An analysis of Coronavirus cases across the world.
The Covid-19 pandemic has changed the world in great ways this year. It has pushed all human beings to focus and unite against one common enemy. It has also changed a lot in the ways we deal with daily life. Many countries have instigated curfews and confinements that have made our time indoors much longer. It is a strange time we live in.
Today I have tried to answer several questions that arose while exploring and combining datasets.
The datasets I have explored are the following :
The ECDC dataset 'https://covid.ourworldindata.org/data/ecdc/full_data.csv'
Which I used for the information it contains on mortality rates from covid-19 around the world.
The countries of the world dataset from Kaggle: 'https://www.kaggle.com/fernandol/countries-of-the-world'
Which I used for the general information it contains on the countries of the world.
The Historical Index of Ethnic Fractionalization Dataset (HIEF) from Harvard: 'https://dataverse.harvard.edu/api/access/datafile/3476857?format=original&gbrecs=true'
Which I used to determine ethnic diversity in countries.
The 'uncover' einstein clinical dataset from Kaggle: 'https://www.kaggle.com/roche-data-science-coalition/uncover'
Which I used for bloodwork results from covid-19 and other patients.
The notebook can be found here: https://github.com/FarahKa/covid_assignment
This blog will follow the order of the answered questions while hopefully contributing insights on the Covid-19 pandemic and our world.
1) Which are the most hit countries, using total deaths?
2) Which are the most hit countries, using total deaths/100000 inhabitants?
3) Are more populous countries hit?
4) What is the correlation between total deaths, habitants and GDP?
5) What is the relationship between total deaths and how diverse a country is?
6) What is the difference between the bloodwork of people with covid and people without covid?
1) Which are the most hit countries, using total deaths?
We start by loading the datasets, then doing primary work to make it possible to merge them. Here we merge the latest coronavirus data with the population info of countries:
recent.rename(columns = {'location':'Country'}, inplace = True)
recent['Country']=recent['Country'].str.strip()
pop['Country']=pop['Country'].str.strip()
df=pd.merge(recent, pop)
df=df.set_index('Country')Which lets us determine the five most hit countries by total death count:
df=df.sort_values('total_deaths', ascending=False)
print("Five most hit countries by total deaths:")
print(df.head()['total_deaths'])Five most hit countries by total deaths: Country United States 40682 Italy 23660 Spain 20453 France 19718 United Kingdom 16060
2) Which are the most hit countries, using total deaths/100000 inhabitants?
However it seems unfair to judge smaller countries and bigger countries on the same scale, which takes us to dividing the death count by 100000 inhabitants. The result is as follows:
Five most hit countries by total deaths/100000 inhabitants: Country San Marino 133.328775 Belgium 54.754440 Spain 50.628942 Andorra 50.561088 Italy 40.699418
We can notice the difference in results, though knowing San Marino's proximity to Italy, it is not surprising.
This led me to the following question:
3) Are more populous countries hit by the coronavirus?
The more people, the bigger the risk, right? We tested the hypothesis as such:
#3) Are more populous countries hit?
print(np.corrcoef(df_nozero.total_deaths, df_nozero.Population))
#it seems there is a slight positive correlation between population and total deaths when we consider all countries
df2=df_nozero.head(50)
print(np.corrcoef(df2.total_deaths, df2.Population))
#but when we consider the 50 most hit countries, the positive correlation spikes up to around 0.7
sns.scatterplot(df2['total_deaths'], df2['Population'])
plt.xlabel("Total Deaths")
plt.ylabel("Population")
sns.regplot(x='total_deaths', y='Population', data=df2)
plt.show()
When we consider the 50 most hit countries, the positive correlation spikes up to around 0.7.
The hypothesis might be correct.
Another question presents itself: Do poorer countries fare worse than richer countries? Or does the Coronavirus have the same kind of impact on countries regardless of GDP?
4) What is the correlation between total deaths, habitants and GDP?
Hypothesis: the lower the GDP, the more poorly a country fares.
#Total deaths and GDP: for all countries with deaths:
sns.scatterplot(df_nozero.death_by_pop, df_nozero['GDP ($ per capita)'])
sns.regplot(df_nozero.death_by_pop, df_nozero['GDP ($ per capita)'])
plt.xlabel("Total Deaths / 100000 inhabitants")
plt.ylabel("GDP ($ per capita)")
plt.show()
print(np.corrcoef(df_nozero.death_by_pop, df_nozero['GDP ($ per capita)']))
A correlation of 0.4 is considered weak to moderate. The hypothesis is disproved since the correlation is positive: Countries with bigger GDP seem to be more hit by the covid19 pandemic.
5) What is the relationship between total deaths and ethnic diversity in a country ?
The hypothesis is the following: If a country is more diverse, there would be more traveling to and from it, which would make it more at risk of a big spike of cases.
We have used the Historical Index of Ethnic Fractionalization Dataset. The Index is the probability that two randomly picked individuals from the population pool would be of different ethnicities.
First we have explored the dataset on its own:
#Which are the most ethnically diverse countries in this dataset?
print("The most ethnically diverse countries in this dataset:")
print(ed_recent.sort_values('EFindex', ascending=False).head(10))
#And the least diverse?
print("The least ethnically diverse countries in this dataset:")
print(ed_recent.sort_values('EFindex').head(10))The results are as follows:
The most ethnically diverse countries in this dataset: EFindex Country Liberia 0.889 Uganda 0.883 Togo 0.880 Nepal 0.860 South Africa 0.856 Chad 0.855 Kenya 0.855 Mali 0.852 Nigeria 0.850 Guinea-Bissau 0.808 The least ethnically diverse countries in this dataset: EFindex Country Japan 0.019 Democratic People's Republic of Korea 0.020 Bangladesh 0.025 Tunisia 0.034 Egypt 0.041 Jordan 0.044 Armenia 0.045 Comoros 0.054 Poland 0.069 Republic of Korea 0.095
We notice the interesting fact that most diverse countries are african.
We then combined the death rates and diversity data to see if there was any correlation. All combinations we did had this kind of appearance:
Which disproves our hypothesis: diversity is uncorrelated with death rate.
The last question comes from the exploration of another dataset containing clinical bloodwork of covid and non-covid patients:
6) What is the difference between the bloodwork of people with covid and people without covid? The attempt is to find the variables in the bloodwork whose mean differs by more than 0.5 between covid patients and other patients.
First we did some exploration by separating sick and healthy patients who were tested for covid 19 (which meant the healthy ones presented the symptoms but did not have the sickness).
sick=clinical.loc[clinical['sars_cov_2_exam_result'] == 'positive' ]
healthy=clinical.loc[clinical['sars_cov_2_exam_result'] == 'negative' ]
sick.describe()
healthy.describe()
Then we searched for variables in the bloodwork whose mean differs by more than 0.5 between covid patients and other patients. patient_age_quantile 1.456424 platelets -0.818720 leukocytes -0.836818 eosinophils -0.558663 monocytes 0.571967 ionized_calcium -0.938920 segmented 0.642866 ferritin 0.641919 pco2_arterial_blood_gas_analysis -0.648399 ph_arterial_blood_gas_analysis 0.630072 po2_arterial_blood_gas_analysis 0.625503 arteiral_fio2 -0.624296 phosphor -0.561041 cto2_arterial_blood_gas_analysis 0.535281 It mignt be useful to consider these variables more closely in the proccess of understanding the disease and diagnosing it.
Insights:
We have learned through this analysis to find new questions to ask about a very important issue, and searching for ways to answer them.
These questions relied mostly on number of deaths which is a variable that does not depend on how much testing there was or whether a country reports accurately.
Hopefully the information gathered can help some understand this disease and where it stems from better.
Thank you for reading and stay safe.
Comments