Understanding Economic and Health Factors Impacting the ...€¦ · 10/04/2020 · as various economic factors that impact COVID -19 Diagnosis, Reporting, and Treatment. Based on

Original Paper AleksandrFarseeva,c,d,Yu-YiChu-Farseevac,QiYanga,DaronBenjaminLoobaITMO University bNational University of Singapore cSoMin.ai Researchdcorrespondingauthor.Email:[email protected]:+(65)912-270-09Address:KronverkskiyProspekt,49,StPetersburg,Russia,197101

Understanding Economic and Health Factors Impacting the Spread of COVID-19 Disease

Abstract Background:The rapid spread of the Coronavirus 2019 disease (COVID-19) haddrastically impacted life all over the world. While some economies are activelyrecoveringfromthispestilence,othersareexperiencingfastandconsistentdiseasespread,compellinggovernmentstoimposesocialdistancingmeasuresthathaveputahaltonroutines,especiallyindenselypopulatedareas.Objective: Aiming at bringingmore light onkey economic and population healthfactorsaffectingthediseasespread,thisinitialstudyutilizesaquantitativestatisticalanalysisbasedonthemostrecentpubliclyavailableCOVID-19datasets.Methods:WehaveappliedPearsonCorrelationAnalysisandClusteringAnalysis(X-MeansClustering)techniquesonthedataobtainedbycombiningmultipledatasetsrelated to country economics, medical system & health, and COVID-19 - relatedstatistics. The resulting dataset consisted of COVID-19 Case andMortality Rates,Economic Statistics, and Population Public Health Statistics for 165 countriesreportedbetween22January2020and28March2020.Thecorrelationanalysiswasconductedwiththesignificancelevelαof0.05.TheclusteringanalysiswasguidedbythevalueofBayesianInformationCriterion(BIC)withthebinvalueb=1.0andthecutofffactorc=0.5,andhaveprovidedastablesplitintofourcountry-levelclusters.Results:Thestudyshowedandexplainedmultiplesignificantrelationshipsbetweenthe COVID-19 data and other country-level statistics. We also identified andstatistically profiled four major country-level clusters with relation to differentaspectsofCOVID-19developmentandcountry-leveleconomicandhealthindicators.Specifically,thisstudyidentifiedpotentialCOVID-19under-reportingtraits,aswellas various economic factors that impact COVID-19 Diagnosis, Reporting, andTreatment. Based on the country clusters, we also described the four diseasedevelopment scenarios, which are tightly knit to country-level economic andpopulationhealthfactors.Finally,wehighlightedthepotentiallimitationofreportingand measuring COVID-19 and provided recommendations on further in-depthquantitativeresearch.

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

The copyright holder for this preprint this version posted June 9, 2020. .https://doi.org/10.1101/2020.04.10.20058222doi: medRxiv preprint

https://doi.org/10.1101/2020.04.10.20058222

http://creativecommons.org/licenses/by-nc/4.0/

Conclusions:Inthisstudy,wefirstidentifiedpossibleCOVID-19reportingissuesandbiasesacrossdifferent countriesandregions.Second,we identified crucial factorsaffectingthespeedofCOVID-19diseasespreadandprovidedrecommendationsonchoosingandoperatingeconomicandhealthsystemfactorswhenanalyzingCOVID-19progression.Particularly,wediscoveredthatthepoliticalsystemandcompliancewithinternationaldiseasecontrolnormsarecrucialforeffectiveCOVID-19pandemiccessation.However,theroleofsomewidely-adoptedmeasures,suchasGHSHealthIndex,mighthavebeenoverestimatedinlieuofmultiplebiasesandunderreportingchallenges.Third,webenchmarkedourfindingsagainstthewidely-adoptedGlobalHealth Security (GHS)model and found that the lattermight be redundantwhenmeasuringandforecastingCOVID-19spread,whileitsindividualcomponentscouldpotentially serve as stronger COVID-19 indicators. Fourth, we discovered fourclusters of countries characterized by different COVID-19 development scenarios,highlighting the differences of the disease reporting and progression in differenteconomic and health system settings. Finally, we provided recommendations onsophisticatedmeasures and research approaches to be implemented for effectiveoutbreakmeasurements,evaluationandforecasting.Wehavesupportedthe latterrecommendationsbyapreliminary regressionanalysisbasedon theour-collecteddataset.We believe that ourworkwould encourage further in-depth quantitativeresearch along the direction as well as would be of support to public policydevelopmentwhenaddressingtheCOVID-19crisisworldwide.Keywords:COVID-19;EconomicFactors;HealthFactors

Introduction The rapid spread of COVID-19 has drastically impacted economies around the

world.On11March2020thediseasewasofficiallyclassifiedasapandemicand,asreported on 24 March 2020, it has infected 440,093, and causing 19,748 deathsworldwide,withthehighestnewcaseintensitiesintheUSA,Spain,Germany,France,Switzerland,SouthKorea,UnitedKingdom(UK),andHubeiProvinceinChina.Inresponsetosuchavolatilesituationintheworld,governmentsandthescientific

communities have been actively studying the underlying principles and possiblereasonsforthediseasespreadandprogression.Forexample,Baiet.al.[1]havefirstdiscovered that COVID-19 could have been possibly transmitted by asymptomaticcarriers,whileWuet.al.[2]conductedalarge-scalestudybasedon72,314confirmedcaseslistingimportantactionablelessonsforothersocietiestoapply.Finally,Martinetal.[42]studiedtheimportanceofsocialdistancingmeasuresbeingappliedtoslowthespreadspeedofCOVID-19spreadpacereduction.Furthermore,theComputerSciencecommunityhasanalyzedthediseasespread

fromastatisticalpointofview.Specifically,in[3],theauthorswitnessedapotentialassociationbetweenCOVID-19mortalityratesandhealth-careresourceavailability,while Chen et.al. [4] discovered a strong statistical relationship between initialemigrationfromWuhanCityandtheinfectionspreadtoothercitiesinChina.Finally,Chinazziet.al.[5]suggestedthattravelrestrictionstoCOVID-19affectedareascouldbe not as effective, as many infected individuals “...have been travelling



https://doi.org/10.1101/2020.04.10.20058222


internationallywithoutbeingdetected...”andassuch,sharperrestrictivemeasuresarenecessarytocurbandtakecontroloftheoutbreak.Eventhoughsignificanteffortshavebeenmadetowardsaproperunderstanding

oftheCOVID-19outbreakfrommultipleperspectives,duetotheconstantlyevolvingpandemic, emerging new information and data sets, and inaccessibility of publiclarge-scale data, literature based on quantitative research on the outbreak is stillrelativelysparse.Inthestudy,wehypothesisethatthespeedofspreadofCOVID-19diseaseistiedtothevariouseconomicandhealthfactors,that,inturn,formacountryprofileaswellasreflectthecountry'sreadinesstoconcurtheCOVID-19pandemic.Tothebestofourknowledge,itisoneofthefirstattemptstobuildamoreholisticview on the COVID-19 development, which hopes to identify and explainrelationshipsbetweenthediseasespreadandvariouseconomicandhealth factorsthroughquantitativeanalysis.

Methods

Dataset Inthisstudy,wehave incorporatedthe“COVID19GlobalForecasting(Week2)”

dataset[6]thatwasreleasedbytheKaggle[33]platform.Thedatasetincludesdailyupdates of the COVID-19 confirmed cases and mortality rates for 173 countriesreported by WHO between 22 January 2020 and 28 March 2020. To study therelationshipsbetweenCOVID-19spreadandvariouseconomic factors,wemergedthe original dataset with “Country Statistics - UNData” dataset [7], “Pollution byCountryforCOVID19Analysis”dataset[8],and“TheWorldBank(Demographics)”dataset by cross-matching country names across data sets. We also merged theoriginaldatasetwith thedatasetobtainedby parsing the “WorldLifeExpectancy”database[9]websiteforobtaininginformationondeathratesfromdifferentchronicdiseases across the world. The selected data indicators were chosen as the keyeconomicandhealth indicatorsavailable inpublic accessaimingatprovisionof amore holistic view into different country profiles with respect to the COVID-19pandemicdevelopment.Specifically,theKaggleplatformisknowntobeoneofthelargestdataintegratorsintheworld,wheretheresearchcommunitycouldsourcethemostrecentandcomprehensivereal-timedataonthelastworld-levelproblems.Atthesametime,theWorldlifeExpectancydatasetsareknowntobethelargestglobalhealth and life expectancy databases, allowing for comprehensive statistics aboutvarioushealth factorswith respect todifferent countries.Finally, theUNdatadataservice was chosen as one of the largest aggregation services for the statisticaldatabasesrelatedtocountryeconomics,providinguswiththeeconomicsstatisticsformostofthecountriesincludedinthesourceKaggledataset.To support this study observation and evaluate our selected indicators against

morewidely-adoptedevaluationsystem,wehavealsoenrichedthedatasetbythedataprovidedinGlobalHealthSecurityIndex(GHS)database,whichwasisclaimedtobe“...thefirstcomprehensiveassessmentofglobalhealthsecuritycapabilitiesin195 countries” [41] and, therefore, could serve as an assessmentmedium for this



https://doi.org/10.1101/2020.04.10.20058222


study.Except for the actual GHS classification, the database also provides indexesmeasuringvariousaspectsofthehealthsystems:● PreventionIndex-Preventionoftheemergenceorreleaseofpathogens;● DetectionandReportingIndex-Earlydetectionandreportingforepidemics

ofpotentialinternationalconcern;● RapidResponseIndex-Rapidresponsetoandmitigationofthespreadofan

epidemic;● HealthSystemIndex-Sufficientandrobusthealthsystemtotreatthesick

andprotecthealthworkers;● CompliancewithInternationalNormsIndex-Commitmentstoimproving

national capacity, financing plans to address gaps, and adhering to globalnorms;

● Risk Environment Index - Overall risk environment and countryvulnerabilitytobiologicalthreats.

Table1:DetailedStatisticsoftheCOVID-19CombinedDatasetDataIndicatorGroup NumberCountries 165Regions 286ChronicDiseaseDeathRateStatistics 32AgeDemographicGroups 4PollutionIndicators 3OtherEconomicFactorStatistics 50GlobalHealthSecurityIndex(GHS)Indicators 14COVID-19RelatedIndicators 2COVID-19ConfirmedCaseSpeedDailyReports 67COVID-19FatalitiesSpeedDailyReports 67After themerging process, the resulting dataset consists of COVID-19Case and

Mortality Rates, Economic Statistics, and Population Health Statistics for 165countriesreportedbetween22January2020and28March2020.Theactualnumberof thedata records in thedataset is286, as therewereCOVID-19statistics in theoriginaldatasetgivenfordifferentregionswithinthesamecountry:54regionsintheUnitedStatesofAmerica,33regions inChina,10regions inCanada,10regions inFrance, 8 regions in Australia, 7 regions in the United Kingdom, 4 regions in theNetherlands, and3 regions inDenmark. Seven countries, namelyBahamas, CongoBrazzaville,CongoKinshasa,Eswatini,Gambia,Taiwan,andVietnamwereexcludedastherewerenoeconomicsandmedicalsystem&healthstatisticsavailablefortheminthemergeddatasets.AmoredetailedstatisticsoftheresultingdatasetareprovidedinTable1.Wehavealsoreleasedthedatasetforpublicuse[10].

Experimental Setup Asmentioned,theprimaryobjectiveofthisresearchistostudytherelationship

betweenthespeedofthediseasespreadandvariouseconomicandhealthfactors.Consideringtheunevenpaceofthedisease’sgeographicalspreadduetoCOVID-19’s



https://doi.org/10.1101/2020.04.10.20058222


long incubation period [11], naturalmigration laws [5] and various government-imposedtravelpolicies[5],itisnotfeasibletodrawtheanalysisbasedontheactualdailyregisteredcaseandfatalityratesavailableintheoriginalKaggledataset[6],butrathernecessarytoperformanadditionaldatapre-processingaimingatestablishingholistic data characteristics reflecting the general worldwide COVID-19 spreadtendencies. Keeping this in mind, we have performed the following data pre-processingsteps:Dataset Combination: Original Dataset [6] was joined by performing Countrymatching to fourauxiliarydata sets [7,8,12,9] asdescribed in thenext sections.Fifteen country names have been replaced with the naming notation used in theoriginaldatasettoperformthesuccessfulmatching.NormalizedDailySpreadSpeedEstimation:Inthisstudy,weanalyzedthelasttwoweeks of reported data from14March 2020 till 28March 2020. To estimate thedisease spread speed of each day in the two-week interval, we subtracted thereportednumberofnewcasesandfatalitiesonthepreviousdayfromthenumberofthecurrentdayandthendividedthisnumbertotheMedianreportednumberduringthepasttwoweeks.

𝑆𝑝𝑒𝑒𝑑(𝑑𝑎𝑦1) =𝑅𝑒𝑝𝑜𝑟𝑡𝑒𝑑(𝑑𝑎𝑦1) − 𝑅𝑒𝑝𝑜𝑟𝑡𝑒𝑑(𝑑𝑎𝑦19:)

𝑀𝑒𝑎𝑛(𝑅𝑒𝑝𝑜𝑟𝑡𝑒𝑑(𝑑𝑎𝑦:)…𝑅𝑒𝑝𝑜𝑟𝑡𝑒𝑑(𝑑𝑎𝑦:>)), 𝑘 = 1…14

,whereReported() is thenumberofConfirmedCasesorFatalitiesreported in thedataset,Mean() is theArithmeticMeanof itsarguments.TheabovenormalizationproceduremitigatestheproblemofunevenspeedofspreadofCOVID-19indifferentgeographicalregionsasittreatseachcountryaccordingtoitsoutbreak“stage”andmakescountrystatisticscomparabletoeachother.Sparse Data Indicator Filtering: As some of the data indicators in the mergeddatasetwerefoundtocontainalargenumberofmissingvalues,whichmightaffectfurther analysis, we excluded data indicators that contained more than 35% ofmissing values. After the sparse data indicator exclusion, the resulting datasetcontained130dataindicators.

Correlation Analysis To determine the relationship between the COVID-19 Spread Speed and other

indicators, we applied Pearson’s product-moment Correlation [13] to 286 datasamples(thecountriesandregionsinthecombineddataset)and130dataindicators(whosedataindicatorsthathaveremainedaftertheSparseDataIndicatorFilteringstep). We then filtered out all non-significant correlation values (α = 0.05) andpresented the obtained results in the form of a correlation semi-matrix forvisualizationpurposes.WelistallthevariablesthathavebeenincorporatedintothecorrelationanalysisinMultimediaAppendix1.TheMin,Max,Mean,Std.Deviation,Variance,Skewness,Kurtosis,andOverallSumstatisticsareprovidedinMultimediaAppendix2.

Cluster Analysis

By adopting correlation analysis, we have determinedmultiple economic andmedicalsystem&healthfactorsthatarestronglyandsignificantlycorrelatedtothe



https://doi.org/10.1101/2020.04.10.20058222


COVID-19 disease spread. We have also witnessed various governments andpopulationbehaviouraltraitspossiblyexplainingthedifferentscenariosofCOVID-19developmentaroundtheworld.

Even though these findings bring more light into the approaches thatgovernmentshave adopted tomitigate the crisis, it is stillunclearwhat the exactdifferencesareintheseapproaches,aswellasinthecountryprofilesaffectedbytheCOVID-19diseasespread.

Aimingatansweringthisquestion,wehavefurtheradoptedaClusteringAnalysistechnique[27] tostudythegroupsofcountries inourdatasetbyseparatingthembasedontheeconomicandmedicalsystem&healthfactors.Ingeneral,clusteringanalysisisawidely-adoptedunsupervisedmachinelearningtechniqueallowingforautomatic discovery of the groups of population samples in a multi-dimensionalspaceofvariables.Suchgroupscouldbethenusedforanin-depthunderstandingoftheworldwidetraitsrelatedtoCOVID-19development,specificallywithrelationtovariouseconomicandhealthsystemfactors.Inthisstudy,wehaveadoptedthe“X-Means”clusteringalgorithm,thathasbeenreportedtobeeffectiveindeterminingthenumberofclustersinthedatasetwithoutnecessarilyhavingapriorassumptiononthenumberofclusters[28].TheX-Meansclusteringwasappliedwiththebinvalueb=1.0andthecutofffactorc=0.5,whichwerefoundempiricallytoproveastablesplitofthedataintoclusters.AimingatavoidingapotentialbiasinclusteringresultsthatcouldbeintroducedbyvarioushumandecisionsmadewhencreatingGHSindex,theclusteringhasbeenperformedinareducedspaceexcludingallGHS-relatedvariables.Insuchaway,weareguaranteedthatGHShasnotimpactedtheclusteringresults,andthereforeGHScouldbeusedasabenchmarkwhenevaluatingtheresultsoftheanalysis.BeingguidedbythevalueoftheBayesianInformationCriterion(BIC),theX-Means clustering algorithm have determined the following four country-levelclusters:

Country Cluster 1: Afghanistan, Angola, Bangladesh, Benin, Bhutan, BurkinaFaso,Cambodia,Cameroon,CentralAfricanRepublic,Chad,CôteD’Ivoire,Djibouti,Equatorial Guinea, Eritrea, Ethiopia, Gabon, Ghana, Guinea, Guinea-Bissau, Haiti,India, Indonesia, Kenya, Liberia, Madagascar, Mali, Mauritania, Mozambique,Namibia,Nepal,Niger,Nigeria,Pakistan,Philippines,Rwanda,Senegal,Somalia,SriLanka,Sudan,Tanzania,Timor-Leste,Togo,Uganda,Zambia,Zimbabwe.

CountryCluster2:Albania,Algeria,Andorra,AntiguaAndBarbuda,Argentina,Armenia, Azerbaijan, Bahrain, Barbados, Belarus, Belize, Bolivia, Bosnia AndHerzegovina,Brazil,Brunei,Bulgaria,CaboVerde,Chile,Colombia,CostaRica,Cuba,Cyprus, Dominica, Dominican Republic, Ecuador, Egypt, El Salvador, Fiji, Georgia,Grenada, Guatemala, Guyana, Holy See, Honduras, Iran, Iraq, Jamaica, Jordan,Kazakhstan, SouthKorea,Kuwait,Kyrgyzstan,Laos,Lebanon,Libya,Liechtenstein,Malaysia, Maldives, Mauritius, Mexico, Moldova, Mongolia, Montenegro, Morocco,Nicaragua,Oman,Panama,PapuaNewGuinea,Paraguay,Peru,Qatar,Romania,SaintKittsAndNevis, SaintLucia, SaintVincentandTheGrenadines, SanMarino, SaudiArabia, Serbia, Seychelles, Singapore, South Africa, Suriname, Syria, Thailand,TrinidadAndTobago,Tunisia,Turkey,Ukraine,UnitedArabEmirates,Uzbekistan,Venezuela;



https://doi.org/10.1101/2020.04.10.20058222


CountryCluster3:Australia(8regions),Austria,Belgium,Canada(10regions),Croatia,Czechia,Denmark,Denmark,Denmark,Estonia,Finland,France(10regions),Germany, Greece, Hungary, Iceland, Ireland, Israel, Italy, Japan, Latvia, Lithuania,Luxembourg, Malta, Monaco, Netherlands (4 regions), New Zealand, NorthMacedonia, Norway, Poland, Portugal, Russia, Slovakia, Slovenia, Spain, Sweden,Switzerland,UnitedKingdom(7regions),Uruguay,US(54regions).

CountryCluster4:China(33regions).Wethentreatedeachclusterassignmentasindependentvariablesandapplied

correlationanalysistouncoverthestatisticalprofilesforeachoftheclusters.

Results

Correlation Analysis Correlation VisualizationMultimediaAppendix3:PearsonProduct-MomentCorrelationBetweenCOVID-19SpreadSpeed,Economics,andHealthFactors.



https://doi.org/10.1101/2020.04.10.20058222


Forvisualizationpurposes,wepresenttheobtainedcorrelationvaluesintheformofacorrelationsemi-matrix(seeMultimediaAppendix3).IntheFigure,whitecirclesdenote a positive correlation, while the black circlesmean that the correlation isnegative.Thesizeofthecircleisproportionaltothecorrelationstrength(thelargerthecircle-thestrongerthecorrelation)andtheabsenceofacircleinacellmeansthattherewasnocorrelationfoundorthecorrelationisnotsignificant.

Correlation Analysis of COVID-19 Confirmed Cases and Fatalities Fromthefirst28linesofthecorrelationsemi-matrix,itcanbeseenthatthereareseveral strong correlations between individual COVID-19 reported statistics. Forexample, it can be seen that there is a strongnegative correlation between theFatality SpeedonMarch 15 (Sunday) andMarch 16 (Monday) aswell as thestrongpositivecorrelationbetweenFatalitySpeedonMarch15(Sunday)andMarch17(Tuesday).Atthesametime,asignificantpositivecorrelationwasalsofound between several subsequent Confirmed Case Speed dates, such as 19(Thursday), 20 (Friday), and 21 (Saturday) March 2020 and 26 (Thursday), 27(Friday),and28(Saturday)March2020.Despite several single negative or positive correlations mentioned above, one

might not find many significant correlations either between consequentmeasurementsofthesamemetric(i.e.ConfirmedCaseSpeedonDifferentDays)orbetweendifferentCOVID-19metrics(i.e.ConfirmedCasevs.Fatalities).

Correlation Analysis of Chronic Diseases and Health Factors Fromthelast39ChronicDiseaseandHealthFactorindicatorsinthelowerpart

ofthecorrelationsemi-matrix,itcanbeseenthattherearemultiplesignificantandconsistent correlations of Chronic Disease Rates with Confirmed Case Speedmeasurements.Forexample,suchindicatorsasSkinCancer(91.7%5-yearsurvivalrate[14]),ProstateCancer(98.6%5-yearsurvivalrate[14]),OvaryCancer(46.5%5-year survival rate [14]),Breast Cancer (89.7% 5-year survival rate [14]), andBladderCancer (77.3%5-year survival rate [14])DeathRateswere found to besignificantlypositivelycorrelatedwiththeCOVID-19SpreadSpeed.Inadditiontotheabovefindings,wewouldalsoliketohighlightthestrongpositivecorrelationof Obesity Rates (especially for Female demographics) to COVID-19 SpreadSpeed.Furthermore,onecanfindthatsuchvariablesasSkinDiseaseDeathRate,Influenza and Pneumonia Death Rate,Diabetes Death Rate,Dementia DeathRate, Alcohol Death Rate are also significantly correlated to COVID-19 SpreadSpeed.WewouldliketoalsohighlightthesignificantpositivecorrelationsofmostoftheabovefactorstoGHSNormandGHSHealthIndexes.

Correlation Analysis of Economic and Other Factors Fromthecorrelationsemi-matrix,itcanbeseenthatsixeconomicattributesare

stronglypositivelycorrelatedtotheCOVID-19SpreadSpeed.First,wefoundastrongpositive correlation between the Number of Health Physicians Per 1000Population,HealthTotalExpense(%ofGDP),andGDPPerCapita(inUSD)totheCOVID-19 disease spread speed. Second, it can also be seen that InternationalMigrantStockPer1000PopulationandInternationalMigrantStock%ofTotal



https://doi.org/10.1101/2020.04.10.20058222


PopulationvariablesarealsostronglyandpositivelycorrelatedtoCOVID-19SpreadSpeedduringthesecondweekof theobservedperiod(22March2020-28March2020).Third,onecannoticeastrongpositivecorrelationoftheServicesandOtherActivities % of Gross Value Added (GVA)measurement to COVID-19 diseasespread.Finally,itcanbeseenthattheGHSDetectIndexvariableisstronglypositivelycorrelatedtoatleastfivemeasurementsofCOVID-19spreadandtwomeasurementsofCOVID-19mortalitytowardstheendoftheobserved14-dayinterval.Atthesametime, it can alsobeobserved that theGHSDemocracy Index is stronglypositivelycorrelated to most of the COVID-19 spread and multiple COVID-19 mortalitymeasures.

Negative and Insignificant Correlations, Mortality Rate Speed AninterestingobservationisthestrongnegativecorrelationofthePopulation

inThousands2017variabletotheCOVID-19SpreadSpeed.Atthesametime,onecan notice that thePopulation Density Per km2 in 2017 does not exhibit anysignificantcorrelationwithCOVID-19.Furthermore, thereadercouldobservethatsuch variables as CO2 Emission Estimates (Million Tonnes Per Capita) andForestedAreaRatioalsoexhibit a strongnegativecorrelation to theCOVID-19SpreadSpeed.Finally,theEstimatesmetricisstronglypositivelycorrelatedtotheLung Cancer Death Rate, which, in turn, is also negatively correlated to theCOVID19SpreadSpeed. Interestingly,sucharelationshipcouldnotbeobservedfromthedatawehave,except for theonlyonenegativecorrelationof theLungDiseaseDeathRatetotheCOVID19FatalitySpeedon23March2020.Moreover,we would like to highlight the International Trade Balance (Million USD) andBalance of Payment Current Account (Million USD) are strongly negativelycorrelatedtotheCOVID-19FatalitySpeedduringthesecondweekoftheobservedperiod(22March2020-28March2020).

Global Health Index (GHS) and Related Indexes Correlations First,fromthecorrelationsemi-matrix,itcanbeseenthatallbutone(GHSIndexRisk)GHSIndexesaresignificantlypositivelycorrelatedtoeachother.Second,it can also be noticed thatmost of theGHS Indexmeasurements are significantlycorrelated to COVID-19 disease spread speed when it comes to the end of theobserved14-day interval.Themost significantpositive correlations8and9,werefound for GHS Norms Index and GHS Democracy Index, respectively. The leastnumberofcorrelations(two)werefoundforGHSHealthIndex.

Cluster Analysis AsitwasoutlinedintheMethodssection,byapplyingX-Meansclusteringtothe

wholedataset,wehavedeterminedfourcountry-levelclustersandfurtherappliedcorrelationanalysistreatingeachclusterassignmentasanindependentvariable.Thecorrelations between each cluster and other economic and health factors arevisualizedinthecorrelationsemi-matrixinMultimediaAppendix3.



https://doi.org/10.1101/2020.04.10.20058222


Correlation Analysis of Country Cluster 1 From the correlation semi-matrix, it can be seen that the countries from the

Cluster1arepositivelycorrelatedtotheCOVID-19ConfirmedCaseSpeedon16,20,and23March2020,whilenegativelycorrelatedon25March2020.

Correlation Analysis of Country Cluster 2 WhenlookingatthecorrelationprofileoftheCountryCluster2,areadercould

immediatelynoticethattheclusterisnotassociatedwithanysignificantCOVID-19correlationsexceptforonepositivecorrelationwithCOVID-19SpreadSpeedon17March2020.Furthermore,onecanalsofindthatotherpositivecorrelationsofthecluster are arguably weak, having its spikes in population growth (see thesignificantpositivecorrelationofPopulationGrowthRateAverageAnnualPercentvariable tocountries inCluster2),Obesity (seesignificantpositivecorrelationsofObesityinFemalePopulationandObesityinMalePopulationvariablestocountriesinCluster2)andDiabetes(seesignificantpositivecorrelationsofDiabetesLevelandDiabetes Death Rate variables to countries in Cluster 2), various heart-relateddiseases(seesignificantpositivecorrelationsofCoronaryHeartDiseaseDeathRateandInflammatoryHeartDiseaseDeathRatevariablestocountriesinCluster2),andreproduction system cancers (see significant positive correlations of CervicalCancerDeathRateandProstateCancerDeathRatevariablestocountriesinCluster2).Wewouldliketonotethatfromthecorrelationsemi-matrixitcanbeseenthatthecountries from the Cluster 2 are strongly negatively correlated to all GHS Indexvariables,exceptfortheGHSRiskIndex.

Correlation Analysis of Country Cluster 3 CountryCluster3isthelargestandalsothemostdiverseclusterthatwehave

discoveredasitincludesmostoftheEuropeanCountriesandallstatesoftheUSthathaveexperiencedspikesinCOVID-19casesoverthepastseveralweeks(seemultiplesignificant positive correlationswith COVID-19 spread speed and fatalities on thecorrelationplot).

From the correlation semi-matrix it can be seen that the countries from theclusteraresignificantlypositivelycorrelated tomultiple factorsassociatedwithmodern developed economies, such as higher GDP Rate (see See significantpositivecorrelationofGDPPerCapitainUSDvariabletocountriesinCluster3),theinvolvement of the population in Services Industry (see significant positivecorrelations of Economy Services and Other Activities Percent of GVA andEmploymentServicesPercentofEmployedvariablestocountriesinCluster3),highRatioofUrbanPopulation(seesignificantpositivecorrelationofUrbanPopulationPercent of Total Population variable to countries in Cluster 3), Health SystemMaturity Level (see significant positive correlations of Health Total ExpenditurePercent of GDP, Health Physicians Per 1000 Population, Life Expectancy at BirthFemale,andLifeExpectancyatBirthMalevariables tocountries inCluster3),andsolid International Migrant Stocks (see significant positive correlations ofInternationalMigrantStockPer1000PopulationandInternationalMigrantStockOfTotalPopulationvariablestocountriesinCluster3).



https://doi.org/10.1101/2020.04.10.20058222


Atthesametime,itcanalsobeseenthatcountriesfromCluster3exhibitastrongpositive correlation with population ageing and its associated diseases (seesignificant positive correlations of Median Population Age, Population Ratio 64+YearsOld,andDementiaDeathRatevariablestocountriesinCluster3),varioustypesofcancers(seesignificantpositivecorrelationsofBladderCancerDeathRate,BreastCancer Death Rate, Colo-rectal Cancer Death Rate, Leukemia Death Rate, OvaryCancer Death Rate, Pancreatic Cancer Death Rate, and Skin Cancer Death RatevariablestocountriesinCluster3)and,correspondingly,urbanpopulation-linkedchronicdiseases(seesignificantpositivecorrelationsofObesityMaleandFemalePopulation,ObesityinFemalePopulation,ObesityinMalePopulation,DrugUseDeathRate, Skin Disease Death Rate, and Alcohol Death Rate variables to countries inCluster3).

Finally, from the correlation semi-matrix, it can be seen that countries fromCluster3arestronglypositivelycorrelatedtoalmostallGHSIndexvariables.

Correlation Analysis of Country Cluster 4 AsChinaistheonlycountryinCluster4anditseconomic,population,and,for

example, pollution statistics are commonly known, we will omit some stronglypositively correlated indicators in this work. Examples of such strong positivecorrelations are Lung Disease Death Rate, Stomach Cancer Death Rate,MalnutritionDeathRateRheumatic,andHeartDiseaseDeathRate,thatisalsostronglynegativelycorrelatedtotheCOVID-19SpreadSpeed.

Discussions Intheprevioussection,wediscoveredthatmultipleeconomicfactorsexhibita

strongrelationshiptothechronicdiseasesacrosstheglobe,and,therefore,itwouldbereasonabletohypothesisethattheycanbeutilizedtocharacterizetheprofilesofthese countrieswith relation to the economic development stage, and ultimately,COVID-19SpreadSpeed.

TogainfurtherinsightsintotherelationshipbetweensucheconomicfactorsandtheCOVID-19diseasespread,inthissectionwewilldiscussthepossiblereasonsforthediscoveredrelationships.

Correlation Analysis

Correlation Analysis of COVID-19 Confirmed Cases and Fatalities: ThenegativecorrelationbetweentheFatalitySpeedonSundayandMonday,aswellas thestrongpositive correlationbetweenFatalitySpeedonSundayandTuesday,could possibly suggest a reporting time-lag on weekends. At the same time, thediscoveredcorrelationsequencestowardstheendoftheweekcouldbeexplainedbythe testing capacity of the medical institutions entailing the situation when testresults fromthebeginningof theweekwerereceivedonlytowardstheendof theweek.Asthisstudydoesnotaimatadetailedanalysisofthelongitudinalpropertiesof the COVID-19 measurement and test procedures [36], we would only like tohighlighttheimportanceandtheinfluenceoftime-relatedmeasurementand



https://doi.org/10.1101/2020.04.10.20058222


testingarrangementaspectsaswellas torecommendfutureresearchalongthisdirection.The absence of significant correlations neither between consequent

measurements of the same disease spread metric (i.e. Confirmed Case Speed onDifferent Days) nor between different COVID-19 metrics (i.e. Confirmed Case vs.Fatalities) reveals that there is no strong linear relation between Confirmed CaseSpeedandFatalitySpeedwithinthe14day-intervalandinthespaceofindependentvariables that have been analyzed in this study. Therefore,we recommend theutilizationofadditionaldatasources,suchashospitalcapacity,testingvolume,internalgovernmentregulationsandpolicies,borderclosure,etc.,forgainingadeeperinsightintotheactualrelationshipbetweenCOVID-19InfectionandFatalitytrends.

Correlation Analysis of Chronic Diseases and Health Factors In order to explain the discovered significant positive correlations between

Chronic Disease Rates and Confirmed Case Speed, it is necessary to consider thefactorsrelatedtocountry-levelchronicdiseasedataindicators.

First, it was previously reported in the literature [17] that COVID-19

developmentandconsequencesmightbedirectlyrelatedtotheoverallhealthstatusofthepopulation,especiallywithregardstotheexistingpre-conditionsaffectingthehuman immune system. Such pre-conditions could further entail various fatalcomplications,suchasCytokineStorm,and,ultimately,affectthecountries’COVID-19ConfirmedCaseSpeedandFatalitySpeedstatistics.Inthisstudy,wecouldhavealsowitnessedsuch relationships inour results. Specifically, the correlationsemi-matrixrevealsseveralpotentialindicatorsofpoorerpopulationimmunityinalargeportionofanalyzedcountries,whichwasreflectedinourcorrelationanalysisresultsbythesignificantpositivecorrelationofe.g.SkinDiseaseDeathRate,InfluenzaandPneumoniaDeathRate,DiabetesDeathRate,DementiaDeathRate,andAlcoholDeathRatetoCOVID-19SpreadSpeed.

Second, a possible indicator of the health systemweaknesses could be found

amongsignificantpositivecorrelationsbetweenCOVID-19SpreadSpeedandvarioustypesofCancersDeathrates,especiallythosecancerswithanaveragehighersurvivalrates in the developed world. In particular, it is reasonable to assume that thecountries exhibiting higher death rates for such “high-survival cancers'' mightexperienceoveralldifficulties inproperandtimelypatient treatment.WhenfacingCOVID-19pandemic,suchcountriesmightnotbealwayswellpreparedforproperpatient isolation and treatment as well, which is essential for COVID-19 diseasespreadcontrol[15,37,38].Correspondingly,insuchcountries,theCOVID-19SpreadSpeedcouldbehigherentailingtheabove-reportedsignificantpositivecorrelation[16].

To summarize, in this section we have discovered two potential traits

affectingthespeedofCOVID-19diseasespread.Thefirstfindingsuggeststhatitcould



https://doi.org/10.1101/2020.04.10.20058222


be possible that the significant correlations between chronic diseases deathratesandCOVID-19SpreadSpeed,especiallythosediseasesthataretightlyknittothehumanimmunesystem,couldreflecttheoverallcountrymedicalsystem&population health status and, therefore, predisposition to infection andcomplication of COVID-19. Even though operationally, the more developedeconomiescould,arguably,respondfastertoCOVID-19outbreak,suchpopulationsmightbealsomoreaffectedbyvariousurban-livingfactors[18,19],that,inturn,could entail COVID-19 health predispositions and skewed disease spreadstatistics.Thesecondfindinghighlightsthepossiblecausalitybetweentheabilityofhealthsystemstocontroltherapidinfectionsdiseaseoutbreaksefficientlyandthespeed of COVID-19 spread. This conclusion is also supported by the negativecorrelations found, for example, between the GHSHealth System andComplianceIndex and the Mortality rates with such diseases as Influenza and Pneumonia.Asa finalnote,wewould like to state that in thisstudywearenotattempting tocomparevarioustypesofCancers,ChronicDiseases,andCOVID-19diseasedirectly,as they are known to be very different in terms of their cause, progression, anddetection/treatmentprinciples.Instead,weareaimingtotreatthemasindependentstatistical variables that, as have been shown above, could both reflect and,potentially,predicttheevolutionpaceoftheCOVID-19disease.

Correlation Analysis of Economic and Other Factors When it comes to explaining the statistical relationships between COVID-19

spread and various Factors describing Economic systems, it is reasonable to startfromthebasiccriteriarelatedtoeconomicstrength,especiallywithrelationtothebasichealthmetrics.Specifically,thestrongpositivecorrelationofHealthPhysicians,HealthTotalExpense,andGDPPerCapitavariablescouldbeattributedtothehigherabilityof the countrieswith strongerhealth systems inperforming timelypatientassessment, diagnosis, and disease reporting. In contrast, countries with weakly-subsidized health systems, many, especially asymptomatic [20], COVID-19 casescould remain unreported bringing the COVID-19 confirmed case statisticsdownandentailingtheinversecorrelationtraitsthatwehavediscoveredfromthedataset.ThefindingalignswellwiththesignificantcorrelationdiscoveredbetweenCOVID-19 disease spread and theGHS Detect Index. Particularly, the correlationssuggestthateconomieswithaveragehigherdiseasedetectionabilitymighthavereportedthehigherCOVID-19increaserates,asopposedtotothecountrieswherethe virus might have been spreading in the community but left undetected andunreported.

InthecasesofInternationalMigrantStockPer1000PopulationandInternationalMigrant Stock % of Total Population variables, the discovered strong positivecorrelations toCOVID-19SpreadSpeedcouldbepossiblyexplainedby thehigherrates of imported cases in countries with larger proportions of the migrantpopulation who often travel abroad or within the country [39] for business andpersonalpurposes.Interestingly,bothvariablesexhibitastrongpositivecorrelationduringthesecondweekoftheobservedperiod(22March2020-28March2020),which could be potentially explained by the travel restrictions imposed by the



https://doi.org/10.1101/2020.04.10.20058222


governmentsduringthatweek,resultinginthesituationwhenmanymigrantswererushingtoreturnbacktotheircountriesofresidencepriortoborderclosures[21,22].Finally,thestrongpositivecorrelationoftheServicesandOtherActivities%ofGrossValueAdded(GVA)measurementcanbeattributedtothemoreintensehumaninteractionratesincountrieswithlargerpopulationsinvolvedintheservicesectorofeconomics,makingtheriskofCOVID-19infectionhigher[23].

Summarizing theabove, again,wewould liketohighlight theimportanceofincorporation in theanalyticsof thedatarelated to theparticular lockdownpoliciesenforcedbythegovernmentsattemptingtocesstheCOVID-19spreadrapidly.Forexample,morecentralizedgovernments,suchasChina,SingaporeandVietnam,couldbeabletoimplementeffectivelockdownsfasterascomparedtolesscentralizedpoliticalsystems.Thisfacilitatestightercontrolandfasterrelieffromthedisease, while, possibly, temporarily sidestepping human rights and privacyconcerns, which reflect similar social concerns during the SARS outbreak [45].Opposingly,theEuropeanUnioncountriestookmuchlongertimetorespondtotheCOVID-19withthenecessarypopulationrestrictionmeasuresandthosemeasureswerefoundtobelesseffectiveintermsofpopulationcompliance[40].Thismightbeduetocountriesoperatingwithdifferentpoliticalgroups,whichhavebeenfoundtohinderaunifiedorequitableapproachwhendealingwithlarge-scalepublichealthconcerns[46].Suchrelationofthecountry'spoliticalsystemandthespeedofeffectiveresponse isalsosupportedby the findingsof this study. Specifically,from the correlation semi-matrix, it can be seen that theGHSdemocracy index iscorrelated to the majority of the COVID-19 spreadmeasures as well as to manyCOVID-19mortalitymeasuresduringtheobserved14-dayinterval.

Negative and Insignificant Correlations, Mortality Rate Speed WhenitcomestoexplainingtheNegativeandInsignificantCorrelationsaswell

as the correlations with Mortality Rate Speed, there are several interesting datarelationshipsthatcanbeobserved.Forexample,thestrongnegativecorrelationofthepopulationsizealongwiththeabsenceofasignificantcorrelationofpopulationdensitywith COVID-19 could be possibly explained by the inexistence of a linearrelationshipbetweenthesefactorsandCOVID-19SpreadSpeed.Asitwaspreviouslyreported [24] and was also observed in this study, the cultural and behaviouralfactors, such ashuman interaction habits, or government-regulated factors andmeasureenforcementviability,suchassocialdistancingenforcementmeasures,couldbeofamuchhigherinfluenceontheabilityofthecountrygovernmenttomanagetheCOVID-19diseaseoutbreak.

Furthermore,thestrongnegativecorrelationoftheForestedAreaRatiotothedisease spread speedcan behypothesizedby thenatural geographical sparsityofpopulationintroducedbytheforestedlandscapeandentailingalimitedinter-humaninteraction.

Atthesametime,thestrongnegativecorrelationoftheCO2Emissionvariablesrequiresadditionalclarifications.OnepossibleexplanationarisesbyalsotakingintoconsiderationthestrongpositivecorrelationofthemetrictotheLungCancerDeathRate, which, in turn, is also negatively correlated to the COVID19 Spread Speed.Precisely,takingintoconsiderationthatthetwometricsmightnotberelateddirectly



https://doi.org/10.1101/2020.04.10.20058222


tothediseasespreadspeed(asCOVID-19diseasegets“...transmittedbetweenpeoplethroughclosecontactanddroplets”[25]and,thus,moredependsontheinter-humanclosecontacts),itisthenreasonabletoassumethatthetwovariablescouldalsobepositivelycorrelatedtotheCOVID-19FatalitySpeedaswecouldexpectmorepatientswithlungpreconditionsinthecountrieswithmorepollutedenvironments.However,sucharelationshipcouldnotbeobservedfromthedataavailableinthisstudy,andthe only one negative correlationof the LungDiseaseDeathRate to the COVID19FatalitySpeedon23March2020couldbeduetothereportingbias.

ThestrongnegativecorrelationofCOVID-19fatalitytotheInternationalTradeBalanceandBalanceofPaymentvariablesobservedduringthesecondweekofthedata (22 March 2020 - 28 March 2020) can be possibly explained as that bothvariables reflect the countries’ ability to manage the spiking COVID-19 diseaseoutbreak:countrieswithstrongereconomiesandmedicalequipmentreservemightbeable toprovidepatientswithnecessarycarewhenbeingpressuredbythehighdaily case numbers, as compared to the economies experiencing a shortage ofresources.

Overall,thehighsparsityofthecorrelationsemi-matrixregardingtheCOVID-19FatalitySpeedmetricsmightpossiblysuggestthat,attheobservedtimeinterval,theCOVID-19dataonFatalityRatemighthavebeennotsufficientformakingconclusive observations on the inter-variable relationships and furtherresearchonmorerecentdataisnecessary.

Global Health Index (GHS) and Related Indexes Correlations Lastbutnotleast,inthisstudy,wewouldliketodiscusstheGHSMetricsandtheir

relationandapplicabilitytotheCOVID-19diseaseoutbreak.First, fromthe strong significantmutual correlationbetweenGHS indexes,we

couldimmediatelynoticeapossibleinsufficientcomprehensivenessofthemetricformeasuring infections disease outbreaks in a multi-factor health and economicsenvironment. Specifically, as it is stated by the authors [40], the index has beenconstructedbasedon“...140questions,organizedacross6categories,34indicators,and 85 sub-indicators to assess a country’s capability to prevent and mitigateepidemicsandpandemics“.Judgingfromthecorrelationanalysisresults,itcanbeseenthatthecategoriesindeedarehighlycorrelatedtoeachotherandthereforemightberedundant.

Second,whenconsideringthespecificGHSIndexcomponents,suchasGHSNormIndex,theyshowtheabilitytoreflecttheCOVID-19diseaseprogressionmostofthe time in the observed interval, and therefore could potentially be a betterindicator forCOVID-19progression forecasting (9Significant correlationswithCOVID-19measurementsourof28possible)ascomparedtothecombinedGHSIndex(6significantcorrelationswithCOVID-19measurementsoutof28possible).The finding highlights the importance of considering the “Compliance WithInternational Norms” factors for COVID-19 pandemic mitigation and forecast,namely: cross-border agreements on public health emergency response;internationalcommitments;completionandpublicationofWHOJEEandtheWorldOrganization for Animal Health (OIE) Performance of Veterinary Services (PVS)



https://doi.org/10.1101/2020.04.10.20058222


Pathwayassessments;financing;andcommitmenttosharingofgeneticandbiologicaldataandspecimens[40].

Finally,wewouldliketohighlightthatthemostCOVID-correlatedvariablewasfound tobe GHSDemocracy Index,again suggesting that countries’ politicalsystemsplayacrucialroledefiningthepaceinwhichcountriescouldadoptandenforcediseasecontrolandpreventionmeasuresand,therefore,mitigatetheCOVID-19 progression. On the contrary, the GHS Health Index was found to be leastcorrelatedtoCOVID-19,revealingthatsuchmetricsashealthcapacityinmedicalinstitutions;medicalcountermeasuresandpersonneldeployment;healthcareaccess;communications with healthcare workers during a public health emergency;infection control practices and availability of equipment; and capacity to test andapprovenewcountermeasuresmightnotbeastrongpredictiveindicatorofCOVID-19pandemicdevelopment.

Toconcludethissection,someoftheGHS-adoptedindicatorswerefoundtobeofa tightrelationtoCOVID-19diseasedevelopmentand, therefore,couldpotentiallyserve as a source of the disease prediction and prevention. However,GHS indexmodel was also found to be simplistic and redundant when being applied toCOVID-19data,highlightingtheimportanceofonlytwovariables(GHSNormIndexand GHS Democracy Score) for COVID-19 outbreak analysis. We, therefore,recommend the adoptionof otherpublic policy andpolitical system-relatedmeasuresaswellasmoresophisticatednon-linearmodelswhenattemptingtoanalyzeandpredicttheCOVID-19pandemicdevelopment.

Cluster Analysis Intheprevioussections,wehavediscussedmultipleeconomicandhealthfactors

that are strongly and significantly correlated to the COVID-19 disease spread.WehavealsowitnessedvariousgovernmentsandpopulationbehavioraltraitspossiblyexplainingthedifferentscenariosofCOVID-19developmentaroundtheworld.Eventhoughtheabove-discoveredfindingsbringmorelightintothewaythatgovernmentscouldadopttomitigatethecrisis,itisstillunclearwhattheexactdifferencesinthecountry profiles affected by the COVID-19 disease spread and how the countrygrouping can be explained. Below, we provide such explanations formost of thediscoveredcorrelationrelationships.

Correlation Analysis of Country Cluster 1 Correlations ThecorrelationsoftheCluster1describedintheResultssectioncouldcharacterizethecountriesfromtheclusterasbelongingtothecategoryofdevelopingworld,whichcould also be observed from the cluster-country member list provided earlier.Therefore,thenon-consistentcorrelationswiththeCOVID-19ConfirmedCaseSpeed(three significant positive correlations and one significant negative correlation),couldthenbeexplainedbypossibletestingandreportingissuesthatfrequentlyoccurwhenfacingworld-scalediseaseoutbreaks[29].Giventhelimitedavailabledatainour COVID-19 dataset regarding COVID-19 reporting procedures in differentcountries,inthiswork,wewouldliketohighlightapossibleunder-reportingissue



https://doi.org/10.1101/2020.04.10.20058222


forthedevelopingworld[34],implyingnotjustbiasedstatisticsofthedatasetsusedforCOVID-19analysis,butalsopossiblewrongperceptionandunderestimationofthe pandemic impact on people lives andworld economies, unavoidably entailinghigher infection/mortality rates and crisis escalation. Consequently, we suggestfurtherin-depthresearchtowardsCOVID-19spreadcharacteristicsinthedevelopingcountries taking into consideration alternative measures of disease progressionevaluationinparallelwithofficialstatistics[44].

Correlation Analysis of Country Cluster 2 Correlations WhenanalyzingthecorrelationprofileoftheCountryCluster2,areadercould

immediately notice that the cluster exhibitsweak positive correlations having itsspikes inpopulationgrowth,Obesity,andDiabetes,variousheart-relateddiseases,and reproduction system cancers. From the observed relationships, we canacknowledgethatthecountriesintheclustercanbecharacterizedbythepopulationoverweight,andcorrespondingly,heart[30]andreproductivecancerproblems[31].Furthermore,italsocanbeseenthattheclusterisnotassociatedwithanysignificantCOVID-19 correlations except for one positive correlationwith COVID-19 SpreadSpeedon17March2020.Apossiblereasonforthecorrelationabsencecouldbethe“noise”inthedatathatisintroducedbytheoperationalchallengesthatthecountriesexperiencewhenmeasuringandreportingtheCOVID-19diseasecases.Forexample,discussingBrazil(amemberoftheCluster2),CostRibeiroet.al.[34],havewitnessedthat“...thenumbersreportedbytheBraziliangovernmentshouldbefarfromtherealsituations.”andsuggestedthat“...theconfirmednumberofcasesmustbemultipliedby a factor of 7.7 to obtain the actual number of infected patients in hospitalconditions.”Inothercountries,forexampleintheUS,themultiplierwasreportedtobeevenhigher, reachingthevalueof8 [35].With suchahighreportingbias, it isreasonabletoassumethattheCOVID-19dataaboutthecountriesfromCluster2could,possibly,beheavilybiased, restricting thestatisticalmethodssuchasCorrelation Analysis to discover significant relationships between COVID-19andtheclustertheybelongto.Thelatterobservationisalsoindirectlysupportedby thediscoverednegative correlationsofCluster2 countries toallGlobalHealthIndexvariablesexceptfortheRiskFactorIndex.Inparticular,thecountriesthatare,conventionally,evaluatedastobebelongingtoa“higherrisk”groupare,surprisingly,exhibitingnocorrelationtoCOVID-19progressionpointsout,oncemore,thatthereisapossiblebiasinevaluationintroducedbythequalityofCOVID-19reporting.

Correlation Analysis of Country Cluster 3 Correlations FromthecorrelationshighlightedintheResults,itcanbeseenthatthecountries

fromtheclusteraresignificantlypositivelycorrelatedtomultiplefactorsassociatedwithtypicalmoderndevelopedeconomies.Atthesametime,itcanalsobeseenthatthesecountriesexhibitastrongpositivecorrelationwithageingpopulationanditsassociated diseases. Taking into consideration these two traits and the multipleobservedcorrelationsofCluster3withCOVID-19variables,wecouldhypothesizethat the Cluster 3members aremostly developed economies and that theirpopulationsmightbealsoinitiallypredisposedtoCOVID-19infectionentailingmultiplestrongpositivecorrelationswithCOVID-19diseasespreadandfatality



https://doi.org/10.1101/2020.04.10.20058222


rates.ThelatterassumptionraisesfromthetwoknownCOVID-19riskfactorsthatarealsotobefoundrelatedtothecountriesfromCluster3,namelyolderpopulationdemographics [32] and existing pre-conditions that could lead to, for example,CytokineStorm[17]orotherhighly-lethalCOVID-19complications.Finally,itcanbeseenthatthecountriesintheCluster3arestronglycorrelatedwithalmostallGHSIndexes,suggestingthatGHSindexingsystem,indeed,tendstoscoredevelopedeconomieshigherandmightnotbeasuitablemetricwhenpredictingdiseaseoutbreaksinthelightofdiseaseunderreporting,politicalsystemdifferences,andthegapsofdevelopmentbetweendifferenteconomicsystems.

Correlation Analysis of Country Cluster 4 Correlations and Possible Limitations WhentalkingabouttheCluster4,whichsolelyconsistsofthedatafromChina,

wewould like to bring the readers’ attention to the possible bias in some of theconclusionsthatwehavedrawnfromthedata in thisstudy.BydrawingaparallelbetweentheobservedpositivecorrelationsofChina-specificchronicdiseasesaswellasthenegativecorrelationwiththeCOVID-19spreadspeed,readerscouldconcludethatdatafromChinamightaffecttheoverallanalysisresultsinrelationtotheshiftofthediseasedevelopmenttimelinebetweenChinaandothercountries.We, therefore, recommend considering such potential time biases in futurequantitative research, as such data pointsmight significantly affect the predictionanalysisresultswhenbeinganalyzedjointlywithotherindicators.Furthermore,wewould also like to highlight the importance of the proper alignment andsynchronizationof thedatathatcomes fromtheregionswith largeterritoriesandspecificdiseasedevelopmenttimelines.Lastly,wewouldliketoreiteratethepossibleshortcomings of our-operated dataset related to the underreporting issuesintroducedbymultiplecountries.Moreprecisely,insomescenarios,thelownumberofreportedCOVID-19casescouldbeexplainedbyunderreportingwhileinothers-by successfully employed COVID-19 controlmeasures. The formermightmisleadgovernmentswhenevaluatingtheperformanceofthelatter,andinverse,thelattermight not have proper feedback on their COVID-19 control measures whenbenchmarking themselves against the former. Not to say that all the above couldpotentiallybiasthestatisticalanalysisresultsandthepredictionmodelsthatcouldattempt to forecast COVID-19 progression based on various economic and healthsystemfactors.

Conclusions In this preliminary research, we have identified the possible underreporting

issue of COVID-19 disease. We have also highlighted four scenarios of COVID-19developmentdeterminedfromthecountry-levelclusteranalysisstudyandoutlinedthe shortcoming of the existing diseasemeasurement approaches, such as GlobalHealthIndex(GHS)scoring.Webelievethatourworkwouldencouragefurtherin-depth quantitative research along thedirection aswell aswould be of support topublicpolicydevelopmentwhenaddressingtheCOVID-19crisisworldwide.

Specifically, we would like to highlight the following key findings andrecommendationsthathavebeenidentifiedinthisresearch:



https://doi.org/10.1101/2020.04.10.20058222


● LongitudinalCOVID-19dataoncasesandmortalityratesalonemightnotbe sufficient for forecasting/predicting COVID-19 situations across theworld and the utilization of additional Economic and Health System datasources.

● Populationimmunityandurban-livingfactorscouldserveasstatisticalindicators reflecting the predisposition of countries to rapid infectiousdiseasespread.

● ReportingisacrucialfactorinunderstandingCOVID-19evolution.Forthelessdevelopedeconomies,manyCOVID-19mighthavebeenleftunreportedor misclassified, while for the more developed economies, the COVID-19disease progression speed might have been reported higher due to well-organizedtestingandreporting.Overallthetestingabilityandthenumberof registered cases factors must be always considered together formakingconclusiveobservationsonCOVID-19progression.

● Thepoliticalsystemwasfoundtobeanothercrucialfactoraffectingthesuccessful implementation of COVID-19 preventive measures, such aslockdownsandborderclosures.GHSDemocracyandNormIndexeswerefound to be of a high relation to COVID-19 development, GHS HealthIndex was found to be redundant and weakly related to COVID-19progressionduringtheobserved14-daytimeinterval.

● GHS index model was found to be simplistic and redundant whenmeasuring and forecasting COVID-19 spread and new morecomprehensive (data-vice and architecture-ice models) models arenecessarytobedeveloped.

Future Work ToencouragefuturestudiesonpredictiveCOVID-19analytics,inthisworkwe

implementedapreliminarytestofRegressionanalysisbyapplyingLinearRegressionModelontoourdatasetsandfittingthemodelforpredictingtheCOVID-19SpreadSpeedduringthelastdayoftheobserved14-daysinterval.Asmanyoftheemployedvariableswerefoundtobecorrelatedtoeachother,inordertoachieveabalancedregressionfitting,priortorunningregressionfittingwehavealsoappliedPrincipalComponentAnalysis(PCA)[43] thathaveresulted inprojectingthedata inanewspace consisting of 132 principal components (determined automatically by PCApreserving100%ofvariance).Theresultsof thetestarepresented inMultimediaAppendix4.Fromtheresults,itcouldbeseenthatthemodulovaluesofatleast53regressioncoefficientswerefoundtobegreaterthan0.01(p<0.05),suggestingtheapplicabilityandthehighpotentialofusingourdatasetinCOVID-19predictiontask.

Infutureworks,weareaimingatestablishingautomotiveMachineLearningandStatisticalframeworksthatwouldbeattemptingtopredictthefuturedevelopmentofCOVID-19 disease based on our COVID-19 dataset.Wewill be also extending the



https://doi.org/10.1101/2020.04.10.20058222


dataset with more dynamic and comprehensive data, such as medical resourceavailability,government-imposedcontrolmeasures,andculture-drivenaspects.

Acknowledgements Aleksandr Farseev and Yu-Yi Chu-Farseeva conceived of the presented idea.Aleksandr Farseev and Qi Yang developed the theory and performed the dataanalysis. Yu-Yi Chu-Farseeva and Daron Benjamin Loo verified the analyticalmethods. Yu-Yi Chu-Farseeva encouraged Aleksandr Farseev to investigatepopulation health-related aspects of the COVID-19 Disease Spread and FatalitiesSpeedandsupervisedthefindingsofthiswork.Allauthorsdiscussedtheresultsandcontributedtothefinalmanuscript.

Conflicts of Interest Wedeclarenocompetinginterests.

Abbreviations COVID-19:Coronavirusdisease2019GDP:GrossDomesticProductGVA:GrossValueAdded

MultimediaAppendix1VariablesParticipatedinCorrelationAnalysis[XLSXfile(MicrosoftExcel),53kB]

MultimediaAppendix2TableofStatistics[XLSXfile(MicrosoftExcel),24kB]

MultimediaAppendix3PearsonProduct-MomentCorrelationBetweenCOVID-19SpreadSpeed,Economics,andHealthFactors[PDFfile,124kB]

MultimediaAppendix4 RegressionCoefficients[XLSXfile(MicrosoftExcel),18kB]

MultimediaAppendix5PearsonProduct-MomentCorrelationBetweenCOVID-19SpreadSpeed,Economics,andHealthFactors[XLSXfile,307kB]



https://doi.org/10.1101/2020.04.10.20058222


References 1. Y.Bai,L.Yao,T.Wei,F.Tian,D.-Y.Jin,L.Chen,M.Wang,Presumedasymptomatic

carriertransmissionofcovid-19.Jama.PMID:320836432. Z. Wu, J. M. McGoogan, Characteristics of and important lessons from the

coronavirusdisease2019(covid-19)outbreakinChina:summaryofareportof72314casesfromtheChinesecenterfordiseasecontrolandprevention.Jama.PMID:32091533

3. Y.Ji,Z.Ma,M.P.Peppelenbosch,Q.Pan,Potentialassociationbetweencovid-19mortalityandhealth-careresourceavailability.TheLancetGlobalHealth.PMID:32109372

4. Z.Chen,Q.Zhang,Y.Lu,Z.Guo,X.Zhang,W.Zhang,C.Guo,C.Liao,Q.Li,X.Han,etal., Distribution of the covid-19 epidemic and correlation with populationemigrationfromWuhan,China..Chinesemedicaljournal.PMID:32118644

5. M.Chinazzi,J.T.Davis,M.Ajelli,C.Gioannini,M.Litvinova,S.Merler,A.P.yPiontti,K.Mu,L.Rossi,K.Sun,etal.,Theeffectoftravelrestrictionsonthespreadofthe2019novelcoronavirus(covid-19)outbreak,Science.PMID:32144116

6. Covid19global forecasting (week2): forecastdaily covid-19 spread in regionsaround world, https://www.kaggle.com/c/covid19-global-forecasting-week-2/data,accessed:2020-03-29.

7. Country statistics - undata dataset,https://www.kaggle.com/sudalairajkumar/undata-country-profiles, accessed:2020-03-29.

8. Pollution by country for covid19 analysist, https://www.kaggle.com/brandonhoeksema/pollution-by-country-for-covid19-analysis, accessed: 2020-03-29.

9. World life expectancy database, https://www.worldlifeexpectancy.com/,accessed:2020-03-29.

10. Covid-19 combined dataset (somin.ai), https://covid19.somin.ai/, accessed:2020-03-29.

11. S.A.Lauer,K.H.Grantz,Q.Bi,F.K.Jones,Q.Zheng,H.R.Meredith,A.S.Azman,N.G.Reich,J.Lessler,Theincubationperiodofcoronavirusdisease2019(covid-19)frompublicly reported confirmed cases: estimation and application. Annals ofinternalmedicine.PMID:32150748

12. Theworldbankdata,https://data.worldbank.org/indicator/SP.POP.0014.TO.ZS,accessed:2020-03-29.

13. D.Freedman,R.Pisani,R.Purves,Statistics(internationalstudentedition),Pisani,R.Purves,4thedn.WWNorton&Company,NewYork.

14. Eer cancer statistics review, 1975-2013,https://seer.cancer.gov/archive/csr/1975_2013/,accessed:2020-03-30.

15. J.Hellewell,S.Abbott,A.Gimma,N.I.Bosse,C.I.Jarvis,T.W.Russell,J.D.Munday,A. J.Kucharski,W. J.Edmunds,F.Sun, et al., Feasibilityof controlling covid-19outbreaks by isolation of cases and contacts. The Lancet Global Health PMID:32119825



https://doi.org/10.1101/2020.04.10.20058222


16. J. C.Wells, A. A.Marphatia, T. J. Cole, D.McCoy, Associationsof economic andgender inequality with global obesity prevalence: understanding the femaleexcess.Socialscience&medicine75(3)(2012)482–490.PMID:22580078

17. P.Mehta,D.F.McAuley,M.Brown,E.Sanchez,R.S.Tattersall,J.J.Manson,Covid-19: consider cytokine storm syndromes and immunosuppression. The Lancet.PMID:32192578

18. V.L.Feigin,C.M.Lawes,D.A.Bennett,C. S.Anderson, Strokeepidemiology: areviewofpopulation-basedstudiesofincidence,prevalence,andcase-fatalityinthelate20thcentury,TheLancetNeurology2003;2(1):43–53PMID:12849300

19. M.Pereira,N.Lunet,A.Azevedo,H.Barros,Differencesinprevalence,awareness,treatment and control of hypertension between developing and developedcountries,Journalofhypertension.2009;27(5);963–975PMID:19402221

20. Z.Hu,C.Song,C.Xu,G.Jin,Y.Chen,X.Xu,H.Ma,W.Chen,Y.Lin,Y.Zheng,etal.,Clinicalcharacteristicsof24asymptomaticinfectionswithcovid-19screenedamongclosecontactsinNanjing,China,ScienceChinaLifeSciences(2020)1–6PMID:32146694

21. Coronavirus: Travel Restrictions, border shutdowns by country,https://www.aljazeera.com/news/2020/03/coronavirus-travel-restrictions-border-shutdowns-country-200318091505922.html,accessed:2020-04-04.

22. W.H.Organization,etal.,Coronavirusdisease2019(covid-19):situationreport,67.

23. J.Hilton,M.J.Keeling,Estimationofcountry-levelbasicreproductiveratiosfornovelcoronavirus(covid-19)usingsyntheticcontactmatrices,medRxiv.

24. S.Merler,M.Ajelli,L.Fumanelli,M.F.Gomes,A.P.yPiontti,L.Rossi,D.L.Chao,I.M.LonginiJr,M.E.Halloran,A.Vespignani,Spatiotemporalspreadofthe2014outbreak of ebola virus disease in liberia and the effectiveness of non-pharmaceutical interventions: a computationalmodelling analysis, The LancetInfectiousDiseases2015;15(2):204–211PMID:25575618

25. W. H. Organization, et al., Rational use of personal protective equipment forcoronavirusdisease(covid-19):interimguidance,27February2020,Tech.rep.,WorldHealthOrganization(2020).

26. T.Beck,Financialdevelopmentandinternationaltrade:Istherealink?,JournalofinternationalEconomics2002;57(1):107–131.

27. X.Wu,S.Hu,A.B.Kwaku,Q.Li,K.Luo,Y.Zhou,H.Tan,Spatio-temporalclusteringanalysisanditsdeterminantsofhand,footandmouthdiseaseinHunan,china,2009–2015,BMCinfectiousdiseases2017;17(1):645.PMID:28946852

28. D. Pelleg, A. W. Moore, et al., X-means: Extending k-means with efficientestimationof thenumberof clusters. InProceedingsof the17th InternationalConf.onMachineLearning,2000;1:727–734.

29. L. Bakalikwira, J. Bananuka, T. Kaawaase Kigongo, D. Musimenta, V. Mukyala,Accountability in the public health care systems: A developing economyperspective,CogentBusiness&Management2017;4(1).

30. A.Keys,Obesityandheartdisease,JournalofChronicDiseases1955;1(4)31. N.M.Maruthur,S.D.Bolen,F.L.Brancati,J.M.Clark,Theassociationofobesity

and cervical cancer screening: a systematic review and metanalysis, Obesity2009;17(2):375–381.



https://doi.org/10.1101/2020.04.10.20058222


32. Q.Ruan,K.Yang,W.Wang,L.Jiang,J.Song,Clinicalpredictorsofmortalityduetocovid-19 based on an analysis of data of 150 patients from Wuhan, China,Intensivecaremedicine2020:1(3).PMID:32253449

33. Kaggle,asubsidiaryofGoogleLLC,isanonlinecommunityofdatascientistsandmachinelearningpractitioners.https://www.kaggle.com/

34. L.C.Ribeiro,A.T.Bernardes,EstimateofunderreportingofCOVID-19inBrazilbyAcuteRespiratorySyndromehospitalizationreports,No010,NotasTécnicasCedeplar-UFMG,Cedeplar,UniversidadeFederaldeMinasGerais,https://EconPapers.repec.org/RePEc:cdp:tecnot:tn010.(2020)

35. Laboratory-ConfirmedCOVID-19-AssociatedHospitalizations,preliminarycumulativeratesasofMay23,2020.https://gis.cdc.gov/grasp/COVIDNet/COVID19_3.html

36. GuestJL,DelRioC,SanchezT.TheThreeStepsNeededtoEndtheCOVID-19Pandemic:BoldPublicHealthLeadership,RapidInnovations,andCourageousPoliticalWill.JMIRPublicHealthSurveill.2020;6(2):e19043PMID:32240972

37. Gilbert,M.,Pullano,G.,Pinotti,F.,Valdano,E.,Poletto,C.,Boëlle,P.Y.,...&Gutierrez,B.(2020).PreparednessandvulnerabilityofAfricancountriesagainstimportationsofCOVID-19:amodellingstudy.TheLancet,395(10227),871-877.PMID:32087820

38. Rodriguez-MoralesAJ,GallegoV,Escalera-AntezanaJP,etal.COVID-19inLatinAmerica:TheimplicationsofthefirstconfirmedcaseinBrazil[publishedonlineaheadofprint,2020Feb29].TravelMedInfectDis.2020;101613.doi:10.1016/j.tmaid.2020.101613PMID:32126292

39. JiaJS,LuX,YuanY,XuG,JiaJ,ChristakisNA.Populationflowdrivesspatio-temporaldistributionofCOVID-19inChina[publishedonlineaheadofprint,2020Apr29].Nature.2020;10.1038/s41586-020-2284-y.doi:10.1038/s41586-020-2284-yPMID:32349120

40. Goniewicz,K.;Khorram-Manesh,A.;Hertelendy,A.J.;Goniewicz,M.;Naylor,K.;Burkle,F.M.,Jr.CurrentResponseandManagementDecisionsoftheEuropeanUniontotheCOVID-19Outbreak:AReview.Sustainability2020,12,3838.

41. Cameron,E.E.,Nuzzo,J.B.,&Bell,J.A.(2019).Globalhealthsecurityindex:Buildingcollectiveactionandaccountability.Baltimore,MD:JohnsHopkins,BloombergSchoolofPublicHealth.Retrievedfromhttps://www.ghsindex.org/wp-content/uploads/2019/10/2019-Global-Health-Security-Index.pdf.

42. Andersen,Martin,EarlyEvidenceonSocialDistancinginResponsetoCOVID-19intheUnitedStates(April6,2020).SocialScienceResearchNetwork.http://dx.doi.org/10.2139/ssrn.3569368

43. Wold,S.,Esbensen,K.,&Geladi,P.(1987).Principalcomponentanalysis.Chemometricsandintelligentlaboratorysystems,2(1-3),37-52.

44. Remuzzi,A.,&Remuzzi,G.(2020).COVID-19andItaly:whatnext?.TheLancet.45. Teo,P.,Yeoh,B.S.,&Ong,S.N.(2005).SARSinSingapore:surveillance

strategiesinaglobalisingcity.HealthPolicy,72(3),279-291.46. Hicken,A.,Kollman,K.,&Simmons,J.W.(2016).Partysystemnationalization

andtheprovisionofpublichealthservices.PoliticalScienceResearchandMethods,4(3),573-594.



https://doi.org/10.1101/2020.04.10.20058222


Documents

Understanding Economic and Health Factors Impacting the ...€¦ · 10/04/2020 · as various economic factors that impact COVID -19 Diagnosis, Reporting, and Treatment. Based on