Real-Time Analysis of Predictors of COVID-19 Infection Spread in Countries in the European Union Through a New Tool

Anikó Balogh, Anna Harman, Frauke KreuterInternational Journal of Public Health 2022 · DOI

TL;DR

This paper introduces "Covid-Predictor-Tracker," a novel interactive tool that uses machine learning to analyze real-time COVID-19 data across European Union countries. It identifies key predictors of infection spread, revealing that vaccination rates and mask usage are consistently important, while also highlighting how country-specific characteristics influence these relationships.

What Problem Does This Paper Address?

The COVID-19 pandemic presented an unprecedented challenge, requiring rapid and effective policy responses. Governments worldwide implemented various non-pharmaceutical interventions (NPIs) and vaccination campaigns, but evaluating their comparative effectiveness and understanding the complex interplay of factors driving infection spread remained difficult. Existing dashboards often provided descriptive statistics but lacked robust analytical capabilities to disentangle predictor importance over time and across diverse national contexts. This paper addresses the need for a dynamic, data-driven tool that can analyze real-time data to identify and rank predictors of COVID-19 spread, thereby supporting evidence-based policy evaluation.

Methods

The core of this research lies in the development and application of the "Covid-Predictor-Tracker" tool. The methodology involves a multi-step approach:

  • Data Integration: Over 100 indicators from diverse sources were combined into two integrated databases: one for time-constant country characteristics and another for time-varying COVID-19 related variables.
  • Predictor Identification (Random Forests): A Random Forest (RF) machine learning algorithm was employed to identify the most important predictors of daily new COVID-19 cases. RF was chosen for its ability to handle complex relationships and its suitability for large, high-dimensional datasets, outperforming traditional time-series models in similar contexts. Time-series cross-validation was specifically used to account for the temporal dependency of the data.
  • Country Clustering (Hierarchical Clustering): To understand how country-specific characteristics influence predictor importance, agglomerative hierarchical clustering was used. Countries were grouped based on time-constant features like population size, healthcare expenditure, cultural participation, and age/gender distribution.
  • Correlation Analysis: Spearman's rank correlations were calculated to assess the relationship between predictor importance ranks within individual countries and their respective country clusters. This helps determine if a country's key predictors align with those of similar nations.

Datasets

The study utilized data from nine sources:

  • Global COVID-19 Trends and Impact Survey (CTIS): Provided behavioral responses like self-reported mask usage and direct contact frequency.
  • Johns Hopkins University (JHU) Coronavirus Resource Center: Supplied daily new infections and deaths.
  • Our World in Data: Offered vaccination data (new and total vaccinations, proportion vaccinated).
  • European Centre for Disease Prevention and Control (ECDC): Contributed country-specific response measures and data on COVID-19 variants.
  • National Centers for Environmental Information: Provided daily average temperatures.
  • Eurostat: Supplied time-constant country characteristics such as population demographics, health expenditures, and cultural participation.

The paper does not introduce a new benchmark dataset but rather integrates existing, publicly available data streams.

Key Insights and Ideas

  • Vaccination and Masking as Consistent Predictors: The study consistently found that the percentage of people vaccinated at least once and self-reported mask usage were among the top predictors of reduced COVID-19 spread across most EU countries. This underscores the dual importance of both individual protective behaviors and population-level immunity.
  • Country Clusters Matter: The analysis revealed that countries with similar characteristics (e.g., Nordic countries, Balkan countries) tend to share similar patterns in the importance of infection predictors. This suggests that policy interventions might need to be tailored not only to the specific intervention but also to the socio-economic and demographic context of the country.
  • Dynamic Nature of Predictors: While vaccination is a strong predictor, its effect can be ambiguous, potentially influenced by the emergence of new variants. Similarly, the impact of NPIs like lockdowns can be complex and vary in their effectiveness depending on the pandemic stage and other factors.
  • Interactive Tool for Policy Support: The development of the Covid-Predictor-Tracker itself is a key contribution. It offers a valuable, open-source platform for public health authorities to dynamically monitor and evaluate the impact of various interventions and factors in near real-time.

Key Results

  • Top Predictors: The five strongest predictors of daily new COVID-19 cases (February 2020 - January 2022) were, in descending order: percentage of people vaccinated at least once, percentage of the Omicron variant (B.1.1.529), average daily temperature, self-reported COVID-like symptoms, and self-reported mask usage.
  • Mask Usage Effectiveness: Mask usage showed a significant negative effect on daily new cases in most countries, often contributing to the deceleration of spread, especially when combined with higher vaccination rates.
  • NPI Effectiveness: The closure of non-essential shops, pubs, daycares, and primary schools were associated with a decline in daily new cases, with daycares and primary school closures being particularly impactful in Balkan countries.
  • Variant Impact: The Omicron variant was a top predictor across most countries, with its emergence leading to the highest increase in cases in Nordic and Mediterranean countries.
  • Vaccination Impact Variation: New vaccinations had a stronger effect in countries with lower vaccination rates and lower healthcare expenditures.
  • Country Cluster Correlations: The correlation between predictor importance ranks within clusters and individual countries ranged from 0.42 to 0.75, indicating that country characteristics play a role in shaping predictor importance.

Limitations and Future Work

  • Granularity of NPI Data: The study acknowledges limitations in the NPI data, noting differences in implementation and enforcement across countries, and the absence of regional-level measures. Delays in reporting and implementation are also potential issues.
  • Specific Country-Level Analysis: While the tool allows for it, this particular article did not delve into highly specific analyses for individual countries or focus on a single predictor theme.
  • Future Directions: The authors suggest several avenues for future work:
  • Transforming the outcome variable (e.g., using growth curve slope estimates).
  • Running models separately for each country and cluster.
  • Exploring spatial models (e.g., Spatial Error Model, Spatial Lag Model) or machine learning extensions like Geographically Weighted Random Forest.
  • Investigating competing machine learning algorithms like Recurrent Neural Networks or Gradient Boosted Machines.
  • Improving NPI data by incorporating more granular information on implementation and enforcement.

Summary updated

Abstract

<b>Objectives:</b> Real-time data analysis during a pandemic is crucial. This paper aims to introduce a novel interactive tool called Covid-Predictor-Tracker using several sources of COVID-19 data, which allows examining developments over time and across countries. Exemplified here by investigating relative effects of vaccination to non-pharmaceutical interventions on COVID-19 spread. <b>Methods:</b> We combine >100 indicators from the Global COVID-19 Trends and Impact Survey, Johns Hopkins University, Our World in Data, European Centre for Disease Prevention and Control, National Centers for Environmental Information, and Eurostat using random forests, hierarchical clustering, and rank correlation to predict COVID-19 cases. <b>Results:</b> Between 2/2020 and 1/2022, we found among the non-pharmaceutical interventions "mask usage" to have strong effects after the percentage of people vaccinated at least once, followed by country-specific measures such as lock-downs. Countries with similar characteristics share ranks of infection predictors. Gender and age distribution, healthcare expenditures and cultural participation interact with restriction measures. <b>Conclusion:</b> Including time-aware machine learning models in COVID-19 infection dashboards allows to disentangle and rank predictors of COVID-19 cases per country to support policy evaluation. Our open-source tool can be updated daily with continuous data streams, and expanded as the pandemic evolves.
BibTeX
@article{balogh2022real,
  author = {Anikó Balogh and Anna Harman and Frauke Kreuter},
  title = {{Real-Time Analysis of Predictors of COVID-19 Infection Spread in Countries in the European Union Through a New Tool}},
  journal = {International Journal of Public Health},
  year = {2022},
  doi = {10.3389/ijph.2022.1604974},
}
Figure