Data-Driven Contagion Risk Management in Low-Income Countries: Machine Learning Applications with COVID-19 in South Asia

Abstract

Objectives To derive and validate a data-driven Contagion Risk Index (CR-Index) at sub-national level for low-income countries – detecting potential infection hotspots – aiding policymakers with efficient mitigation planning for COVID-19 and future epidemics and pandemics. Methods We utilize daily district-level COVID-19 data (positive cases and deaths) from South Asia (India, Pakistan, and Bangladesh) from 2020–2022 to derive the CR-Index – founded on commutable disease spreadability vectors across four domains: urbanization, informality, migration, and health infrastructure. We validated CR-Index based risk-zoning by utilizing time-series regressions and machine learning (ML) estimates (Random Forests and a battery of cross-validation) for predictive accuracy. Results Regressions demonstrate a strong association between the CR-Index and sub-national COVID-19 epidemiology data. ML driven validation show strong predictive support for the CR-Index that can distinguish districts with high-risk COVID-19 cases/deaths for more than 85% of the time. Conclusion Our proposed simple and replicable CR-Index is an easily interpretable tool that can help low-income countries to prioritize resource mobilization (such as vaccination roll-out or free in-home test-kits) to contain the disease spread and associated crisis management, with global relevance and applicability.