National age and coresidence patterns shape COVID-19 vulnerability Albert Estevea,b,1, Iñaki Permanyera, Diederik Boertiena, and James W. Vaupelc a Center for Demographic Studies, Centres de Recerca de Catalunya, 08193 Bellaterra, Spain; bGeography Department, Universitat Autònoma de Barcelona, 08193 Bellaterra, Spain; and cInterdisciplinary Centre on Population Dynamics, University of Southern Denmark, 5000 Odense, Denmark Edited by Douglas S. Massey, Princeton University, Princeton, NJ, and approved June 8, 2020 (received for review May 8, 2020) Based on harmonized census data from 81 countries, we estimate infections). Lower rates of household transmission would reduce how age and coresidence patterns shape the vulnerability of this number of indirect deaths proportionally. The direct effect countries’ populations to outbreaks of coronavirus disease 2019 depends on the age structure of the population; the indirect effect (COVID-19). We estimate variation in deaths arising due to a sim- hinges on the size and age structure of households. Combined, they ulated random infection of 10% of the population living in private show how, all else equal, national age and coresidence patterns alter households and subsequent within-household transmission of the the vulnerability of a country to COVID-19 outbreaks. virus. The age structures of European and North American coun- The expected direct death rates per 100,000 individuals range tries increase their vulnerability to COVID-related deaths in gen- from 19 in South Sudan to 120 in Italy. Together with Italy, three eral. The coresidence patterns of elderly persons in Africa and southern European countries—Greece, Portugal, and Spain— parts of Asia increase these countries’ vulnerability to deaths in- rank among the top six, followed by the rest of Europe and North duced by within-household transmission of COVID-19. Southern America. Latin American countries form a homogenous cluster European countries, which have aged populations and relatively lower than the European and North American cluster. Asian high levels of intergenerational coresidence, are, all else equal, the countries spread all over the range, with estimates as high as 81 most vulnerable to outbreaks of COVID-19. In a second step, we in South Korea and as low as 23 in Jordan. African countries estimate to what extent avoiding primary infections for specific age tend to experience the lowest direct death rates. Where the el- groups would prevent subsequent deaths due to within-household derly comprise a large portion of the population, the direct effect transmission of the virus. Preventing primary infections among the is high, whereas direct deaths are much lower where the elderly elderly is the most effective in countries with small households and are vastly outnumbered by younger people. little intergenerational coresidence, such as France, whereas confining Mortality due to intrahousehold contagion (right-hand segment of a younger age groups can have a greater impact in countries with large bar in Fig. 1) does not follow the same order, because coresidence and intergenerational households, such as Bangladesh. patterns differ widely across countries, even among those countries with similar age structures (4–7). The ratio between indirect and direct demography | households | COVID-19 | aging | global effects is a simple indicator of the importance of coresidence patterns, in particular, of the elderly, the most vulnerable group. For European T he coronavirus disease 2019 (COVID-19) pandemic currently confronts nearly all of the world’s countries. A growing number of governments are enforcing or recommending home quarantines and North American countries, direct and indirect deaths are roughly equal. In Latin America, indirect deaths could approximately double the number of direct deaths. The ratio between potential indirect and to contain the spread of the virus. As the virus can be transmitted direct deaths in Asia ranges from 1.3 (South Korea) to 3.7 (Laos). In outside and within households, the effects of such measures will Africa, indirect deaths would be 3 to 4 times the number of direct depend on the number of transmissions that take place outside and deaths. Such variation is closely associated with cross-national varia- within the household. Evidence shows that the risk of severe disease tion in coresidence patterns and, more specifically, with the number and mortality increases sharply with age (1, 2). Therefore, the age and age of the persons with whom elderly people reside. structure of the population—what proportion are young or old— Despite differences in ratios, the combined death rate (direct and the structure of coresidency—how big are households and how plus indirect) reproduces a broad regional pattern similar to the old are their members—are two key factors that determine the one observed in direct mortality, but with variation in the specific vulnerability of countries to outbreaks of COVID-19, and how ef- position of countries. For instance, countries with similar direct fective general and age-specific household confinement policies can death rates, such as France and Spain, show remarkably different be in reducing mortality after an outbreak (3). indirect rates, due to higher levels of intergenerational cor- esidence in Spain. Countries with similar indirect death rates, Results such as Italy and China, have quite distinct direct death rates, Fig. 1 provides estimates of the number of deaths from COVID- due to differences in their age structure. 19 per 100,000 individuals if countries were to experience an outbreak of COVID-19 of equal size, more specifically, a ran- Preventing Primary Infection of Specific Age Groups. Debates exist dom infection of 10% of the population living in private about the role that specific age groups, and particularly children, households (i.e., excluding people living in old-age homes). play in the transmission of the virus (8). In addition, countries Results are shown for 81 countries, covering all regions of the world, and are solely based on census-based microdata on age and coresidence patterns combined with age-specific infection Author contributions: A.E. and J.W.V. designed research; A.E., I.P., and D.B. performed fatality ratios (2). These ratios were not available to us for males research; D.B. analyzed data; and A.E., I.P., D.B., and J.W.V. wrote the paper. and females separately, but uniform adjustments of age-specific The authors declare no competing interest. death rates by sex lead to very similar results. The left-hand This open access article is distributed under Creative Commons Attribution-NonCommercial- segment of each bar provides an estimate of direct mortality of NoDerivatives License 4.0 (CC BY-NC-ND). individuals who catch the disease in a 10% random infection of Data deposition: Information about data, methods, and code are available at https:// the population (primary infections). The right-hand segment www.medrxiv.org/content/10.1101/2020.05.13.20100289v1. of the bars shows the additional deaths that would occur if all 1 To whom correspondence may be addressed. Email: aesteve@ced.uab.es. other members of the household become infected too (secondary First published June 23, 2020. 16118–16120 | PNAS | July 14, 2020 | vol. 117 | no. 28 www.pnas.org/cgi/doi/10.1073/pnas.2008764117 BRIEF REPORT Italy countries chosen to illustrate diverse interactions between age Greece structures and coresidence patterns. The first column shows the Portugal France total number of deaths per 100,000 population if no age group is Austria Spain excluded from primary infection. The other columns indicate the Hungary total number of direct and indirect deaths per 100,000 population if Romania Belarus primary infections could be averted for a particular age group. Poland Ukraine Deaths due to primary infection are reduced most in all Switzerland countries if infections of individuals aged 65 y or more are pre- Canada Russia vented. There is much more variation in how much avoiding pri- Uruguay South Korea mary infections of specific age-groups affects the number of indirect United States deaths that arise due to secondary infection (within-household Slovenia Ireland transmission). In France, indirect deaths are simulated to go Cuba Argentina down considerably if no person aged 65 y or older were to be Trinidad and Tobago infected directly. This indicates that elderly persons primarily China Chile live with other persons aged 65 y or older in France. These Panama coresidence patterns also imply that avoiding primary infections Jamaica Brazil for other age groups has relatively little effect on deaths that Costa Rica Mexico emerge due to within-household transmission of the virus in Vietnam France. At the other extreme, there are countries, such as Dominican Republic El Salvador Bangladesh, where preventing direct infections of the elderly Peru Ecuador would barely reduce indirect deaths and where avoiding the Colombia primary infection of children or adults aged 19 y to 49 y has a Thailand Iran larger impact on indirect deaths. This result is explained by the Morocco South Africa high levels of intergenerational coresidence of the elderly, to- SOCIAL SCIENCES Lesotho gether with the fact that children comprise a large share of the Turkey Indonesia total population in these countries. Other countries fall between Haiti Nepal these two extremes, with the United States being similar to Bangladesh France, and Ghana and South Africa resembling Bangladesh. Botswana Kyrgyz Republic Some cases combine elements from both extremes, such as Italy, Paraguay where confining both the elderly and individuals aged 19 y to 49 y Venezuela Ghana reduces indirect deaths. However, in none of the scenarios do in- Bolivia Philippines direct deaths go down considerably. This illustrates the double Sierra Leone challenge that countries such as Greece, Italy, Portugal, and Spain Nicaragua Guatemala face: The combination of an aged population with intergenerational Cambodia Malaysia residence leads to high estimated death rates due to COVID-19 but Zimbabwe also makes preventing deaths due to within-household transmission Laos Togo of the virus particularly challenging. Egypt Tanzania Honduras Discussion Sudan Senegal In confronting COVID-19, epidemiologists should analyze and Kenya Malawi policy makers should consider how age structure and coresidence Rwanda patterns in their countries can shape the number of infections Liberia Mongolia and deaths. Differences in age structures put countries at dif- Burkina Faso ferent risk; a less considered factor, coresidence patterns, mod- Cameroon Benin ulates this risk. In our simulations, which can be considered Jordan Mali baseline scenarios before accounting for specific national poli- Ethiopia Europe & North America Uganda South & Central America cies, the proportion dying per 100,000 population is 240 in Italy Mozambique and 71 in Mozambique (Fig. 1). In contrast, under the same Zambia Asia South Sudan Africa assumptions, the proportion of the population that becomes in- Papua New Guinea fected is 28% in Italy and 44% in Mozambique. Because of 0 50 100 150 200 250 300 different age and coresidence patterns, Italy is confronted with Direct and indirect deaths per 100,000 more deaths per capita than Mozambique but fewer infections. Fig. 1. COVID-19 vulnerability to national age and coresidence patterns. The effectiveness of policies in one country compared with an- Estimated number of direct (dark) and indirect (light) deaths per 100,000 other country should be evaluated in light of different baseline individuals if 10% of the population living in private households were to be vulnerabilities. Our results are applicable to persons living in infected by COVID-19 at random. Analyses based on most recent available private households. Had we been able to systematically consider data from either 2000 or 2010 census rounds. persons living in old-age homes, country differences in vulnera- bility could change based on the share and health status of the population living in institutions. In countries where the elderly have adopted age-specific policies such as school closures and form a large part of the population and primarily live with their extended confinement of the elderly in their homes. This generational peers, avoiding the primary infection of elderly provokes the question to what extent preventing primary in- people will considerably reduce direct deaths and will also pre- fection of certain age groups would reduce the number of vent indirect deaths due to within-household transmission of the deaths that can arise due to within-household transmission of virus. In countries where the elderly form a small part of the the virus. population but live together with young people, indirect deaths The bars in Fig. 2 provide information about the expected through infection within households can outnumber direct mortality from direct and indirect deaths if primary infections for deaths. Therefore, avoiding primary infections of the elderly will specific age groups could be averted. Results are shown for 10 be less effective in reducing deaths, because the elderly might Esteve et al. PNAS | July 14, 2020 | vol. 117 | no. 28 | 16119 Italy France Russia South Korea United States 250 200 150 100 Direct and indirect deaths per 100,000 50 0 China Mexico South Africa Bangladesh Ghana 250 200 150 100 50 0 None 0-18 19-49 50-64 65+ None 0-18 19-49 50-64 65+ None 0-18 19-49 50-64 65+ None 0-18 19-49 50-64 65+ None 0-18 19-49 50-64 65+ Fig. 2. Estimated number of direct (dark) and indirect (light) deaths per 100,000 individuals if primary infections of specific age groups are avoided. Data are from 2010 census round. Individuals from each age group who were selected in the 10% random draw are recoded as not infected before calculating direct deaths and simulating within household transmission. still get infected by younger household members. In such cases, from 2000 were used but adjusted to the age structure of 2010 using data from measures that reduce or avoid within-household transmission of the 2010 census available at the National Bureau of Statistics of China. For South the virus to the elderly become relatively more important to Korea, data come from Statistics Korea. Mortality is determined by age-specific reduce mortality due to COVID-19. COVID-19 infection fatality ratios (2). For further information about data, methods, and code, see https://www.medrxiv.org/content/10.1101/2020.05.13.20100289v1. Materials and Methods Data on age structure and coresidence come from the most recent high-quality ACKNOWLEDGMENTS. We thank A. Turu for help with the harmonization harmonized population microdata samples from Integrated Public Use Microdata and graphic representation of the data. This analysis was supported Series (IPUMS) (9) on individuals living in private households (i.e., individuals living by Grants ERC-2014-StG-637768, ERC-2019-COG-864616, and RTI2018- in collective dwellings such as old-age homes are excluded). For China, IPUMS data 096730-B-I00. 1. World Health Organization, Coronavirus disease 2019 (COVID-19): Situation report, 51. https:// 6. D. S. Reher, Family ties in Western Europe: Persistent contrasts. Popul. Dev. Rev. 24, www.who.int/emergencies/diseases/novel-coronavirus-2019/situation-reports. Accessed 2 April 203–234 (1998). 2020. 7. A. Esteve, D. S. Reher, R. Treviño, P. Zueras, A. Turu, Living alone over the life 2. R. Verity et al., Estimates of the severity of coronavirus disease 2019: A model-based course: Cross-national variations on an emerging issue. Popul. Dev. Rev. 46, 169–189 analysis. Lancet Infect. Dis. 20, 669–677 (2020). (2020). 3. J. B. Dowd et al., Demographic science aids in understanding the spread and fatality 8. P. Zimmermann, N. Curtis, Coronavirus infections in children including COVID-19: An rates of COVID-19. Proc. Natl. Acad. Sci. U.S.A. 117, 9696–9698 (2020). overview of the epidemiology, clinical features, diagnosis, treatment and prevention 4. United Nations, Living Arrangements of Older Persons around the World, (United options in children. Pediatr. Infect. Dis. J. 39, 355–368 (2020). Nations, 2005). 9. Minnesota Population Center, Integrated Public Use Microdata Series, In- 5. S. Ruggles, M. Heggeness, Intergenerational coresidence in developing countries. ternational: Version 7.2. https://doi.org/10.18128/D020.V7.2. Accessed 20–31 Popul. Dev. Rev. 34, 253–281 (2008). March 2020. 16120 | www.pnas.org/cgi/doi/10.1073/pnas.2008764117 Esteve et al.