Volume 55 - Article 17 | Pages 491–532  

Estimating COVID-19 excess mortality during and after the pandemic: A Bayesian model, with an application to New Zealand

By John Bryant, Kim Dunstan, Pubudu Senanayake, Lucianne Varn, Junni Zhang

Abstract

Background: COVID-19 excess deaths are a standard measure of the effects of COVID on mortality. They are defined as the difference between actual death counts and the counts that would have been expected in the absence of the pandemic. One way to identify the full effect of the COVID pandemic on mortality is to calculate excess deaths over the entire course of the pandemic and subsequent return to normality. However, existing methods for deriving expected death counts are not reliable for periods longer than one to two years.

Objective: We develop a new model for expected deaths that can be used over periods longer than one to two years. We apply the model to estimating monthly excess deaths in New Zealand from February 2020 to December 2025.

Methods: Building on models originally developed for long-term mortality forecasting, we construct a Bayesian hierarchical model for forecasting expected monthly death counts, disaggregated by age and sex. The model is designed to capture long-term and short-term shifts in underlying trends and age–sex-specific mortality rates. This flexibility allows us to fit the model to 22 years of historical data. The model also accommodates sparse, confidentialized data.

Conclusions: The model captures the key features of the data in the years leading up to the pandemic, lending credibility to the associated estimates of expected and excess deaths. Although the excess death estimates are subject to considerable uncertainty, they confirm that New Zealand experienced relatively few excess deaths during the pandemic, and suggest that by 2025 the country had returned to pre-pandemic mortality trends.

Contribution: We build on existing mortality forecasting methods to develop a new Bayesian hierarchical model. The new model captures complex mortality dynamics, allowing it to forecast expected deaths, and estimate excess COVID deaths, over periods longer than one to two years.

Author’s Affiliation

• John Bryant - Bayesian Demography Limited, New Zealand EMAIL
• Kim Dunstan - Statistics New Zealand, New Zealand EMAIL
• Pubudu Senanayake - Statistics New Zealand, New Zealand EMAIL
• Lucianne Varn - Statistics New Zealand, New Zealand EMAIL
• Junni Zhang - Peking University, China EMAIL

Similar articles in Demographic Research

The contribution of forecast uncertainty to lifespan uncertainty
Volume 55 - Article 14    | Keywords: cohort mortality, lifespan, lifespan inequality, lifespan variation, mortality forecasting, variance decomposition

A personal perspective on the history of the World Fertility Survey and The Demographic and Health Surveys Program: An interview with Tom Pullum
Volume 55 - Article 6    | Keywords: Demographic Health Surveys, fertility, fertility, health, mortality, surveys, World Fertility Survey

Collecting data on HIV/AIDS mortality during household surveys: A randomized validation study in Malawi
Volume 54 - Article 41    | Keywords: data quality, excess mortality, HIV/AIDS, mortality, siblings, social desirability bias, surveys

Life expectancy in China and the contribution of regional dynamics
Volume 54 - Article 39    | Keywords: composition, decomposition, life expectancy, mortality

Partnering in turbulent times: Hooking up, dating, and romantic relationship formation in college, 2019–2024
Volume 54 - Article 37    | Keywords: college students, COVID-19, gender, partner choice, relationship formation, romantic love, sexual orientation