2021-12-16-世界卫生组织-Methodology_for_estimating_global_excess_deaths_associated_with_COVID-19_11页_171kb

Methodology Summary: Estimating Global Excess Deaths from COVID-19

This document outlines the methodology for estimating global excess deaths due to COVID-19, focusing on handling data gaps and providing standardized analyses.

Introduction

• Excess mortality is defined as deaths above expected levels based on pre-pandemic trends, serving as an objective measure due to challenges in reported COVID-19 data (e.g., incomplete certification, overwhelmed health systems).
• Estimation covers the period January 2020 to June 2021, incorporating deaths from direct and indirect pandemic effects.

Methods for Estimating All-Cause Deaths

1. Expected Deaths Calculation:
   • For countries with historical weekly/monthly data, use negative-binomial models with time-series smoothing (e.g., Penalized Cyclic Cubic Regression) based on 2015-2019 data.
   • For countries with only annual data, forecast annual deaths using Random Walk models and disaggregate monthly using temperature proxies.
   • For data-sparse countries, predict deaths using Poisson models with covariates (e.g., COVID-19 cases, positivity rates, containment indices) and Bayesian posterior simulations.

Methods for Deriving Sex and Age Patterns

1. Disaggregation:
   • Generate expected and observed death rates by 5-year age bands and sex using 2020 data from reported sources.
   • Handle data-lacking countries by:
     • Applying K-means clustering to group countries based on features like mean age at death, population age, and excess mortality indicators.
     • Extrapolating cluster-specific sex-age mortality ratios into normal distributions, adjusted for country clusters.
   • Propagate uncertainty through resampling of Poisson counts and bootstrapped distributions, ensuring consistency with overall mortality projections.

Uncertainty Propagation

• Uncertainty from models and data gaps is propagated through repeated sampling of Poisson counts, Poisson excess mortality parameters, and smooth draws, generating country-specific distributions and credible intervals. This provides a range of plausible values conditional on total predicted deaths and expected death levels.

This methodology aims for comparability across countries while accounting for data limitations and pandemic impacts.