# Mortality risk during the COVID-19 pandemic is shaped by human development

## TL;DR

The study shows that socioeconomic factors, not just reported cases, strongly influence pandemic mortality, using machine learning to uncover hidden patterns.

## Contribution

A novel framework using compressed National Framework Conditions (cNFCs) better explains pandemic mortality than traditional metrics.

## Key findings

- The machine learning model explained nearly half of global excess mortality variance.
- cNFCs had a stronger impact on mortality predictions than reported case numbers.
- Socioeconomic factors like labor force age and health spending shaped mortality outcomes.

## Abstract

During the global COVID-19 pandemic (2020–2021), excess mortality varied substantially across countries. Notably, upper-middle-income countries experienced greater variability in excess mortality than both low- and high-income countries, despite reporting fewer COVID-19 cases than high-income countries but more than low-income countries. This disconnect between case numbers and mortality suggests more complex structural vulnerabilities. Socioeconomic conditions and healthcare system performance, collectively referred to as National Framework Conditions (NFCs), are likely key determinants of pandemic outcomes. However, the specific relationship between these factors and excess mortality remains poorly understood.

We constructed a predictive model of excess mortality using reported COVID-19 case counts and a wide array of NFCs derived from the World Development Indicators (WDI), employing a tree-based machine learning method (XGBoost). To reduce dimensionality, we applied a non-linear method (e-Isomap), extracting latent components called compressed National Framework Conditions (cNFCs). We applied SHapley Additive exPlanations (SHAP) values to estimate the feature importance and quantify the contribution of each cNFC.

Our machine learning model explained nearly half of the global variance in excess mortality (\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document}$$R^2$$\end{document}R2: median 49.7; interquartile range (IQR): 10.9). SHAP analysis revealed that cNFCs contributed most strongly to model predictions of excess mortality (SHAP: median 8.1; IQR 1.2), followed by pandemic indicators, such as reported COVID-19 cases (SHAP: median 6.4; IQR 0.7). Using explainable artificial intelligence (XAI), we further identified how interconnected socioeconomic conditions, including labor force participation age and health spending, shaped mortality outcomes.

Our findings demonstrate that cNFCs outperform conventional epidemiological or preparedness metrics, in explaining cross-country differences in COVID-19 excess mortality during 2020–2021. By capturing latent socioeconomic structures, the cNFC framework reveals systemic vulnerabilities that reported COVID-19 cases and other indicators fail to detect. This approach offers a new perspective on structural resilience and pandemic preparedness.

The online version contains supplementary material available at 10.1186/s44263-026-00255-0.