Obesity as a driver of international differences in COVID‐19 death rates

Abstract Aim To determine what proportion of the inter‐country variation in death rates can be explained in terms of obesity rates and other known risk factors for coronavirus disease 2019 (COVID‐19). Materials and Methods COVID‐19 death rates from 30 industrialized countries were analysed using linear regression models. Covariates modelled population density, the age structure of the population, obesity, population health, per capita gross domestic product (GDP), ethnic diversity, national temperature and the delay in the government imposing virus control measures. Results The multivariable regression model explained 63% of the inter‐country variation in COVID‐19 death rates. The initial model was optimized using stepwise selection. In descending order of absolute size of model coefficient, the covariates in the optimized model were the obesity rate, the hypertension rate, population density, life expectancy, the percentage of the population aged older than 65 years, the percentage of the population aged younger than 15 years, the diabetes rate, the delay in imposing national COVID‐19 control measures, per capita GDP and mean temperature (with a negative coefficient indicating an association between higher national temperatures and lower death rates). Conclusions A large proportion of the inter‐country variation in COVID‐19 death rates can be explained by differences in obesity rates, population health, population densities, age demographics, delays in imposing national virus control measures, per capita GDP and climate. Some of the unexplained variation is probably attributable to inter‐country differences in the definition of a COVID‐19 death and in the completeness of the recording of COVID‐19 deaths.

and health-related covariates with a view to identifying explanatory factors underlying the inter-country variation in mortality rates. Government responses to the pandemic have varied greatly. Some countries, such as Norway and Denmark, imposed restrictions on the movement and association of citizens soon after the first case was identified in each country. 1 Sweden, on the other hand, has largely relied on voluntary public health measures to control the spread of infection. 2 Higher mortality rates have been reported among older adults 3 and those with certain pre-existing health conditions, including obesity, 4-6 diabetes 7,8 and hypertension. 9 Studies have also highlighted smoking as a risk factor for COVID-19 mortality. 3 There is conflicting evidence around the significance of ethnicity as a risk factor for mortality, which may be confounded by socioeconomic and medical disadvantages experienced by ethnic minorities. 10,11 There is also evidence of an interaction between the effects of obesity and ethnic group. 12 It is widely acknowledged that greater population density facilitates the spread of COVID-19. 13 Because of the prolonged incubation period shown among children, some have suggested that they facilitate viral spread, although this is contentious. 14,15 Another factor implicated in the spread of respiratory viruses, such as COVID-19, is the climate: there is growing evidence that COVID-19 spreads more readily in temperate latitudes compared with hotter climates. 16 Informed by this evidence, model covariates were drawn from the domains of age distribution, obesity and health, population density, socioeconomic factors and climate.
In addition to these demographic and environmental factors, a covariate was included, which measured how quickly governments responded to the coronavirus pandemic via severe restrictions on person-to-person contacts.

| MATERIALS AND METHODS
To be included in the study countries had to have a population of at least 2 million and a per capita gross domestic product (GDP) of at least $27,000 (US) per year. Countries for which data were considered to be unreliable for data transparency reasons were excluded, for example, China and Russia. The selection of countries was made prior to data collection and data analysis. The following 30 countries were included in the analysis: Australia, Austria, Belgium, Canada, the Czech Republic, Denmark, Finland, France, Germany, Greece, Hungary, Ireland, Italy, Japan, Lithuania, Malaysia, the Netherlands, New Zealand, Norway, Poland, Portugal, Slovakia, Slovenia, South Korea, Spain, Sweden, Switzerland, Turkey, the UK and the United States. These countries make up 15% of the world's population.
The date of the first COVID-19 case in each country was retrieved from the World Health Organization (WHO) website. 17 The number of COVID-19 deaths in each country up to 26 July 2020 was found from the John Hopkins University coronavirus resource centre. 18 National populations and life expectancy data were sourced from the United Nations. 19,20 Data were retrieved from the World Bank regarding the percentage of each population living in urban settlements, 21 the percentages of the population aged younger than 15 years and older than 65 years, 22,23 as well as the percentage of the population aged 20-79 years with diabetes. 24 Per capita GDP and adult obesity rates were sourced from the Central Intelligence Agency (CIA) World Factbook. 25,26 The percentage of the adult population (aged ≥18 years) with hypertension, defined as a systolic blood pressure of 140 mmHg or higher, or a diastolic blood pressure of 90 mmHg or higher, was recorded from WHO data. 27 Smoking rates were retrieved from the World Population Review, 28  control measures once compulsory restrictions were adopted that applied to the whole country and included at least two of the following: a limit on public gatherings to 30 people or fewer; the closure of public buildings, including non-essential shops, restaurants, bars, theatres and cinemas; the closure of schools; a requirement for the public to stay at home, except for medical reasons, exercise, essential shopping or work that could not be carried out remotely; a requirement for the public to observe social distancing. Determining when countries met these criteria for national COVID-19 control measures required the consultation of a large number of sources. Most of these are accessible from the Wikipedia pages on the COVID-19 pandemic in specific countries. 33 Full details of the sources consulted are provided in the supporting information (Appendix S1). The delay in introducing national COVID-19 control measures was taken to be the time from when the first case was reported to when these conditions were first met. For two countries, Sweden and South Korea, these criteria for national COVID-19 control measures were never met. The delay in introducing national control measures was therefore considered to extend to 26 July 2020, the date up to which COVID-19 deaths were counted.
The ethnic diversity index is defined as the probability that two randomly chosen individuals in a population come from different ethnic groups. 32 For an ethnically homogenous population the index is zero, while for a population with many ethnic groups it increases towards a value of one. covariates. This was carried out using the stepAIC procedure in the MASS library for R. 35,36 The covariate 'delay in introducing national COVID-19 control measures' differs from the others in that it was the result of government decisions. A further model (Model 2) was fitted that omitted this covariate from Model 1.
Countries were ranked by the model residuals from Models 1 and 2. These show the variation in death rates, which is not explained by the covariates in a given model. These rankings were compared with a ranking of countries by raw death rates.
Analyses were carried out in R 3.6.3. 35

| RESULTS
The date of the first case, the date upon which national COVID-19 control measures were introduced, the delay in introducing national control measures, and the COVID-19 death rate per million population per year are shown in Table 1. Summary statistics for all variables are given in Table 2.
The correlation between obesity rate and COVID-19 death rate (deaths per million population per year) was 0.153 for the raw death rate and 0.297 for the logged death rate. The initial multivariable model including all covariates had a coefficient of determination (multiple R 2 ) of 0.629, which indicates that T A B L E 1 Date of first case, date upon which national measures were introduced, national measures delay and death rate per million population per year (raw and logged)  Table 4. Where a country is higher in ranking (C) than in ranking (B), this indicates that a comparatively long delay in introducing national COVID-19 control measures contributed to increasing the death rate.
For example, Sweden, which never introduced national COVID-19 control measures as defined in this analysis, is ranked nine places higher in ranking (C) than in ranking (B), and the United States, which had a comparatively long delay of 78 days in introducing COVID-19 control measures, is ranked seven places higher in ranking (C) than in ranking (B).

| DISCUSSION
In the final multivariable model, the factor most strongly associated with COVID-19 death rate is the obesity rate, followed by the hypertension rate then population density. Factors related to the age structure of the population show somewhat weaker associations, with higher life expectancy and a higher percentage of the population aged older than 65 years being associated with higher death rates, as would be expected given the greater vulnerability of older people to COVID-19. A higher proportion of children aged younger than 15 years in the population is also associated with a higher death rate.
The role of children in spreading COVID-19 remains unclear, with recent evidence suggesting that children play less of a role than adults in transmitting the virus. 37 However, it is possible that children have Mean temperature ( C)  One strength of this analysis is the inclusion of a wide range of covariates resulting in a model that accounts for 63% of the intercountry variation in COVID-19 death rates. The countries studied comprise 15% of the global population, and therefore this represents a large proportion of the world's wealthier nations.
However, the sample size of 30 countries is small for linear regression analysis. This makes the results of the regression models less robust than where a larger sample size is available. Another limitation is that data were analysed at country level while there is considerable regional heterogeneity in the pandemic within countries. Some of the unexplained inter-country variation in death rates is probably a result of modelling the outcome and covariates at this comparatively high level.
In conclusion, a considerable proportion of the international variation in COVID-19 death rates in wealthier countries can be explained by obesity rates, population density, age distribution, hypertension, diabetes rates and national temperature, all factors which are outside government control, at least in the short term.
However, how quickly governments imposed virus control measures was also an influential factor on death rates, and this should inform policy in the case of future viral epidemics. There is also a case for strengthening policies aimed at tackling obesity, diabetes and hypertension, because an additional benefit of such public health measures may be that populations have greater resistance to future respiratory virus epidemics. The relationships between obesity, diabetes and COVID- 19 are not yet fully understood. 40 Further studies are needed to clarify how obesity, diabetes and hypertension influence death rates from respiratory epidemics such as COVID-19.