Using hierarchical clustering analysis to evaluate COVID-19 pandemic preparedness and performance in 180 countries in 2020

Abstract

Objective To rank and score 180 countries according to COVID-19 cases and fatality in 2020 and compare the results to existing pandemic vulnerability prediction models and results generated by standard epidemiological scoring techniques.Setting One hundred and eighty countries’ patients with COVID-19 and fatality data representing the healthcare system preparedness and performance in combating the pandemic in 2020.Design Using the retrospective daily COVID-19 data in 2020 broken into 24 half-month periods, we applied unsupervised machine learning techniques, in particular, hierarchical clustering analysis to cluster countries into five groups within each period according to their cumulative COVID-19 fatality per day over the year and cumulative COVID-19 cases per million population per day over the half-month period. We used the average of the period scores to assign countries’ final scores for each measure.Primary outcome The primary outcomes are the COVID-19 cases and fatality grades in 2020.Results The United Arab Emirates and the USA with F in COVID-19 cases, achieved A or B in the fatality scores. Belgium and Sweden ranked F in both scores. Although no African country ranked F for COVID-19 cases, several African countries such as Gambia and Liberia had F for fatality scores. More developing countries ranked D and F in fatality than in COVID-19 case rankings. The classic epidemiological measures such as averages and rates have a relatively good correlation with our methodology, but past predictions failed to forecast the COVID-19 countries’ preparedness.Conclusion COVID-19 fatality can be a good proxy for countries’ resources and system’s resilience in managing the pandemic. These findings suggest that countries’ economic and sociopolitical factors may behave in a more complex way as were believed. To explore these complex epidemiological associations, models can benefit enormously by taking advantage of methods developed in computer science and machine learning.