The total disease burden due to acute COVID-19 in the Netherlands was overwhelmingly determined by premature mortality (> 99% of DALY is YLL), in particular from age 35 and up (Fig.
2). The disease burden was unequally distributed over age, with half of the total burden experienced by persons aged 80+ years and with comparatively little burden among persons under 50 years old. The absolute disease burden grew more slowly between our two analysis periods (increasing by 33%), although the estimated cumulative incidence of infection had greatly increased (by 66%). The DALY/1000 infected person measure for the first analysis period (which approximately corresponds to the first wave) was four-fold that estimated for the rest of the year (Supplemental Materials, Fig. S3). This is most likely due to changes in the age-distribution of infected persons (as demonstrated by successive PICO rounds [
12]), plus a contribution from improvements in COVID-19 patient prognosis, with as consequence a favourable impact on the risk of severe or fatal outcomes.
Using the relative disease burden measure (DALY/100,000 population), we could compare the per-capita burden between different strata of the population. The (age-aggregated) estimated burden experienced by healthcare workers (approximately 1400 DALY/100,000; Supplemental Materials, Fig. S4) was an order of magnitude lower than the burden experienced by the oldest segment of the population (e.g., 18,500 DALY/100,000 for the age-groups 85–89 years and older; Fig.
3). Although analysis of testing data between June and October 2020 in the Netherlands showed that the occupation sectors catering, public transportation and contact professions had relatively high positivity rates [
26], this did not appear to translate to an increased disease burden for these occupations.
Comparison with estimated COVID-19 burden in other countries
It is important to set the Netherlands estimates into the European and international context. To date DALY estimates using the COVID-19 burden protocol [
6] have been produced for Scotland, Germany and Malta for 2020 [
31‐
32]. We could therefore compare the COVID-19 burden in the Netherlands to that estimated for these three countries. The disease burden per 100,00 population in the Netherlands was estimated at 1640 DALY (95% CI: 1620–1670). Table
3 shows how this figure compares with other countries' estimates using similar approaches. We note that interpretation of YLD differences between countries should recognise differences in testing policies and behaviour, and for interpreting YLL differences one must consider potential differences in COVID-19 death under-reporting rates. Among the four countries, Scotland has reported the highest per capita COVID-19 burden – this estimate includes burden due to post-acute consequences – and Germany reported the lowest per-capita burden. The per capita burden estimate for the Netherlands is 4.5 times greater than for Germany, in part due to differences in normative life expectancies. When YLL for Germany is also calculated using the GBD-2019 tables, YLL is 1.5 times higher (A. Wengler, pers. comm.), increasing the relative disease burden from 368 to 542 DALY/100,000. Although GBD-2019 life expectancy values were used for both the Scotland and Netherlands estimates, YLL/100,000 is 13% lower in the Netherlands as compared to Scotland, despite the fact that the number of COVID-19 fatal cases per 100,000 was quite similar in the two countries. This suggests that the average age at death is younger in Scotland. In summary, mortality appears to be the main driver of these between-country differences. Given that the Netherlands, Scotland and Germany have broadly similar demographics, differences in the DALY per 100,000 measure reflect relative success in protecting the elderly and vulnerable segment of the population from SARS-CoV-2 infection.
Table 3
Between-country comparison of COVID-19 disease burden
Netherlands [2020] | Yesa (evidence synthesis) | GBD-2019 | Statistics Netherlands registered (confirmed + suspected) | No | 1640 (1620–1670) | 1% |
Scotland [2020] | Yes (SEIR modelling) | GBD-2019 | Death registry (confirmed only or confirmed + suspected) | Yes, limited | 1770–1980 | 2% |
Germany [2020] | No (notified positives only) | Germany 2016/2018 | Deaths among notified cases | No | 368 | 0.7% |
Malta [7 Mar 2020—31 Mar 2021] | Yes (notified positives adjusted for underascert.) | GBD-2019 | Daily COVID-19 bulletins issued by Malta Ministry of Health | Yes, limited | 1086b | 5% |
Our estimates covered the calendar year 2020, to facilitate comparison with the COVID-19 burden in other countries, and with the routinely reported burden of other infectious diseases in the Netherlands. However, although the peak of the second wave (based on notification data) was in December 2020, the end of this wave occurred around the end of January [
1], and mortality among persons infected during the last part of the second wave would be observed until approximately the end of February. Therefore, based on published mortality figures [
33] we estimated the additional YLL until the end of the second wave (in January 2021), and also when including the associated fatal cases (1 January until 28 February 2021). These were 56,200 (95% CI: 53,800 -58,600) and 91,700 (95% CI: 88,700–94,900) DALY in January 2021 and January/February 2021, respectively.
As we have shown, for meaningful across-country and between-disease DALY comparisons, the same life-table must be used in the calculation of YLL. Furthermore, we have chosen to use the aspirational life expectancy approach, for which age at death is the only relevant factor. When DALY are used to inform public health decision-making, it is important that certain subpopulations (whether defined by a higher prevalence of comorbidities, lifestyle risk factors, or degree of socio-economic deprivation) are not disadvantaged for receipt of prevention or treatment interventions because they have a lower expected remaining life expectancy than other subpopulations [
34].
Strengths of this study include making use of all relevant data sources to estimate the disease burden, and the adoption of a developed protocol for estimation of the COVID-19 disease burden. We have identified the following limitations. First, the total disease burden for the period until 31 December 2020 presented here is known to underestimate the true burden because health outcomes following the resolution of acute infection (i.e., sequelae, often known as 'long COVID') have not yet been included. Current knowledge regarding post-COVID-19 syndrome is that it can be described as constellations of symptoms affecting different physiological systems that can vary in severity and duration [
35], but early estimates indicate its contribution to the total disease burden is on the order of 1–3% [
30,
32]. As more data on progression risk, severity, and duration come available [
36,
37], the current estimates can be revised to include the burden attributable to the long-term sequelae of SARS-CoV-2 infection.
Second, the estimated relative disease burden per occupation category must be interpreted with caution as there were limitations to the data sources and the consequent possible analyses. For a given occupation category, the relative disease burden was estimated for the entire analysis period and is not necessarily indicative of the recent burden; for instance, widespread availability of personal protective equipment and other risk-reducing measures may mean that the proportion of burden experienced by healthcare workers over the last half of the year was likely much reduced. Related to this point, DALY per occupation category was derived using the distribution of notified cases over the entire year, thus aggregating together periods of relatively 'open' society with periods in which strict measures were in place. The procedure also combined periods in which there was non-universal access to testing (i.e., before 1 June 2020, priority was given to severe/hospitalised cases) and/or priority testing for certain occupations, such as healthcare workers and the education sector, and so the distribution of occupation categories among notified cases is influenced by access to testing; with periods in which there were minimal public health restrictions in place, with (near) universal access to testing.
When strict measures were in place, some occupations could not be practiced; for others, contact patterns and ensuing transmission risk in the workplace setting might be quite different. As an example, the proportion in category 'education' will not be fully representative of the normal term-time situation with in-person teaching, due to the (partial) continuation of online teaching after 1 June 2020, and the school holiday period. A further assumption is that the occupation provided in a notified case's Osiris record applied throughout the analysis period (i.e., person was not (temporarily) inactive in their occupation, and did not become unemployed). In addition, our approach did not take into account possible variation in the risk of severe disease and/or mortality by occupation, because the occupation distribution (per age-group) is applied to the total burden (for that age-group). For instance, if (conditioning on age) healthcare workers have better underlying health and therefore better prognosis [
38], or are more skilled in risk perception and personal health management, compared with other occupations, then both the absolute and relative disease burden will have been overestimated for this occupation category.
Third, our estimate of the cumulative SI incidence depends on the age-specific attributable risk derived symptomatic proportion. This method estimates the proportion of infections for which symptoms can uniquely be attributed to SARS-CoV-2 infection and as such represents an lower bound for the true proportion; mild symptoms that also occur with other afflictions (e.g., common cold, hay fever) are thus discounted. This would lead to an underestimation of YLD, but would have a very small impact on DALY due to the overwhelming contribution of YLL. Finally, disability duration post-hospital/ICU discharge (time until recovery) was not estimated or included, which would also lead to an underestimation of YLD, and improvements in treatment over time, potentially leading to short hospital stay, were not considered.
We have presented the real-world disease burden estimates, i.e., as derived from infections that occurred during a period in which (partial) lockdown measures were in place more often than not. We did not attempt to calculate counterfactuals – what would the disease burden have been if no measures were imposed? How much could the burden have been reduced if stricter measures were taken, or at earlier stages of the epidemic? Although such alternative scenarios are potentially useful for evaluation and future planning, for a number of infectious agents – whether the cause of large historical outbreaks or endemic situations—widespread population interventions were not been considered feasible, and so the best use of disease burden estimates is to inform planning and prioritisation based on the data generated by real-world situations.
The primary focus of this work is the morbidity and mortality
directly caused by SARS-CoV-2. The impact of health care displaced or delayed by COVID-19 patients (i.e., the indirect impact of the pandemic) has been calculated to be on the order of 34,000 to 50,000 lost healthy-life years (QALY) up to 31 August 2020 [
39]. In addition, the imposition of non-pharmaceutical control measures such as social distancing and lockdown has almost certainly had a toll on mental health, the burden of which still needs to be estimated.