Welcome

The goal of this dashboard is to monitor district-specific time trends in the total mortality (all causes of death combined) in the federal state of Rhineland-Palatinate, Germany, starting with the COVID-19 pandemic.

This is a joint project by the Institute of Medical Biostatistics, Epidemiology and Informatics (IMBEI) at the University Medical Center of the Johannes Gutenberg-University Mainz together with the Statistical Office Rhineland-Palatinate . The section gives more information on the motivation behind surveillance of all-cause mortality, and lists contributors.

We use different statistical methods to compare current all-cause mortality against mortality during the years 2015-2019 to judge whether mortality is currently higher (excess mortality) or lower (mortality deficit) than may be expected based on historical data. Our focus is to use data with a high spatial resolution (NUTS-3 level: district), and fine-grained age-standardization to make estimates comparable between countries and time periods despite different demographics. The section contains more information about data sources and statistical methods. The following publication provides an application and more detailed methods: Wollschläger et al. Explaining the age-adjusted excess mortality with COVID-19-attributed deaths from January 2020 to July 2021. Bundesgesundheitsblatt 2021. doi 10.1007/s00103-021-03465-z.

Data availability: 2024-09-30

Usage

  • I - Green area: navigation between sections
    • Dashboard: Map and timeline for different types of information.
    • Mortality rates: More results from modeling mortality rates using different methods.
    • Input data: More information on input data used for modeling mortality rates.
    • About: Background information on the mortality monitoring project, people, data sources, and methods.
  • II - Orange area: Navigation to select the type of information that should be displayed on the map and in the timeline.
    • Standardized mortality rate: Directly standardized mortality rate with respect to a reference population (Germany 2011 by default)
    • SMR 2020-2024 vs. 2015-2019: Standardized Mortality Ratio comparing observed deaths against the expected counts for 2020 based on the average mortality rate from 2015-2019
    • Excess mortality 2020-2024: Estimates of excess mortality / mortality deficit based on predicted counts from a multivariable negative binomial regression model fit to data from 2015-2019
    • Population: Raw population counts
    • Mortality: Raw mortality counts
    • Corona: RKI data on Corona infections and deaths
    • Influenza: RKI data on seasonal influenza activity
    • Deprivation: GISD regional socioeconomic deprivation index
    • Weather: Average and extreme temperatures, rain, humidity
  • III - Blue area: Dashboard map of Rhineland-Palatinate with color-representation of selected information. Clicking a district selects it for data display. Clicking twice outside of any district selects Rhineland-Palatinate. Scroll to zoom, click-and-drag to pan map area.
  • IV - Pink area: Dashboard timeline of selected information in the district selected on the map (or in Rhineland-Palatinate if no district is selected). The diagram has interactive elements.
  • V - Yellow area: More controls to fine tune methods that affect multiple analyses (e.g., choice of reference population for directly standardized mortality rates, time resolution month vs. week).

Methods

Age standardized mortality rate 2015-2024 / 100,000 / year

Method

Age standardized mortality rate 2015-2024 / 100,000 / year

Standardized mortality ratio (SMR) all-cause mortality 2020-2024 vs. average 2015-2019

‌

Method

Standardized mortality ratio (SMR, indirect age standardization) comparing observed deaths in 2020-2024 to the expected number of deaths based on the stratified average mortality rate 2015-2019. Excess mortality (absolute counts) is the difference between observed and expected number of deaths. Data source: Statistical Office Rhineland-Palatinate . Population and mortality data from 2015-2018 come from a different source than data from 2019-2024. See -About- for details.

Standardized mortality ratio (SMR) all-cause mortality 2020-2024 vs. average 2015-2019

Q-Q plot p-values against uniform distribution

Model display options

Method

Estimates of excess mortality / mortality deficit based on expected counts from a multivariable negative binomial regression model fit to data from 2015-2019 and taking into account age group, sex, average temperature, influenza activity, regional socioeconomic deprivation, calendar year, and calendar month. Multiple data sources: Population and mortality data from 2015-2019 come from a different source than data from 2020-2024. See the -About- section for details.

Model fit results

Population Rhineland-Palatinate

Population Rhineland-Palatinate

‌

Monthly counts from 2015-2018 are linear interpolations between adjacent yearly counts with a reference date of December 31st. Data source for the years 2015-2018: Statistical Office Rhineland-Palatinate . Data source for the years 2019-2024: KommWis GmbH - EWOIS Gemeindestatistik. See -About- for details.

Mortality Rhineland-Palatinate

Mortality Rhineland-Palatinate

‌

Data source: Statistical Office Rhineland-Palatinate . See -About- for details.

Seasonal influenza activity 2015-2024

Seasonal influenza activity 2015-2024

‌

Data source: Robert Koch-Institut (RKI) SurvStat@RKI 2.0 https://survstat.rki.de/ . See -About- for details.

Weather by district 2015-2024

Weather by district 2015-2024

‌

Data source: Agricultural meteorology service Rhineland-Palatinate - Dienstleistungszentrum Ländlicher Raum. See -About- for details.

GISD regional deprivation index

GISD regional deprivation index

‌

Data source: GitHub - Article . See -About- for details.

Data sources

Population

  • Statistical Office Rhineland-Palatinate
  • 2015-2023: Official population statistics
    • Stratified by district, year, sex, age group (5 years)
    • Verified annual population count for reference date December 31st
  • 2019-2024: EWOIS municipality statistics (KommWis GmbH)
    • Stratified by district, month, sex, age group (varies, 1-10 years)
    • Subject to changes after verification by the Statistical Office

Mortality

  • Statistical Office Rhineland-Palatinate
  • 2015-2020: Official mortality statistics
    • Stratified by district, month/week, sex, age group (5 years)
    • Deaths are attributed to the location of residence, not the location of registration.
    • Full quality control.
  • 2020-2024: Incoming data (Bureaus of Vital Statistics)
    • Stratified by district, month/week, sex, age group (5 years)
    • Deaths are attributed to the location of residence, not the location of registration.
    • Initial plausibility checking done, but late corrections from registration offices as well as exchange of deaths with other federal states not carried out.

SARS-CoV-2 infections / COVID-19 deaths

Seasonal influenza activity

GISD regional socioeconomic deprivation index

Weather

Geocoding

Methods

Estimating excess mortality / mortality deficit

Excess mortality and mortality deficit both designate current observed mortality that can be regarded unusual in light of past mortality when considering all relevant boundary conditions that are not unique to the current situation.

Here, excess mortality refers to current observed mortality that is higher than expected. The expected or baseline mortality is defined as the corresponding average mortality from previous years while taking into account major influences on mortality that may change over time. Conversely, we refer to it as a mortality deficit when the current observed mortality is lower than expected. Excess mortality and mortality deficit are thus meant to capture the mortality that can be attributed to the entirety of particular conditions associated with the COVID-19 pandemic.

On the population level, all cause mortality rates depend on the demographic structure as well as on external influences such as temperature or a severe flu season. Mortality may also follow slow secular trends that reflect changes in nutrition, lifestyle factors, and medical treatment efficacy for common diseases.

Age is the primary risk factor for death with sex-specific dependencies, e.g., due to breast cancer. Therefore, the absolute number of deaths needs to be assessed in the context of total population size as well as its distribution across age and sex. In addition to slow demographic change towards an older population, there may be more short-term fluctuations at the district level and in specific age groups following local economic development. In addition, post-2015 immigration changed population demographics in Germany.

Therefore, comparing mortality trends across time, and especially across regions to identify unusual mortality rates requires methods to account for different demographic structures and external influences. To this end, we pursue three approaches to analyze mortality trends in Rhineland-Palatinate:

Standardized Mortality Rates / 100,000 / Year

  • Direct age standardization
    • Standard population default: Germany 2011
    • Population data with last observation carried forward when mortality data is newer than the last available population data
  • Mortality stratified by district, month, sex, age group (10 years)
  • Mortality rate in a given month per 100,000 is expressed as the corresponding yearly mortality rate by scaling with (365.25/days in month) or (365.25/7)
  • Confidence interval using the method by Dobson

Standardized Mortality Ratio (SMR) 2020-2024 vs. 2015-2019

  • Indirect age standardization
    • Ratio of observed to expected deaths in 2020-2024 based on respective averages of the age- (10 year groups) and sex-stratified mortality rates from 2015-2019
    • Excess mortality / mortality deficit is calculated as the observed minus the expected mortality.
  • Mortality stratified by district, month, sex, age group (10 years)
  • p-value based on Poisson test with mid-p correction
  • Confidence interval based on Poisson test

Excess Mortality 2020-2024

  • Regression modeling of mortality rates taking into account relevant covariates that reflect population demographics as well as external influences. The regression model is fit to district-level data from 2015-2019 and is used to derive district-level predictions for 2020-2024. The difference from the observed mortality to the expected mortality (= predicted from the regression model) is the excess mortality / mortality deficit.
  • Negative-Binomial GLM for all cause mortality
    • Using Bayesian model fitting with Hamilton Monte Carlo ( Stan / brms )
  • Covariates
    • Numeric calendar date
    • Numeric calendar month
      • Scaled to [0,2*pi] - cyclic (sin, cos)
    • Sex, reference group: male
    • Age group (10 years), reference group: 30-39 years
    • Mean average daily temperature
      • Piecewise linear spline, 1 internal knot at 18 degrees Celsius
    • Seasonal influenza activity
    • GISD regional socioeconomic deprivation index
    • Offset: log person-time (population*days per month)
      • Population data with last observation carried forward when mortality data is newer than the last available population data
  • Prediction
    • Excess mortality (mortality deficit) = observed count - predicted count
    • Prediction uncertainty based on 25%/75% (2.5%/97.5%) quantiles of posterior-predictive distribution from all 10.000 MCMC draws, taking into account uncertainty in model parameter estimates as well as expected random variation in counts following a negative binomial distribution.
    • Excess mortality can also be displayed after direct age standardization based on a reference population (Germany 2011 by default). The result is not a true rate since excess mortality can be negative, representing mortality deficit. The result has no confidence intervals since their calculation requires nonnegative counts.

Different data sources for population counts

For 2015-2023, population data is available as the verified annual count at the reference date of December 31st. Monthly counts are derived from annual counts by linear interpolation between adjacent end-of-year counts, stratified by district, sex, and age group. For 2019 onward, preliminary monthly counts are also available. Since official registration of residents may be processed with an unknown delay, monthly counts have some uncertainty.

For 2019-2023, population data is available both as the verified end-of-year counts, and as preliminary monthly counts. The observed differences between preliminary monthly counts and linearly interpolated annual counts are minor relative to population size. With a Poisson GLM regression model, we calibrated the 2019-2023 monthly counts to the linearly interpolated end-of-year counts using age group as a covariate and log monthly population count as offset. Predictions from this regression model were used as the estimate of the monthly population counts for 2019-2024 in all analyses.

Influenza activity corrected for public holidays

Weekly seasonal influenza activity can be biased downward on national holidays due to low reporting when physician practices are closed. For Christmas and Easter, we correct for this effect with the following heuristic:

If the Influenza activity in the calendar week of the national holiday is lower than both the preceding and the following calendar week, this dent is replaced by linear interpolation of the values for the adjacent calendar weeks.

Average district-specific monthly temperature

To calculate the district-specific average monthly temperature, we use daily recordings from all meteorological stations maintained by Agricultural meteorology service Rhineland-Palatinate and by DWD . The daily mean temperature is averaged over all stations located in a district over days of the month. Since there is no station located in district Zweibrücken, the temperature is imputed by taking the average temperature of the remaining districts in the Landesregion Pfalz, i.e., Südwestpfalz, Kusel, Kaiserslautern.

Background

Project

Mortality monitoring during the COVID-19 pandemic in Rhineland-Palatinate

Wollschläger D, Schmidtmann I, Fückel S, Blettner M, Gianicolo E. Explaining the age-adjusted excess mortality with COVID-19-attributed deaths from January 2020 to July 2021. Bundesgesundheitsblatt 2021. doi 10.1007/s00103-021-03465-z.

Wollschläger D, Fückel S, Blettner M, Gianicolo E. Übersterblichkeit im Kontext der COVID-19-Pandemie in Deutschland. Die Kardiologie 2024. doi 10.1007/s12181-024-00666-z

Summary

The total impact of COVID-19 on public health includes short-term and long-term effects as well as direct and indirect effects on population mortality. A lack of testing can lead to undercounting of SARS-CoV-2 infections. Therefore, total COVID-19 associated mortality cannot be solely derived from the number of individual deaths that were explicitly attributed to COVID-19 when the deceased had a confirmed SARS-CoV-2 infection. In contrast, uncertainty in cause-of-death assignment and multi-causal pathways to death can lead to overreporting of COVID-19 deaths. Furthermore, there may be changes in health-related behavior, and in delivery of health services because of the pandemic. This may result in indirect effects on mortality by changing mortality patterns from other causes of death.

Therefore, monitoring of spatial and temporal patterns of all-cause mortality as well as of shifts in individual causes of death is required to assess the total impact of COVID-19 on mortality.

Background

  • Deaths caused by COVID-19 may go undetected, and may be wrongly attributed to other causes of death because infected persons were not tested for SARS-CoV-2. Lack of testing would result in underestimation of COVID-19 mortality. Deaths that are a late complication of COVID-19 disease long after hospitalization for the infection may not be attributed to COVID-19.
  • Cause-of-death attribution can be uncertain when multiple causal factors contribute to the death of a person. To an unknown extent, deaths caused by COVID-19 may affect patients of old age or with severe co-morbidities who had limited life expectancy even without COVID-19. On the population level, this would lead to overestimating the net effect of COVID-19 on long-term population mortality because COVID-19 mortality would be partially compensated by a delayed reduction in mortality from competing risks of death (harvesting effect).
  • Registration of mortality due to any cause is independent from SARS-CoV-2 testing and from uncertainty in cause-of-death assignment. It can therefore be carried out objectively and consistently across regions. Uncertainty with respect to all-cause mortality usually concerns only the precise date of death, and attribution to the location of residence vs. location of the reporting institution.
  • The COVID-19 pandemic may have had a number of indirect effects on population mortality that are not reflected in mortality directly caused by COVID-19:
    • Perceived increased risk of SARS-CoV-2 infection in hospitals may have made people reluctant to seek medical attention for early symptoms of potentially fatal disease. An increase in severe diseases initially diagnosed in later stages would lead to delayed higher mortality from causes of death such as cancer, myocardial infarction or stroke.
    • Allocation of hospital capacity towards treatment of COVID-19 patients may have led to disruptions in health care for other patients, potentially affecting mortality. For example, delayed surgeries may have reduced the immediate mortality, but may have increased delayed mortality.
    • Non-pharmaceutical interventions (social distancing) have led to changes in behavior with reduced mobility which may have affected the frequency of accidents.
    • Psychological distress due to perceived health threats, social isolation, or financial burden following reduced income may have affected the frequency of deaths of despair. Conversely, reduced pressure due to work or school combined with more spare time for family members, hobbies, and sport may have had beneficial effects.
    • Less industrial activity and less traffic may have resulted in reduction of air pollution levels.

For the given reasons, it can be informative to complement case-based mortality statistics for COVID-19 with an analysis of spatial and temporal trends in all-cause mortality at the population level irrespective of attribution to COVID-19. This allows to assess the sum of the direct and indirect effects of the pandemic on public health with respect to mortality.

The results of this study will help understand the severity of the COVID-19 pandemic for population mortality in Rhineland-Palatinate on a NUTS-3 level (district), and will shed light on the question whether there was under- or overreporting in official statistics on COVID-19 mortality.

Research questions

  • Are reported COVID-19 case fatality numbers consistent with population mortality in the same time period, or is there evidence for an undetected excess mortality associated with COVID-19?
  • Is there evidence for a temporal shift in mortality in the sense of early excess mortality due to COVID-19 that is compensated by a delayed reduction in mortality due to other causes of death?
  • Are there indirect effects of the COVID-19 pandemic via non-pharmaceutical interventions or changes in health-related behavior on mortality due to individual causes of death other than COVID-19?

Project contributors

People

Software

Funding

Other resources on mortality monitoring

National

European