key: cord-1026110-tpxqj2do
authors: Esteve, A.; Permanyer, I.; Boertien, D.; Vaupel, J. W.
title: National age and co-residence patterns shape covid-19 vulnerability
date: 2020-05-16
journal: nan
DOI: 10.1101/2020.05.13.20100289
sha: 2f4d28ae0c0170b4ce8b6719e0e4fedef5ec6124
doc_id: 1026110
cord_uid: tpxqj2do

Based on harmonized census data from 81 countries, we estimate how age and co-residence patterns shape the vulnerability of countries' populations to outbreaks of covid-19. We estimate variation in deaths arising due to a simulated random infection of 10% of the population living in private households and subsequent within-household transmission of the virus. The age-structures of European and North American countries increase their vulnerability to covid-related deaths in general. The co-residence patterns of elderly persons in Africa and parts of Asia increase these countries' vulnerability to deaths induced by within-household transmission of covid-19. Southern European countries, which have aged populations and relatively high levels of intergenerational co-residence are, all else equal, the most vulnerable to outbreaks of covid-19. In a second step, we estimate to what extent avoiding primary infections for specific age-groups would prevent subsequent deaths due to within-household transmission of the virus. Preventing primary infections among the elderly is the most effective in countries with small households and little intergenerational co-residence such as France, whereas confining younger age groups can have a greater impact in countries with large and inter-generational households such as Bangladesh.

The covid-19 pandemic currently confronts nearly all of the world's countries. A growing number of governments are enforcing or recommending home quarantines to contain the spread of the virus. As the virus can be transmitted outside and within households, the effects of such measures will depend on the number of transmissions that take place outside and within the household. Evidence shows that the risk of severe disease and mortality increases sharply with age (1, 2) . Therefore, the age structure of the population-what proportion are young or old-and the structure of co-residency-how big are households and how old are their members-are two key factors that determine the vulnerability of countries to outbreaks of covid-19, and how effective general and age-specific household confinement policies can be in reducing mortality after an outbreak (3) . Figure 1 provides estimates of the number of deaths from covid-19 per 100,000 individuals if countries were to experience an outbreak of covid-19 of equal size, more specifically, a random infection of 10% of the population living in private households. Results are shown for 81 countries, covering all regions of the world and are solely based on census-based micro-data on age and co-residence patterns combined with age-specific infect fatality ratios (2) . The left-hand segment of each bar provides an estimate of direct mortality of individuals who catch the disease in a 10% random infection of the population (primary infections). The right-hand segment of the bars shows the additional deaths that would occur if all other members of the household become infected too (secondary infections). Lower rates of household transmission would reduce this number of indirect deaths proportionally. The direct effect depends on the age structure of the population; the indirect effect hinges on the size and age structure of households. Combined, they show how, all else equal, national age and co-residence patterns alter the vulnerability of a country to covid-19 outbreaks. The expected direct death rates per 100,000 individuals range from 20 in South Sudan to 136 in Italy. Together with Italy, three Southern European countries-Greece, Portugal and Spain-rank among the top five, followed by the rest of Europe and North America. Latin American countries form a homogenous cluster lower than the European and North American cluster. Asian countries spread all over the range with estimates as high as 96 in South Korea and as low as 25 in Jordan. African countries tend to experience the lowest direct death rates. Where the elderly comprise a large portion of the population, the direct effect is high, whereas direct deaths are much lower where the elderly are vastly outnumbered by younger people.

Mortality due to intra-household contagion (right-hand segment of a bar) does not follow the same order because co-residence patterns differ widely across countries, even among those countries with similar age structures (4, 5, 6, 7) . The ratio between indirect and direct effects is a simple indicator of the importance of co-residence patterns, in particular of the elderly, the most vulnerable group. For European 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 direct deaths in Asia ranges from 1.3 (South Korea) to 3.2 (Bangladesh). In Africa, indirect deaths would be three to four times the number of direct deaths. Such variation is closely associated with cross-national variation in co-residence patterns and, more specifically, with the number and age of the persons with whom elderly people reside.

Despite differences in ratios, the combined death rate (direct plus indirect) reproduces a broad regional pattern similar to the one observed in direct mortality but with variation in the specific position of countries. For instance, countries with similar direct death rates, such as France and Spain, show remarkably different indirect rates due to higher levels of intergenerational coresidence in Spain. Countries with similar indirect death rates, such as Italy and China, have quite distinct direct death rates, due to differences in their age structure.

Debates exist about the role that specific age-groups, and particularly children, play in the transmission of the virus (8) . In addition, countries have adopted age-specific policies such as school closures and extended confinement of the elderly in their homes. This provokes the question to what extent preventing primary infection of certain age groups would reduce the number of deaths that can arise due to within-household transmission of the virus. The bars in Figure 2 provide information about the expected mortality from direct and indirect deaths if primary infections for specific age groups could be averted. Results are shown for 10 countries chosen to illustrate diverse interactions between age structures and co-residence patterns (see Figure S3 for all countries). The first column shows the total number of deaths per 100,000 population if no age group is excluded from primary infection. The other columns indicate the total number of direct and indirect deaths per 100,000 population if primary infections could be averted for a particular age group. Deaths due to primary infection are reduced most in all countries if infections of individuals aged 65 or more are prevented. There is much more variation in how much avoiding primary infections of specific age-groups affects the number of indirect deaths that arise due to secondary infection (within household transmission). In France, indirect deaths are simulated to go down considerably if no person aged 65 or older were to be infected directly. This indicates that elderly persons primarily live with other persons aged 65 or older in France ( Figure S4 , Figure S5 ). These coresidence patterns also imply that avoiding primary infections for other age groups have relatively little effect on deaths that emerge due to within household transmission of the virus in France. At the other extreme, there are countries such as Bangladesh where preventing direct infections of the elderly would barely reduce indirect deaths and where avoiding the primary infection of children or adults aged 19-49 has a larger impact on indirect deaths. This result is explained by the high levels of intergenerational co-residence of the elderly together with the fact that children comprise a large share of the total population in these countries ( Figure S4 , Figure S5 ). Other countries fall between these two extremes with the United States being similar to France, and Ghana and South Africa resembling Bangladesh. Some cases combine elements from both extremes, such as Italy where both confining the elderly and individuals aged 19-49 reduce indirect deaths. However, in none of the scenarios do indirect deaths go down considerably. This illustrates the double challenge that countries such as Greece, Italy, Portugal and Spain face: the combination of an aged population with inter-generational residence leads to high estimated death rates due to covid-19 but also makes preventing deaths due to within-household transmission of the virus particularly challenging.

In confronting covid-19, epidemiologists should analyze and policymakers should consider how age structure and co-residence patterns in their countries can shape the number of infections and deaths. Differences in age structures put countries at different risk; a less considered factor, coresidence patterns, modulates this risk. In our simulations, which can be considered baseline scenarios before accounting for specific national policies, the proportion dying per 100,000 population is 275 in Italy and 80 in Mozambique ( Figure 1 ). In contrast, if 10% of the population is infected at random and their household members become infected too, the proportion of the population that becomes infected is 28% in Italy and 44% in Mozambique ( Figure S2 ). Because of different age and co-residence patterns, Italy is confronted with more deaths per capita than Mozambique but fewer infections. The effectiveness of policies in one country compared with another country should be evaluated in light of different baseline vulnerabilities.

In countries where the elderly form a large part of the population and primarily live with their generational peers, avoiding the primary infection of elderly people will considerably reduce direct deaths and will also prevent indirect deaths due to within household transmission of the virus. In countries where the elderly form a small part of the population but live together with young people, indirect deaths through infection within households can outnumber direct deaths. Therefore, avoiding primary infections of the elderly will be less effective in reducing deaths because the elderly might still get infected by younger household members. In such cases, measures that reduce or avoid within household transmission of the virus to the elderly become relatively more important to reduce mortality due to covid-19.

Data on age structure and co-residence come from high-quality harmonized census data from IPUMS (9) on individuals living in private households (i.e. individuals living in collective dwellings such as old-age homes are excluded). Mortality is determined by age-specific covid-19 death rates (2) . [See supplementary materials (SM) and figs. S7-11 for more details on methods, data, and other robustness checks].

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 16, 2020. . https://doi.org/10.1101/2020.05.13.20100289 doi: medRxiv preprint Figure 1 . Covid-19 vulnerability to national age and co-residence patterns.

Estimated number of direct (dark) and indirect (light) deaths per 100,000 individuals if 10% of the population living in private households were to be infected by covid-19 at random.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 16, 2020. . https://doi.org/10.1101/2020.05.13.20100289 doi: medRxiv preprint 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.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 16, 2020. . https://doi.org/10.1101/2020.05.13.20100289 doi: medRxiv preprint

We provide the following supplementary information l that the editors might consider useful to provide to reviewers. If published, this material (which includes replication code) will be made available on the authors' own website. Figure S1 . Analysis with 20% primary infections In our main analysis, we randomly simulated a primary infection rate of 10%. In robustness checks, we simulated a 20% infection rate. Figure S1 displays results from this robustness check.

It can be observed that countries with large households move down the distribution of death rates. This is because in several African countries infecting 10% of the population at random already leads to the majority of the population being infected if all household members get infected too. Simulating additional primary infections therefore leads to relatively fewer additional secondary infections (and deaths) as compared to countries with smaller households.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 16, 2020. . https://doi.org/10.1101/2020.05.13.20100289 doi: medRxiv preprint Figure S2 . Estimates of primary and secondary infections instead of deaths Figure S2 displays the simulated infection rates per 100,000 population in our main analysis. As 10,000 individuals are infected in all countries, primary rates do not vary. The secondary rates vary and the sum of direct and indirect rates divided by 10,000 reflects the average household size in each country. Note that a random primary infection of 10% of the population leads in several countries to the majority of the population being infected after secondary infection.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 16, 2020. . https://doi.org/10.1101/2020.05.13.20100289 doi: medRxiv preprint Figure S3 . Estimated number of direct (dark) and indirect (light) deaths per 100,000 individuals if primary infections of specific age groups are avoided.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 16, 2020. . https://doi.org/10.1101/2020.05.13.20100289 doi: medRxiv preprint Figure S3 . Estimated number of direct (dark) and indirect (light) deaths per 100,000 individuals if primary infections of specific age groups are avoided. (continuation)

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 16, 2020. . https://doi.org/10.1101/2020.05.13.20100289 doi: medRxiv preprint Figure S3 . Estimated number of direct (dark) and indirect (light) deaths per 100,000 individuals if primary infections of specific age groups are avoided. (continuation)

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 16, 2020. . https://doi.org/10.1101/2020.05.13.20100289 doi: medRxiv preprint Figure S3 . Estimated number of direct (dark) and indirect (light) deaths per 100,000 individuals if primary infections of specific age groups are avoided. (continuation)

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 16, 2020. . https://doi.org/10.1101/2020.05.13.20100289 doi: medRxiv preprint Analysis of deaths by household type Figure S4 . Direct deaths by 100, 000; by living arrangements.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 16, 2020. . https://doi.org/10.1101/2020.05.13.20100289 doi: medRxiv preprint Figure S5 . Indirect deaths by 100, 000; by living arrangements.

As the number of indirect deaths depends on whom the elderly live with, we provide more insight into the household structures of persons who are simulated to die due to infection with the virus. Figure S4 shows the household structure of the persons who were simulated to die after primary infection (direct deaths). It becomes clear that many of the persons who die due to primary infection are persons aged 65+ who live alone or with another person (probably their partner) in European and North American countries. In Africa and Asia, the majority of direct deaths occur within households where persons above 65+ live with three other persons or more. Figure S5 further completes this picture by showing the household structure of persons who are estimated to die due to secondary infection. These numbers underline that in Europe and North America many indirect deaths occur in households where persons aged 65+ live with one other person (most likely their partner). In France and Switzerland this is the most common household type among those simulated to die due to secondary infection. Not only does this appear to reduce the number of estimated indirect deaths, it might also make confining the elderly in their households a more effective strategy to further reduce indirect deaths. In contrast, in Asia and Africa the majority of indirect deaths occur in households where a person over 65 lives in a household with four or more members. As indirect deaths comprise the lion's share of estimated deaths in these countries, reducing mortality in Africa and Asia will depend on how effectively secondary infections within large households can be prevented.

Analysis adjusting for sex Initial reports indicated that men die more often after being infected with Covid-19 than women. Fatality/infection ratios were not available to us by age for males and females . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 16, 2020. . https://doi.org/10.1101/2020.05.13.20100289 doi: medRxiv preprint separately. Because men are underrepresented in older age groups, it is not straightforward to recalculate age and sex specific fatality ratios from the separate pieces of information. We therefore decided to only adjust for age in the main analysis and to check how robust these results were as follows. The Chinese Center for Disease Control and Prevention reported that fatalities for men were 2.8% of infections and for women 1.7%

(http://weekly.chinacdc.cn/en/article/id/e53946e2-c6c4-41e9-9a9b-fea8db1a8f51). In robustness checks, we therefore adjusted age-specific fatality rates by sex such that the male death rate was 1.24 times the rate for the population as a whole and the female death rates was 0.76 times the population rate. Figure S6 shows little change in terms of ranking of countries. There is a slight reduction in deaths because of the under-representation of men in older age groups, which makes the above calculation adjust overall fatality rates downward. Figure S6 . Comparison of direct and indirect death rates per 100,000 using age-adjusted and age-sex-adjusted fatality rates . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 16, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 16, 2020. . https://doi.org/10.1101/2020.05.13.20100289 doi: medRxiv preprint Figure S7 . Comparison of direct death estimates using original and updated fatality rates . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 16, 2020. . https://doi.org/10.1101/2020.05.13.20100289 doi: medRxiv preprint Figure S8 . Comparison of indirect death estimates using original and updated fatality rates

Only a subset of datasets included individuals living in group quarters. For our purposes, this is especially relevant in the case of residences for the elderly. Therefore, we compared changes in estimates when including them in the subset of countries for which data was available. In some datasets large households are split into various 1-person households. These cases are excluded from both sets of estimates. Figure S9 shows how including group quarters leads to small upward adjustments in countries like Italy and Switzerland where many old people live in group quarters. In other counties small . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 16, 2020. . https://doi.org/10.1101/2020.05.13.20100289 doi: medRxiv preprint downward adjustments are observed as many young people live in group quarters (e.g. Jamaica). Figure S9 . Comparison of total death estimates including and excluding group quarters

The analysis relies on simulating the random infection of 10,000 individuals out of 100,000 in each sample. This implies that running the analysis several times can lead to slightly different estimates. To investigate the sensitivity of estimates to this issue, we re-ran the analysis 1000 times for Spain. Figure S10 and S11 display the distribution of estimated direct and indirect . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 16, 2020. . https://doi.org/10.1101/2020.05.13.20100289 doi: medRxiv preprint deaths per 100.000 persons resulting from these 1000 repetitions. It can be observed that the estimated number of deaths varies across random draws, but these changes are relatively small.

Even though this might affect the specific ranking of countries that are very similar in terms of the total simulated deaths, no major changes of countries in the death distribution is likely to arise from this issue. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 16, 2020. To adjust for changes in the age structure we used data on age by year from the 2010 census as available at (accessed 31/03/2020): https://www.census.gov/datatools/demo/idb/region.php?T=10&RT=0&A=both&Y=2010&C=CH&R=

We first ran the analysis using the 2000 census and calculated individuals' risk of direct and indirect death by age. This risk was calculated for each age in years, but grouped for those aged 80-84 and 85 or older to gain more reliable estimates. We subsequently took the age . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 16, 2020. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 16, 2020. drop gq ************************************ ***** The following code serves to select households until 100,000 individuals are reached ************************************ **** Most datasets are self-weighting according to IPUMS (Can be found in "Microdata sample characteristics" once clicking on a specific sample **** For several samples a stratified selection procedure has been used for data collection, or other methods that require the inclusion of household weights. **** For these sample, we created x duplicates of each individual case, x being the household weight. To avoid an enormous dataset, we divided weights by 100 or 10 depending on the minimum weight (if minimum weight<100 we used 10)

********************** after expanding the data we created unique household numbers for each duplicate. This all to create a pool of households from which the sample is drawn to reach 100.000 households . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 16, 2020. . https://doi.org/10.1101/2020.05.13.20100289 doi: medRxiv preprint sort sample serial pernum by sample serial pernum: gen dupno=_n tostring serial, g(hhnr)

tostring dupno, g(d)

gen hhid=hhnr+"d"+d drop serial hhnr d dupno dupl rename hhid serial ********************************************************** *** To randomize the draw we generated a random number between 0 and 1 for each first person in the household. is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 16, 2020. . https://doi.org/10.1101/2020.05.13.20100289 doi: medRxiv preprint by sample serial: egen cut=min(ccc) *** drop all households with a minimum person rank above 100.000. This will generate samples of 100.000 persons are a few persons more drop if cut>100000 *** drop unnecessary variables drop order randomize rand2 rand3 ccc cut ************************************************************* ***** Save the selected sample append using "SouthKorea.dta" label define sample_lbl 201007 "United States 2010", add label define sample_lbl 1111 "South Korea", add **************************** **** For Italy, ages above 75 are grouped together. Our analysis depends death rates that vary across 10-year groups but **** above 80 there is no further distinction in death rates. We therefore randomly divided individuals for Italy across the 70s and 80s groups **** based on the distribution observed in ISTAT's Multiscopo 2009: Among individuals aged 75+ 55.9% is aged 80+ . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 16, 2020. *** In robustness checks we adjusted for sex using a uniform formula: *** replace deathrate=deathrate*1.24 if sex==1 *** replace deathrate=deathrate*0.76 if sex==2 ********************** *** The following code randomly selects individuals who are infected . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 16, 2020. . https://doi.org/10.1101/2020.05.13.20100289 doi: medRxiv preprint ******* insert amount of absolute cases infected after local cases. 10000 represents 10%; In robustness checks we infected 20000. replace finalinfect`y'=1 if infected==1 *** generate the total number of primary plus secondary infections per 100.000 in each country by sample: egen totinf`y'=total(finalinfect`y') ********* calculate the amount of deaths per 100.000 in each country *** individual chance of dying by multiplying with 1 the death rates if a person is infected gen deathchance`y'=deathrate*finalinfect`y' **** sum all individual death chances within each sample sort sample by sample: egen amtdeath`y'=total(deathchance`y') **** generate integers for presentation purposes . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 16, 2020. . https://doi.org/10.1101/2020.05.13.20100289 doi: medRxiv preprint gen amtdeathint`y'=round(amtdeath`y') } **** generate estimate of direct and indirect deaths per 100.000 gen direct=amtdeathint0 gen indirect=amtdeathint100-amtdeathint0 *** indicator for total deaths per 100.000 gen totdeath=amtdeathint100 *** drop unnecessary variables drop random order infhh amtinfhh infran infchance0 totinf0 amtdeath0 amtdeathint0 infchance100 totinf100 amtdeath100 amtdeathint100 **** numbers for Figure 1 in Table. Graph bar to see all is ok. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 16, 2020. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 16, 2020. . https://doi.org/10.1101/2020.05.13.20100289 doi: medRxiv preprint gen i1=ih1-dh1 gen i2=ih2-dh2 gen i3=ih3-dh3 gen i4=ih4-dh4 ***** crappy graphs to check things worked, sorted by direct deaths.

graph bar dh0 dh1 dh2 dh3 dh4 if sample!=51201101, stack over(sample, sort(direct)) graph bar i0 i1 i2 i3 i4 if sample!=51201101, stack over(sample, sort(direct)) ****************************************************************************** **** ************* ANALYSIS OF AVOIDING PRIMARY INFECTIONS OF AGE GROUPS ************** ****************************************************************************** **** label define ag 0 "0-18 years" 1 "19-49 years" 2 "50-64 years" 3 "6+ years" , replace label values agegroups ag . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 16, 2020. replace finalinfect`y'=1 if infected==1 **** total infections in sample sort sample by sample: egen totinf`y'=total(finalinfect`y') ********* amount of deaths in sample gen deathchance`y'=deathrate*finalinfect`y' sort sample by sample: egen amtdeath`y'=total(deathchance`y') gen amtdeathint`y'=round(amtdeath`y') } *** generate death rates for each age counterfactual gen direct`m'=amtdeathint0 gen totdeath`m'=amtdeathint100 gen indirect`m'=amtdeathint100-amtdeathint0 drop infected infhh amtinfhh infran infchance0 finalinfect0 totinf0 deathchance0 amtdeath0 amtdeathint0 infchance100 finalinfect100 totinf100 deathchance100 amtdeath100 amtdeathint100 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 16, 2020. . https://doi.org/10.1101/2020.05.13.20100289 doi: medRxiv preprint } ****************************************************************************** ************************************************ ***** The following code prepares the dataset to create a long-format containing all counterfactual estimates for each country ***** this code should be adjusted if more age categories are used replace indirect=indirect3 if order==5 label define ord 1 "All" 2 "Aged 0-18" 3 "Aged 19-49" 4 "Aged 50-64" 5 "Aged 65+" 6 "Aged 65+", replace label values order ord . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 16, 2020. . https://doi.org/10.1101/2020.05.13.20100289 doi: medRxiv preprint ***** keep only those rows with information keep if order<6 ****************************************************************************** *************************************************** *** some examples of results by age graph bar direct indirect if sample==380201101 | sample==201007, stack over(order) over(sample, sort(totdeath)) save "200401 age.dta", replace ***** create a dataset with results by age for all countries is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 16, 2020. . https://doi.org/10.1101/2020.05.13.20100289 doi: medRxiv preprint

COVID-19): situation report, 51

Esimates of the severity of COVID-19 disease

Demographic science aids in understanding the spread and fatality rates of COVID-19

United Nations, Living Arrangements of Older Persons around the World. (United Nations

Intergenerational coresidence in developing countries

Family ties in Western Europe: persistent contrasts

Living Alone over the Life Course: Cross-National Variations on an Emerging Issue Popul

Coronavirus infections in children including COVID-19: An overview of the epidemiology, clinical features, diagnosis, treatment and prevention options in children

We thank A. Turu for help with the harmonization and graphic representation of the data. This analysis was supported by the following grants: ERC-2014-StG-637768, RTI2018-096730-B-I00.