Fig 1.
Geographical distribution of COVID-19 cases (per 1 million population) for 1,020 countries/regions worldwide.
(A–F) Monthly patterns for the cumulative number of COVID-19 cases on January 31, 2020 (A), February 29, 2020 (B), March 31, 2020 (C), April 30, 2020 (D), May 31, 2020 (E), and June 30, 2020 (F) based on the cumulative number of day-to-day COVID-19 cases since December 2019. See S3 Video. The map was prepared using shapefile reprinted from a freely available database (GADM; www.gadm.org).
Fig 2.
The distribution of COVID-19 cases across biome types based on the relationship between mean temperature and annual precipitation.
Biome classification is based on the scheme by Whittaker [20]. (TR) tropical rain forest; (TS) tropical seasonal forest/savanna; (TE) temperate rain forest; (SD) subtropical desert; (TD) temperate deciduous forest; (WS) woodland/shrubland; (TG) temperate grassland/desert; (BF) boreal forest, (TU) tundra. Colors indicate the number of COVID-19 cases (per 1 million population) and also contours of climatic regions with ≥1000 cases per 1 million population. (A–F) Monthly patterns for the cumulative number of COVID-19 cases on January 31, 2020 (A), February 29, 2020 (B), March 31, 2020 (C), April 30, 2020 (D), May 31, 2020 (E), and June 30, 2020 (F) based on the cumulative number of day-to-day COVID-19 cases since December 2019. Arrows indicate the location of Wuhan in China. See S4 Video.
Fig 3.
Patterns for the cumulative number of COVID-19 cases (per 1 million population) in relation to country type.
Based on the pattern of increasing COFVID-19 case numbers, individual countries were classified into four types (A–D): (A) Type A, countries that had a peak in the number of COVID-19 cases per week before the middle of April and had more than 1,000 COVID-19 cases per 1 million population; (B) type B, countries that exhibited an increase in the number of COVID-19 cases per week after the middle of June and had more than 1,000 COVID-19 cases per 1 million population; (C) type C, countries that had a peak in the number of COVID-19 cases per week before the middle of April and had less than 1,000 COVID-19 cases per 1 million population; (D) type D, countries that exhibited an increase in the number of COVID-19 cases per week after the middle of June and had less than 1,000 COVID-19 cases per 1 million population. The map was prepared using shapefile reprinted from a freely available database (GADM; www.gadm.org).
Fig 4.
Standardized regression coefficients and the partial coefficient of determination (r2) of each explanatory factor in the regression model explaining the cumulative number of COVID-19 cases (per 1 million population).
(A–F) Values for the period from December 2019 to January 31, 2020 (A), February 29, 2020 (B), March 31, 2020 (C), April 30, 2020 (D), May 31, 2020 (E), or June 30, 2020 (F). Temp, mean temperature; Temp2, squared mean temperature; Prec, mean monthly precipitation; Pop dens, population density; Visitor, relative amount of foreign visitors per population; GDP, gross domestic product per person; BCG, BCG vaccination effect as defined by the first PCA axis summarizing five variables related to BCG vaccination (see the Methods section for details); Malaria, relative malaria incidence; Age, relative proportion of the population aged ≥65 years; First cases, number of days from case onset. The regressions were conducted using ordinary least squares analyses. Vertical lines represent the 95% confidence intervals of parameters. Closed symbols indicate the significance of explanatory variables (p < 0.05). The coefficient of determination (R2) for the overall model is also shown. A nonlinear modeling analysis was also conducted using the random forest method with the same set of response and explanatory variables and the same covariates; the results of this parallel analysis are shown in S2 Fig.
Fig 5.
Coefficients of determination (adjusted R2) of the regression model explaining the cumulative number of COVID-19 cases (per 1 million population) from December, 2019 to June 30, 2020.
(A) Overall coefficient of determination of the regression model; (B) coefficient of partial determination (r2) for each explanatory variable in the model. The results shown are based on data starting from January, 2020, because the number of cases in December 2019 was insufficient for this analysis.
Fig 6.
Time-series pattern of the standardized regression coefficients of the model explaining the cumulative number of COVID-19 cases (per 1 million population) from December 2019 to June 30, 2020.
Vertical lines represent the 95% confidence intervals of parameters. The results are based on data starting from January 2020 because the number of COVID-19 cases in December 2019 was insufficient for this analysis.
Fig 7.
Residual pattern of the regression model predicting the number of COVID-19 cases (per 1 million population) for 1,020 countries/regions across the globe and for 47 prefectures in Japan.
The map was prepared using shapefile reprinted from a freely available database (GADM; www.gadm.org).
Table 1.
Drivers of the COVID-19 spread in relation to the country types.
Country types were defined by the patterns of COVID-19 spread (cases per 1 million population) (see Fig 3). Type A, countries that had a peak in the number of COVID-19 cases per week before the middle of April and had more than 1,000 COVID-19 cases per 1 million population; type B, countries that exhibited an increase in the number of COVID-19 cases per week after the middle of June and had more than 1,000 COVID-19 cases per 1 million population; type C, countries that had a peak in the number of COVID-19 cases per week before the middle of April and had less than 1,000 COVID-19 cases per 1 million population; and type D, countries that exhibited an increase in the number of COVID-19 cases per week after the middle of June and had less than 1,000 COVID-19 cases per 1 million population. The statistical significance of differences between the country types was tested by a Bonferroni’s multiple comparison test. Different letters indicate the values that are significantly different (p < 0.05) from each other.