Only a few months after its disclosure by Chinese health authorities, SARS-CoV-2 had spread worldwide. By late winter of 2020, the World Health Organization (WHO) had designated the disease caused by the virus, COVID-19, to be a worldwide epidemic [1]. As can be seen from the effects of the approximately 80 million infections reported by the end of 2020, COVID-19 can manifest as mild, influenza-like symptoms or, far more seriously, as a severe and often deadly respiratory disease with pneumonia.

From the outset of the COVID-19 pandemic, the public has been exposed to numerous speculations about the degrees to which age and various underlying morbidities may amplify the risk of intensifying the severity of infection with SARS-CoV-2. Authoritative sources such as the US Centers for Disease Control and Prevention (CDC) [2] have issued warnings. Conditions cited by the CDC as increasing risk include cancer, chronic kidney disease, obesity, coronary disease, type 2 diabetes mellitus (DM), and sickle cell disease. The CDC also warns that asthma, hypertension, and liver disease, among other conditions, may subject a person to increased risk. In some countries, such as the United States, the incidence of COVID-19 has been more prevalent in some ethnic groups than others [3], leading to speculations that this disparity may be due to biology rather than behavior. Such differences are not unknown; for example, sickle cell disease is most commonly found among persons whose ancestors come from Africa and Mediterranean countries, where malaria is a prevalent affliction [4].

As many of the diseases cited by the CDC are more common in persons in late middle age and older, a common warning early in the pandemic was that SARS-CoV-2 presented a particular danger to persons over 50 years of age. In the initial wave of cases in China [5] and in the strong wave of cases in Italy, the probability of death due to COVID-19 was judged to be a strong function of a patient’s age, being only a few percent for those aged less than 50 years and rising to nearly 20% for patients older than 80 years. The large number of fatalities [6] in care homes in New York, the United Kingdom, and elsewhere have fueled speculations about the risks that comorbidities frequently seen in older people will increase the fatality rate of COVID-19. An alternative explanation is the decrease in immune functions with aging [7].

Why is COVID-19 more dangerous to older than to younger persons? Complicating the answer to this question, the actual mortality rate of COVID-19 remains highly uncertain, as the prevalence of asymptomatic and unreported infections has been estimated to be from 2 to 5 times greater than that of infections with clearly defined symptoms. An early exemplary source of testing-based data was provided by the passengers aboard the Princess Line cruise ship, the Diamond Princess, on which half of the passengers who tested positive for COVID-19 were asymptomatic or at least presymptomatic [8]. To some degree, that uncertainty may explain the very wide distributions of reported (or apparent) rates of mortality (case fatality rates) of COVID-19 in different countries, ranging from <0.03% (Singapore) to almost 30% (Yemen). Moreover, in most (but not all) countries, by December 2020, the integrated average case fatality rate had declined significantly from the high levels seen in March and April of 2020.

For a less anecdotal (and less speculative) assessment of risk factors for serious consequences of COVID-19, a data-driven examination of worldwide national statistics seems to be in order, with the goal of identifying strong correlations of mortality due to COVID-19 with other potential comorbidities and even with ethnically specific biological and economic factors. Based on a global investigation of the statistical correlations on a nation-by-nation basis between the statistical data concerning reported COVID-19 fatalities and potential comorbidities, this paper presents a set of calculations of linear and multivariate correlations that may influence the severity of the infection.

To gain confidence in this statistical approach, one can plot two variables for which one may expect to see a correlation: GDP-PPP and median age (Figure 2). Here, the linear correlation is quite high (0.625). Closer examination of Figure 2 suggests a limitation of considering only linear correlations. The countries circled in red show a strong correlation, while those in the green ellipses show scarcely any correlation of a nation’s wealth with the age of its population. Clearly, a refinement of the statistical approach is needed. Identifying the data underlying each point with each country’s region in Figure 3 reveals that median age and national wealth are essentially uncorrelated for European nations but strongly correlated for countries in Africa and Asia. Regional grouping was thus adopted throughout this study.

To illustrate the utility of this refinement, Figure 3 shows the correlation between deaths per 100,000 persons due to malnutrition and national wealth measured by GDP corrected for purchasing power (GDP-PPP). The relatively strong (0.455) correlation is driven by the high rates of malnutrition in Africa, Central America, and the less economically advantaged countries of Asia. No such effect is apparent in Europe.

From the outset of the pandemic, national health authorities warned the public about the increased risk of mortality for persons 60 years of age and older. Figure 4 shows an early example of the basis for these warnings in the data provided by the UK Office of National Statistics in September 2020 [12] and also in reference [13]. The UK government website notes several caveats: (1) the figures include deaths of nonresidents of the United Kingdom; (2) they are based on the date that a death was registered rather than when it occurred; (3) they are provisional and use the tenth edition of the International Classification of Diseases for definitions of the coronavirus (COVID-19). Again, the question arises of why the severity of COVID-19 infection should be a function of age.

From these data, one might expect a strong correlation between the apparent national case fatality rate and the median age of a country’s population. Even accepting the hypothesis of universality for the data in Figure 4, one should first multiply these rates by the demographics of a nation’s population normalized to the UK population grouped into the same age bins. Such a plot (Figure 5) shows a surprising result. The overall linear correlation is negative, –0.181, partially due to the disparity among the regions: –0.258 for the Americas, 0.052 for Asia, 0.141 for Europe, 0.02 for Africa, and –0.608 for the Middle East and Central Asia.

Rather than plotting the COVID-19 case fatality rate versus the national median age, one might examine the dependence on the percentage of the population of people aged 65 years or greater. In that case, the overall correlation (–0.081) is negative, consistent with reference [14]; however, this is mostly the result of regional variations, with a larger but still relatively low correlation (approximately 0.19) in Europe and Africa.

As a measure of the influence of the age of a population on SARS-CoV-2 contagion, the national rate of confirmed cases of COVID-19 per 1 million persons with respect to the percentage of the population aged older than 65 years (Figure 6) displays a moderate correlation of 0.447.

One may hypothesize that the “care home effect,” as in, the large numbers of deaths seen in nursing homes in Italy, the United Kingdom, and the US state of New York, was more the result of overcrowding and poor hygienic practices compounded with the general infirmity and the reduced immune function of nursing home residents than of any extreme dependence of the lethality of COVID-19 on individual, specific, underlying medical disorders. The linear correlations of age with potential causal factors, shown in Figure 7, suggest the strength of candidate cofactors to explain the care home effect. In addition to specific cofactors, the care home effect also reflects a generally very weakened physical condition of many occupants of care homes, which could render any pneumonia-inducing disease potentially lethal. The data in Figure 7 show no evidence that age alone influences the probability of a person becoming infected with SARS-CoV-2.

Figure 7 may explain what appears to be a startling result, namely, the globally negative correlation of the COVID-19 case fatality rate with the age of the national population. The negative value is due to the strong correlations between the national median age and the combination of adjusted GDP (0.64), national health care expenditures (0.48), and the WHO Universal Health Care Index. Nations with the oldest populations are generally those that are the wealthiest and in which health care services are the most robust, thus reducing the level of mortality.

In contrast with infections due to SARS-CoV-2, fatalities from influenza-induced pneumonia (Figure 8) are highly correlated (–0.652) with the median age of the population. The correlation also displays a strong regional dependence. The correlation is negative for the same reasons previously explained for COVID-19.

This result for influenza suggests the hypothesis that because COVID-19 typically presents as a severe respiratory disease, the severity of COVID-19 infection may correlate with the incidence of asthma. The global value (Figure 9) is small but not negligible (0.165), largely driven by the strong correlation (0.68) in the Middle East.

For asthma as a cofactor, the contrast with influenza-related pneumonia is striking. A relatively high overall correlation of 0.594 for influenza was observed in all regions. Hence, any reference to COVID-19 as a “flu-like” infection or as a “superflu” is grossly misleading.

The CDC issued a warning early in 2020 that obesity represented a comorbidity that could potentially lead to severe consequences of a COVID-19 infection. However, once again, the actual national data (Figure 10) essentially display no correlation (–0.017) of a country’s COVID-19 case fatality rate with the percentage of its population that is considered obese. A better metric of national obesity may be the average BMI (in kg/m²) of the population. With BMI as the metric of the national prevalence of obesity, the correlation increases to 0.052, which is still very small. Moreover, that figure may itself be misleading when comparing regions, as the correspondence between BMI and body fat percentage varies considerably from country to country (10% to 20%).

The contribution of obesity to the outcome of other pulmonary disorders is significantly different from that of COVID-19, as is displayed in Figure 11. Obesity does have a significant correlation (0.516) with the risk of contracting SARS-CoV-2 infection, although not with the apparent outcome of the infection. Observation of increased risk of infection (although not its outcome) was previously reported in [13]. Reference [14] reports an increased risk of infection (0.329) for people with chronic kidney disease. That correlation of risk of infection is not seen in the statistics of this study, which consistently found a temporally increasing negative correlation (–0.046). However, as shown in Figure 10, this study does confirm an increased risk of mortality (0.269) for persons with chronic kidney disease who do develop COVID-19.

One might speculate that as a chronic respiratory disorder involving pulmonary airways, asthma would increase the seriousness of the consequences of COVID-19 and its induced pneumonias; this analysis shows no such significant correlation (0.053). Examining the correlation of COVID-19 mortality with other lung diseases also showed a very small correlation (0.013). In contrast, the relationship of influenza-induced pneumonias with asthma and other lung diseases presents correlations that are quite high, 0.594 and 0.348, respectively. With respect to their effects on patients with underlying conditions, influenza and COVID-19 are very different diseases.

Another early warning to persons with underlying conditions concerned DM. That suspicion is echoed by the strong increase of incidence of diabetic conditions with age. Whether one measures the incidence of diabetes by deaths due to diabetes or to the reported national rates of diabetes in adults (20-79 years of age), the correlation with COVID-19 mortality is similarly low (0.109). In otherwise healthy persons, diabetes does not appear to be a significant risk factor with respect to the serious complications of infection with SARS-CoV-2.

Figure 10 and Table 2 summarize the linear correlations and their time variations, respectively, of the COVID-19 case fatality rate with underlying medical and economic conditions (green bars in the figure) considered herein. As the percentage of the population over 65 years of age correlates at best weakly with the apparent COVID-19 case fatality rate, one may surmise that poor health care management played a very large role in the care-home effect.

Figure 10 shows a strong negative correlation of the case fatality rate with both COVID-19 tests per million persons and with the number of cases per million persons. More tests mean earlier detection, more detection of mild and weakly symptomatic cases, and better triage followed by earlier and more effective clinical treatments.

Before investigating cross-correlations for root causes, one should perform a multivariate analysis of the COVID-19 case fatality rate against a common trio of risk factors commonly found in patients in nursing and convalescent homes–namely DM, hypertension, and coronary disease. For that trio, the coefficient of multiple correlation is 0.171, which is not negligible but is unlikely to be the root cause of the care home effect. Computing the correlation of DM, hypertension, and coronary disease with deaths due to influenza and its associated pneumonia yields a stronger correlation of 0.359. Replacing hypertension with asthma in the DM, hypertension, and coronary disease trio reduces the coefficient of multivariate correlation for COVID-19 mortality to 0.121. In contrast, analogous analysis for influenza increases the multiple correlation coefficient to 0.627, demonstrating once again (see Table 3) that influenza and COVID-19 are very different diseases.

Other calculations of multivariate correlations with the apparent national mortality rates of COVID-19 are presented in Table 4.

The previous section argues and Figure 12 illustrates that there is a striking contrast between the correlations of COVID-19 with those of influenza/pneumonia with respect to other potential underlying conditions.

Although obesity appears to be correlated with SARS-CoV-2 contagion, it appears uncorrelated with the outcome of COVID-19 infections, contrary to the findings of reference [15]. Understanding the correlations of obesity calls for a deeper look at the relationship of obesity with the conditions that show the most influence. Already, in the case of contagion, regional differences represent a substantial fraction of the apparent effect. The regional differences could be due to factors such as national median age, or they may be influenced by national wealth reckoned in terms of per capita GDP-PPP, as shown in Figure 13.

As is the case with asthma, DM (Figure 14) shows significant correlations with several medical and economic conditions, such as age, household size, and mortality due to influenza/pneumonia. Once again, no correlation with COVID-19 mortality (red bar) is evident.

A key assumption of this study is the high degree of country dependence of the COVID-19 case fatality rate. Even though the case fatality rate has fallen dramatically in many countries with rates originally greater than 10%, after nearly one year of the pandemic, the disparity by country and by region remains large, ranging over one order of magnitude, as illustrated in Figure 15.

The size of the regional data sets is obviously much smaller and the uncertainties in computed correlations are much higher than those of the aggregated world data. However, examining the regional dependence of the COVID-19 case fatality rates on the most commonly cited comorbidities is instructive. (See Table S1 of Multimedia Appendix 1.)

The differences in the magnitudes, outcomes, and characteristics of the waves of infections among national subregions with roughly equivalent medical factors indicates that economics and public health policies makes a significant difference in the severity of SARS-CoV-2 infection. This section examines dependencies on GDP-PPP, average household size, percentage of the population living in slums, percentage of the population living in cities, health expenditures per capita, and the WHO UHC index.

Figure 3 has already shown an example of economic impact on medical outcomes; the per capita GDP-PPP has a strong influence (–0.446) on the rate of deaths due to malnutrition. That observation is hardly surprising. One might also ask whether per capita GDP-PPP would have a similar impact on mortality due to COVID-19 infections. The distribution of COVID-19 mortality with national wealth shows essentially no correlation (–0.059). The politics of poverty does not, of itself, explain the observed national rates of COVID-19 mortality.

The distribution of contagion of SARS-CoV-2 over the global data set is noticeable and positive (0.299). However, as shown in Figure 16, that value is entirely driven by the strong dependence of the rising contagion on rising income in African countries. If one removes the African countries from the sample, the correlation disappears (0.028).

The differing behavior in Africa, which can be observed in Figure 16 (a), may be due to the increase in urbanization with increasing national wealth. One might further suspect that the increase in urbanization is also likely to increase the fraction of the national population living in slums. In fact, the percentage of people living in slums actually decreases with the urbanized fraction of the population.

The correlation of economic and policy factors with contagion (measured in confirmed COVID-19 cases per 1 million persons and apparent COVID-19 mortality) is presented in Table 5. As the mortality rate varies in time and seems to decline as the pandemic progresses (at least in the northern hemisphere), the mortality rate was benchmarked on December 30, 2020. The surprising negative correlation in contagion with the percentage of the urban population living in slums is likely due to the trend in Africa that the smaller the fraction of the population living in cities, the more likely it is that they live in slums (World Bank data).

An examination by region of the impact of economic cofactors in the COVID-19 case fatality rate is shown in Figure 17. The negative correlations with national wealth and with national health care expenditures are to be expected. Nonetheless, these effects are weaker in Africa than in other regions. More detailed investigation of these effects would require examination of underlying conditions on a country-by-country basis.

The correlation with respect to GDP is explained by the correlation of the GDP with the percentage of the population aged over 65 years. The substantial correlation of contagion with testing results from the obvious fact that the more one looks, the more one sees. The correlation of contagion with the percentage of urban population is due to the cross-correlation of GDP with percentage of urban population (0.648) and the high cross-correlation of urban population with testing for COVID-19 (0.497). The values for average health care expenditures and the UHC index of the WHO are similarly explained. The data that underlie the value of case fatality rate versus the percentage of the urban population that live in slums appears in Figure 18.