3.1. Excess All-Cause Mortality during COVID-19
The performances of benchmark mortality based on the three-year average estimation and the SARIMA-based estimation are displayed in Figure 1
. Average errors in most countries (based on the three-year average estimation) were below 0.02, while the average errors in SARIMA-based estimations were around 0.01. The percentages of average errors lower than 0.1 were approximately 0.9 for the three-year average estimation, while the percentages for SARIMA-based estimations were higher in most countries. Thus, we concluded that SARIMA-based estimation performed better than three-year average estimation in benchmark mortality estimation. The excess all-cause mortality during COVID-19 was further calculated based on the difference between the observed all-cause mortality and the benchmark all-cause mortality estimations based on the SARIMA model.
shows the observed and benchmark all-cause mortality rates over the 10th–52nd weeks of 2020 in 16 studied countries. Significant peaks of increasing death count in observed mortality rates were noted in comparison with the benchmark mortality in many countries. For instance, drastically increased mortality rates were seen in the differences between observed and benchmark data during the 20th to 30th weeks in Chile. Meanwhile, several peaks were observed during the 10th to 20th weeks, 30th to 35th weeks, and 40th to 52nd weeks in Belgium. On the other hand, continuously-increasing observed mortality rates (compared with the benchmark) were observed through the 10th–52nd weeks in the United States, indicating the large number of deaths caused by the COVID-19 outbreaks there.
shows temporal weekly mortality trends during the 10th–52nd weeks in 16 countries, in order to further distinguish the excess all-cause mortality. Diverse temporal mortality patterns were observed in different countries. In 12 countries—Chile, Croatia, the Czech Republic, France, Hungary, Lithuania, Luxembourg, Netherlands, Poland, Spain, Switzerland and the United States—the variations observed in excess all-cause mortality were consistent with those of COVID-19 mortality. Essentially, the death count of excess all-cause mortality was significantly larger than COVID-19 mortality. This indicated that the outbreaks of COVID-19 had a drastic influence on increasing all-cause mortality. On the other hand, additional peaks of increasing death count for excess all-cause mortality were noted in four countries: Belgium, Germany, Norway and Portugal. The inconsistency between all-cause and COVID-19 mortality can be explained by the potential impact of COVID-19. For instance, studies have reported an increased risk of death in hospitalized COVID-19 patients with cancers, fever, or acute respiratory distress syndrome [36
summarizes the COVID-19 and excess all-cause death counts and the percentages of changes for all 16 countries. Chile and Luxembourg showed 12.94% and 26.52% decreases from COVID-19 to excess mortalities, respectively. In all remaining countries, increasing death counts were observed, ranging from 1.81% to 373.60%. This indicated the significant discrepancy between COVID-19 and excess mortalities, suggesting the need to investigate excess all-cause mortality during COVID-19 and the potential impacts of environmental and socioeconomic conditions in facilitating control of the virus.
3.2. Air Pollution/Human Activity Impacts Analysis
summarizes the minimum, maximum, mean, and standard deviation values of air pollution data, human activity data, meteorological data, and excess all-cause mortality. On this basis, the impacts of air pollution and human activity were estimated on weekly excess all-cause mortality and COVID-19 mortality using the quasi-likelihood Poisson-based GAM model. The performances of GAM with lag effects of 0, 1, and 2 weeks were evaluated based on R square values. As shown in Figure 4
, the performances of models based on COVID-19 mortality were relatively higher than those based on excess all-cause mortality. Moreover, no significances were observed among the models with different lag effects, with most values within 0.5–0.9. Model performances also varied among different countries; Croatia obtained relatively higher R square values than others. Considering that the R square values of most countries were higher than 0.5, the performances of models based on both excess and COVID-19 mortality were ensured.
To distinguish the impacts of air pollution and human activities on both excess all-cause and COVID-19 mortalities, relative risks were calculated for each factor with a 95% confidence interval. Figure 5
shows the relative risks of weekly NO2
on the death count variation with lag effects of 0,1, and 2 weeks. For the estimation based on the excess all-cause mortality, relative risks varied from 0.99 to 1.05 in most countries. Lag effects showed less impact on the relative risk estimation of NO2
in most countries. However, the lag effects (1 or 2 weeks) in Belgium indicated a decreasing trend of NO2
on the increasing excess mortality, while less impact was observed with no lag effect. For the estimation based on COVID-19 mortality, the relative risks of NO2
were approximately within 1 to 1.08, suggesting that increasing NO2
levels corresponded to increasing COVID-19 mortality in most countries. Similar to the excess mortality-based estimation, the relative risks in Belgium also revealed negative patterns with lag effects. Compared with the COVID-19 mortality estimation, the relative risks of NO2
estimated by excess mortality were lower, suggesting less impact of NO2
when estimating potential environmental factors. On the other hand, despite the lockdown policies—which significantly reduced air pollutant emissions—the death counts of both excess and COVID-19 mortality were still positively related to ground-level NO2
shows the relative risks of PM2.5
based on excess all-cause and COVID-19 mortalities. While the ranges of relative risks were approximately within 0.96 to 1.04 (based on excess mortality), the relative risks based on COVID-19 narrowed down to 0.97–1.02. This indicated that the impact of PM2.5
on COVID-19 mortality has been overestimated compared with excess mortality. For instance, in Lithuania, Luxembourg, Portugal and the United States, no significant relative risks were observed with a 2-week lag effect based on excess mortality, whereas positive and negative relative risks from 0.99 to 1.01 were estimated in those countries based on COVID-19 mortality with a 2-week lag effect. Overall, lower levels of PM2.5
corresponded to increasing death counts in most countries. Additionally, 1- or 2-week lag effects increased these potential risks.
displays the influence of the frequencies of park visits on mortality variation. In particular, the relative risks in Chile showed significant differences from those in other countries. While relative risks—ranging from 0.995 to 1.01, based on COVID-19 mortality with no lag effect—revealed an overall positive trend between park visits and death counts, lower frequencies of park visits were related to increasing death counts when estimated based on excess mortality with lag effects of 0,1, and 2 weeks and when estimated based on COVID-19 mortality with lag effects of 1 and 2 weeks. A negative association was also revealed in the relationship between park visits and COVID-19 mortality in Lithuania. For other countries, higher relative risks of park visits were related to excess all-cause mortality with negative association—compared with those based on COVID-19 mortality with lag effects of 0,1, and 2 weeks. Specifically, a 1-unit decrease in park visits corresponded to approximately 1% increase in death count for excess mortality.
reveals the relative risks of workplace visitation to the excess all-cause and COVID-19 mortalities. Similar to park visits, the relative risks based on excess all-cause mortality were higher than those based on COVID-19 mortality. In terms of the excess all-cause mortality estimation, the relative risks in Chile showed a positive association between workplace visits and death count, ranging from 1 to 1.03. On the other hand, relative risks in other countries ranged from approximately 0.99 to 1 with lag effects of 0,1, and 2 weeks. This indicated a negative association between workplace visits and death counts in most countries. For the estimation based on COVID-19 mortality, a lesser influence was noted for workplace visits (approximately 0.995 to 1.005). The relative risks based on COVID-19 mortality also varied among studied countries, changing between negative and positive trends with lag effects in several countries, including Poland, Spain, and Luxembourg. On the other hand, consistent positive trends of relative risks were shown among different lag effects in some countries, including Chile and Portugal.
displays the average country-level relative risks (without 95% confidence interval) of weekly NO2
, park visits, and workplace visits on COVID-19 and excess mortality, with lag effects of 0, 1, and 2 weeks. Generally, positive relationships were revealed between weekly NO2
and COVID-19/excess mortality, while negative associations were shown between weekly PM2.5
, park visits, workplace visits and COVID-19/excess mortality with lag effects of 0, 1, and 2 weeks. Mean relative risks of four variables (calculated based on the average relative risk with lag effects of 0, 1, and 2 weeks) showed values of 1.0171, 0.998, 0.9981, and 0.9979, associated with COVID-19 mortality, and values of 1.0154, 0.9981, 0.9977, and 0.9977, associated with excess all-cause mortality. Regardless of the positive or negative directions, the data suggested that weekly NO2
levels were associated with lower relative risks for excess mortality than COVID-19 mortality—and that weekly park and workplace visits were associated with higher relative risks for excess mortality than COVID-19 mortality.