Belgian COVID-19 Mortality, Excess Deaths, Number of Deaths per Million, and Infection Fatality Rates (8 March - 9 May 2020) ============================================================================================================================ * Geert Molenberghs * Christel Faes * Jan Aerts * Heidi Theeten * Brecht Devleesschauwer * Natalia Bustos Sierra * Toon Braeye * Françoise Renard * Sereina Herzog * Patrick Lusyne * Johan Van der Heyden * Herman Van Oyen * Pierre Van Damme * Niel Hens ## Abstract **Objective** Scrutiny of COVID-19 mortality in Belgium over the period 8 March – 9 May 2020 (Weeks 11-19), using number of deaths per million, infection fatality rates, and the relation between COVID-19 mortality and excess death rates. **Data** Publicly available COVID-19 mortality (2020); overall mortality (2009 – 2020) data in Belgium and demographic data on the Belgian population; data on the nursing home population; results of repeated sero-prevalence surveys in March-April 2020. **Statistical methods** Reweighing, missing-data handling, rate estimation, visualization. **Results** Belgium has virtually no discrepancy between COVID-19 reported mortality (confirmed and possible cases) and excess mortality. There is a sharp excess death peak over the study period; the total number of excess deaths makes April 2020 the deadliest month of April since WWII, with excess deaths far larger than in early 2017 or 2018, even though influenza-induced January 1951 and February 1960 number of excess deaths were similar in magnitude. Using various sero-prevalence estimates, infection fatality rates (IFRs; fraction of deaths among infected cases) are estimated at 0.38 - 0.73% for males and 0.20 – 0.39% for females in the non-nursing home population (non-NHP), and at 0.79 – 1.52% for males and 0.88 – 1.31% for females in the entire population. Estimates for the NHP range from 38 to 73% for males and over 22 to 37% for females. The IFRs rise from nearly 0% under 45 years, to 4.3% and 13.2% for males in the non-NHP and the general population, respectively, and to 1.5% and 11.1% for females in the non-NHP and general population, respectively. The IFR and number of deaths per million is strongly influenced by extensive reporting and the fact that 66.0% of the deaths concerned NH residents. At 764 (our re-estimation of the figure 735, presented by “Our World in Data”), the number of COVID-19 deaths per million led the international ranking on May 9, 2020, but drops to 262 in the non-NHP. The NHP is very specific: age-related increased risk; highly prevalent comorbidities that, while non-fatal in themselves, exacerbate COVID-19; larger collective households that share inadvertent vectors such as caregivers and favor clustered outbreaks; initial lack of protective equipment, etc. High-quality health care countries have a relatively older but also more frail population [1], which is likely to contribute to this result. **Thumbnail summary: What this paper adds** *COVID-19 mortality and its relation to excess deaths, case fatality rates (CFRs), infection fatality rates (IFRs), and number of deaths per million are constantly being reported for a large number of countries globally*. *This study adds detailed insight in the Belgian situation over the period 8 March – 9 May 2020 (Week 11–Week 19)*. *Belgium has virtually no discrepancy between COVID-19 reported mortality (confirmed and possible cases) and excess mortality. This, combined with a high fraction of possible cases that is COVID-19 related [2] provides a basis for using all COVID-19 cases and thus not only the confirmed ones, in IFR estimation*. *Against each of the years from 2009 and 2019 and the average thereof, there is a strong excess death peak in 2020, which nearly entirely coincides with confirmed plus possible COVID-19 cases. The excess death/COVID-19 peak rises well above seasonal fluctuations seen in the first trimester during the most recent decade (induced in part by seasonal influenza). In the second week of April 2020, twice as many people died than in the corresponding week of the reference year. April 2020 was the deadliest month of April since WWII, although January 1951 and February 1960 saw similar figures. More recently, in the winter of 2017-2018, there was 4*.*6% excess mortality in Belgium (70,215 actual deaths; 3093 more than the Be-MOMO-model prediction). In the winter of 2016-2017, there was an excess of 3284 deaths (4*.*9% excess mortality)* [https://epistat.wiv-isp.be/docs/momo/Be-MOMO%20winter%202017-18%20report_FR.pdf](https://epistat.wiv-isp.be/docs/momo/Be-MOMO%20winter%202017-18%20report_FR.pdf). *At 764 (our estimate), the number of COVID-19 deaths per million leads the international ranking, but drops sharply to 262 in the non-nursing home population*. *CFR is not a good basis for international comparison, except as a tool in estimating global infection fatality rates [2]. These authors used asymptotic models to derive IFR as a limit of CFR. CFR is strongly influenced by testing strategy, and in several studies the delay between case confirmation and deaths is not accounted for. The handling of possible cases is ambiguous at best. We do not consider it here*. *Bias and precision in estimation of IFR is influenced by difficulties surrounding the estimation of sero-prevalence, such as sensitivity and specificity of the tests used [3], time to IgM and in particular IgG seroconversion [4], and potential selection bias occurring in data from residual sample surveys. A sensitivity analysis is undertaken by augmenting one primary with three auxiliary estimates of sero-prevalence*. *Because in Belgium there is a very close agreement between excess mortality on the one hand and confirmed and possible COVID-19 cases combined on the other, and because an international study [2] suggested that a fraction as high as 0*.*9 of possible cases could be attributable to COVID-19 [5], it is a reasonable choice to use all COVID-19 cases in IFR estimation. This encompasses a large fraction of deaths occurring in nursing homes. The IFR values obtained align with international values [2]. Using various sero-prevalence estimates, IFRs across all ages are estimated at 0*.*38 - 0*.*73% for males and 0*.*20 – 0*.*39% for females in the non-nursing home population (non-NHP), and at 0*.*79 – 1*.*52% for males and 0*.*88 – 1*.*31% for females in the entire population. Estimates for the NHP range from 38 to 73% for males and over 22 to 37% for females. The IFRs rise from nearly 0% under 45 years, to 4*.*3% and 13*.*2% for males in the non-NHP and the general population, respectively, and to 1*.*5% and 11*.*1% for females in the non-NHP and general population, respectively*. *The IFR is strongly influenced by extensive death cases reporting and the fact that 66*.*0% of the deaths concerned NH residents. Apart from a strong age-related gradient, also for each age category, IFRs are substantially higher in males than in females Because of these dependencies, IFRs should be considered in an age, gender, and sub-population specific manner. The same proviso is made for the number of deaths per million*. *An important such population is the NHP because of a specific cocktail: age-related increased risk; highly prevalent comorbidities that, while non-fatal in themselves, exacerbate COVID-19; larger collective households that share inadvertent vectors such as caregivers; initial lack of protective equipment, etc. High-quality health care countries have a relatively older but also more frail population [1], which might contribute*. ## Introduction Belgium’s per million COVID-19 related mortality has been reported the highest worldwide over the period April – May 2020. For example, as reported on May 13, 2020 at [https://ourworldindata.org/coronavirus-data](https://ourworldindata.org/coronavirus-data), the figure is 756 for Belgium, versus 91 for Germany, 414 for France, 511 for Italy, 482 for the UK, 249 for the US, and 328 for Sweden. Because of its relative nature, this measure appears to be objective; nevertheless, it requires scrutiny. To this end, we examine COVID-19 reported mortality over the period 8 March – 9 May 2020, and place it against the background of excess mortality in Belgium. The study period is chosen such that there is a sufficiently long data cleaning period, leading to accurate death counts. This allows one to gauge whether there is evidence for over-, under- or sufficiently accurate reporting of COVID-19 cases. Using data on the number of COVID-19 deaths and sero-prevalence estimates based on data from a repeated cross-sectional serological survey [3], infection fatality rates (IFRs) and number of deaths per million (DPM) are estimated, overall, in relation to age and sex, for the total population, the nursing home population (NHP), and the non-NHP. ## Data and Methods ### Covid-19 mortality The Belgium-based institute for health, Sciensano, reports daily COVID-19 mortality figures ([https://epistat.wiv-isp.be/covid/](https://epistat.wiv-isp.be/covid/)). These daily data were extracted on 5 June 2020 and then binned to form age category by week mortality tables for each of the sexes and for the period 8 March - 9 May 2020 (Week 11 – Week 19); age categories (in years) are 0-24, 25-44, 45-64, 65-74, 75-84, and 85+. These six categories are used throughout the analyses. Of the 8732 deaths reported, 3136 (3072) are male (female) and have age category reported. Only 2 men and 1 woman are of unknown age, whereas 11 people have their age but not their sex reported. However, 2510 have neither age nor sex available, of which 96% stem from deaths in the NHP of Flanders, one of the three Belgian Regions. For everyone, the date and hence week of death is reported. Missing data redistribution methods are used to classify all data in an age-week-sex table [6]. Cases with neither age nor sex observed are redistributed in an ad-hoc fashion over the proper week, so as to match the age-sex distribution observed from excess mortality (defined as observed deaths in 2020 minus the average over the 2009 – 2019 period). Redistribution is done into the 65+ age groups only, even though there is a small number of deaths in the age range 45-64. In addition, two sub-populations are considered that jointly comprise the NHP deaths: (a) deaths occurring in NH (4494 deaths, with 1989 complete records), and (b) NH residents that died in hospitals (1276 deaths, with 705 complete records). Based upon considering various re-distribution methods (details not reported), the imprecision engendered by missingness is ignored, because it is dominated by uncertainty in sero-prevalence estimation, the latter of which is taken into account by precision estimation and statistical sensitivity analysis. ### Overall mortality Weekly mortality per sex and age category, for the years 2009 – 2019 (complete) and 2020 (until early May 2020) originate from the National Register. Statistics Belgium, the national statistical institute, processes these deaths and integrates them in Demobel, its demographic data warehouse. Open data by district (NUTS 3) can be found in [3]: [https://statbel.fgov.be/en/open-data/number-deaths-day-sex-district-age](https://statbel.fgov.be/en/open-data/number-deaths-day-sex-district-age). Using the years 2009 – 2019 combined, a weekly average profile (termed reference year) is obtained, with pointwise corresponding 99% prediction bands. ### Population sizes The Belgian population sizes (situation 1 January 2020), by age category and sex, are taken from Statistics Belgium (Demobel), based on National Register data: [https://statbel.fgov.be/en/themes/population/structure-population](https://statbel.fgov.be/en/themes/population/structure-population). ### Sero-prevalences Based on Herzog et al. [3] age-category specific sero-prevalences referring to April 2020, are used; details on sensitivity and specificity of the tests used can be found there. Data over the age range 0-101 years of age are available. A re-analysis of the data used in [3] provided estimates and confidence intervals for the age bins used in this paper. These estimates use the population structure for Belgium in 2020, as forecast by the Federal Plan Bureau ([www.plan.be](https://www.plan.be)). Bias and precision in the determination of sero-prevalence depends on sensitivity and specificity of the tests used [1], time to IgG seroconversion [4], and potential selection bias occurring in data from residual samples surveys. A sensitivity analysis is undertaken by augmenting this primary with three auxiliary estimates of sero-prevalence for the general population, with in addition several forecast of the sero-prevalence in NH starting from extensive test results ([https://covid-19.sciensano.be/sites/default/files/Covid19/COVID-19\_Daily%20report_20200526%20-%20FR.pdf](https://covid-19.sciensano.be/sites/default/files/Covid19/COVID-19_Daily%20report_20200526%20-%20FR.pdf)). ### Estimated number of COVID-19 cases These are calculated by multiplying the age-sex-subgroup (NHP/non-NHP) population sizes with the corresponding sero-prevalences, for various sero-prevalence estimates. Assuming an approximately constant sero-prevalence over the month of May, this is considered a sensible approach, even though no delay-adjustment is done. ### Infection fatality rates From the sero-prevalence and population size, per category, the actual number of infected cases is estimated. The IFR is calculated as the ratio of the number of deaths (confirmed and possible) over the number of infected cases. This is done per age and sex category, for the general population, the NHP, and the non-NHP. Given the uncertainty in the sero-prevalence estimates, delta-method-based confidence intervals supplement the primary IFR estimates. Case fatality rates (CFR), defined as the number of confirmed deaths (confirmed and possible) over the number of confirmed cases, ideally delay-distribution adjusted, will not be examined in this manuscript. ### Statistical software The data analysis was performed using SAS Software, GAUSS, and R; visualizations were made using Vega. Python scripts to reproduce the analyses will be available at [https://www.uhasselt.be/DSI](https://www.uhasselt.be/DSI). ## Results ### COVID-19 Mortality In line with international findings [7], the number of deaths strongly increases with age. It is difficult to compare sexes in absolute terms, because the higher number of deaths in the female 85+ group, for example, is offset by the fact that the number of males in the 85+ category is less than half in size of the female category. Summary data are given in Table 1 and depicted in Figure 1. Table 2 presents the figures for the NH population. Because of the large fraction of incomplete records, the redistribution over age categories in Table 2 (32.0% incomplete in terms of age and/or sex), is subject to uncertainty. There are more incomplete records in the NH sub-population than in the general population, because the dataset of NH residents who died in hospitals is currently not directly extractable from the hospital-deaths dataset, for which age and sex are virtually complete. View this table: [Table 1.](http://medrxiv.org/content/early/2020/06/20/2020.06.20.20136234/T1) Table 1. All confirmed and possible COVID-19 deaths in Belgium (March 8 – May 9, 2020; weeks 11 – 19), by week, age, and gender, 2524 cases with only partial information on age or gender are redistributed over the 65+ strata. ![Figure 1.](http://medrxiv.org/https://www.medrxiv.org/content/medrxiv/early/2020/06/20/2020.06.20.20136234/F1.medium.gif) [Figure 1.](http://medrxiv.org/content/early/2020/06/20/2020.06.20.20136234/F1) Figure 1. Graphical representation of the age and sex specific COVID-19 mortality curves in Belgium over the period 8 March – 9 May 2020. View this table: [Table 2.](http://medrxiv.org/content/early/2020/06/20/2020.06.20.20136234/T2) Table 2. Confirmed and possible COVID-19 deaths within the Belgian nursing home population (March 8 – May 9, 2020; weeks 11 – 19), by week, age, and gender, 3076 cases with only partial information on age or gender are redistributed over the 65+ strata. ### Excess deaths Figures 2-3 display mortality in each of the years 2009-2019, as well as on average over these years. The excess in 2020 is apparent; the peak rises well over the seasonal variation seen in the first trimester of 2009-2019, and lies clearly outside the 99% pointwise prediction bands. Mortality was high in the winter seasons of 2011-2012, 2012-2013, 2014-2015, 2016-2017, and 2017-2018, unlike in 2013-2014 and 2015-2016. Then, in the second week of April 2020, twice as many people died than on average over 2009-2019. April 2020 was the deadliest month of April since WWII, although January 1951 and February 1960 saw similar figures. Detailed long-term mortality studies for Belgium have been reported elsewhere [5,8]. More recently, in the winter of 2017-2018, there was 4.6% excess mortality in Belgium (70,215 actual deaths; 3093 more than the Belgian Mortality Monitoring (Be-MOMO) model prediction). In the winter of 2016-2017, there was an excess of 3284 deaths (4.9% excess mortality) ([https://epistat.wiv-isp.be/docs/momo/Be-MOMO%20winter%202017-18%20report_FR.pdf](https://epistat.wiv-isp.be/docs/momo/Be-MOMO%20winter%202017-18%20report_FR.pdf); [https://epistat.wiv-isp.be/momo/](https://epistat.wiv-isp.be/momo/)) ![Figure 2.](http://medrxiv.org/https://www.medrxiv.org/content/medrxiv/early/2020/06/20/2020.06.20.20136234/F2.medium.gif) [Figure 2.](http://medrxiv.org/content/early/2020/06/20/2020.06.20.20136234/F2) Figure 2. Belgium. Excess deaths (overall). Grey curves refer to years 2009 – 2019; the black curve is the average over 2009 – 2019, with dashes 99% pointwise prediction bands; the red curve refers to 2020. ![Figure 3.](http://medrxiv.org/https://www.medrxiv.org/content/medrxiv/early/2020/06/20/2020.06.20.20136234/F3.medium.gif) [Figure 3.](http://medrxiv.org/content/early/2020/06/20/2020.06.20.20136234/F3) Figure 3. Belgium. Excess deaths (by age and sex category). Grey curves refer to years 2009 – 2019; the black curve is the average over 2009 – 2019; the red curve refers to 2020. That the peak is strongly driven by the older age category is clear from Figure 3. ### Covid-19 Mortality Versus Excess Deaths In the above, reference was made to Belgium’s extensive COVID-19 death reporting, as also internationally noted. Figure 4 compares COVID-19 mortality with excess deaths in Belgium, from the beginning of year until the end of April 2020. ![Figure 4.](http://medrxiv.org/https://www.medrxiv.org/content/medrxiv/early/2020/06/20/2020.06.20.20136234/F4.medium.gif) [Figure 4.](http://medrxiv.org/content/early/2020/06/20/2020.06.20.20136234/F4) Figure 4. Belgium. Weekly COVID-19 versus excess mortality. The red curve represents overall observed weekly mortality in Belgium. The grey curve is the reference-year mortality (average of 2009 – 2019). The black curve is reference + COVID-19 mortality. The coloured bands sketch the fractions taken in reference years by major causes of death. ([https://www.healthybelgium.be/en/health-status/mortality-and-causes-of-death/overall-mortality-by-cause](https://www.healthybelgium.be/en/health-status/mortality-and-causes-of-death/overall-mortality-by-cause)) As we saw in Figure 2, mortality in January and February 2020 was below the average over 2009-2019, although coherent with the prediction interval. Then, the peak emerges and, most important for this study is the near coincidence of excess and COVID-19 mortality. ### Covid-19 Number of Deaths per Million Inhabitants Now that the close agreement between COVID-19 and excess mortality has been established, it is useful to consider the number of deaths per million inhabitants (DPM). Belgium’s DPM has been reported as the highest worldwide over the period April – May 2020. For example, according to [https://ourworldindata.org/coronavirus-data](https://ourworldindata.org/coronavirus-data) (accessed on June 16, 2020), the figures are: 735 for Belgium on 9 May 2020, versus 500 for Italy, 88 for Germany, 402 for France, 460 for the UK, 233 for the US, and 314 for Sweden. Table 3 displays age- and sex-specific DPM, for the general Belgian population, the population without nursing homes, as well as the NHP. Because the Belgian mortality data have been cleaned and verified between the end of the study period (May 9, 2020) and final analysis (June 16, 2020), the overall figure of 764 in Table 3 differs slightly from the “Our World in Data” figure of 735. The population fractions in NHP are ([http://www.ima-aim.be](http://www.ima-aim.be)): 1.0% in 65-74, for both males and females; 3.2% [M] and 5.7% [F] in 75-84; 14.0% [M] and 27.7% [F] in 85+. View this table: [Table 3.](http://medrxiv.org/content/early/2020/06/20/2020.06.20.20136234/T3) Table 3. Number of COVID-19 deaths per million inhabitants in Belgium on 9 May 2020, per age and sex category: (a) for the general population; (b) [restricted to non-nursing home population]; (c) (restricted to nursing home population) From Table 3 we deduce a strong age and important sex effect. But what is most striking is the extent of the impact in the NHP. For the non-NHP, the overall figure goes down to 262, which internationally might not stand out, but then reliable figures are needed of other countries’ NHP/non-NHP as well. For a nursing home population of around 1% of the total population, this effect is striking. It is clear from Table 3 that the overall number is not very informative, but rather an age, sex, and population-specific breakout is necessary. For a coherent interpretation, the IFR need to be considered as well. ### Infection Fatality Rates IFR are displayed in Table 6, along with supporting quantities (population sizes, estimated number of cases and sero-prevalences) in Tables 4 and 5. We use in the numerators not only (lab) confirmed cases, but also possible cases. This would create difficulty for CFR estimation (as the denominator would be lab-confirmed cases only), but is a sensible choice for IFR estimation, especially because confirmed plus possible COVID-19 deaths nearly coincide with the excess death rates in Belgium (Figure 4). This coincidence is not a proof for the fact that all excess deaths are COVID-19 related, although it has been reported internationally that around 90% of possible cases are proper COVID-19 [2]. While it may be possible, for example, that some excess deaths are related to other factors, such as lockdown-induced stress, the plausible assumption is made that this effect on mortality is minor. Further examination is warranted as soon as the cause-specific mortality database becomes available, typically after a three-year interval. We make no claims regarding non-mortality related effects. View this table: [Table 4.](http://medrxiv.org/content/early/2020/06/20/2020.06.20.20136234/T4) Table 4. Belgian population (1 January 2020) and estimated COVID-19 cases (around 1 May 2020), by sex and age category (restricted to nursing home population; because of uncertainty about the sero-prevalence, the number of estimated cases is represented as a range over values obtained by several estimates).The nursing home population is estimated based on IMA proportions. View this table: [Table 5.](http://medrxiv.org/content/early/2020/06/20/2020.06.20.20136234/T5) Table 5. Belgium. Estimated sero-prevalences (95% confidence intervals) by Herzog et al. [3]; and based on nowcasting of the 31 March 2020 figures. In the first case, sero-prevalence in the NHP is replaced by a range, to reflect uncertainty and likely higher prevalence. View this table: [Table 6.](http://medrxiv.org/content/early/2020/06/20/2020.06.20.20136234/T6) Table 6. Belgium. Infection fatality rates (%), with 95% confidence intervals, by sex and age category: (a) for the overall population; (b) [restricted to non-nursing home population]; (c) {restricted to nursing home population}.Quantities are based on sero-prevalence estimates in the left-hand column of Table 5. Because of uncertainty about sero-prevalence in the NHP, the confidence interval is replaced by a statistical sensitivity analysis range. Age- and sex-specific IFRs are presented in Table 6, based on the sero-prevalence estimates of Herzog et al. [3] for the non-NHP. For the NHP, these sero-prevalences are replaced by a range, defined by the values based on [3] and 4%/0.70*3=17.1%, where 4% is the percentage of positive tests obtained from extensive testing in nursing homes ([https://covid-19.sciensano.be/sites/default/files/Covid19/COVID-19\_Daily%20report\_20200526%20-%20FR.pdf](https://covid-19.sciensano.be/sites/default/files/Covid19/COVID-19\_Daily%20report\_20200526%20-%20FR.pdf)), 0.70 reflects test sensitivity and 3 is an ad-hoc scale factor. Of note, bias and precision in IFR estimates are influenced by issues surrounding the estimation of sero-prevalence, such as sensitivity and specificity of the tests used [3], time to IgG seroconversion [4,11,12] detectability [11] and clearance [12], and potential selection bias occurring in data from residual sample surveys. The overall values are re-estimated and reported in Table 7 based on three alternative sero-prevalence estimates, to enable a statistical sensitivity analysis, based on three alternative sero-prevalence estimates. The first rests upon the use of delay distributions and projected epidemic evolution, using sero-prevalence figures of 31 March 2020 thus projected forward until about 1 May 2020. The second uses Sciensano’s blood donor based sero-prevalence estimate of 4.7% (Belgian Red Cross data; [https://covid-19.sciensano.be/sites/default/files/Covid19/COVID-19\_Weekly%20report\_20200529%20-%20FR\_0.pdf](https://covid-19.sciensano.be/sites/default/files/Covid19/COVID-19_Weekly%20report_20200529%20-%20FR_0.pdf)). The third one uses an ad-hoc correction to the Herzog et al. [3] estimate, based on a prolonged onset of detectability of IgG; the sero-prevalences are inflated by a factor 1.5 and consequently the IFR deflated. The IFRs in Tables 5 and 6 combined reflect sampling variability and allow to gauge the impact of biased estimation in sero-prevalence. View this table: [Table 7.](http://medrxiv.org/content/early/2020/06/20/2020.06.20.20136234/T7) Table 7. Belgium. Statistical sensitivity analysis of infection fatality rates across all ages (%): (I) using nowcasting starting from March 2020 sero-prevalence survey; (II) (using Scienano’s blood donors prevalence of 4.7%, based on Belgian Red Cross data); (III) [applying a 150% inflation factor to the March and April 2020 survey based sero-prevalence]. Figures are shown for three population: (a) for the overall population; (b) [restricted to non-nursing home population]; (c) {restricted to nursing home population}. Figure 5 displays the IFR represented in Table 6, for the entire population, as well as for the non-NHP. ![Figure 5.](http://medrxiv.org/https://www.medrxiv.org/content/medrxiv/early/2020/06/20/2020.06.20.20136234/F5.medium.gif) [Figure 5.](http://medrxiv.org/content/early/2020/06/20/2020.06.20.20136234/F5) Figure 5. Age and sex specific COVID-19 IFR in Belgium, with average IFR across ages, for each sex, for the general population and for the non-NHP. Based on the various sero-prevalence estimates, IFRs across all ages are estimated at 0.38 - 0.73% for males and 0.20 – 0.39% for females in the non-nursing home population (non-NHP), and at 0.79 – 1.52% for males and 0.88 – 1.31% for females in the entire population. Estimates for the NHP range from 41 to 73% for males and over 24 to 37% for females. The IFRs rise from nearly 0% under 45 years, to 4.3% and 13.2% at 85+ for males in the non-NHP and the general population, respectively, and to 1.5% and 11.1% for females in the non-NHP and general population, respectively. In the NHP, the IFR may well be above 50% in males and about 30% in females of 85+. The IFR is strongly influenced by extensive death cases reporting and the fact that 66.0% of the deaths concerned NH residents. Apart from a strong age-related gradient, also for each age category, IFRs are substantially higher in males than in females. The average, obtained from pooled figures, shows the reverse ordering in the overall population (1.2% in males and 1.3% in females), although the difference is non-significant. In the non-NHP, the effect is reversed (0.6% in males, 0.3% in females). This phenomenon, known as Simpson’s paradox, stems from the fact that age-specific IFRs are lower in females, but the age-specific population size in older age groups is larger in females (e.g., 200,000 females versus 100,000 males in 85+). Like with DPM, also the IFR strongly differs between the NHP and the non-NHP. A cautionary remark is in place. The NHP is relatively small and it is exactly in this population that age and sex suffers from incompletes. The more reliable IFR estimates, therefore, are for the overall and non-NHP populations, because of stable denominators, but for the NHP, the age and/or sex specific values should be interpreted qualitatively, in terms of ranges, only. In the NHP the overall IFR climbs to 28 – 45%. The difference between male and female IFR for 85+ (to a lesser extent also in 75-84) appears to be more pronounced in the non-NHP than in the general population. This is another instance of Simpson’s paradox: At 85+, the female population is twice in size the male population, but in nursing homes, there are four times as many females than males. Thus, the general population at 85+ is mixed quite differently over NH and non-NH settings, with the NHP the more vulnerable fraction. ## Discussion Some of Belgium’s figures, are higher than those reported by other countries. A single metric that underscores this is the country-specific number of deaths per million. It is nevertheless simplistic and requires careful qualification. There are several contributing causes. (1) The definition of COVID-19 mortality monitoring varies between countries. Few countries have decided to count possible cases and/or nursing home deaths, which underestimates the DPM and biases international comparisons. If Belgium had reported only confirmed and possible death cases, the DPM would have been halved. (2) The timing of the epidemic plays a role; figures should be compared relative to a well-defined baseline (e.g., 50 days since the first day at which the DPM exceeded 1.0) rather than calendar time. This would produce, for example: Belgium, 8 May, 726; Italy, 24 April, 423; Germany, 12 May, 90; France, 7 May, 443; UK, 7 May, 443; US, 11 May, 240; and Sweden, 10 May, 319. (3) The size and geographical dispersion of a country arguably play a role. Large countries with loosely connected regions and/or low population density might see a much slower increase of sero-prevalence but with large regional differences, than a compact and well-connected country. For example, Sweden’s population density is 14 times smaller than Belgium’s. (4) Multiple entry points and major transmission events such as carnival festivities play a role. Spring break in Belgium took place in week 9 for the entire country, whereas in some countries they are spread over two (the Netherlands) or four (Sweden) weeks. Belgium had several clusters simultaneously whereas, for example, in the Netherlands the virus was introduced in the south while the north was relatively spared. The presence of communities with Italian heritage that traveled back from spring breaks is a contributing factor. (5) The varying measures taken by national and regional authorities to fight the epidemic arguably are relevant. (6) An effect that does not show in the overall sero-prevalence, but may account for large differences in DPM, is the age distribution of a country, because of the steep age gradient in IFR, but also in the DPM itself. A relatively older population will lead to considerably more deaths, for constant IFR structure. As reported by Wyper et al.[1], in Belgium 13.1% of the population is 70+ (the European continent extremes being Israel for 7.7% and Italy for 16.4%), among whom further 42.2% are 80+ (with extremes 31.5% in Czechia and 44.5% in France). Larger elderly fractions may have various demographic reasons. To some extent, one of them is that countries with high-quality health care facilities are accommodating to a somewhat more frail elderly population, with underlying comorbidities such as high blood pressure and diabetes that are known to be risk factors for COVID-19 mortality. This point is partially underscores by the higher DPM and IFR in the NHP, for given age and sex. (7) Directly related to this, but worth separate mention, is that the epidemic has been very severe in the NHP, which shows in both the DPM and IFR, suggesting a nuanced explanation. While figures should be interpreted with caution, the IFR in the 85+ NHP appears to be roughly 15 times that in the non-NHP, pointing to increased frailty and higher prevalence of underlying comorbidities. Also, the effect of vectors, such as caregivers, should not be underestimated and protection and preventive measures taken in view of possible future outbreaks. In summary, the very large DPM in the NHP versus the non-NHP, when compared within a given age and sex group, arguably results from a larger sero-prevalence, in combination with an increased IFR. All in all, the outbreak in Belgian nursing homes was extremely serious, in line with international findings [13]. A more detailed study and further international comparison is urgent, as well as the implementation of targeted non-pharmaceutical interventions, while awaiting promising pharmaceutical development. For the general population, the steep age-related gradient in mortality, expressed in IFR or number of deaths per million, contributes useful information to policymakers for differential non-pharmaceutical interventions. It is difficult to compare COVID-19 figures to these of countries that have a less extensive reporting strategy, in particular when the gap between excess deaths and COVID-19 mortality is large, such as in the Netherlands, Italy, or Austria [14]. Excess mortality across countries is a better base for comparison. This is definitely true in view of the very strong difference in DPM for the nursing home population as opposed to the general population, even though the figures are generally very high for the older ages groups. Countries that underreport deaths in nursing homes are therefore not a basis for comparison. Arguably, excess mortality is a better basis for comparison. Currently, EuroMOMO allows for this by means of the Z-score, a useful metric that indicates how unusual mortality is over a given period, relative to average mortality in that same period. It does allow for within-country comparisons only. For example, countries with less extreme variations in the reference period will have a smaller variance and for the same deviation in the epidemic period a larger Z, compared to a country with more variation in the reference period. It may thus be useful to supplement it with other metrics, such as the excess mortality rate, relative to reference mortality. For instance, in the second week of April 2020, mortality is about double the average over 2009 – 2019. Because of its inherent limitations, especially dependence on testing strategy, CFR is a flawed metric for international comparisons [7,15], and should be used with extreme caution. When compared, the delay distribution between confirmation and death should be taken into account [16], also to accommodate under-reporting of cases. The IFR is preferred, even though care should be taken when estimating sero-prevalence, and its sources for bias and uncertainty quantified. This suggests the use of statistical sensitivity analysis, along with the reporting of interval estimates. COVID-19-related mortality figures suggest the seriousness of the epidemic. Belgium’s April 2020 mortality was the highest among all months of April since WWII. In the week of April 5, 2020, COVID-19 mortality was twice as high as long-term average-mortality for that week. ## Data Availability Data can be obtained from the authors upon reasonable request. ## Summary of limitations The large fraction of missing age and sex data adds uncertainty to age and sex specific estimates, especially in those pertaining to the NHP. Should more complete data become available, future adjustments will be possible. The determination of sero-prevalence is naturally surrounded with uncertainty, especially in the NHP, for reasons related to survey sampling and at this point unfolding virological knowledge. ## Competing interest declaration *“All authors have completed the Unified Competing Interest form at [www.icmje.org/coi\_disclosure.pdf](http://www.icmje.org/coi_disclosure.pdf) (available on request from the corresponding author) and declare that (1) PVD reports research grants from GSK Biologicals, Pfizer, SANOFI, Merck, Themis, Osivax, J&J and Abbott, grants from The Bill & Melinda Gates Foundation, PATH, Flemish Government, and European Union, outside the submitted work; (2) none of the other authors has anything to disclose*. ## Role of study sponsors The sero-prevalence study of which the results are used in this manuscript has been sponsored by the University of Antwerp’s Research Fund. ## Other declarations The investigators were independent from the funders; all authors had full access to the data and can take responsibility for the integrity of the data and the accuracy of the data analysis; the lead author affirms that the manuscript is an honest, accurate, and transparent account of the study being reported; that no important aspects of the study have been omitted. Data sharing: data and software code used for the tabular and graphical displays in this study are publicly available from [https://www.uhasselt.be/DSI](https://www.uhasselt.be/DSI)]. This works reflects the views of the authors and not necessarily the official position of the institutions they belong to. ## Acknowledgements We are grateful for the ability to use open data on COVID-19 mortality and cause-specific mortality (Sciensano, Belgium), general mortality and population figures (Statistics Belgium, Demobel; National Register). The data providers hold no responsibility for the analyses reported in this manuscript. We thank Sciensano colleagues Sophie Quoilin, Katrien Tersago, Dominique Van Beckhoven, Nina Van Goethem, and others, for suggesting relevant data sources from among their publicly available data, for useful comments and critical reflections on the analysis strategy, and for comments on earlier drafts of the manuscript. Particular thanks go out to Sciensano colleagues Sara Dequeker and Eline Vandael from the Nursing Homes Surveillance Team. First author GM thanks Barbara Debusschere for suggesting an earlier version of this project. This project has received funding from the European Union’s Horizon 2020 Research and Innovation Programma – Project EpiPose (No 101003688). * Received June 20, 2020. * Revision received June 20, 2020. * Accepted June 20, 2020. * © 2020, Posted by Cold Spring Harbor Laboratory This pre-print is available under a Creative Commons License (Attribution-NonCommercial-NoDerivs 4.0 International), CC BY-NC-ND 4.0, as described at [http://creativecommons.org/licenses/by-nc-nd/4.0/](http://creativecommons.org/licenses/by-nc-nd/4.0/) ## References 1. 1.Wyper GMA, Assunção RMA, Cuschieri S, Devleeschauwer B, et al. Population vulnerability to COVID-19 in Europe: a burden of disease analysis. Arch Public Health 2020; 78:47. 2. 2.Grewelle RE, De Leo GA. Estimating the global infection fatality rate of COVID-19. Preprint 2020. [https://www.medrxiv.org/content/10.1101/2020.05.11.20098780v1](https://www.medrxiv.org/content/10.1101/2020.05.11.20098780v1) 3. 3.Herzog S, De Bie J, Abrams S, Wouters I, Ekinci E, Patteet L, Coppen A, De Spiegeleer S, Beutels P, Van Damme P, Hens N, Theeten H. Seroprevalence of IgG antibodies against SARS coronavirus 2 in Belgium: A prospective cross-sectional study of residual samples. Preprint. doi: [https://doi.org/10.1101/2020.06.08.20125179](https://doi.org/10.1101/2020.06.08.20125179) 4. 4.Borremans B, Gamble A, Prager KC, et al. Quantifying antibody kinetics and RNA shedding during early-phase SARS-CoV-2 infection. Preprint 2020. [https://osf.io/evy4q](https://osf.io/evy4q) 5. 5.Bustos Sierra N, Bossuyt N, Braeye T, et al. All-cause mortality supports the COVID-19 mortality figures in Belgium. Submitted for publication. 6. 6.Molenberghs G, Kenward MG. Missing Data in Clinical Studies. New York: Wiley 2007. 7. 7.Khafaie MA, Rahim F. Cross-country comparison of case fatality rates of COVID-19/SARS-CoV-2. Osong Public Health and Research Perspective 2020; 11:74-80. 8. 8.Eggerickx T, Sanderson JP, Vandeschrick C. Mortality in Belgium from nineteenth century to today. Quetelet J 2020; 8:7-59. 9. 9.Russell TW, Hellewell J, Jarvis CI, et al. Estimating the infection and case fatality ratio for coronavirus disease (COVID-19) using age-adjusted data from the outbreak on the Diamond Princess cruise ship, February 2020. Euro Surveill. 2020: 25. doi: 10.2807/1560-7917 [published online 26 March 2020]. [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.2807/1560-7917&link_type=DOI) 10. 10.Russell TW, Hellewell J, Abbott S, et al. Using a delay-adjusted case fatality ratio to estimate under-reporting. Preprint. [https://fondazionecerm.it/wp-content/uploads/2020/03/Using-a-delay-adjusted-case-fatality-ratio-to-estimate-under-reporting-\_-CMMID-Repository.pdf](https://fondazionecerm.it/wp-content/uploads/2020/03/Using-a-delay-adjusted-case-fatality-ratio-to-estimate-under-reporting-_-CMMID-Repository.pdf) [published online 22 March 2020, retrieved 1 May 2020] 11. 11.Grzelak L, Temmam S, Planchais C, et al. SARS-CoV-2 serological analysis of COVID-19 hospitalized patients, pauci-symptomatic individuals and blood donors. Preprint 2020. [https://www.medrxiv.org/content/10.1101/2020.04.21.20068858v1.full.pdf](https://www.medrxiv.org/content/10.1101/2020.04.21.20068858v1.full.pdf) 12. 12.Huang AT, Garcia-Carreras B, Hitchings MDT, et al. A systematic review of antibody mediated immunity to coronaviruses: antibody kinetics, correlates of protection, and association of antibody responses with severity of disease. medRxiv preprint doi: 10.1101/2020.04.14.20065771. [Abstract/FREE Full Text](http://medrxiv.org/lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6NDoiQUJTVCI7czoxMToiam91cm5hbENvZGUiO3M6NzoibWVkcnhpdiI7czo1OiJyZXNpZCI7czoyMToiMjAyMC4wNC4xNC4yMDA2NTc3MXYxIjtzOjQ6ImF0b20iO3M6NTA6Ii9tZWRyeGl2L2Vhcmx5LzIwMjAvMDYvMjAvMjAyMC4wNi4yMC4yMDEzNjIzNC5hdG9tIjt9czo4OiJmcmFnbWVudCI7czowOiIiO30=) 13. 13.ECDC Public Health Emergency Team, Danis K, Fonteneau L, et al. High impact of COVID-19 in long-term care facilities, suggestion for monitoring in the EU/EEA, May 2020. Euro Surveill 2020 25(22). [https://doi.org/10.2807/1560-7917.ES.2020.25.22.2000956](https://doi.org/10.2807/1560-7917.ES.2020.25.22.2000956) 14. 14.The Economist. Tracking covid-19 excess deaths across countries. [https://www.economist.com/graphic-detail/2020/04/16/tracking-covid-19-excess-deaths-across-countries](https://www.economist.com/graphic-detail/2020/04/16/tracking-covid-19-excess-deaths-across-countries) 15. 15.Porcheddu R, Serra C, Kelvin D et al. Similarity in case fatality rates (CFR) of COVID-19/SARS-CoV-2 in Italy and China. J Infect Dev Ctries 2020; 14: 125-128. doi: 10.3855/jidc.12600. [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.3855/jidc.12600&link_type=DOI) [PubMed](http://medrxiv.org/lookup/external-ref?access_num=32146445&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2020%2F06%2F20%2F2020.06.20.20136234.atom) 16. 16.Baud D, Qi X, Nielsen-Saines K, et al. Real estimates of mortality following COVID-19 infection. Lancet 2020 doi: 10.1016/S1473-3099(20)30195-X [published online 12 March 2020] [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/S1473-3099(20)30195-X&link_type=DOI) [PubMed](http://medrxiv.org/lookup/external-ref?access_num=32171390&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2020%2F06%2F20%2F2020.06.20.20136234.atom)