Skip to main content

Table 2 Details of the presented approaches for excess mortality estimation during the COVID-19 pandemic

From: Assessing excess mortality in times of pandemics based on principal component analysis of weekly mortality data—the case of COVID-19

Study Geography Demographics Data Method Main Results Consideration of Long-term Trends in Mortality Quantification of Stochasticity Dealing with Correlations among time series
Magnani et al. 4,433 Italian municipalities No sex distinction
Two age groups
Daily death numbers from January 1st to April 15th, 2015-2020
Yearly population estimates for January 1st, 2015–2019
Daily Death numbers attributed to COVID-19 by lab testing in 2020
Mean mortality rates by calendar days for 2015–2020 are computed
Daily mortality rate estimates in 2020 are computed and assumed to follow a Poisson distribution
Relative risk estimates with 95% CIs are computed relative to the baseline mean rates for 2015–2019
Statistically significant increase in mortality rates between early-March and mid-April, 2020 in Italy
Only significant for North and parts of Central Italy
Only significant for persons above age 59, except for Lombardia (both age groups significant)
None Poisson assumption for observed deaths in 2020 None
Michelozzi et al. 19 Italian cities Two sexes
Four age groups
Daily death numbers from January 1st, 2015 to April 18th, 2020 Mean death numbers by calendar days for 2015–2020 are computed
Daily death numbers implicitly assumed to be Gaussian, 95% PIs of baseline data are computed
Comparison of observed death numbers with PIs
Statistically significant excess mortality in Italy between mid-March and mid-April, 2020
In the North, statistically significant excess mortality for males for all investigated age groups (age 15 and older), for females for age groups 65 and older
In the Center and South only statistically significant excess mortality for very old females (above age 84) and males (above age 74)
None Prediction assumed Gaussian with a standard deviation of past five years None
NYC DOHMH New York City None Daily death numbers with a lab-confirmed COVID-19 infection in 2020 Daily death numbers are predicted by OLS fit of the Serfling model to daily death numbers of baseline period
Comparison of observed with expected daily death numbers
Comparison of COVID-19 associated deaths with overall excess mortality estimate
Excess mortality after March 10th, with a large share of confirmed or suspected COVID-19 cases Serfling model None None
EUROMOMO 22 EU-28 countries
2 German federal states
No sex distinction
Seven age groups
Weekly all-cause death numbers since week 1, 2016 Poisson model fit to baseline data
95% PIs derived from the GLM model
Comparison of observed weekly death numbers to 95% PIs
Statistically significant excess mortality in all countries for age groups 45+ between circa calendar weeks 11 and 19, 2020; death numbers above threshold in calendar weeks 13–15 for persons aged 15–44 years; no significant excess mortality among children
Significant excess mortality observed in Belgium, France, Ireland, Italy, Netherlands, Portugal, Spain, Sweden, Switzerland, and the UK
No significant excess mortality in Austria, Denmark, Estonia, Finland, Greece, Hungary, Luxembourg, Malta, Norway, and the two German federal states
Poisson model PIs from the GLM model None
Ours 19 countries in Europe and the Middle East Two sexes
Four age groups
Weekly all-cause mortality rate estimates since week 2, 2000 by HMD
Daily official COVID-19 associated deaths in 2020 by ECDC
Principal component analysis on all 152 time series of logit mortality rates simultaneously
OLS fit of logistic SARIMA model to past course of first PC
Monte Carlo simulation of weekly forecasts of all PCs
Retransformation of simulations to simulations of age-sex- and country-specific mortality rates
Multiplication of simulations with population estimates for 2020
Derivation of nonparametric PIs for forecasts of death numbers
Comparison of observed all-cause and COVID-19 associated weekly death numbers with PIs of forecasts
No significant excess mortality for persons below age 65, slightly significant excess mortality among females aged 65-74 years, statistically significant excess mortality among males above age 64 and females above age 74 years in calendar weeks 13–16, 2020
Statistically significant overall excess mortality only observable for Belgium, France, Netherlands, Scotland, Sweden, Poland, and Spain; results for Switzerland inconclusive; no significant excess mortality in Austria, Estonia, Finland, Hungary, Israel, Latvia, Lithuania, Norway, Portugal, Slovakia, and Slovenia
Logistic trend fitted to first principal component time series SARIMA models and Monte Carlo simulation of principal component time series with nuisance derived from 20 years of past data, leading to stochastic estimates of all mortality rate series Correlations among age groups and countries completely covered by principal component analysis