Belgian Mortality Monitoring Be-MOMO

This graph presents the observed all-cause mortality in Belgium outputted from the Be-MOMO model over the last five years. The daily observed numbers are the average of the day, the 3 preceding and the 3 subsequent days. Due to delays in death registration, the counts for the most recent weeks are preliminary and numbers are visualized with a 3 week time lag. Separate curves for Flanders, Wallonia and Brussels are available using the top right-hand corner of this graph. You are free to use the results produced by this application on the sole condition that the source is mentioned (cf. Footnote).

Environmental Risk Factors

This graph shows the observed all-cause mortality in Belgium and the curves of the environmental risks such as meteorological data (temperature) and air pollution data (ozone, PM10 and PM2.5). It allows to visualize the correlations between mortality and extreme temperatures (above 25°C or below 0°C) and high air-pollutant concentrations (ozone above 100µg/m³ (max 8-hour mean), PM10 above 45µg/m³ and PM2.5 above 15µg/m³; corresponding to WHO thresholds of 2021). The vertical light-orange bands highlight the days with statistically significant excess mortality. All values are smoothed with a 7-day centered moving average. Meteorological data are provided by the Royal Meteorological Institute of Belgium (RMI). Ozone, PM10 and PM2.5 data are provided by The Belgian Interregional Environment Agency (IRCEL-CELINE).

Abbreviations:
Exp. Mort.
Expected Mortality
Obs. Mort.
Observed Mortality
Tmin
Minimum temperature
Tmax
Maximum temperature
PM10
Particulate Matter < 10 µm
PM2.5
Particulate Matter < 2.5 µm
How to play with this graph
  • – Hover lines with your mouse to view numbers
  • – Click & drag with your mouse to zoom on a period
  • – Right-click to reset default zoom
  • – Hover legend items to highlight the lines
  • – Click legend items to hide/show the lines

COVID-19

This graph depicts the observed all-cause mortality in Belgium, with the reported COVID-19 related deaths (Source) subtracted and colored in orange. The resulting blue area represents the remaining deaths from all other causes. When the total number of daily deaths exceeds the upper or lower limits of the prediction interval predicted by the modelling (green dashed lines), there is a significant excess or under-mortality.

During the first wave of COVID-19 (from 1 March to 21 June 2020), an unusual peak in deaths was observed, almost entirely attributed to COVID-19 related deaths. This first peak of COVID-19 deaths occurred around 8 April 2020. A second unusual peak in August 2020 was attributed to a heat wave. From 31 August 2020 to 14 February 2021 (second wave of COVID-19), a third unusual peak in deaths was observed (6 November 2020), again primarily attributed to COVID-19 related deaths. During these first two COVID-19 waves, the percentage of excess mortality was relatively high, but then dropped drastically for the subsequent waves.

The year 2021 was marked by three epidemic waves of COVID-19 and a brief heat episode. In 2022, there were five COVID-19 epidemic waves, a very hot summer, and two influenza epidemics. The last peak in deaths observed in December 2022 coincides with an increase in respiratory infections (influenza, RSV, COVID-19), cold temperatures, and higher concentrations of fine particles.

As of 30 June 2023, data collection for COVID-19 deaths via epidemiological surveillance has stopped, so this figure will no longer be updated. More information about the COVID-19 surveillance and related reports can be found here.

Be-MOMO Project

In Belgium, surveillance of all-cause mortality is carried out on a weekly basis by the service of Infectious Diseases Epidemiology of Sciensano. The mortality monitoring model is designed to serve as a tool for rapid detection and quantification of unusual mortality which might result from disease epidemics, such as influenza or extreme environmental conditions, such as heat waves. A timely assessment of the impact on mortality may be useful to guide or reinforce new or existing public health measures, such as influenza vaccinations and the heat action plan. Moreover, mortality monitoring can be used to evaluate possible effects of public health measures by comparing periods before and after the implementation of the intervention.

Analyses

Data

Data are updated on a weekly basis, except population sizes for which the official numbers at the 1st of January are used. Mortality data are provided by the National Register and the population data by Statistics Belgium. The mortality file contains information on all deaths registered by Belgian municipalities during the previous week (up until Saturday midday). The data comprises the date of birth, date of death, gender, nationality, place of residence and place of death. The causes of death are unknown. Because of a considerable variation in the rapidness of death registration (ranging from a few days to many weeks after the actual date of death), figures for recent periods are incomplete. Around 97% of mortality data are available after two weeks. Deaths taking place abroad are removed from the analyses as they are assumed to be unrelated to concurrent meteorological or environmental conditions in Belgium.

Statistical Methods

Observed deaths are aggregated by day. To be able to detect and quantify important increases in mortality, observed death counts are compared to 2 types of reference lines, obtained by modelling past 5-year mortality data:

  • Expected deaths are the model predictions and represent normal/average mortality. They are used for the calculation of the excess number of deaths (observed – expected).
  • The threshold is the upper limit of the prediction interval around expected mortality, calculated by a 2/3-power transformation to correct for skewness in the Poisson distribution (Farrington et al, 1996). Threshold values represent critical mortality levels and are used to detect unusual or significant mortality. The confidence level for the upper threshold was chosen as the optimal compromise between sensitivity and specificity of alert detection. It was set at 99.5% for daily-level data.

The statistical model is a modification of the overdispersed poisson Farrington model, originally developed for the detection of infectious diseases outbreaks based on weekly disease counts (Farrington et al, 1996). The model was adapted in order to be applicable to both daily- and weekly-level mortality data. While the original method limits the amount of reference data by using only historical data from similar weeks, a sine and cosine wave component was added to capture the seasonal pattern of mortality. This enables modelling the complete 5-year time series and reduces random variation in the predicted baseline, especially for daily-level data. The Be-MOMO model was adapted on 14 June 2021 following the 2020 excess mortality associated with the COVID-19 epidemic (EN - FR - NL).

An automated analysis procedure is implemented since 2018 in R, software for statistical computing (previously with Stata version 13). Statistical methods and performance of the Be-MOMO system are described in more detail in Cox et al (2010) and in the last Be-MOMO report.

Cox B, Wuillaume F, Van Oyen H, Maes S. Monitoring of all-cause mortality in Belgium (Be-MOMO): a new and automated system for the early detection and quantification of the mortality impact of public health events. International Journal of Public Health 2010, 55(4):251-259.

Publications