copyright holder for this preprint this version posted June 23, 2020 . . https://doi.org/10.1101 linked agammaglobulinemia have recovered from COVID-19, suggesting that B-cell response may not be necessary for the recover from this disease (Soresina et al., 2020) . While other dysfunction of immunity, such as age-related T-cell dysfunction, may be involved in the severity of COVID-19 (Minato, Hattori, & Hamazaki, 2020) , the mechanism should be carefully investigated. Along with immune dysfunction, old individuals tend to provoke unwanted inflammation, which may contribute to the severity of COVID-19 (Mueller, McNamara, & Sinclair, 2020) . This tendency can be seen in SARS-CoV-infected old nonhuman primate (Smits et al., 2010) . Also, reported that older age is correlated with the higher salivary viral load, which is highest during the first week after symptom onset. Furthermore, as in the case of obese patients, decreased respiratory function of elderly could also be the reason. In elder individual, both forced expiratory volume in one second (FEV1) and forced vital capacity (FVC) decrease dramatically (Falaschetti et al., 2004) , to almost half of their lifetime maximum values, especially in males (Leem et al., 2019) . We identified a clear sex-dependent difference in the Spearman's correlation coefficients between the number of deaths per million and obesity-related parameters. This is in line with the fact that males are more susceptible to severe outcomes from COVID-19 than females (Montopoli et al., 2020; Grandi , Facchinetti, & Bitzer, 2020) . It has been proposed that estrogen may be protective against COVID-19 in females (Grandi , Facchinetti, & Bitzer, 2020) , while androgens may worsen COVID-19 in males (Montopoli et al., 2020) . It is also possible that smoking, which is more common in males globally, may play a role in worsening COVID-19 (Grandi , Facchinetti, & Bitzer, 2020) . Also, females generally have more fear to COVID-19 and tend to take more preventive behaviors than males (Yıldırım, 2020) Face masks are considered to be effective in preventing transmission of COVID-19 to others, and also have a protective effect in preventing transmission from others. A recent simulation study showed that universal masking at 80% adoption suppresses COVID-19 deaths significantly more than maintaining a lockdown. Comparison between the validated modeling results and empirical data from Asian regions showed an almost perfect correlation between early universal masking and successful suppression of COVID-19 outbreaks (Kai et al., 2020) . Also, Zhang et al. (2020) reported that face mask alone could have reduced the number of infections by over 78,000 in Italy from April 6 to May 9 and over 66,000 in New York City from April 17 to May 9, claiming that the failure in containing the propagation of COVID-19 pandemic worldwide is largely attributed to the unrecognized importance of airborne virus transmission. Likewise, the mask non-wearing rate in mid-March was found to be the strongest predictor for the . 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. The copyright holder for this preprint this version posted June 23, 2020. . https://doi.org/10.1101/2020.06.22.20137745 doi: medRxiv preprint (Li, 2005) when wearing face masks depending on the temperature and humidity. The small size of universal face masks may be a simple reason for the correlation. There was a significant positive correlation between age ≥ 80 (male) and mask non-wearing rate in mid-March. Due to the lack of data on mask wearing rates by age, this cannot be fully discussed either. It has been reported that age was correlated with fear and were related to preventive behaviors (Yıldırım, 2020) . Face mask wearing in public correlates with the proportion of people who are afraid of COVID-19, except for Scandinavian countries and United Kingdom (Smith, 2020). These facts are inconsistent with the correlation found in this study. Further data is required for detailed investigation. The mask non-wearing rate in mid-March alone explained up to 72% of variations of the number of deaths per million. The fact that the face mask non-wearing rate in mid-March, the early phase of the pandemic, more strongly predicted the total number of deaths per population over the entire threemonth course of the pandemic than the face mask non-wearing rate in late April to early May may suggest that face mask wearing from the early phase of the pandemic may be very important for suppressing COVID-19-related deaths, if mask wearing is assumed to be an independent risk factor. In the lasso regression, four factors, BMI (male), age ≥ 80 (male), the mask non-wearing rate (mid-March), and the mask non-wearing rate in late April to early May were chosen, and the observed and predicted model of number of deaths per million showed high correlation (Pearson's coefficient = 0.918, P = 1.635e-09). Although this has not been proven at this time to be the cause of the differences in COVID-19 mortality by country, it may be a useful method for predicting COVID-19 mortality. The death rate of the country in the early stages of a pandemic could be further predicted if factors such as the timing and number of infected people entering the country from abroad are taken into account. • Baseel C. 15 Rude Things Not to Do on Trains in Japan. 9 Sept. 2019, japantoday.com https://japantoday.com/category/features/lifestyle/15-rude-things-not-to-do-on-trains-in-japan • Brochu P, Bouchard M, Haddad S. Physiological daily inhalation rates for health risk assessment . 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. The copyright holder for this preprint this version posted June 23, 2020 . . https://doi.org/10.1101 in overweight/obese children, adults, and elderly. Risk Anal. 2014;34 (3) . 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. The copyright holder for this preprint this version posted June 23, 2020. . https://doi.org/10.1101/2020.06.22.20137745 doi: medRxiv preprint . 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. The copyright holder for this preprint this version posted June 23, 2020. . 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 June 23, 2020. . https://doi.org/10. 1101 Pandemic: SEIR and Agent Based Models, Empirical Validation, Policy Recommendations, submitted 22 Regression modeling of the North East Atlantic Spring Bloom suggests previously unrecognized biological roles for V and Mo What does and does not correlate with COVID-19 death rates medRxiv 2020.06.09 World Bank Open Data Longitudinal decline in lung function: a community-based cohort study in Korea Association of country-wide coronavirus mortality with demographics, testing, lockdowns, and public wearing of masks Predictors of COVID-19 incidence, mortality, and epidemic growth rate at the country level medRxiv 2020.05.15 Mask Appeal: The Addiction of Surgical Masks in Japan Effects of wearing N95 and surgical facemasks on heart rate, thermal stress and subjective sensations Antibody responses to SARS-CoV-2 in patients with COVID-19 Clinical and immunological assessment of asymptomatic SARS-CoV-2 infections Physiology and pathology of T-cell aging Why obesity, hypertension, diabetes, and ethnicities are common risk factors for COVID-19 and H1N1 influenza infections. Authorea Evaluating the determinants of COVID-19 mortality: A cross-country study • Tibshirani R (2011) Regression shrinkage and selection via the lasso: a retrospective Temporal profiles of viral load in posterior oropharyngeal saliva samples and serum antibody responses during infection by SARS-CoV-2: an observational cohort study Uncovering Japan's Coronavirus X-Factor Matters to the World The pathogenesis and treatment of the `cytokine storm' in COVID-19 The Effect of Wearing a Face Mask on Body Temperature The impacts of vulnerability, perceived risk, and fear on preventive behaviours against COVID-19 Identifying airborne transmission as the dominant route for the spread of COVID-19 Obesity Paradox" in acute respiratory distress syndrome: asystematic review and meta-analysis