Impact of Pandemic on European Well-Being: Visualizing Scenarios from the SHARE Database

International Journal of Environmental Research and Public Health

Article Impact of Pandemic on European Well-Being: Visualizing Scenarios from the SHARE Database

Aurea Grané * , Irene Albarrán and David E. Merchán

Statistics Department, Universidad Carlos III de Madrid, 28903 Getafe, Spain; [email protected] (I.A.); [email protected] (D.E.M.) * Correspondence: [email protected]

Abstract: The objective of this study is to analyse the effect of a pandemic shock on the well-being of the European population aged 50 or over. Data comes from wave 7 of the Survey of Health, Ageing and Retirement in Europe (SHARE), carried out in 28 countries and representing over 170 million aged individuals in Europe. We start by designing two indicators in order to capture the risk of being unhealthy and economically vulnerable; next, we combine them with socio-demographic information and obtain the vulnerability profiles by means of the k-prototypes clustering algorithm. Subsequently, we design a shock similar to the COVID-19 pandemic and measure its effects on the vulnerability profiles. The results suggest that the average level of economic and health vulnerability is relatively low, although levels differ across European regions, with the most vulnerable being Eastern European countries. It was observed that the shock most affected countries with a greater proportion of individuals initially deemed vulnerable in terms of mental and physical health, as well as countries where tourism and retail sectors are the most vital for their economies. These findings led us to conclude that public policies should be differentiated by European regions, and Governments must establish action plans in order to better meet the physical and mental health Citation: Grané, A.; Albarrán, I.; needs of their citizens, as well as addressing monetary poverty and financial difficulties. Merchán, D.E. Impact of Pandemic on European Well-Being: Visualizing Keywords: clustering analysis; k-prototypes; mixed data; pandemic; SHARE survey; vulnerability Scenarios from the SHARE Database. profiles; well-being Int. J. Environ. Res. Public Health 2021, 18, 4620. https://doi.org/10.3390/ ijerph18094620 1. Introduction Academic Editor: Paul B. Tchounwou According to Eurostat [1] the percentage of population in the European Union older Received: 26 March 2021 than 50 years of age is 40.9%. The World Health Organization (WHO) estimates that Europe Accepted: 24 April 2021 will have one of the oldest populations in the world by 2050, with 30% of the population Published: 27 April 2021 over 60 years old; additionally, life expectancy will be close to 80 years [2]. This means that older people will constitute a proportionally large population and will demand inclusion, Publisher’s Note: MDPI stays neutral as well as socioeconomic and health care. For these reasons, European institutions are with regard to jurisdictional claims in designing policies that include the older population. For example, in the framework of the published maps and institutional affil- UN’s 2030 Agenda for Sustainable Development, Sustainable Development Goals (SDGs) iations. were laid out, some of which were related to the older population. For instance, Goal 1 “End poverty in all its forms everywhere” and Goal 8 “Decent work and economic growth” support the older population with policies related to retirement, tax-funded minimum pension, education for older workers, social assistance; Goal 2 “End hunger achieve food Copyright: © 2021 by the authors. security and improved nutrition, and promote sustainable agriculture” aims to improve Licensee MDPI, Basel, Switzerland. nutrition, i.e., through a diet with vitamins, minerals and proteins; Goal 3 “Ensure healthy This article is an open access article lives and promote well-being for all at all age” aims to guarantee a health system that distributed under the terms and cares about complex and chronic disease using a prevention, promotion, treatment and conditions of the Creative Commons rehabilitation approach. Other goals like quality education (Goal 4), gender equality Attribution (CC BY) license (https:// (Goal 5), and sustainable cities and communities (Goal 11) aim to reduce the vulnerability creativecommons.org/licenses/by/ of this group. 4.0/).

Int. J. Environ. Res. Public Health 2021, 18, 4620. https://doi.org/10.3390/ijerph18094620 https://www.mdpi.com/journal/ijerph Int. J. Environ. Res. Public Health 2021, 18, 4620 2 of 26

At present, we are suffering a pandemic due to SARS-CoV-2 (COVID-19). Although this virus can affect any age group, it affects older people with greater intensity, especially those who have pre-existing medical diseases such as lung or heart disease, diabetes or cancer [3]. Therefore, the risk for severe illness with COVID-19 increases with age, with older adults at highest risk (The Center for Disease Control and Prevention (CDC), 2021. https: //www.cdc.gov/coronavirus/2019-ncov/need-extra-precautions/older-adults.html (accessed on 25 March 2021)). The Center for Disease Control and Prevention (CDC) [4] mentioned that in USA “people in their 50s are at higher risk for severe illness than people in their 40s. Similarly, people in their 60s and 70s are, in general, at higher risk for severe illness than people in their 50s. The greatest risk for severe illness from COVID-19 is among those aged 85 or older.” In March 2021, the CDC posted a graph where it was observed that in USA, there were a total of 1117.3 hospitalisations per 100,000 individuals over 50 years of age. Health authorities all over the world are warning older people that they are at a higher risk of more serious and possibly fatal symptoms associated with COVID-19. Mortality data from the Oxford COVID-19 Evidence Service indicated a risk of mortality of 3.6% for people in their 60s, which increases to 8.0% and 14.8% for people in their 70s and 80s (Oxford COVID-19 Evidence Service, University of Oxford, https://www.cebm.net/ oxford-covid-19-evidence-service/ (accessed on 25 March 2021)). The severity and outcome of COVID-19 largely depends on a patient’s age. Adults over 65 years of age represent 80% of hospitalisations and have a 23-fold greater risk of death than those under 65 [5]. COVID-19 has so far killed more than 2,750,000 people (2,758,733 by the time this paper was written), with the majority of deaths (73%) occurring in people over the age of 65 (Worldometer, 2020. https://www.worldometers.info/coronavirus/ coronavirus-age-sex-demographics/ (accessed on 25 March 2021)) (See Ref. [6]) COVID 19 has caused a marked increase in deaths worldwide, but with signiﬁcant variation between countries and causes of death. The total amount of excess mortality will also depend on the age structure of a population. Countries with age structures weighted towards an older population will experience higher mortality those with an age structure weighted towards a younger population. This pandemic has challenged healthcare systems around the world in a way that has not been seen in modern times. Older people are the most affected by the pandemic as a group with the highest risk of hospitalisation and death (Williamson E, Walker AJ, Bhaskaran KJ et al. (2020) OpenSAFELY: factors associated with COVID-19-related hospital death in the linked electronic health records of 17 million adult NHS patients. https://www. medrxiv.org/content/10.1101/2020.05.06.20092999v1 (accessed on 25 March 2021)) [7,8]. Research is a central component of the global response to COVID-19 [9], as the pandemic has had a huge impact on our ability to design and conduct research relevant to older people, not just COVID-19. So, one of the biggest challenges for senior research is currently COVID-19. This pandemic is disrupting lives and communities in the United States and around the world [10,11]. As the impacts of the disease continue, the economic consequences of the pandemic and the responses to it begin to manifest themselves, as jobs are lost, businesses close and ﬁnancial markets falter. While the implications of the current economic downturn for older adults are only just beginning to emerge, past experience suggests that the consequences could be devastating for many [12]. Elderly people and those with pre-existing chronic health conditions may be at higher risk of developing severe health consequences from COVID-19. In Europe, this is of particular relevance to aging populations living with noncommunicable diseases, multimorbidity and frailty [13]. This pandemic constitutes an extraordinary global health, social and economic chal- lenge. The impact on people’s mental health and the psychological well-being of vulnerable groups (health workers, outpatients or the elderly among others) is expected to be high [14]. People who have or had symptoms related to COVID-19 or various disadvantaged socio- Int. J. Environ. Res. Public Health 2021, 18, 4620 3 of 26

economic backgrounds are at significantly higher risk of general psychiatric disorders and loneliness. It is important to consider the broader psychological impact of the pandemic on a wider population. During the first wave, almost a third of the population was affected by COVID-19 in various ways and to different degrees [15]. The COVID-19 outbreak has evolved into a global public health crisis with an increasing numbers of people affected psychologically in various ways and to varying degrees (see [15–17] among others). This pandemic, as well as the public health and emergency measures implemented to reduce contagion, have profoundly changed people’s lifestyles and are believed to be a threat to physical and mental wellbeing. Factors such as the long duration of quarantine, fears of infection, inadequate information, financial, and job loss have been associated with a greater negative psychological impact [14,18,19]. According to [18], depression is a normal reaction to a sudden worsening in living conditions, involving isolation and uncertainty about economic expectations and life expectations in general. With these facts in mind, we can infer that older people are vulnerable in economic, social, educational, gender, and health terms, but they may be more affected by health issues like pandemics. Therefore, it is important to generate knowledge about the condition of vulnerability of the elderly. For that reason, the main objective of our research is to analyse the impact of a pandemic on the well-being of European people older than 50 years of age. For this purpose, we designed two indicators that can capture the risk of being unhealthy and economically vulnerable and combine them with socio-demographic information in order to obtain vulnerability profiles. The paper proceeds as follows. In Section2 we describe the data used for this study, the methodology to construct the indicators and the process to obtain the vulnerability profiles. Section2 also contains the design of the pandemic shock. In Section3, we analyse the results by visualising the indicators, the vulnerability profiles and the pandemic shock. Finally, we conclude in Section4.

2. Data and Methods In this section, we describe the construction of two indicators which are useful to measure the economical and health vulnerability of the elderly. They are based on information available in the SHARE dataset. Next, we present the process to obtain the vulnerability profiles, and finally, we explain the design of the pandemic shock to be applied to the SHARE dataset, with the aim of studying its effect on the vulnerability profiles and detect possible deficiencies on national social protection systems.

2.1. Descriptive Analysis and Design of Indicators 2.1.1. Data In this work, we use data from the Survey of Health, Ageing and Retirement in Europe (SHARE), which is a cross-national and multidisciplinary panel database of microdata related to topics like health, socio-economic status and social and family networks of about 170 million individuals aged 50 or older. See also http://www.share-project.org/home0 .html for more details (accessed on 1 April 2020) on this project. In particular, we use SHARE survey—Wave 7. This survey was implemented using the web in 2017 in Austria, Germany, Sweden, Netherlands, Spain, Italy, France, Denmark, Greece, Switzerland, Belgium, Israel, Czech Republic, Poland, Luxembourg, Hungary, Portugal, Slovenia, Estonia, Croatia, Lithuania, Bulgaria, Cyprus, Finland, Latvia, Malta, Romania, and Slovakia. This last wave contains questions about activity, physical health, retrospective health care, retrospective accommodation, behavioural risks, retrospective children, retrospective employment, cognitive functions, children, employment, and pensions, ﬁnancial transfer, health care, household income, housing, mental health, and social support. In this work, we use data from wave 7 to achieve the proposed objective. The target of this survey are all persons aged 50 or more, with regular domicile in one of the aforementioned countries. People who are incarcerated, hospitalised or out of the country, unable to speak or have moved to an unknown address were excluded from the Int. J. Environ. Res. Public Health 2021, 18, 4620 4 of 26

sample [20]. Since it is a longitudinal survey, all the respondents interviewed in previous waves were part of the sample of wave 7. The survey has calibrated cross-sectional weights, which were computed by country to match the size of national target populations. For each country, a logit specification of the calibration function and calibration margins that consider gender and age groups were used. This information for the calibration was taken from the EUROSTAT regional database. For our research we only considered European countries, so we removed Israel from the database and kept only those individuals that had a calibration value. After that, the dataset of study consists of 73,274 complete cases that represent 171,585,757 European citizens. Regarding survey information, we use 30 variables recoded in the modules of De- mographics, Physical Health, Mental Health, Health Care, Employment and Pensions, Housing, Household Income, Consumption, Activities and Expectations. We split the variables into two groups, that we call descriptive variables and thematic blocks. The first group consists of sociodemographic information and contains variables such as country, gender, age group, marital status, and education. These variables do not contribute to the indicators, but will be used to create the vulnerability profiles. In this paper, we consider two thematic blocks: one related to economics and the other to health. The variables included in these blocks contribute to the indicators and are described below. Due to the sample design of this survey and the weight of each individual, we used the R-package survey by Lumley [21] to perform the statistical analysis.

2.1.2. Indicators Our first aim is to measure the condition of vulnerability of the elderly by means of two indicators: the first is related to economic situation and the second to health condition. To build them, we use the information available on SHARE dataset and follow the protocol described in Grané et al. [22], designed to create indicators from mixed data. In particular, we first identify those variables in the dataset that measure relevant aspects to be studied and group them in thematic blocks. Next, in each thematic block, variables are grouped in subthemes, binarised following the recommendations of related literature, always respecting the same polarisation, and each particular subindicator is obtained by aggregation. Finally, indicators are created by adding the corresponding subindicators and rescaled to 0–10; value 0 means no vulnerability at all and value 10 indicates maximum level of vulnerability. Note that all variables have the same contribution to each particular indicator. Thus, grouping variables in subthemes is optional, although it may help to better interpretation. This design has several advantages: First, we can use both numerical and categorical variables, which can be measured on different scales; second, the binarisaton of variables is useful to unify scales and to identify different patterns; finally, indicators are additively constructed, so that relevant new variables can be added if necessary. In what follows, we give a detailed description of the construction of the indicators.

Economic Indicator This indicator is compounded by four subindicators: Monetary poverty, Unemploy- ment, Economical Distress, Weekly consumption. Monetary poverty subindicator measures whether the individual household is in a situation of monetary poverty. It takes value 1 if the individual income is less than the 60% of the national median income after social transfers, otherwise takes the value 0. The threshold is computed by Eurostat for each European country and we took this value for the year 2018. The individual income is computed with the variable thinc2, which calculates the total household net income, and we normalise it by considering the household composition according to OECD modiﬁed scale [23] (This assigns a value of 1 to the household head, of 0.5 to each additional adult member and of 0.3 to each child under 14 years old). The Unemployment subindicator measures whether the individual is unemployed. It takes value 1 when the individual is unemployed and 0 otherwise. We use variable cjs that Int. J. Environ. Res. Public Health 2021, 18, x 5 of 26

60% of the national median income after social transfers, otherwise takes the value 0. The threshold is computed by Eurostat for each European country and we took this value for the year 2018. The individual income is computed with the variable thinc2, which calculates the total household net income, and we normalise it by considering the household composition according to OECD modified scale [23] (This assigns a value of 1 to the household head, of 0.5 to each additional adult member and of 0.3 to each child under 14 years old). The Unemployment subindicator measures whether the individual is unemployed. It takes value 1 when the individual is unemployed and 0 otherwise. We use variable cjs that asks for the current job situation. We compute this subindicator because these people do not receive income, and this could cause them to be unable to buy food, save or pay Int. J. Environ. Res. Public Health 2021,for18, 4620their health or retirement. 5 of 26 The Financial Distress subindicator lets us know whether the individual has enough income to be able to make ends meet. To create this subindicator, we used the variable fdistressasks for the that current is computed job situation. from We compute the question this subindicator “thinking because of your these household’s people do total monthly income,not receive would income, you say and that this your could household cause them is toable be unableto make to ends buy food,meet”, save the or possible pay for outcomes are their health or retirement. 1-WithThe great Financial difficulty, Distress subindicator 2-With some lets usdifficulty, know whether 3-Fairly the individual easily and has enough4-Easily. In this case, theincome sub toindicator be able to takes make value ends meet. 1 when To create the answer this subindicator, is 1-With we gr usedeat difficulty the variable or 2-With some difficulty,fdistress that and is computed it takes value from the0 otherwise. question “thinking of your household’s total monthly income,Finally, would you the say Weekly that your consumption household is able tosub makeindicator ends meet measures”, the possible vulnerability outcomes in terms of are 1-With great difficulty, 2-With some difficulty, 3-Fairly easily and 4-Easily. In this case, howthe subindicator often the takesindividual value 1 consume when the answers legumes, is 1-With beans, great eggs, difficulty meat, or 2-With fish or some poultry. To create thisdifficulty, subindi andcator, it takes we value use 0 otherwise. two variables: the first one, legeggs, measures how often the in- dividualFinally, consume the Weeklys consumptionlegumes, beans subindicator or eggs, measures while vulnerability the second in termsone, ofmeat how, measures how often the the individual individual consumes consume legumes,s meat, beans, fish eggs, or poultry. meat, fish For or poultry. both variables, To create this the possible out- comessubindicator, are Not we use applicable two variables:, Less the than first one,oncelegeggs, a weekmeasures, Once how a week often, the Twice individual a week, 3–6 times a consumes legumes, beans or eggs, while the second one, meat, measures how often the week,individual and consumes Every day meat,. We fish assign or poultry. a numerical For both variables, value for the possibleeach possible outcomes outcome are (from 0 to 4)Not and applicable, add both Less variables than once obtaining a week, Once a new a week, one Twice that atakes week, values 3–6 times between a week, 0 and 8. Using thisand Everyinformation day. We we assign define a numerical a threshold value in for which each possible if the value outcome is less (from than 0 to or 4) equal to 2, we considerand add both the variablesindividual obtaining as vulnerable, a new one which that takes means values that between in the 0 andweekly 8. Using consumption sub- this information we define a threshold in which if the value is less than or equal to 2, weindicator consider the the outcome individual of as this vulnerable, individual which is means 1 or less. that in the weekly consumption subindicatorIn Figure the outcome1, we can of thissee individual a bar plot is 1for or less.each subindicator with their distributions. The onlyIn sub Figureindicator1, we can where see a barthe plot group for eachof vuln subindicatorerable people with their is greater distributions. than Thethe nonvulnerable isonly that subindicator of financial where distress. the group This of vulnerable may happen people because is greater thanit is thea self nonvulnerable-reported question, also Pettinicchiis that of financial & Börsch distress.-Supan This may[24] happenmention because using it SHARE is a self-reported data that question, formerly also self-employed Pettinicchi & Börsch-Supan [24] mention using SHARE data that formerly self-employed retirees have have lower lower incomes incomes and they and report they high report levels high of financial levels of distress. financial distress.

Figure 1. Economic subindicators distributions. Figure 1. Economic subindicators distributions. The economic indicator is obtained by adding the previous subindicators and rescaling to 0–10. The indicator mean value is 2.33, the standard deviation is 0.01 and the median is equal to 2.5. This means that, in general, Europe has a low level of economic vulnerability, since there are low observations with the maximum levels of vulnerability. This makes sense because Europe is recognised as a continent with a strong economy with a market Int. J. Environ. Res. Public Health 2021, 18, x 6 of 26

The economic indicator is obtained by adding the previous subindicators and rescaling to 0–10. The indicator mean value is 2.33, the standard deviation is 0.01 and the median is equal to 2.5. This means that, in general, Europe has a low level of economic vulnerability, since there are low observations with the maximum levels of vulnerability. This Int. J. Environ. Res. Public Health 2021, 18, 4620 6 of 26 makes sense because Europe is recognised as a continent with a strong economy with a market of 27 countries that make up the European Union. Figure 2 shows the distribution ofof 27the countries economic that make indicator. up the European Union. Figure2 shows the distribution of the economic indicator.

Figure 2. Economic indicator distribution. Figure 2. Economic indicator distribution. Health Indicator This indicator evaluates the vulnerability in terms of four subindicators: General Healthhealth, Mental Indicator health, Physical health, and Health care. The General health subindicator is compounded by two variables. The first one is sphus whichThis measures indicator Self-perceived evaluates health using the de US vulnerability scale; this scale includes in Excellent, terms of four subindicators: General Good, Fair, Poor, Very Poor. In order to design the subindicator, we consider as vulnerable health,those individuals Mental that answerhealth, Fair, Physical Poor or Very Poor, health, which meansand theHealth subindicator care. takes value 1 whenever those responses take place. The second variable used is chronic, which is a numericalThe variableGeneral that counts health the number subindicator of chronic diseases; is compounded in this case we used as by a two variables. The first one is sphusthreshold which 2 chronic measures diseases following Self the-perceived research of Costa-Font health et al. using [25]. Thus, de if US the scale; this scale includes Excel- interviewee has 2 or more chronic diseases, they are considered as a vulnerable (it takes lentthe value, Good 1 in the, Fair binary, transformation). Poor, Very We Poor. added these In twoorder binary to variables design to compute the subindicator, we consider as vul- nerablethe General those health subindicator. individuals that answer Fair, Poor or Very Poor, which means the subin- The Mental health subindicator measures the vulnerability in terms of three vari- dicatorables. First, takes eurod usesvalue the EURO 1 whenever depression scale those 0–12 to responses measure whether take the indi-place. The second variable used is vidual feels depressed or not. We binarise this numerical variable following the work of chronicCosta-Font, etwhich al. [25], inis which a numerical they classified avariable respondent withthat a 4counts or less on thethe EURO number of chronic diseases; in this casedepression we scaleused as aas 0 (not a threshold depressed/not vulnerable),2 chronic else diseases a 1 (depressed/vulnerable). following the research of Costa-Font et al. The second variable used is lifesat, which is a numerical variable that uses a scale between [25]0 and. 10,Thus, where if 0 is the completely interviewee dissatisfied andhas 10 2 completely or more satisfied; chronic in this case,diseases, we they are considered as a vul- classify as vulnerable those people that answer in the scale a value less than 7. The last nerablevariable used (it was takes lifehap, the a categorical value variable 1 in that the measures binary the life transformation). happiness through We added these two binary variablesthe question: “toHow compute often, on balance, the do General you look back onhealth your life sub with aindicator sense of happiness?. ” in which the possible outcomes are Often, Sometimes, Rarely, Never. In this case, we classifiedThe a person Mental as vulnerable health if his/her sub responsesindicator were Rarelymeasures or Never, the which vulnerability means in terms of three varia- that the binarised variable takes value 1 when these responses occur. For all the above, the bles.subindicator First, is theeurod sum of uses the three the binarised EURO variables. depression scale 0–12 to measure whether the individual feelsThe depressed Physical health or subindicator not. We considers binar fiveis variablese this relatednumerical to mobility variable limita- following the work of Costa- tions, limitations with activities of daily living and the body mass index (BMI). Mobility Fontlimitations et al. issues [25] are measured, in which using twothey variables classified Gali and Mobility, a respondent the former classifies with a 4 or less on the EURO depres- speopleion scale as being as limited a 0 or (not not limited depressed/not in mobility, while the vulnerable), latter measures how else many a diffi- 1 (depressed/vulnerable). The second variable used is lifesat, which is a numerical variable that uses a scale between 0 and 10, where 0 is completely dissatisfied and 10 completely satisfied; in this case, we classify as vulnerable those people that answer in the scale a value less than 7. The last variable used was lifehap, a categorical variable that measures the life happiness through the question: “How often, on balance, do you look back on your life with a sense of happiness?” in which the possible outcomes are Often, Sometimes, Rarely, Never. In this case, we classified a person as vulnerable if his/her responses were Rarely or Never, which means that the binarised variable takes value 1 when these responses occur. For all the above, the subindicator is the sum of the three binarised variables. The Physical health subindicator considers five variables related to mobility limitations, limitations with activities of daily living and the body mass index (BMI). Mobility limitations issues are measured using two variables Gali and Mobility, the former classifies people as being limited or not limited in mobility, while the latter measures how many

Int. J. Environ. Res. Public Health 2021, 18, x 7 of 26

Int. J. Environ. Res. Public Health 2021, 18, 4620 7 of 26 difficulties the person has within a range between 0 and 10. We consider a person as vulnerable if he or she has at least one mobility limitation. Likewise, we used adl and iadl variables,culties the personwhich has are within used ato range measure between the 0 andfunctionality 10. We consider in terms a person of personal as vulnerable and instru- mentalif he or sheactivities has at leastof daily one mobilitylife. Both limitation. variables Likewise, are numeric, we used the adl first and one iadl with variables, a range be- tweenwhich are0 and used 6, to while measure thethe second functionality one takes in terms values of personalbetween and 0 and instrumental 9. Like the activities previous variables,of daily if life. the Both person variables has at are least numeric, one limitation the first one is classified with a range as betweenvulnerable. 0 and The 6, while last variable usedthe second for this one particular takes values sub betweenindicator 0 and is 9. the Like BMI, the previousa very popular variables, index if the personwhose has categories areat least established one limitation by the is classifiedfollowing as ranges: vulnerable. less The than last 18.5 variable (underweight), used for this from particular 18.5 to 24.9 (normalsubindicator weight), is the from BMI, 25 a veryto 29.9 popular (overweight) index whose and greater categories than are 30 established (obesity). byWe the binarised following ranges: less than 18.5 (underweight), from 18.5 to 24.9 (normal weight), from the BMI defining as vulnerable those individuals who are underweight, overweight or 25 to 29.9 (overweight) and greater than 30 (obesity). We binarised the BMI defining as obese.vulnerable For thoseall the individuals above, the who sub areindicator underweight, is theoverweight sum of the or five obese. binar Foris alled thevariables. above, the subindicatorFinally, the is Health the sum care of the sub fiveindicator binarised measures variables. how the elderly population has been takenFinally, care of the and Health has used care subindicator the health system. measures To how create the elderlythis sub populationindicator, haswe beenuse the var- iabletaken doctor care of that and measures has used thehow health many system. times a Toperson create has this seen subindicator, or talked to we a medical use the doctor duringvariable the doctor last thatyear. measures For this howvariable, many we times classify a person as vulnerable has seen or people talked towho a medical have not seen ordoctor spoken during to thea doctor last year. within For this the variable, last year we or classify who have as vulnerable seen or peoplespoken who to a have doctor not at least 52seen times or spoken a year, to athat doctor is, on within average the last, once year a or week. who have Therefore, seen or spokenwe consider to a doctor as vulnerable at peopleleast 52 timesthose awho year, do that not is, oncare average, about once their a week.health Therefore,, as well weas those consider who as vulnerableappear to have a people those who do not care about their health, as well as those who appear to have a medical problem. The second variable considered for this subindicator is the number of medical problem. The second variable considered for this subindicator is the number of nightnight staysstays inin hospital hospital during during the the last last year, year, for whichfor which the selected the selected threshold threshold was one. was We one. We addedadded thesethese two two binary binary variables variables to computeto compute the Healththe Health care subindicator. care subindicator. The distributiondistribution of of each each subindicator subindicator is shown is shown in Figure in Figure3. 3.

Figure 3. Health subindicators distribution. Figure 3. Health subindicators distribution. The health indicator is created by adding the previous health subindicators and rescalingThe to health 0–10. In indicator Figure4, weis created depict the by health adding indicator the previous distribution. health The meansubindicators value and rescalingof this indicator to 0–10. is 3.28,In Figure the standard 4, we depict deviation the ishealth 0.01 and indicator the median distribution. value is equal The mean to 3. value ofConsidering this indicator the scale is 3.28, of this the indicator, standard the deviation previous measuresis 0.01 and allow the usmedian to understand value is that, equal to 3. in general, Europe has low levels of health vulnerability. However, 25% of the population Considering the scale of this indicator, the previous measures allow us to understand that, is concentrated between levels 5 and 10 of vulnerability. A possible explanation is that inhealth genera carel, expenditureEurope has among low levels European of health countries vulnerability. is different. However, For instance, 25% according of the population to isEurostat concentrated [26], in 2017, between the health levels care 5 expenditureand 10 of vulnerability. per capita in Switzerland A possible was explanation 8474 euros, is that healthwhile in care Bulgaria, expenditure it was 550 among euros. European countries is different. For instance, according to Eurostat [26], in 2017, the health care expenditure per capita in Switzerland was 8474 euros, while in Bulgaria, it was 550 euros.

Int. J. Environ. Res. Public Health 2021, 18, x 8 of 26 Int. J. Environ. Res. Public Health 2021, 18, 4620 8 of 26

Figure 4. Health indicator distribution. Figure 4. Health indicator distribution. 2.2. Construction of the Profiles The SHARE survey contains sociodemographic questions related to gender, age, 2.2.marital Construction status, and education. of the Here, Profiles we combine the information of these variables with the economic and health indicators to get the vulnerability profiles. For this purpose, we use a clusteringThe SHARE algorithm which survey is able contains to cope with largesociodemographic datasets of weighted and questions mixed related to gender, age, maritaldata. Due tostatus, the sampling and design, education. the SHARE Here dataset, we includes combine a weighting the variable, information so of these variables with that each individual represents a group of different size of the target population. the Aeconomic very popular and clustering health algorithm indicators is k-Means, to which get is the fast andvulnerability easy to interpret, profiles. For this purpose, we usebut does a clustering not work with algorithm categorical data which [27]. One is possibilityable to couldcope be with to use large k-Medoids datasets of weighted and mixed or Partitioning Around Medoids (PAM) with Gower’s distance. However, this algorithm data.is not efficientDue to with the a large sampling datasets [28 design,]. To circumvent the SHARE the dimensionality dataset problem, includes a weighting variable, so thatClustering each Large individual Applications represents (CLARA) can be a used. group This methodof different uses a small size sample of ofthe target population. the dataset and applies PAM to generate the medoids and compute the final clusters. Since the firstA sample very may popular not be representative, clustering the algorithm method takes several is k- samples,Means, repeats which the is fast and easy to interpret, butclustering does process not severalwork times with and categorical the final result isdata selected [27] considering. One possibility the minimal could be to use k-Medoids average dissimilarity. or PartitioningIn this work, we preferAround to use Medoids the k-Prototypes (PAM) clustering with algorithm Gower’s by Huang distance. [29], However, this algorithm isbecause not efficient it presents severalwith advantagesa large datasets compared to[28] the. CLARA: To circumvent (1) It works with the the dimensionality problem, Clus- entire dataset instead of a sample, (2) It optimises the cost function using all the dataset, tering(3) CLARA Large takes aApplications sample of the whole (CLARA) dataset, but the can larger be the used. dataset, This the larger method the uses a small sample of the sample size. Hence, if the sample is large, thousands of objects (as it is our case), the datasetefficiency ofand CLARA applies will be affected PAM [29 to]. generate the medoids and compute the final clusters. Since the Asfirst with sam any partitioningple may method,not be the representative, number of clusters must the be method provided. Duetakes to several samples, repeats the the large size of the dataset, we decide to use the “elbow” method, instead of the average clusteringsilhouette width process or other time-consuming several times criteria. and This the method final consists result of running is selected the considering the minimal averagealgorithm with dissimilarity. a varying k and calculating the cost function for each run. Next, cost function values are plotted in a line-graph and at the point where there is a turning point, or “elbow”,In this the improvement work, we in theprefer cost function to use levels the off. k That-Prototypes is to say, the improvement clustering algorithm by Huang [29], becauseby adding anit additional presents grouping several in the advantages data is marginal, andcompared the added complexityto the CLARA: of (1) It works with the understanding the profiles outweighs any benefits given. entireOn thedataset k-prototypes instead algorithm. of a sample, (2) It optimises the cost function using all the dataset, (3) CLARALet X1, X2, ...takes, Xp be ap numericalsample variablesof the availablewhole in dataset, the dataset but and X thep+1, Xlargep+2, r the dataset, the larger the ... , X be m-p categorical variables. Then, the dissimilarity between two individ- samplem size. Hence, if the sample is large, thousands of objects (as it is our case), the effi- ciency of CLARA will be affected [29]. As with any partitioning method, the number of clusters must be provided. Due to the large size of the dataset, we decide to use the “elbow” method, instead of the average silhouette width or other time-consuming criteria. This method consists of running the algorithm with a varying k and calculating the cost function for each run. Next, cost function values are plotted in a line-graph and at the point where there is a turning point, or “elbow”, the improvement in the cost function levels off. That is to say, the improvement by adding an additional grouping in the data is marginal, and the added complexity of understanding the profiles outweighs any benefits given. On the k-prototypes algorithm. Let 푋1, 푋2, . . . , 푋푝 be p numerical variables available in the dataset and 푋푝+1, 푋푝+2, . . . , 푋푚 be m-p categorical variables. Then, the dissimilarity between two indi- 풑 viduals 퐱풊, 퐱풋 ∈ ℝ is measured by a combination between the Euclidean and the Ham- ming distance: 푝 푚 2 푑(퐱풊, 퐱풋) = ∑(푥푖ℎ − 푥푗ℎ) + 휆 ∑ 훿(푥푖ℎ, 푥푗ℎ) (1) ℎ=1 ℎ=푝+1

Int. J. Environ. Res. Public Health 2021, 18, 4620 9 of 26

p uals xi, xj ∈ R is measured by a combination between the Euclidean and the Hamming distance:

p m 2 d xi, xj = ∑ xih − xjh + λ ∑ δ xih, xjh (1) h=1 h=p+1

where δ(., .) is the Hamming distance and λ ≥ 0 is a parameter that measures the influence of numerical and categorical variables. When λ = 0 the algorithm only considers the numerical ones. This algorithm was implemented by Szepannek [30] in the clustMixType R-package. This package contains the function kproto that performs the following procedure: 1. Initialisation with random cluster prototypes, 2. For each observation do: (a) Assign observation to its closest prototype according to distance d (.,.) (b) Update cluster prototypes by cluster-specific means/modes for all variables 3. As long as any observations have swapped their cluster assignment in 2 or the maximum number of iterations has not been reached: repeat from 2 ([30], p. 201). In this work, we introduce a particularity to the k-prototypes algorithm so that it can cope with weighted datasets. Thus, we modified function kproto accordingly with the help of the R-package (R Foundation for Statistical Computing, Vienna, Austria) survey by Lumley [21], which contains standard statistical techniques for weighted data. More details are given in the AppendixA.

2.3. Simulation of a Pandemic Shock With the COVID-19 pandemic in 2020, we observed different shocks to health and economic systems. In developed countries, this crisis was different from previous crises, such as ﬁnancial crisis of 2008, because it began with a health issue, that is, the economic crisis did not go directly to market failures, but was a consequence of having a quarantine in which economic sectors such as manufacturing, tourism and retail were affected. Therefore, to simulate a situation similar to COVID-19 pandemic, we design a shock that starts affecting a health variable and deﬁne which individuals are affected using proxy variables. Note that the surveys does not contains the information about which individuals are infected with COVID-19 to shock the features of those individuals directly. National entities and organisations analysed what kind of population have more probability of being affected by this virus. For example, CDC [31] shows two groups of people at increased risk for severe illness: older adults and people with underlying medical conditions. According to WHO [32], that second group is formed by people affected by noncommunicable disease (NCDs) like cancer, diabetes, chronic respiratory disease, and cardiovascular disease. During the pandemic, one of the recommendations for vulnerable population was isolation; some governments recommended implementing the telework to protect health and employment, but there are some jobs that cannot implemented in these ways, leading to risk of unemployment. On the other hand, the lockdown policies affected the economy, and many small and medium-sized enterprises (SMEs) have gone bankrupt, leading to an increase in the unemployment rate in Europe. Jones et al. [33] shows that in the United Kingdom 19% of workers were furloughed, in Germany this proportion was 23%, 29% in Italy and 14% in France. Some publications mentioned the effects of this pandemic on mental health. For instance, Pieh et al. [34] studied the effects of Coronavirus on mental health in Austria. They found increases in depression, insomnia, anxiety and a decreases of quality life. Javed et al. ([35], p. 2) mentioned that “physical distancing due to the COVID-19 outbreak can have drastic negative effects on the mental health of the elderly and disables individuals”, these effects are stress, anxiety and depression. Schoeder [36] used the SHARELIFE survey Int. J. Environ. Res. Public Health 2021, 18, 4620 10 of 26

to estimate the effect of involuntary job loss on health. He classiﬁed the job loss in two Int. J. Environ. Res. Public Health 2021, 18, x types, the ﬁrst one as exogenous to the individual, like a plant closures, and the10 second of 26 one due to lay-offs. In his work, he found that the probability to be depressed increases when a man loses his job due to an exogenous cause. On the other hand, women had poorer general health; they reported more chronic conditions, and the probability to be obese or The previous overweightliterature increases.allows us to understand possible ways in which to design the The previous literature allows us to understand possible ways in which to design the shock. In this case, the shock is sequential, but all the changes of the variables occur at the shock. In this case, the shock is sequential, but all the changes of the variables occur at the same time. The shocksame design time. The is shockshown design in Figure is shown 5 and in Figure explain5 and it explain below. it below.

Figure 5. ShockFigure design. 5. Shock design.Variables Variables and sequence and sequence considered considered to simulatesimulate a pandemica pandemic shock. shock.

The first step is to consider people who have chronic diseases. This variable produces The first step is to consider people who have chronic diseases. This variable produces an effect on the unemployed and self-perceived health subindicators in the following way: an effect on the unemployedwe consider and that thoseself-perceived people who health had chronic subindicators diseases lost in their the job.following Also those way: people we consider that thosewith chronic people diseases who had now chronic they would diseases perceive lost that their healthjob. Also is worse. those people with chronic diseases nowThe coronavirus they would has perceive not only affectedthat their the jobshealth of peopleis worse. with chronic diseases, but The coronavirusalso has jobs innot certain only economicaffected sectorsthe job suchs of as people tourism with and retail.chronic Therefore, diseases, we supposebut that people who worked in these sectors lose their jobs, due to this shock. We select the also jobs in certainindividuals economic affected sectors by saidsuch shock, as tourism by means and of the retail. International Therefore, Standard we Classification suppose of that people who workedOccupations in these (ISCO sectors 88) code lose related their to tourism jobs, due and retailto this sector. shock. We select the individuals affected byThe said unemployed shock, by do means not have of the resourcesInternational to be ableStandard to make Classification ends meet and to of Occupations (ISCOconsume 88) thecode same related quantity to oftourism foods like and meat retail and legumes.sector. For this reason, after the shock, those people are more vulnerable. Thus, individuals who become unemployed due to the The unemployedcoronavirus do not andhave before the theresources shock were to be able able to make to make ends ends meet, meet but now and are to struggling. consume the same quantityFor this of reason, food theys like are meat also consideredand legumes vulnerable,. For this which reason means, after that thethe subindicator shock, those people are moreFinancial vulnerable. distress takes Thus, the individuals value 1. Likewise, who people become who unemployed become unemployed due to due the to the coronavirus and beforeCoronavirus the shock and before were the able shock to hadmake a value ends 0 inmeet, the weekly but now consumption are struggling. subindicator, For this reason, theynow are after also the shockconsidered this last vulnerable, value changed which to 1. means that the subindicator Finally, the pandemic and those previous shocks could affect happiness and life Financial distress takessatisfaction the value and increase 1. Likewise, depression people and obesitywho become or overweight, unemployed as we mentioned due to the before. Coronavirus and beforeFor this the case, shock we consider had a twovalue simultaneous 0 in the weekly shocks. Theconsumption first, regarding sub theindicator, unemployed, now after the shockwith this financial last value distress changed and vulnerable to 1. in terms of weekly consumption, will have a value Finally, the pandemic1 in the life and happiness, those previous depression, shocks and life could satisfaction affect variables.happiness On and the life other sat- hand, isfaction and increase depression and obesity or overweight, as we mentioned before. For this case, we consider two simultaneous shocks. The first, regarding the unemployed, with financial distress and vulnerable in terms of weekly consumption, will have a value 1 in the life happiness, depression, and life satisfaction variables. On the other hand, individuals with chronic diseases and with a worse self-perceived health, also will have a value 1 in those three variables. The approach that we use to define the shock includes the idea that a pandemic crisis is different from an economic crisis, because, in this case, the shock starts with a health variable instead of an economic variable.

Int. J. Environ. Res. Public Health 2021, 18, x 11 of 26

3. Results In this section, we present the results of applying the construction of the indicators, Int. J. Environ. Res. Public Health 2021,the18, 4620 vulnerability profiles and the pandemic shock to the SHARE11 of 26 dataset. We start by visualising the indicators according to different groups established by sociodemographic in- formationindividuals with. Next, chronic we diseases visual andis withe the a worse vulnerability self-perceived profiles health, also across will have Europe a and study the effect valueof the 1 in pandemic those three variables. shock on them. The approach that we use to define the shock includes the idea that a pandemic crisis is different from an economic crisis, because, in this case, the shock starts with a health variable3.1. Visual insteadising of an the economic Vulnerability variable. Indicators 3. ResultsIn Figure 6, we depict the boxplots of these indicators by gender. The distribution of the Inecono this section,mic indicator we present the between results of applyingmale and the construction female are of the different. indicators, For females, we can see the vulnerability profiles and the pandemic shock to the SHARE dataset. We start by visualisingthat 50% the of indicators the observations according to different are concentrated groups established around by sociodemographic 2.5, whereby the quartiles (Q1, Q2, Q3)information. are equal. Next, weWhile, visualise the the distribution vulnerability profiles of the across economic Europe and indicator study the is negatively skewed, Q1 takeseffect of value the pandemic 0 and shock Q3 on is them. equal to 2, then 50% of the male population is located between 0 and3.1. Visualising 2. We infer the Vulnerability that for Indicators both genders there are not many people with high economic vulnerability,In Figure6 nevertheless, we depict the boxplots females of these are indicators more economically by gender. The distribution vulnerable of than males. These gen- the economic indicator between male and female are different. For females, we can see derthat 50% differences of the observations occur are because concentrated of around the gender 2.5, whereby employment the quartiles (Q1, and Q2, pay gap. The European CommissionQ3) are equal. While, [37] the estimated distribution of for the economic2017 that indicator the gender is negatively employment skewed, Q1 gap was equal to 11%, takes value 0 and Q3 is equal to 2, then 50% of the male population is located between and0 and the 2. We gender infer that pay for both gap genders was equal there are to not 16%, many for people example, with high in economic Spain this gap is equal to 44.9% vulnerability,for females nevertheless over 64 femalesyears areold. more For economically the health vulnerable indicator, than males. we Thesedo not observe differences in thegender median differences between occur because males of the and gender females. employment Both and paymedian gap. The values European are around 3, with a posi- Commission [37] estimated for 2017 that the gender employment gap was equal to 11%, tivelyand the genderskewed pay gapdistribution was equal to 16%,for females for example, and in Spain a symmetric this gap is equal distribution to 44.9% for males. Also, there arefor females possible over 64outliers, years old. Forwhich the health correspond indicator, we to do the not observemost differencesvulnerable in the cases. Van Oyen et al. [38] median between males and females. Both median values are around 3, with a positively explainsskewed distribution that this for femalescan happen and a symmetric because distribution women for males.have Also,long there life are expectancy, which makes possiblethem reach outliers, ages which where correspond more to thediseases most vulnerable and mobility cases. Van limitations Oyen et al. [38 appear.] explains that this can happen because women have long life expectancy, which makes them reach ages where more diseases and mobility limitations appear.

Figure 6. Boxplot6. Boxplot economic economic and health and indicators health by gender. indicators by gender. Next, we compare the distribution of our indicators by marital status. In Figure7, we observeNext, in thewe economic compare indicator the thatdistribution single people of have our a greater indicators vulnerability by marital level status. In Figure 7, we with a positively skewed distribution; in this case the median is around 2.5. People with a partnerobserve have in a lower the leveleconomic of vulnerability indicator and the distributionthat single is negatively people skewed. have These a greater vulnerability level with a positively skewed distribution; in this case the median is around 2.5. People with a partner have a lower level of vulnerability and the distribution is negatively skewed. These differences occur because people with a partner may have more resources available for the household due to their jobs or pensions. Also, in the plot, we observe that people with a partner have better health than singles. The distribution for the people with a partner is positively skewed with a median of 2, while the distribution for single people is symmetric. People who live alone have more health risks due to physical condition: limited mobility, obesity and heavy smokers, according to Holt-Lunstad et al. [39], and also due to mental conditions like depression, stress and anxiety.

Int. J. Environ. Res. Public Health 2021, 18, 4620 12 of 26

differences occur because people with a partner may have more resources available for the household due to their jobs or pensions. Also, in the plot, we observe that people with a partner have better health than singles. The distribution for the people with a partner is Int. J. Environ. Res. Public Health 2021, 18, x 12 of 26 positively skewed with a median of 2, while the distribution for single people is symmetric. People who live alone have more health risks due to physical condition: limited mobility, obesity and heavy smokers, according to Holt-Lunstad et al. [39], and also due to mental conditions like depression, stress and anxiety.

Figure 7. Boxplot economic and health indicators by marital status. Figure 7. Boxplot economic and health indicators by marital status. In Figure8, both indicators are plotted by age group. Regarding the economic indicator, we observe that although all groups have the same median value, the distribution for people over 80Inyears Figure old is8, positively both indicators skewed, and are this plotted group presents by age the group. greatest vulnerabilityRegarding the economic indicator,levels. Onwe the observe other hand, that for although the health indicator,all groups it is have observed the that same vulnerability median level value, the distribution forincreases people with over age. The80 years median old value is for positively people under skewed 60 years, oldand is 2,this while group that for presents the greatest people over 80 years old is 5. However, in the ﬁrst two age groups, there are some outliers, vulnerabilityindicating a high levels. vulnerability On the level. other Likewise, hand, thefor shape the health of the distributionindicator, changes it is observed that vulner- abilityamong groups,level increases for example, with the distribution age. The formedian the youngest value group for people is positively under skewed, 60 years old is 2, while thatmeanwhile, for people the distribution over 80 ofyears the oldest old groupis 5. However, is symmetric. in Above, the first we mentioned two age thatgroups, there are some life expectation has a negative correlation with health because the Healthy Life Years (HLY) outliers,are less, that indicating is, people tend a to high have vulnerability more limitations and level. chronic Likewise, diseases as the life goes shape on. of the distribution changesIn Figure among9, we comparegroups, both for indicators example, according the distribution to years of education. for the The youngest groups group is positively skewed,for years ofmeanwhile, education were the selected distribution using the quartilesof the oldest and adding group two groupsis symmetric. for the Above, we men- people who never went to school and for those who were still studying. For the economic tionedvulnerability that life indicator, expectation we observe has differences a negative in terms correlation of the distribution with health shape, because but the Healthy Life Yearsnot in the(HLY) median. are In less, general, that the is, vulnerability people tend levels to tend have to decreasemore limitatio with the yearsns and of chronic diseases as lifeeducation. goes on. The distribution for the two ﬁrst groups (people who never went to school and people that studied at most 9 years) are very similar, both groups reaching higher values of vulnerability. For the group who studied between 10 and 12 year, the values of vulnerability are concentrated around the median. The distribution is negatively skewed for those that studied more than 12 years, the 50% of the population older than 50 years old had economic vulnerability levels between 0 and 2.5. Also, people who said to be still studying have similar values of vulnerability than those with more years of education. The previous behaviour happens because the probability of being unemployed is reduced with more education [40], in addition there is a positive relation between the inequality years of education and inequality in income [41]. Regarding the health indicator, we observed more differences among groups. People who never went to school tend to have

Figure 8. Boxplot of economic and health indicators by age group.

Int. J. Environ. Res. Public Health 2021, 18, x 12 of 26

Figure 7. Boxplot economic and health indicators by marital status.

In Figure 8, both indicators are plotted by age group. Regarding the economic indicator, we observe that although all groups have the same median value, the distribution for people over 80 years old is positively skewed, and this group presents the greatest vulnerability levels. On the other hand, for the health indicator, it is observed that vulnerability level increases with age. The median value for people under 60 years old is 2, while that for people over 80 years old is 5. However, in the first two age groups, there are some Int. J. Environ. Res. Public Health 2021, outliers,18, 4620 indicating a high vulnerability level. Likewise, the shape of the13 of distribution 26 Int. J. Environ. Res. Public Health 2021, 18, x 13 of 26 changes among groups, for example, the distribution for the youngest group is positively skewed, meanwhile, the distribution of the oldest group is symmetric. Above, we men- greatertioned valuesthat life of expectation health vulnerability. has a negative When thecorrelation years of with education health increase, because the the levels Healthy Life ofYears vulnerabilityIn (HLY)Figure 9,are decreasewe less, compare that until is, both those people indicators people tend who accordingto have studied more to moreyears limitatio thanof education. 12ns years. and Thechronic With groups these diseases as for years of education were selected using the quartiles and adding two groups for the results,life goes we on. can conclude that there are not many differences between people who studied anpeople undergraduate who never degreewent to or school postgraduate and for those degree who in termswere still of health studying. vulnerability. For the economic vulnerability indicator, we observe differences in terms of the distribution shape, but not in the median. In general, the vulnerability levels tend to decrease with the years of education. The distribution for the two first groups (people who never went to school and people that studied at most 9 years) are very similar, both groups reaching higher values of vulnerability. For the group who studied between 10 and 12 year, the values of vulnerability are concentrated around the median. The distribution is negatively skewed for those that studied more than 12 years, the 50% of the population older than 50 years old had economic vulnerability levels between 0 and 2.5. Also, people who said to be still studying have similar values of vulnerability than those with more years of education. The previous behaviour happens because the probability of being unemployed is reduced with more education [40], in addition there is a positive relation between the inequality years of education and inequality in income [41]. Regarding the health indicator, we observed more differences among groups. People who never went to school tend to have greater values of health vulnerability. When the years of education increase, the levels of vulnerability decrease until those people who studied more than 12 years. With these results, we can conclude that there are not many differences between people who studied an undergraduate degree or postgraduate degree in terms of health vulnerability. The impact between education and health have been explained in two ways. The first is related to the economic indicator, in which we observed that more years of education implies more income; thus the more income, the more money that can be spent in health care. The second way is through mental and physical health. Zimmerman and Woolf [42] noted that people with more education develop more skills like reading, math and science that are related with better health, and also, people with higher education tend to do more exercise and healthy activities. Figure 8. Boxplot of economic and health indicators by age group. Figure 8. Boxplot of economic and health indicators by age group.

Figure 9. Boxplot of economic and health indicators by years of education. Figure 9. Boxplot of economic and health indicators by years of education. The impact between education and health have been explained in two ways. The ﬁrst Finally, we compute the average value of each indicator per country and plot these is related to the economic indicator, in which we observed that more years of education values in a map to analyse both indicators across Europe (see Figures 10 and 11). implies more income; thus the more income, the more money that can be spent in health care.When The second we plot way the is through average mental value for and the physical economic health. indicator Zimmerman per country, and Woolf we [ can42] notedclearly that see peoplegroups with of countries. more education We observe develop that more the Nordic skills like countries reading, have math low and le sciencevels of economic vulnerability, followed by the Central European region with levels around 2.

Int. J. Environ. Res. Public Health 2021, 18, x 14 of 26

Int. J. Environ. Res. Public Health 2021, 18, 4620 14 of 26

Int. J. Environ. Res. Public Health 2021, 18, x 14 of 26 Southern and Eastern Europe have similar levels of vulnerability, around 2.5. Finally, the Baltic region has the highest levels of vulnerability. that are related with better health, and also, people with higher education tend to do more exercise and healthy activities. Southern and Eastern Europe have similar levels of vulnerability, around 2.5. Finally, the Finally, we compute the average value of each indicator per country and plot these Balticvalues region in a map has to the analyse highest both levels indicators of vulnerability. across Europe (see Figures 10 and 11).

Figure 10. Average value of economic indicator per country.

Likewise, we analyse the health indicator across Europe. Like the economic indicator, Nordic countries and Switzerland have the lowest levels of vulnerability. The next in the ranking are countries like France, Greece, Italy, and Portugal. On the other hand, the re- gionsFigure with 10. Average highest value level of of economic vulnerability indicator are per Baltic country. and Balkan regions. Figure 10. Average value of economic indicator per country.

Likewise, we analyse the health indicator across Europe. Like the economic indicator, Nordic countries and Switzerland have the lowest levels of vulnerability. The next in the ranking are countries like France, Greece, Italy, and Portugal. On the other hand, the regions with highest level of vulnerability are Baltic and Balkan regions.

Figure 11. Average value of health indicator per country. When we plot the average value for the economic indicator per country, we can clearly Finally, we want to understand the relationship between the two indicators in each see groups of countries. We observe that the Nordic countries have low levels of economic country. For that purpose, we plot their mean values by country in Figure 12, where four vulnerability, followed by the Central European region with levels around 2. Southern and groups of countries can be observed. In the lower left corner are located the Nordic coun- Eastern Europe have similar levels of vulnerability, around 2.5. Finally, the Baltic region tries, those with the lowest vulnerability level, according to the indicators created from has the highest levels of vulnerability. the SHARE database. After that, in the centre of the graph are located two groups, the first Likewise, we analyse the health indicator across Europe. Like the economic indicator, one with Southern European countries like Portugal, Greece, Cyprus, Malta, Italy, Spain, Nordic countries and Switzerland have the lowest levels of vulnerability. The next in Figureand two 11. Baltic Average countries value likeof health Slovenia indicator and Sl perovakia, country. those countries have an average value the ranking are countries like France, Greece, Italy, and Portugal. On the other hand, the between 3 and 3.5 in the health indicator and values around 2.5 in the economic indicator. regions with highest level of vulnerability are Baltic and Balkan regions. The secondFinally, group we want located to inunderstand the centre is the compounded relationship by Francebetween and the Finland two indwithicators a very in each Finally, we want to understand the relationship between the two indicators in each country. For that purpose, we plot their mean values by country in Figure 12, where four country. For that purpose, we plot their mean values by country in Figure 12, where groups of countries can be observed. In the lower left corner are located the Nordic countries, those with the lowest vulnerability level, according to the indicators created from the SHARE database. After that, in the centre of the graph are located two groups, the first one with Southern European countries like Portugal, Greece, Cyprus, Malta, Italy, Spain, and two Baltic countries like Slovenia and Slovakia, those countries have an average value between 3 and 3.5 in the health indicator and values around 2.5 in the economic indicator. The second group located in the centre is compounded by France and Finland with a very

Int. J. Environ. Res. Public Health 2021, 18, 4620 15 of 26

four groups of countries can be observed. In the lower left corner are located the Nordic Int. J. Environ. Res. Public Health 2021, 18, x 15 of 26 countries, those with the lowest vulnerability level, according to the indicators created from the SHARE database. After that, in the centre of the graph are located two groups, the ﬁrst one with Southern European countries like Portugal, Greece, Cyprus, Malta, Italy, Spain, and two Baltic countries like Slovenia and Slovakia, those countries have an average similarvalue between score 3 in and both 3.5 in indicators the health indicator, near to and 2 in values the aroundeconomic 2.5 in indicator the economic and 3.25 in the health indicator.indicator. The In second the third group group located we in the found centre countries is compounded like by Austria, France and Belgium, Finland Germany, Luxem- with a very similar score in both indicators, near to 2 in the economic indicator and 3.25 bourg,in the health and indicator.Czech Republic, In the third which group we are found character countriesised like by Austria, having Belgium, low levels of economic vulnerability,Germany, Luxembourg, but a and scor Czeche greater Republic, than which 3.5 are in characterised the health by indicator. having low levelsFinally, there is a group locatedof economic in vulnerability,the top right but corner, a score greater which than basically 3.5 in the healthconsists indicator. of Baltic Finally, and there Balkans states, whose is a group located in the top right corner, which basically consists of Baltic and Balkans countriesstates, whose have countries higher have scores higher scores than than the theothers others in in both indicators. indicators.

Figure 12. Economic vs. health indicator. Figure 12. Economic vs. health indicator. With the previous groups, we analyse the causes of these differences in the economic and healthWith indicators. the previous In Table groups,1, we report we the analyse mean and the median causes values of these of the differences subindi- in the economic cators and indicators according to the groups described above. We can see that Southern andEuropean health and indicators. Eastern regions In have Table similar 1, vulnerabilitywe report levelsthe mean of unemployment, and median although values of the subindica- torsthe Eastern and indicators region is more according vulnerable dueto tothe the groups monetary described poverty and above. ﬁnancial distress.We can see that Southern EuropeanNorthern European, and Eastern Scandinavian regions regions, have and similar Switzerland vulnerability have the lowest levels levels of of unemployment, alt- monetary poverty according with the SHARE data, but the unemployment reported in houghthe Northern the Eastern European region countries is is more higher. vulnerable For the health due indicator, to the Eastern monetary European poverty and financial distress.countries haveNorthern a great differenceEuropean, compared Scandinavian to the other regions, regions in and all the Switzerland subindicators; have the lowest lev- elhowever,s of monetary these differences poverty are more according noticeable with in the the physical SHARE and mental data, health but the subindica- unemployment reported tors. The Northern European region has higher physical health vulnerability than Southern inEuropean, the Northern Scandinavian European regions andcountries Switzerland. is higher. On the otherFor the hand, health the mean indicator, values Eastern European countriesof the health have care subindicator a great difference of Scandinavian compared region and to Switzerlandthe other areregion highers in than all the subindicators; however,those of Southern these and differences Northern European are more countries, noticeable which means in the that physical the population and ofmental health subin- these regions have spoken to or seen a medical doctor more times or have stayed more dicators.nights at the The hospital. Northern Similar Europeanresults to those region previously has described higher are physical observed health using vulnerability than Southernthe median European, value. It is shown Scandinavian that at least 50%regions of the and Southern, Switzerland. Northern and On Eastern the other hand, the mean valuesEuropean of population the health are vulnerablecare subindicator in terms of ﬁnancialof Scandinavian distress. Also, region the median and value Switzerland are higher of the health indicator is the highest for the Eastern European region, while the median thanvalue ofthose mental of health Southern for the Southern,and Northern Northern European and Eastern Europeancountries, regions which is higher means that the popula- tionthan correspondingof these regions one for have Scandinavian spoken countries to or seen and Switzerland. a medical doctor more times or have stayed more nights at the hospital. Similar results to those previously described are observed using the median value. It is shown that at least 50% of the Southern, Northern and East- ern European population are vulnerable in terms of financial distress. Also, the median value of the health indicator is the highest for the Eastern European region, while the median value of mental health for the Southern, Northern and Eastern European regions is higher than corresponding one for Scandinavian countries and Switzerland.

Table 1. Mean (median) value of each indicator and subindicator per European region.

Weekly Financial Unemployed Poverty Economic Region Consumption Distress Subindicator Subindicator Indicator Subindicator Subindicator Southern European 0.04 (0) 0.22 (0) 0.01 (0) 0.83 (1) 2.73 (2.5) Eastern European 0.04 (0) 0.29 (0) 0.00 (0) 0.91 (1) 3.07 (2.5) Northern European 0.03 (0) 0.16 (0) 0.01 (0) 0.56 (1) 1.86 (2.5) Scandinavian and 0.01 (0) 0.16 (0) 0.01 (0) 0.37 (0) 1.37 (2.5) Switzerland

Int. J. Environ. Res. Public Health 2021, 18, 4620 16 of 26

Table 1. Mean (median) value of each indicator and subindicator per European region.

Weekly Financial Unemployed Poverty Economic Int. J. Environ. Res. Public Health 2021, 18,Region x Consumption Distress 16 of 26 Subindicator Subindicator Indicator Subindicator Subindicator Southern 0.04 (0) 0.22 (0) 0.01 (0) 0.83 (1) 2.73 (2.5) European Eastern 0.04 (0) 0.29 (0) 0.00 (0) 0.91 (1) 3.07 (2.5) European General Health Mental Health Health Care Physical Health Health NorthernRegion 0.03 (0) 0.16 (0) 0.01 (0) 0.56 (1) 1.86 (2.5) European Subindicator Subindicator Subindicator Subindicator Indicator SouthernScandinavian European 0.65 (1) 1.26 (1) 0.25 (0) 1.68 (2) 3.13 (3) and 0.01 (0) 0.16 (0) 0.01 (0) 0.37 (0) 1.37 (2.5) SwitzerlandEastern European 0.85 (1) 1.57 (2) 0.33 (0) 2.06 (2) 3.95 (4) General Mental Physical Northern European 0.67 (1)Health Care 1.15 (1) Health0.23 (0) 1.87 (2) 3.22 (3) Region Health Health Health Subindicator Indicator ScandinavianSubindicator and Subindicator Subindicator Southern 0.46 (0) 0.89 (1) 0.26 (0) 1.49 (1) 2.56 (2) Switzerland0.65 (1) 1.26 (1) 0.25 (0) 1.68 (2) 3.13 (3) European Eastern 0.85 (1) 1.57 (2) 0.33 (0) 2.06 (2) 3.95 (4) 3.2.European Visualising the Vulnerability Profiles Northern 0.67 (1) 1.15 (1) 0.23 (0) 1.87 (2) 3.22 (3) EuropeanHere, we cluster the individuals in groups with similar characteristics. To do so, we Scandinavian combineand the0.46 economic (0) 0.89 and (1) health 0.26 (0) vulnerability 1.49 (1) indicators 2.56 (2) with the sociodemographic in- formationSwitzerland via a modified version of the k-Prototypes clustering algorithm. In order to se- l3.2.ect Visualising the number the Vulnerability of clusters Profiles we run the algorithm for k = 2, 3, …, 10. In Figure 13 we show Here, we cluster the individuals in groups with similar characteristics. To do so, we thecombine cost the function, economic and healthwhere vulnerability an “elbow” indicators can with be the sociodemographicobserved around k = 3 or k = 4. Finally, after investigatinginformation via a modified having version both of the k-Prototypes3 or 4 profiles, clustering algorithm.we selected In order tok = 4, since k = 3 led to too broad select the number of clusters we run the algorithm for k = 2, 3, ... , 10. In Figure 13 we results.show the cost function, where an “elbow” can be observed around k = 3 or k = 4. Finally, after investigating having both 3 or 4 profiles, we selected k = 4, since k = 3 led to too broad results.

Figure 13. Scree plot. Figure 13. Scree plot.

To define the profiles in terms of vulnerability, we analyse the distribution of each variable in each cluster. The distributions are shown in Figure 14. Cluster 1 has low levels of economic vulnerability, with a maximum value close to 5. The health indicator has many instances with a vulnerability level of 2, but with some outliers with both very low and high levels of vulnerability. This cluster is characterised by females living with a partner, under 60 years old and with more than 12 years of education. For cluster 2, the boxplot of the economic indicator shows that about 50% of the population has a vulnerability level between 0 and 2, but that there are people with higher level of economic vulnerability than in cluster 1. The distribution of the health indicator shows that at least half of the population has vulnerability levels between 2 and 3, with some outliers that reach higher levels, like 8. The profile of this cluster are males living with a partner, between 60 and 70 years old and more than 12 years of education. The economic indicator in the cluster 3 has a positively skewed distribution with a median around 3, but 50% of the population is distributed between 3 and 5 level, also the values of this indicator reached values of 9 and it has some outliers with value 10 in the scale. For the health indicator in the same cluster, we find a median value of 5 and a mean of 5.21, both values are the highest among the clusters. In addition, this cluster is characterised by single females over 70 years old with at most 9 years of education. Finally, in cluster 4, we observe a similar distribution in the economic indicator to that of cluster 3. The health indicator has a negatively skewed distribution with a median of 4 and a mean of 3.7. Likewise, 50% of the observations have a vulnerability level between 2 and 5. The profile of this cluster are males living with a partner, under 60 years old and with 10–12 years of education.

Int. J. Environ. Res. Public Health 2021, 18, x 17 of 26

In conclusion, we can say that the most vulnerable profile is that found in cluster 3, since it presents both the highest levels of the health and economic indicators. The least Int. J. Environ. Res. Public Health 2021, 18, 4620 17 of 26 vulnerable profile is that found in cluster 1. Clusters 4 and 2 define those profiles with a medium level of vulnerability. In summary, we can establish the following profiles sorted from the least to the most vulnerable: Profile 1 (Cluster 1), Profile 2 (Cluster 2), Profile 3 (ClusterTo deﬁne 4), and the Profile proﬁles 4 (Cluster in terms 3). of vulnerability, we analyse the distribution of each variable in each cluster. The distributions are shown in Figure 14.

Figure 14. Distribution ofeach variable per cluster.

ClusterFigure 1 has14. Distribution low levels of of each economic variable vulnerability, per cluster. with a maximum value close to 5. The health indicator has many instances with a vulnerability level of 2, but with some outliersNext, with we both compute very low the and proportion high levels of of the vulnerability. population This per cluster profile is tocharacterised analyse which by femalescountries living have with the amost partner, vulnerable under 60 people. years oldWe andplot withthese more maps than in Figure 12 years 15. ofThe education. countries withFor more cluster population 2, the boxplotin the least of thevulnerable economic profile indicator (Profile shows 1) are that Denmark about (48.4%), 50% of Swe- the populationden (41.2%), has Switzerland a vulnerability (39.0%), level betweenLuxembourg 0 and (35.8%), 2, but that Belgium there are(35.50%), people and with Germany higher level(34.22%). of economic Meanwhile, vulnerability Baltic and than Balkans in cluster states 1. Thehave distribution less population of the in health this profile, indicator for showsinstance, that Romania at least half (11.41%), of the population Latvia (11.72%), has vulnerability Hungary (12.16%), levels between and Croatia 2 and (12.93%). 3, with someAlso, outliers in Profile that 2 reachthere is higher a high levels, concentration like 8. The of proﬁleScandinavian of this clusterpopulation, are males around living 30%, with a partner, between 60 and 70 years old and more than 12 years of education. The economic indicator in the cluster 3 has a positively skewed distribution with a median around 3, but 50% of the population is distributed between 3 and 5 level, also the values of this indicator reached values of 9 and it has some outliers with value 10 in Int. J. Environ. Res. Public Health 2021, 18, 4620 18 of 26

the scale. For the health indicator in the same cluster, we find a median value of 5 and a mean of 5.21, both values are the highest among the clusters. In addition, this cluster is characterised by single females over 70 years old with at most 9 years of education. Finally, in cluster 4, we observe a similar distribution in the economic indicator to that of cluster 3. The health indicator has a negatively skewed distribution with a median of 4 and a mean of 3.7. Likewise, 50% of the observations have a vulnerability level between 2 and 5. The profile of this cluster are males living with a partner, under 60 years old and with 10–12 years of education. In conclusion, we can say that the most vulnerable profile is that found in cluster 3, since it presents both the highest levels of the health and economic indicators. The least vulnerable profile is that found in cluster 1. Clusters 4 and 2 define those profiles with a medium level of vulnerability. In summary, we can establish the following profiles sorted from the least to the most vulnerable: Profile 1 (Cluster 1), Profile 2 (Cluster 2), Profile 3 (Cluster 4), and Profile 4 (Cluster 3). Next, we compute the proportion of the population per profile to analyse which countries have the most vulnerable people. We plot these maps in Figure 15. The countries with more population in the least vulnerable profile (Profile 1) are Denmark (48.4%), Swe- den (41.2%), Switzerland (39.0%), Luxembourg (35.8%), Belgium (35.50%), and Germany (34.22%). Meanwhile, Baltic and Balkans states have less population in this profile, for instance, Romania (11.41%), Latvia (11.72%), Hungary (12.16%), and Croatia (12.93%). Also, in Profile 2 there is a high concentration of Scandinavian population, around 30%, but also countries as Greece (34.2%), Italy (32.4%), France (29.79%), Switzerland (29.7%), and Belgium (29.4%). In this profile the Baltic and Balkans states have a low concentration of population. This trend changes in Profile 3, where Eastern European countries have around 33% of their population, followed by countries like Poland (29.2%), Malta (28.1%), Estonia (27.5%), Czech Republic (26.5%), and Portugal (24.8%). On the other hand, Denmark, Sweden, and Switzerland have 6.8%, 10.8% and 14.3% of population within this cluster, respectively. Finally, concerning Profile 4, countries with more vulnerable population are Latvia (41.1%), Romania (20.1%), Hungary (36.68%), Estonia (36.3%), and Lithuania (36.0%), followed by Eastern European countries, like Poland (31.4%), and Southern Eu- ropean countries, like Portugal (31.0%), Italy (30.1%) and Spain (28.9%). In these profiles the top 5 of countries with less vulnerable population are Denmark (11.9%), Switzerland (16.8%), Sweden (17.7%), Belgium (19.3%), and France (20.7%). Looking at the maps presented in Figure 15, we clearly observe that Eastern European countries are the most vulnerable, followed by Southern European countries. In third place are Central European countries, and finally, Scandinavian countries and Switzerland. Note that these findings are based on the geographical distribution of profiles 4 and 1, which are obtained from the available information in SHARE wave 7. Notice that some European countries (like Norwegian, Ireland, United Kingdom) were not included in the survey, thus, the inclusion of such countries might modify the previous findings.

3.3. Visualising the Pandemic Shock Next, we apply the shock designed in Section 2.3 and again compute the economic and health indicators to analyse the differences. The distribution of the economic indicator moves to the right, as it has more observations with values of 7.5 and 10. This means that there are more vulnerable people. Also, we observed that 50% of the population now is located between 2.5 and 5, the median does not change, and the mean is 2.91 instead of 2.33. The Q1 value is equal to the Q3 value before the shock, this means that the pandemic had an economic impact. See panel (a) of Figure 16. Int. J. Environ. Res. Public Health 2021, 18, x 18 of 26

but also countries as Greece (34.2%), Italy (32.4%), France (29.79%), Switzerland (29.7%), and Belgium (29.4%). In this profile the Baltic and Balkans states have a low concentration of population. This trend changes in Profile 3, where Eastern European countries have around 33% of their population, followed by countries like Poland (29.2%), Malta (28.1%), Estonia (27.5%), Czech Republic (26.5%), and Portugal (24.8%). On the other hand, Den- mark, Sweden, and Switzerland have 6.8%, 10.8% and 14.3% of population within this cluster, respectively. Finally, concerning Profile 4, countries with more vulnerable population are Latvia (41.1%), Romania (20.1%), Hungary (36.68%), Estonia (36.3%), and Lith- uania (36.0%), followed by Eastern European countries, like Poland (31.4%), and Southern European countries, like Portugal (31.0%), Italy (30.1%) and Spain (28.9%). In these profiles the top 5 of countries with less vulnerable population are Denmark (11.9%), Switzer- land (16.8%), Sweden (17.7%), Belgium (19.3%), and France (20.7%). Looking at the maps presented in Figure 15, we clearly observe that Eastern Euro- pean countries are the most vulnerable, followed by Southern European countries. In third place are Central European countries, and finally, Scandinavian countries and Switzer- land. Note that these findings are based on the geographical distribution of profiles 4 and

Int. J. Environ. Res. Public Health 20211,, 18 which, 4620 are obtained from the available information in SHARE wave 7. Notice that 19some of 26 European countries (like Norwegian, Ireland, United Kingdom) were not included in the survey, thus, the inclusion of such countries might modify the previous findings.

Int. J. Environ. Res. Public Health 2021, 18, x 19 of 26

2.33. The Q1 value is equal to the Q3 value before the shock, this means that the pandemic had an economic impact. See panel (a) of Figure 16. The health indicator also changes; the distribution moves to the right, but it still has a positively skewed distribution. As in the economic indicator, the median value does not change, but the mean is 3.86 instead of 3.2. Nevertheless, the Q3 is equal to 6, while before the shock this value was equal to 5 and after the shock a vulnerability level of 10 is not considered as an outlier. Finally, with a pandemic 50% of the European population older

than 50 years old presents vulnerability levels between 2 and 6. See panel (b) of Figure 16. FigureFigure 15. 15. ProportionProportion of of population population per per country country in in each each profile. proﬁle.

(a) Economic indicator (b) Health indicator

FigureFigure 16.16. Bar-plotBar-plot andand boxplotboxplot ofof economiceconomic andand healthhealth indicatorsindicators beforebefore andand afterafter thethe pandemic.pandemic.

Next, we analyse how the countries were affected by the shock. In particular, we study how the groups sketched in Figure 12 are modified by the pandemic shock. These changes are illustrated in Figure 17. Before the shock Denmark, Sweden, and Switzerland’s economic levels were around 1.5 and health levels around 2.5, while after the shock, they reach vulnerability levels around 2 and 3.25 in the economic and health indicators, respectively. This represents the highest growth rate in vulnerability levels after the shock. This is explained by the fact that before the pandemic these countries did not have a large population within the vulnerability groups, however, the after the pandemic shock those with the previously explained characteristics, such as chronic conditions, working sector and so on (see Figure 5) became vulnerable. Nevertheless, these three countries are still far apart on the scale of vulnerability from the other groups of countries. The group comprising France, Finland, Luxembourg, Austria, Belgium, Germany, and Czech Republic experiment a high growth in their vulnerability levels. This means that these countries are more sensitive to health shocks that affect the mental health of their population. For example, before the shock, in Germany, the mean value of the mental health subindicator was 1.12, while after the shock, this value was 1.77. Meanwhile, the

Int. J. Environ. Res. Public Health 2021, 18, x 20 of 26

health indicator value preshock was 3.44, and after the shock it was 4.18. It is easy to see that the difference in the value of the health indicator is almost entirely explained by the variation in the mental health subindicator. A third group was made up of Slovenia and Southern European countries like Spain, Portugal, Italy, Greece, and Malta. These countries still have greater economic vulnerability than Scandinavian and Central European countries, but the differences seem to remain in absolute terms. Additionally, countries like Spain and Greece present a great growth in their levels of health vulnerability, around 25.3% and 23.9%, respectively. The economy of these countries depends on sectors highly affected by the pandemic like tourism; this Int. J. Environ. Res. Public Health 2021,could18, 4620 increase the effect of the pandemic on their health vulnerability. 20 of 26 Finally, Eastern European countries move to the upper right corner but with a smaller magnitude. For instance, countries such as Romania, Hungary, and Bulgaria are more economically vulnerable after the shock, but the change is only of 3.6%, 8.2% and 9.89%, The health indicator also changes; the distribution moves to the right, but it still has a positivelyrespectively. skewed Likewise, distribution. these countries As in the have economic the highest indicator, levels theof health median vulnerability, value does notbut change,also their but growth the mean rate iss are 3.86 the instead lowest of among 3.2. Nevertheless, European countries the Q3 is; for equal example, to 6, while the growth before therate shock of Romania, this value Latvia, was equaland Bulgaria to 5 and is after 7.8%, the 8.9% shock and a vulnerability10.1%, respectively. level of Although 10 is not consideredEastern European as an outlier. countries Finally, have withless variation a pandemic in their 50% economic of the European and health population vulnerability, older thanthese 50 countries years old are presents still the vulnerability most vulnerable. levels Actually, between 2this and is 6.not See a good panel result (b) of Figurebecause 16 it. meansNext, that we in analysethe pre howshock the world countries these were countries affected were by thevery shock. vulnerable, In particular, there weis a study high howinequality the groups in economic sketched and in Figurehealth terms12 are between modiﬁed Eastern by the pandemicEuropean shock.countries These and changes the rest areof Europe. illustrated in Figure 17.

Figure 17. Relationship between economic and health indicator per country before and after the shock. Figure 17. Relationship between economic and health indicator per country before and after the shock. Before the shock Denmark, Sweden, and Switzerland’s economic levels were around 1.5 andWith health the new levels indicators around, we 2.5, recalculate while after the the vulnerability shock, they profiles reach vulnerabilityto analyse their levels var- aroundiation after 2 and the 3.25 shock. in theTo economiccompute the and new health profile indicators, composition, respectively. we use the This predict represents func- thetion highest of the clustMixType growth rate inR- vulnerabilitypackage, and levelsthen we after calculate the shock. the proportion This is explained of population by the factthat thatbelongs before to each the pandemic profile and these we plot countries it on a did map. not In have Figure a larges 18 and population 19 we illustrate within thethe vulnerabilitychanges in the groups, geographical however, distribution the after of the the pandemic most and shock least vulnerable those with profiles, the previously respec- explainedtively. characteristics, such as chronic conditions, working sector and so on (see Figure5 ) becameAfter vulnerable. the shock, Nevertheless, all countries thesehave more three population countries are in stillthe farmost apart vulnerabl on thee scaleprofile. of vulnerabilityNevertheless, from Scandinavian the other groupscountries of countries.and Switzerland remain the least vulnerable popu- lationThes. Southern group comprising European France, countries Finland, are more Luxembourg, vulnerable Austria, after the Belgium, shock. W Germany,e observed and a Czechhigh change Republic in experiment the proportion a high of growth population in their in countries vulnerability like levels. Portugal This (7.1%), means Spain that these(6.2%), countries and Greece are more(6.5%). sensitive The proportion to health of shocks the most that vulnerable affect the population mental health in northern of their population.European countries For example, also grew before rapidly the shock,after the in shock, Germany, for instance, the mean Belgium’s value of growth the mental rate healthwas of subindicator7.3%, Germany’s was 1.12,7.17%, while Luxemburg’s after the shock, 6.9%, thisand valueCzech was Republi 1.77.c’s Meanwhile, 6.7%. Finally, the healthEastern indicator European value countries preshock have was more 3.44, vulnerable and after population the shock it but was the 4.18. changes It is easy of in to these see that the difference in the value of the health indicator is almost entirely explained by the variation in the mental health subindicator. A third group was made up of Slovenia and Southern European countries like Spain, Portugal, Italy, Greece, and Malta. These countries still have greater economic vulnerability

than Scandinavian and Central European countries, but the differences seem to remain in absolute terms. Additionally, countries like Spain and Greece present a great growth in their levels of health vulnerability, around 25.3% and 23.9%, respectively. The economy of these countries depends on sectors highly affected by the pandemic like tourism; this could increase the effect of the pandemic on their health vulnerability. Finally, Eastern European countries move to the upper right corner but with a smaller magnitude. For instance, countries such as Romania, Hungary, and Bulgaria are more economically vulnerable after the shock, but the change is only of 3.6%, 8.2% and 9.89%, respectively. Likewise, these countries have the highest levels of health vulnerability, but Int. J. Environ. Res. Public Health 2021, 18, x 21 of 26 Int. J. Environ. Res. Public Health 2021, 18, 4620 21 of 26

countries were less abrupt. In countries like Romania the vulnerable proportion grew by 1.8%,also their in Latvia growth by rates1.9%, are and the in lowest Bulgaria among by 2.2%. European countries; for example, the growth rate of Romania, Latvia, and Bulgaria is 7.8%, 8.9% and 10.1%, respectively. Although Eastern European countries have less variation in their economic and health vulnerability, these countries are still the most vulnerable. Actually, this is not a good result because it means that in the preshock world these countries were very vulnerable, there is a high inequality in economic and health terms between Eastern European countries and the rest

Int. J. Environ. Res. Public Health 2021, of18, Europe.x 21 of 26 With the new indicators, we recalculate the vulnerability profiles to analyse their variation after the shock. To compute the new profile composition, we use the predict function of the clustMixType R-package, and then we calculate the proportion of population that be- countrieslongs to each were profile less abrupt. and we In plot countries it on a map. like Romania In Figures the 18 vulnerableand 19 we illustrateproportion the grew changes by 1.8%,in the in geographical Latvia by 1.9%, distribution and in Bulgaria of the most by 2.2%. and least vulnerable profiles, respectively.

Figure 18. Geographical distribution of the most vulnerable profile before and after the shock.

The effect on the least vulnerable profile is similar. The proportion of population in these profiles decreases in each country. For instance, the proportion of population of Denmark and Sweden in this profile decreases around 7 and 5 percentage points, respectively, while the proportion of population of Romania and Latvia in this profile decreases around 0.5 and 0.9 percentage points, respectively. However, the effect of the pandemic is different among Eastern European countries; before the shock, the country with the lowest proportion of population in this profile was Romania (before the shock 11.4% and after the shock 10.8%), while afterwards it was Hungary (before the shock 12.1% and after Figure 18. Geographicalthe shock distribution10.4%). of the most vulnerable proﬁle before and after the shock. Figure 18. Geographical distribution of the most vulnerable profile before and after the shock.

Figure 19. Geographical distribution of the least vulnerable profile proﬁle before and after the shock. After the shock, all countries have more population in the most vulnerable proﬁle. The previous results allow us to understand that a health shock like a pandemic with Nevertheless, Scandinavian countries and Switzerland remain the least vulnerable popula- economic and health consequences affect all European countries, but it affects more those tions. Southern European countries are more vulnerable after the shock. We observed a countries with high levels of vulnerability in physical and mental health. Additionally, we high change in the proportion of population in countries like Portugal (7.1%), Spain (6.2%), observe two types of effects, i.e., countries that were previously more vulnerable do not and Greece (6.5%). The proportion of the most vulnerable population in northern European experiment a high change in vulnerability, while countries with medium vulnerability countries also grew rapidly after the shock, for instance, Belgium’s growth rate was of profiles present a higher variation after the shock and their proportion of population in 7.3%, Germany’s 7.17%, Luxemburg’s 6.9%, and Czech Republic’s 6.7%. Finally, Eastern theEuropean most vulnerable countries haveprofile more tends vulnerable to increase. population but the changes of in these countries were less abrupt. In countries like Romania the vulnerable proportion grew by 1.8%, in Figure 19. GeographicalLatvia by distribution 1.9%, and inof Bulgariathe least vulnerable by 2.2%. profile before and after the shock.

The previous results allow us to understand that a health shock like a pandemic with economic and health consequences affect all European countries, but it affects more those countries with high levels of vulnerability in physical and mental health. Additionally, we observe two types of effects, i.e., countries that were previously more vulnerable do not experiment a high change in vulnerability, while countries with medium vulnerability profiles present a higher variation after the shock and their proportion of population in the most vulnerable profile tends to increase.

Int. J. Environ. Res. Public Health 2021, 18, 4620 22 of 26

The effect on the least vulnerable profile is similar. The proportion of population in these profiles decreases in each country. For instance, the proportion of population of Denmark and Sweden in this profile decreases around 7 and 5 percentage points, respectively, while the proportion of population of Romania and Latvia in this profile decreases around 0.5 and 0.9 percentage points, respectively. However, the effect of the pandemic is different among Eastern European countries; before the shock, the country with the lowest proportion of population in this profile was Romania (before the shock 11.4% and after the shock 10.8%), while afterwards it was Hungary (before the shock 12.1% and after the shock 10.4%). The previous results allow us to understand that a health shock like a pandemic with economic and health consequences affect all European countries, but it affects more those countries with high levels of vulnerability in physical and mental health. Additionally, we observe two types of effects, i.e., countries that were previously more vulnerable do not experiment a high change in vulnerability, while countries with medium vulnerability profiles present a higher variation after the shock and their proportion of population in the most vulnerable profile tends to increase. Future research and public health policies need to pay special attention to vulnerable populations including women, old people, the unemployed and those who have or had COVID-19 symptoms among others [15]. Due to the large proportion of the population affected by COVID-19, public health policies need to take health consequences into account [15]. For instance, the increase in loneliness related to social isolation during COVID-19 is linked to long-term health outcomes including all-cause mortality [43,44]. According to [45] this analysis of European countries’ vulnerability to the COVID- 19 pandemic has identified the areas in which nations need to bolster their investments in order to alleviate the severity of the impact. Both domestic and EU funds should be allocated to these areas in order to ensure that their citizens are safe from the disease and, above all, protected from the related economic consequences.

4. Discussion The elderly population in Europe constitutes a growing vulnerable group that is facing some economic and health challenges. Nowadays, the COVID pandemic is affecting the world health, economy, and daily life, being older people and those with NCDs the most affected. The objective of this work was to analyse the effects of a health shock on the older population’s well-being. In the first part of this study, we used information from a survey applied to the Euro- pean population aged 50 or older to design indicators that would allow us to measure the risk of being economically vulnerable and unhealthy. The first indicator was compounded by four subindicators: Monetary Poverty, Unemployment, Weekly Consumption, and Financial Distress, while the second one included Mental Health, Physical Health, Health Care, and General Health measures. Using these indicators, we concluded that the risk of vulnerability in Europe is low, because the mean value of the economic indicator is 2.33 and the median is 2.5, while the mean and median of the health indicator are 3.28 and 3, respectively—all values in a range from 0 to 10. However, the risk of vulnerability is not homogenous in the European territory, as it varies substantially among European regions. For instance, the Scandinavian countries and Switzerland have the lowest vulnerability levels in both indicators, while the Eastern European countries have the highest levels. The difference between regions is explained by the financial distress, the monetary poverty, and their vulnerability in mental and physical health. Next, we combine the indicators with sociodemographic information in order to obtain the vulnerability profiles. This was done by using clustering techniques that allowed the use of quantitative variables, like the indicators, and categorical variables, such as gender, marital status, age groups, and education. More specifically, we used the k-prototypes clustering algorithm and modified it so that it can cope with weighted datasets. We found Int. J. Environ. Res. Public Health 2021, 18, 4620 23 of 26

four profiles, where the most vulnerable profile is characterised by single women over 71 years old and with few years of education, more specifically with at most 9 years of education. Furthermore, Eastern European countries have the largest population within the most vulnerable profile, around a 38% of their population is older than 50 years old. These results allowed us to achieve our objective, nevertheless, it was necessary to measure the effect of the pandemic. At the time this work was written, the effects of the COVID are still not clear, therefore, we designed a shock based on the available literature. This shock began affecting those individuals with chronic diseases. From this variable an effect was triggered onto variables such as individuals’ self-perception of health, life satisfaction, depression, happiness, and BMI. Additionally, the shock affected economic variables such as unemployment, which was most impactful in the tourism and retail sectors. After the shock, we observed that a pandemic with the previous characteristics had effects in the health and economics of all European countries, which means their well-being is lower; however, it mostly affected countries where their population is more vulnerable in terms of physical and mental health. On the other hand, countries where the strongest economic sectors were affected, the negative effect on the health indicator was greater, for example, Spain and Greece. Although the relative effect of the shock was lower in the Eastern European countries, the levels of economic and health vulnerability were high compared to the other European regions. This means that there is an inequality in Europe that should be reduced if European authorities want to reach the UNs 2030 Agenda for Sustainable Development. The work presented herein is relevant, because it allowed us to characterise the most vulnerable populations, which is useful for public policies design and to better understand how these policies should be oriented differently in each region, according to the characteristics found in this study. For instance, it is important to design policies aimed towards mental and physical care, because these are variables that not only affect the Eastern European countries, but also, the Southern and Northern European countries. Likewise, this document suggests the study of public policies that help to reduce the financial distress in the Eastern European countries, since this was a variable in which there were notable differences with other European regions. This work highlights the importance of indicators as instruments for measuring real phenomena, and the use of these for decision-making and the development of effective policies related to socioeconomic and health issues. Finally, this work provides a methodology for the creation of profiles and the measure- ment of the effects of shocks through clustering techniques that allow the use of numerical and categorical variables, as well as calibrated cross-sectional weights. Future studies should measure the real Coronavirus effects on the European population and the magnitude in which the vulnerability profiles were affected. After we designed our shock, SHARE designed a questionnaire related to COVID-19, in which they asked questions related to the variables used in our shock, for example, health and health behaviour, mental health, infections and health care, changes in work and economic situation. With this information, it will be possible not only to measure the general effect, but also the specific effects of the variables that compounded the subindicators and the indicators. In this way, public policies will be more effective in alleviating the most extreme situations of the vulnerable population.

5. Limitations The statistical analysis presented in this paper is limited to the information available during the ﬁrst COVID-19 wave, regarding health and wellbeing status in the EU. This information was collected in wave 7 SHARE dataset. As a result, this limitation constitutes recoded variables, surveyed individuals and countries that agreed to participate. Int. J. Environ. Res. Public Health 2021, 18, 4620 24 of 26

The analysis might be extended to other geographical areas with more diverse characteristics in order to conﬁrm the conclusions drawn here, in cases where similar datasets to the SHARE database exist.

Author Contributions: Conceptualization, A.G. and I.A.; methodology, A.G., I.A.; software, D.E.M.; validation, A.G., I.A. and D.E.M.; formal analysis, D.E.M.; investigation, D.E.M.; resources, I.A.; data curation, I.A., D.E.M.; writing—original draft preparation, D.E.M.; writing—review and editing, A.G., I.A.; visualization, I.A.; supervision, A.G.; project administration, A.G.; funding acquisition, A.G. All authors have read and agreed to the published version of the manuscript. Funding: This research was funded by the Spanish Ministry of Economy and Competitiveness, grant number MTM2014-56535-R. Institutional Review Board Statement: Not applicable. Informed Consent Statement: Not applicable. Data Availability Statement: Original data can be downloaded from http://www.share-project.org/ (accessed on 1 April 2020). Conﬂicts of Interest: The authors declare no conﬂict of interest.

Appendix A. Dataset and R Code The k-prototypes method combines numerical and categorical variables using two types of measures, the Euclidean distance for numerical variables and Hamming distance for categorical variables. And this last term is controlled by λ. Then, these are the main steps in the function that we modified using the R-package “Survey”. The first step was to adapt the definition of parameter λ in Szepannek [30] to the weighted context, that is, the calculus of the variance for the numeric and categorical variables used to compute λ. In the case of the numeric variable instead of computing the variance using the function var, we used the function svyvar that return the population variance, then as in the unweighted version we compute the mean of the variance vector. For the case of categorical variables, we had to compute the frequency table for each categorical variable using the function svytable but using a for loop that allows the program to loop through each of the variables. As with the numerical variables, we calculated the mean of the vector variance. The final step was to keep both results and divided them to obtain λ as in the unweighted case. The first prototypes and the measure between them are computed in the same way as the original function. We changed the way in which the function computes the new prototypes, in the case of numerical variable we compute the population mean using the svymean. For the categorical case, we create a function that returns the levels with the highest frequency in each cluster. It is important to mention that in this case we have to use a subset of the initial survey design because we have now subsamples (clusters) that have not been defined previously, so the computation that we are doing only works for a part of the survey. The same idea is used for the final update of the prototypes. Also, we use the survey package to compute the size of each cluster, the sum over all within clusters distance (withinss) and the sum over all clusters (tot.withinss). The dataset and the R code used to obtain the results of this study are located on the next website: https://davidmerchan04.github.io/Impact-of-pandemic-on-European-well- being/ (accessed on 20 September 2020). In this github project, you will find: • SHARE_W7.zip: File with the Dataset used to the TFM Final code R: R file with to code used to compute all the results of the TFM. Int. J. Environ. Res. Public Health 2021, 18, 4620 25 of 26

References 1. Eurostat. Population by Age Group. Available online: https://ec.europa.eu/eurostat/databrowser/view/TPS00010/default/ table (accessed on 1 June 2020). 2. World Health Organization. World Report on Ageing and Health; World Health Organizations: Geneva, Switzerland, 2015. 3. Johns Hopkins Medicine. Health 2020. Available online: https://www.hopkinsmedicine.org/health/conditions-and-diseases/ coronavirus/coronavirus-caregiving-for-the-elderly (accessed on 1 June 2020). 4. Centers for Disease Control and Prevention. Your Health. Available online: https://www.cdc.gov/coronavirus/2019-ncov/need- extra-precautions/older-adults.html#:~{}:text=Among%20adults%2C%20the%20risk%20for,or%20they%20may%20even%20 die (accessed on 16 August 2020). 5. Mueller, A.L.; McNamara, M.S.; Sinclair, D.A. Why does COVID-19 disproportionately affect older people? Aging 2020, 12, 9959–9981. [CrossRef][PubMed] 6. World Health Organization. Coronavirus Disease 2019 (COVID-19). Situation Report; WHO: Geneva, Switzerland, 2020. 7. Richardson, S.J.; Carroll, C.B.; Close, J.; Gordon, A.L.; O’Brien, J.; Quinn, T.J.; Rochester, L.; Sayer, A.; Shenkin, S.D.; Van der Velde, N.; et al. Research with older people in a world with COVID-19: Identification of current and future priorities, challenges and opportunities. Age Ageing 2020, 49, 901–906. [CrossRef][PubMed] 8. Onder, G.; Rezza, G.; Brusaferro, S. Case-fatality rate and characteristics of patients dying in relation to COVID-19 in Italy. JAMA 2020, 323, 1775–1776. [CrossRef][PubMed] 9. World Health Organisation. World Health Organisation COVID-19 Strategy Update; World Health Organisation: Geneva, Switzer- land, 2020. Available online: https://www.who.int/publications-detail/covid-19 (accessed on 26 March 2021). 10. Centers for Disease Control and Prevention. Severe Outcomes among Patients with Coronavirus Disease 2019 (COVID-19); Morbidity and Mortality Weekly Report; Centers for Disease Control and Prevention: Atlanta, GA, USA, 2020; Volume 69, pp. 343–346. 11. Van Dorn, A.; Cooney, R.E.; Sabin, M.L. COVID-19 exacerbating inequalities in the US. Lancet 2020, 395, 1243–1244. [CrossRef] 12. Li, Y.; Mutchler, J.E. Older Adults and the Economic Impact of the COVID-19 Pandemic. J. Aging Soc. Policy 2020, 32, 477–487. [CrossRef][PubMed] 13. Wyper, G.M.A.; Assunção, R.; Cuschieri, S.; Devleesschauwer, B.; Fletcher, E.; Haagsma, J.A.; Hilderink, H.B.M.; Idavain, J.; Lesnik, T.; Von Der Lippe, E.; et al. Population vulnerability to COVID-19 in Europe: A burden of disease analysis. Arch. Public Health 2020, 78, 1–8. [CrossRef][PubMed] 14. Bueno-Notivol, J.; Gracia-García, P.; Olaya, B.; Lasheras, I.; López-Antón, R.; Santabárbara, J. Prevalence of depression during the COVID-19 outbreak: A meta-analysis of community-based studies. Int. J. Clin. Health. Psychol. 2021, 21, 100196. [CrossRef] [PubMed] 15. Li, L.Z.; Wang, S. Prevalence and predictors of general psychiatric disorders and loneliness during COVID-19 in the United Kingdom. Psychiatry Res. 2020, 291, 113267. [CrossRef][PubMed] 16. Qiu, J.; Shen, B.; Zhao, M.; Wang, Z.; Xie, B.; Xu, Y. A nationwide survey of psychological distress among Chinese people in the COVID-19 epidemic: Implications and policy recommendations. Gen. Psychiatry 2020, 33, e100213. [CrossRef][PubMed] 17. Huang, Y.; Zhao, N. Generalized anxiety disorder, depressive symptoms and sleep quality during COVID-19 outbreak in China: A web-based cross-sectional survey. Psychiatry Res. 2020, 288, 112954. [CrossRef] 18. Huremović,D. Social distancing, quarantine, and isolation. In Psychiatry of Pandemics; Springer: Berlin/Heidelberg, Germany, 2019; pp. 85–94. 19. Pfefferbaum, B.; North, C.S. Mental Health and the Covid-19 Pandemic. N. Engl. J. Med. 2020, 383, 510–512. [CrossRef] 20. SHARE. Release Guide 7.1.0; MEA: Beirut, Lebanon, 2020. 21. Lumey, T. Analysis of Complex Survey Samples (survey) Version 4.0.; The Comprehensive R Archive Network (CRAN). 2020. Available online: http://r-survey.r-forge.r-project.org/survey/ (accessed on 1 June 2020). 22. Grané, A.; Albarrán, I.; Lumley, R. Visualizing inequality in health and socioeconomic wellbeing in the EU: Findings from the share survey. Int. J. Environ. Res. Public Health 2020, 17, 7747. [CrossRef][PubMed] 23. Hagenaars, A.J.M.; De Vos, K.; Zaidi, M.A. Poverty Statistics in the Late 1980s: Research Based on Micro-Data; Office for Official Publications of the European Communities: Luxembourg, 1994. 24. Pettinicchi, Y.; Börsch-Supan, A. The economic situation of fomerly self-employed workers. In Health and Socieconomic Status over the Life Course: First Results From SHARE Waves 6 and 7; De Gruyter: Berlin, Germany, 2019; pp. 120–127. 25. Costa-Font, J.; Jimenez-Martin, S.; Vilaplana, C. Does Long-Term Care Subsidisation Reduce Unnecessary Hospitalisations; CESifo Working Paper Series; CESifo: Munich, Germany, 2017. 26. Eurostat. Health Care Expenditure by Provider. Available online: https://ec.europa.eu/eurostat/statistics-explained/index. php/Healthcare_expenditure_statistics#Healthcare_expenditure_by_provider (accessed on 1 September 2020). 27. Singh, A.; Yadav, A.; Rana, A. K-means with three different distance metrics. Int. J. Comput. Appl. 2013, 67, 13–17. [CrossRef] 28. Izenman, A.J. Modern Multivariate Statistical Techniques; Springer Science and Business Media LLC.: Berlin/Heidelberg, Germany, 2008. 29. Huang, Z. Extensions to the k-Means algorithm for clustering large data sets with categorical values. Data Min. Knowl. Discov. 1998, 2, 283–304. [CrossRef] 30. Szepannek, G. Clustmixtype: User-friendly clustering of mixed-type data in R. R J. 2019, 10, 200. [CrossRef] Int. J. Environ. Res. Public Health 2021, 18, 4620 26 of 26

31. Centers for Disease Control and Prevention. Your Health. Available online: https://www.cdc.gov/coronavirus/2019-ncov/ need-extra-precautions/people-at-increased-risk.html (accessed on 25 June 2020). 32. WHO. Information Note on COVID-19 and NCDs; WHO: Geneva, Switzerland, 2020. 33. Jones, L.; Palumbo, D.; Brown, D. BBC. Available online: https://www.bbc.com/news/business-51706225 (accessed on 30 June 2020). 34. Pieh, C.; Budimir, S.; Probst, T. The effect of age, gender, income, work, and physical activity on mental health during coronavirus disease (COVID-19) lockdown in Austria. J. Psychosom. Res. 2020, 136, 110186. [CrossRef] 35. Javed, B.; Sarwer, A.; Soto, E.; Mashwani, Z.-U.-R. Impact of SARS-CoV-2 (Coronavirus) Pandemic on Public Mental Health. Front. Public Ment. Health 2020, 8, 1–4. [CrossRef] 36. Schoroeder, M. Jobless Now, Sick Later? Investigating the Long-term Consequences of Involunatary Job Loss on Health. Adv. Life Course Res. 2013, 18, 5–15. [CrossRef][PubMed] 37. European Commission. Equal Pay? Time to Close the Gap; European Commission: Brussels, Belgium, 2019. 38. Van Oyen, H.; Nusselder, W.; Jagger, C.; Kolip, P.; Cambois, E. Gender differences in healthy life years within the EU: An exploration of the health-survival paradox. Int. J. Public Health 2013, 58, 143–155. [CrossRef] 39. Holt-Lunstad, J.; Smith, T.B.; Baker, M.; Harris, T.; Stephenson, D. Loneliness and social isolation as risk factors for mortality: A meta-analytic review. Perspect. Psychol. Sci. 2015, 10, 227–237. [CrossRef][PubMed] 40.N úñez, I.; Livanos, I. Higher education and unemployment in Europe: An analysis of the academic subject and national effects. High. Educ. 2009, 59, 475–487. [CrossRef] 41. Rodríguez-Pose, A.; Tselios, V. Education and income inequality in the regions of the european union. J. Reg. Sci. 2009, 49, 411–437. [CrossRef] 42. Zimmerman, E.; Woolf, S.H. Virginia commonwealth university understanding the relationship between education and health. NAM Perspect. 2014, 4.[CrossRef] 43. Killgore, W.D.; Cloonan, S.A.; Taylor, E.C.; Dailey, N.S. Loneliness: A signature mental health concern in the era of COVID-19. Psychiatry Res. 2020, 290, 113117. [CrossRef] 44. Steptoe, A.; Shankar, A.; Demakakos, P.; Wardle, J. Social isolation, loneliness, and all-cause mortality in older men and women. Proc. Natl. Acad. Sci. USA 2013, 110, 5797–5801. [CrossRef] 45. Marti, L.; Puertas, R. European Countries’ Vulnerability to COVID-19: Multicriteria Decision-Making Techniques. Econ. Res. Ekon. Istraživanja 2021.[CrossRef]