Workforce Sick-Leave as an Early Indication of Increasing COVID-19 Hospitalizations: Sick-Leave Compared with COVID-19 Hospitalizations

The highest correlation coefficients between sick-leave and all or ICU-only hospitalizations were during the delta period: for the younger age group (18-66 years) being 0.90 and 0.75 for all and ICU-only hospitalizations respectively, for the elderly age group (67+ years) these were 0.91 and 0.93. The lowest correlation coefficients were observed during the total period for the 18-66 years old group, at 0.35 and 0.16 for all and ICU-only hospitalizations. As previously mentioned, we considered the lag where the correlation coefficient is highest, as the optimal lag time. The optimal lag time between sick-leave and hospitalizations and ICU admissions varied per period, the shortest being during omicron with an optimal lag of -1 (i.e., the sick-leave incidence one week before COVID-19 hospitalizations showing the highest correlation coefficient), and longest being lag of -4 during the alpha and delta periods. Generally, both the correlation coefficient and optimal lag time within a period were quite similar between age groups and hospitalization types.

Discussion

Trends in sick-leave were correlated with and preceded all COVID-19 hospitalizations as well as ICU-only admissions. The optimal lag time varied by SARS-CoV-2 variant time period, but was always negative. This implies that sick-leave might offer a 1-to-4week earlier indication of increasing hospitalizations, potentially providing early insight into increasing pressures on hospital care as well as on ICU care.

The correlations of the total period were much lower than the correlations for the separate variant periods. This suggests that the correlation between work sick-leave and COVID-19 hospital and ICU-only admissions is dependent on the type of dominant SARSCoV-2 variant, which differed in infectiousness and severity, population immunity, both infection-induced and by vaccination, and seasonal influences. As for vaccination coverage, there were four different vaccination campaigns during the study period. The first and second dose programme began at the end of the wildtype period (January 2021), where first the elderly and healthcare workers were prioritized for vaccination before other vaccine recipients. Vaccination continued through the alpha period (until July 2021). In November 2021 (the end of the delta period), the programme for the booster vaccination started, and in March and September 2022 (during the omicron period) specific groups (people aged 60 years and older and specific high-risk groups of any age) were invited to receive the booster vaccination [7]. The differing immunity over the study period is expected to explain the lower correlation coefficients for the total and alpha periods. Throughout the total study period, the SARS-CoV-2 seroprevalence (i.e. the percentage of individuals having antibodies) in the population changed from zero to over 95% [8]. Because vaccination primarily protected against severe disease and related hospitalization and to a lesser extent against infection [9,10], correlation coefficients will be impacted when they are calculated over a period that overlaps with changing population immunity. As the vaccine was rapidly introduced into a largely naive population during the alpha period, the same effect can be expected during this period specifically. We found higher correlation coefficients when restricting the alpha period to a shorter window where persons aged 18-66 were not yet vaccinated [7] (Table 3), compared to the entire alpha period correlation coefficient. These were also more similar to the correlation coefficients in other periods.

Table 3: Correlation coefficients at optimal lags between sickleave and all COVID-19 hospital and ICU-only admissions specifically during the adjusted alpha period (1 February 2021 to 10 April 2021, during which persons aged 18-66 were not yet vaccinated) for ages 18-66.

During all variant periods, the correlation coefficients were comparable with those reported between sick-leave and the less severe indicator of SARS-CoV-2 infection notifications, but the lag time to hospitalizations was longer [1]. The optimal lag times per variant-period were larger than those reported between sick-leave absenteeism and infections at -1 (omicron) to -4 (alpha) compared to +1 to -2 during the same periods [1]. Since COVID-19 hospitalizations can only occur after infection, a longer time between sickleave and hospitalizations is expected than between sick-leave and SARS-CoV-2 infection notification date. Additionally, an infection pattern where first younger age-groups were infected, who then infected older age-groups [11], who were much more likely to be hospitalized, would increase the lag time. However, the level of delay was shorter during the omicron period, which had a lag time between sick-leave and hospitalizations of -1. Infection pressure was much higher during that time and a previous study showed that the (mostly) milder endpoint of SARS-CoV-2 infection notifications were actually leading sick-leave [1] by one week during omicron circulation.

Uniquely, we could study the association during periods (wildtype, alpha and delta specifically) where mostly a single respiratory viral pathogen circulated (SARS-CoV-2), as the control measures in place during large parts of the COVID-19 pandemic reduced the circulation of other pathogens [12] (only RSV surveillance showed RSV circulation from summer 2021 onwards [13]; influenza virus circulation re-started during the omicron period). We cannot rule out entirely however that other infections or factors played a role in sick-leave during the COVID-19 epidemic. Sick-leave is a syndromic indicator and is thus not specific to any (respiratory) infectious disease. When multiple pathogens are co-circulating, which is common outside of the COVID-19 pandemic period, work sickleave may not largely reflect the effect of a single pathogen.

The study data had multiple strengths. The coverage of the working population (at ~11%), the inclusion of all hospitals in The Netherlands [5] and the availability of the sick-leave data throughout the study period. These data are consistently collected, and can be used when other surveillance systems, such as specific testing, general practitioner sentinel networks or wastewater surveillance are not available or disrupted. Laboratory diagnostic testing specifically may require time to develop or expand capacity for when a new pathogen arises. Another strength is that COVID-19 hospitalizations were not as directly impacted by changing behaviour of people and changing policies as testing was during the epidemic. This makes the correlation with sick-leave possibly more reliable throughout the study period compared to the correlations with SARS-CoV-2 infection notifications. Although during periods where the pressure on hospitals was very high, especially during the first months of the epidemic, the number of hospitalizations may have been lower than what would have been the case when there would have been enough capacity. Additionally, sick-leave was likely influenced by changing measures imposed to control the pandemic, such as the advice to work from home [14]. For jobs where work from home was not possible, sick-leave would increase despite symptoms being mild. For jobs where work from home was possible, the opposite might occur if infections were mild and sick-leave would not increase despite possible infections. During certain periods, multiple industries were directly affected by measures taken by the government. For example, entertainment and hospitality industries were effectively shut down multiple times throughout the pandemic. These situations are likely to have impacted the number of sick-leave reports, possibly reducing the actual correlation.

In conclusion, sick-leave absenteeism was associated with both all COVID-19 hospitalizations and with COVID-19 ICU-only admissions in a period wherein control measures were in place. To understand the association between sick-leave and SARI hospitalizations outside of pandemic situations, it would be beneficial to investigate this association throughout post-COVID-19 pandemic respiratory seasons. To help shed light on the additional use of sick-leave data as a routine syndromic surveillance source for viral respiratory infections it would also be beneficial to compare sickleave data to other timely data sources, such as citizen science data or sewage surveillance. One such citizen science initiative in the Netherlands that could be used for this comparison continuously collects data on multiple circulating viral respiratory infections in the general population [15].

Acknowledgements

We would like to acknowledge the effort of all Dutch hospitals to record their COVID-19 patients daily and additionally all the ICUs that also collected the comprehensive clinical data of the COVID-19 patients to NICE-database. Without their effort this study would not have been possible. We also like to thank all employees of NICE Research & Support for the technical development of the COVID-19 Online Module, processing the data, and support under high work pressure.

Conflict of Interest

• Femke Jongenotter: none declared.
• Martijn G. Keet: none declared.
• Bram W. Wisse: is employed by HumanTotalCare.
• Nicolette F. de Keizer and Ferishta Bakhshi-Raiez: the department they work for received funding from the Ministry of Health, Welfare and Sport to process ICU and hospital data due to COVID-19 admissions. The NICE foundation pays the department for processing clinical and patient outcome data from all Dutch ICUs into quality improvement information.
• Dylan W. de Lange: is secretary of the board of the NICE foundation.
• Dave A. Dongelmans: is chair of the NICE foundation.
• Arianne B. van Gageldonk-Lafeber: none declared.
• Liselotte van Asten: none declared.

Funding

This study was funded from the regular budget of the Centre for Infectious Disease Control made available by the Ministry of Health, Welfare and Sport, the Netherlands.

Data Availability Statement

The sick-leave dataset underlying this study are not publicly available as they are owned by a third party (HumanTotalCare, HTC). On reasonable request the dataset used for this study can be requested from HTC. The metadata of the SARS-CoV-2 dataset underlying this study are publicly available at RIVMdata.

References

1. Keet MG, Boudewijns B, Jongenotter F, van Iersel S, van Werkhoven CH, et al. (2024) Association between work sick-leave absenteeism and SARS-CoV-2 notifications in the Netherlands during the COVID-19 epidemic. Eur J Public Health 34: 497-504.
2. Acosta N, Dai X, Bautista MA, Waddell BJ, Lee J, et al. (2023) Wastewater-based surveillance can be used to model COVID-19-associated workforce absenteeism. Sci Total Environ 900: 165172.
3. HumanTotalCare (2024) ArboNed and HumanCapitalCare are occupational health services and both a part of HumanTotalCare. ArboNed mainly focuses on the SME segment and HumanCapitalCare on large enterprises. These occupational health services service over 65000 employers and over 1200000 employees in the Netherlands.
4. CBS. Working Population: Statistics Netherlands; 2022.
5. Geubbels E, Backer JA, Bakhshi-Raiez F, van der Beek R, van Benthem BHB, et al. (2023) The daily updated Dutch national database on COVID-19 epidemiology, vaccination and sewage surveillance. Sci Data 10: 469.
6. Reukers DFM, van Asten L, Hooiveld M, Jongenotter F, de Lange MMA, et al. (2023) Surveillance of acute respiratory infections in the Netherlands: winter 2022/2023. SARS-CoV-2, influenza virus, RSV and other respiratory viruses. RIVM; 2023. Report No.: 2023-0378.
7. Valk A, van Meijeren DL, Smorenburg N, Neppelenbroek NJM, van Iersel S, et al. (2021) COVID-19 vaccination coverage in the Netherlands in 2021. RIVM; 2022. Report No.: 2022-0055.
8. Vos ERA, van Hagen CCE, Wong D, Smits G, Kuijer M, et al. (2024) SARS-CoV-2 Seroprevalence Trends in the Netherlands in the Variant of Concern Era: Input for Future Response. Influenza Other Respir Viruses 18: e13312.
9. Huiberts AJ, de Gier B, Hoeve CE, de Melker HE, Hahne SJM, et al. (2023) Vaccine effectiveness of primary and booster COVID-19 vaccinations against SARS-CoV-2 infection in the Netherlands from July 12, 2021 to June 6, 2022: A prospective cohort study. Int J Infect Dis 133: 36-42.
10. Niessen FA, Knol MJ, Hahne SJM, VECTOR study group, Knol MJ (2022) Vaccine effectiveness against COVID-19 related hospital admission in the Netherlands: A test-negative case-control study. Vaccine 40: 5044-5049.
11. Kiem CT, Bosetti P, Paireau J, Crepey P, Salje H, et al. (2021) SARSCoV-2 transmission across age groups in France and implications for control. Nat Commun 12: 6895.
12. Reukers DFM, van Asten L, Brandsema PS, Dijkstra F, Hendriksen JMT, et al. (2021) Annual report Surveillance of COVID-19, influenza and other respiratory infections in the Netherlands: winter 2020/2021. RIVM; 2021. Report No.: 2021-0133.
13. Reukers DFM, van Asten L, Brandsema PS, Dijkstra F, Hendriksen JMT, et al. (2022) Annual report Surveillance of COVID-19, influenza and other respiratory infections in the Netherlands: winter 2021/2022. RIVM; 2022. Report No.: 2022-0098.
14. Fiorini LA (2024) Remote Workers’ Reasons for Changed Levels of Absenteeism, Presenteeism and Working Outside Agreed Hours During the COVID-19 Pandemic. SAGE Open 14: 21582440241240636.
15. RIVM. Infectieradar: RIVM; 2025.