Abstract
Background We aimed to estimate vaccine effectiveness (VE) against COVID-19 mortality, and to explore whether an increased risk in non-COVID-19 mortality exists in the weeks following a COVID- 19 vaccine dose.
Methods National registries of causes of death, COVID-19 vaccination and long-term care reimbursements were linked by a unique identifier using data from 1 January 2021 to 31 January 2022. We used Cox regression with calendar time as underlying time scale to, firstly, estimate VE against COVID-19 mortality after primary and first booster vaccination, per month since vaccination and, secondly, estimate risk of non-COVID-19 mortality in the 8 weeks following a first, second or booster dose, adjusting for birth year, sex and country of origin.
Results VE against COVID-19 mortality was >90% for all age groups two months after completion of the primary series. VE gradually decreased thereafter, to around 80% at 7-8 months post-primary series for most groups, and around 60% for elderly receiving a high level of long-term care and for people aged 90+ years. The risk of non-COVID-19 mortality was lower or similar in the 8 weeks following a first booster dose compared to no vaccination, first or second dose, respectively, for all age and long-term care groups.
Conclusion COVID-19 vaccination greatly reduced the risk of COVID-19 mortality and no increased risk of death from other causes was seen at the population level.
Background
COVID-19 vaccination programmes aim to reduce severe disease and death from COVID-19. Therefore, risk groups for severe outcomes of COVID-19 were prioritized for COVID-19 vaccination in many countries, including the Netherlands. Monitoring of vaccine safety has revealed that the majority of known side effects of COVID-19 vaccines are mild and self-limiting. Severe adverse events are very rare (European Medicines Agency, 2022). Nonetheless, fatal consequences of severe adverse events after COVID-19 vaccination have been reported, such as thrombosis combined with thrombocytopenia after vaccination with viral vector vaccines (Hafeez 2021, Andrews et al, 2022a). Especially for people not belonging to risk groups for severe COVID-19, continuous monitoring of the risk-benefit ratio of vaccination is of utmost importance.
We aimed to analyze associations between COVID-19 vaccination and mortality for the Dutch population during the study period 1 January 2021 to 31 January 2022 in two ways. First, vaccine effectiveness (VE) against death from COVID-19 was estimated. Second, risk of death from causes other than COVID-19 during the weeks immediately following a first, second or booster dose of vaccination was compared to that of the vaccination status before receipt of the dose.
Methods
Study population
The national COVID-19 vaccination registry CIMS was linked to population data from Statistics Netherlands (CBS, the Hague), within the secure remote access research platform provided by Statistics Netherlands. The CIMS dataset was linked to the nationwide population register, a dataset containing causes of death of deceased persons and long-term care (LTC) reimbursement data, and country of origin, using a unique personal identifier. All registered inhabitants of the Netherlands on 1 January 2021, born in or before 2009, were included in the cohort. All results are based on calculations by the authors using non-public microdata from Statistics Netherlands.
Vaccination status
With each administered COVID-19 vaccine in the Netherlands, the recipient is asked for consent for central registry of the vaccination in database CIMS. Around 93% of persons receiving primary vaccination and 95% of persons receiving booster vaccination gave consent for registration. The date, product and series label (primary or booster dose) from CIMS was used to define start and end dates of vaccination statuses per person. As COVID-19 vaccination in the Netherlands commenced on January 6, 2021, each person started with the status ‘unvaccinated’ at 1 January 2021. Persons without registered vaccinations in CIMS retained the status ‘unvaccinated’. Vaccinees who did not give informed consent could not be distinguished from those who were not vaccinated.
For the VE analyses, vaccination status was defined as follows. Status “Primary series – partly” started on the day of the first administered dose. Fourteen days after the second dose of Comirnaty (Pfizer/BioNtech), Spikevax (Moderna) or Vaxzevria (AstraZeneca/ Oxford University), or 28 days after a single dose of Janssen vaccine, status “Primary series – completed” started. Seven days after receipt of a booster dose, status “Boosted” commenced. For 0.5% of all people with vaccinations registered in CIMS, only a booster dose but no primary series was registered (presumably because no consent was given for registration of the primary series). In these cases, the first 6 days after receipt of the booster dose were marked as “Primary series – completed” (and as “Boosted” thereafter).
For the analyses regarding risk of death following vaccination, exposure statuses were defined as follows. The eight weeks immediately following vaccination were defined as ‘risk periods’ and each week was analyzed separately. However, after a first dose of an mRNA vaccine, the first five weeks were analyzed as risk periods, because (generally) after 5 weeks, the second dose was administered. Unvaccinated person-time was used as reference for the analysis of risk of death after the first dose of COVID-19 vaccine. For the relative risk of death after a second dose, person-time at more than 14 days after a first dose was used as reference. For the relative risk of death after booster vaccination, person-time at least 3 months after a second dose was used as reference. Persons who had received a 3-dose primary series due to an immunocompromising condition (N = 133.380) were excluded from the reference group in the analysis of risk of death after booster vaccination, as high death rates in this especially vulnerable population might deflate hazard ratio estimates after the booster.
Date and cause of death
Dates of death were retrieved from the population registry. Underlying cause of death, as reported by a physician on death registry forms and coded centrally by Statistics Netherlands, was available as ICD- 10 codes. ICD-10 codes U07.1 and U07.2 were defined as COVID-19 mortality, which was used as the outcome measure for the VE analyses. All other codes were defined as non-COVID-19 mortality, which was used as the outcome measure for the analyses regarding risk of death following vaccination.
Year of birth, sex, country of origin
Year of birth, sex and country of origin were retrieved from the population registry. Age was estimated as 2021 minus birth year. If a person was born abroad, their country of origin was the country of birth. If a person was born in the Netherlands but their mother was born abroad, their country of origin was defined as the mother’s country of birth. If only the father was born abroad, their father’s country of birth was the person’s country of origin. Countries of origin were categorized into: The Netherlands, Europe excluding the Netherlands, Turkey, Morocco, Surinam, Dutch Caribbean, Indonesia, Other Africa, Other Asia, Other America and Oceania, and Unknown.
Long-term care use
Start and end dates and types of LTC use (partly) reimbursed by insurance were available through the CAK at individual level. Person-time was stratified according to LTC due to physical, sensory or mental disabilities (disability care, DC), and high-level LTC for people aged 70 years or older. Long-term elderly care was further stratified into two broad groups, based on the level of care needs (i.e. ‘care profiles’ within the Dutch LTC system): LTC low 70+ (profiles 1-4) and LTC high 70+ (profiles 5-10). While LTC for elderly people with certain care profilescan be given either at home or in institutions, the LTC low 70+ group is most likely to receive care at home and the LTC high 70+ group is most likely to receive care in a nursing home. This variable was time-dependent as care use could commence or end during the study period.
Statistical analysis
VE was estimated using Cox proportional hazards regression with calendar time as underlying time scale and vaccination status as time-dependent exposure. The outcome was death from COVID-19. Every person’s follow-up time started on 1 January 2021 and ended on date of death, date of emigration or 31 January 2022, whichever came first. VE was calculated as (1- hazard ratio (HR)) *100%, with unvaccinated person-time as reference. Vaccination statuses “Primary series – completed” and “Boosted” were stratified into months since vaccination. The proportional hazards assumption was met, as per visualization of Schoenfeld residuals over calendar time using R package ‘ggsurvminer’ (V0.4.9.; Kassambara, 2021). Sex, year of birth (as natural cubic spline with 4 degrees of freedom) and country of origin (in 10 categories) were included as covariates in the models.
The relative risk of death in the weeks following vaccination was estimated using similar models as the VE analysis, the only differences being the definition of the exposure status (risk periods after vaccination) and the outcome (non-COVID-19 deaths).
All analyses were stratified based on LTC use and birth year into the following strata: LTC high 70+ (born before 1951), LTC low 70+ (born before 1951), disability care (DC, all ages), other persons born before 1931 (90+ years), other persons born between 1931 and 1950 (70-89 years), other persons born between 1951 and 1970 (50-69 years), and other persons born between 1971 and 2009 (12-49 years). Analyses were repeated with all-cause mortality as secondary outcome.
As per the CBS guidelines to prevent disclosure of groups or persons, all numbers smaller than 10 and incidences based on less than 10 observations are not presented, VE estimates higher than 99% are replaced by “>99%” and HR estimates lower than 0.1 are replaced by “<0.1”.
Results
Vaccine effectiveness against COVID-19 mortality
Table 1 shows characteristics of the study population, stratified into persons who did not die during the study period, persons who died of COVID-19, persons who died of other causes, and persons with a date of death but with a missing cause of death. Because vaccination status and LTC use are time-dependent, they are not included in Table 1. Older people, men, and persons with the Netherlands as country of origin were represented more in the group that died of COVID-19 compared with the group that survived the study period. Among the people who died, older people, men, and persons with a country of origin outside of the Netherlands died relatively more often of COVID-19 than of other causes. Numbers of person-days per vaccination status and stratum are included in Table S1.
Study population characteristics, per outcome. *: Due to small numbers, the country of origin is not presented for the population that died with no registered cause of death.
Figure 1 shows, per stratum of LTC use and birth cohort, the 4-week moving average of COVID-19 deaths per week (A) and incidence per 100,000 person-days (B), by vaccination status. The number and incidence of COVID-19 deaths was lowest in the 12-49 years age group and in people receiving disability care (DC). Throughout the study period, the incidence of COVID-19 deaths was highest in unvaccinated elderly. The number of COVID-19 deaths in the beginning of 2021 was highest in unvaccinated elderly. At the end of 2021, the number of COVID-19 deaths was highest in vaccinated elderly as, due to the high vaccination coverage among elderly, the absolute number of vaccinated persons was much higher than the number of unvaccinated persons.
Four-week moving average of number of COVID-19 deaths per week (A) and incidence per 100.000 person-days (B), per stratum of long-term care use and birth cohort. Data points based on fewer than 10 deaths are not shown.
Figure 2 shows, per stratum of LTC use and birth cohort, the estimates of VE against COVID-19 mortality, which were very high (>90%) in the first two months after completion of the primary series. VE decreased gradually to around 80% at 7-8 months post-primary series for most groups. However, for the group born before 1931 (age 90+) and the group receiving a high level of LTC aged 70 years or older, VE decreased to around 60% at 7-8 months after completion of the primary series. VE estimates for months 9 and 10 post-primary series are lower, but these are based on fewer person-days as most people had received a booster vaccination before that time. For the group born between 1971 and 2009 (12-49 years old), the models did not converge for 6 or more months after primary vaccination due to low numbers of COVID-19 deaths. In this group, VE remained >90% up to 5 months post- primary vaccination. In the first two months after booster vaccination, VE was high in most groups (>93%); in the 90+ and 70+ high LTC groups the booster VE was >85%.
Vaccine effectiveness (VE) estimates against COVID-19 mortality, compared to unvaccinated, with 95% confidence interval, per stratum of long-term care (LTC) use and birth cohort. VE >99% are shown in grey without confidence interval. Underlying numbers are shown in Supplementary Table S1.
Figure S1 shows the results of the secondary VE analysis, with all-cause mortality as outcome. This figure shows the same trends as Figure 1, but with lower VE, as COVID-19 comprised a minority of all deaths in the study period.
Risk of non-COVID-19 mortality after vaccination
Figure 3 shows, per stratum of LTC use and birth cohort, the 4-week moving average of non-COVID- 19 deaths per week (A) and incidence per 100.000 person-days (B), per vaccination status. The absolute number of non-COVID deaths was highest among vaccinated persons from the period when this group became eligible for vaccination, and thus when the coverage increased (earlier in 2021 for the oldest ages and LTC recipients, and mid-2021 for the youngest age groups). The incidence of non-COVID-19 deaths per 100.000 person-days was highest in unvaccinated people for most of the study period.
Four-week moving average of number of non-COVID-19 deaths per week (A) and incidence per 100.000 person-days (B), per stratum of long-term care use and birth cohort. Data points based on fewer than 10 deaths are not shown.
Figure 4 shows, per stratum of LTC use and birth cohort, the HR estimates for risk of non-COVID death in the 5 weeks following a first dose of mRNA vaccine (A) and the 8 weeks following a first dose of vector vaccine (B), compared to unvaccinated person-time. For the first mRNA vaccination, nearly all estimates were significantly lower than 1, consistent with a reduced risk of non-COVID-19 death in these 5 weeks. After the first dose of vector vaccine, most HR estimates were also below 1, although these estimates had broader confidence intervals as these vaccines were not used as much in the Netherlands, except for the group of 60-64-year-olds who were specifically offered Vaxzevria initially. Figure 5 shows the HR estimates for risk of non-COVID-19 death following a second dose in the primary series, using person-time at least 2 weeks after a first dose as reference. Similarly, most estimates were below 1, except for the youngest age groups where after 4 weeks post-second dose the HR approached 1, consistent with no association between vaccination and non-COVID-deaths. Likewise, in the 8 weeks following booster vaccination, compared to person-time at least 3 months post-second dose, all HR estimates were below 1 (Figure 6).
Hazard ratio (HR), with 95% confidence interval, for risk of non-COVID-19 mortality after a first dose of mRNA (A) or vector (B) vaccine compared to unvaccinated, per stratum of long-term care use and birth cohort. HR <0.1 are shown in grey without confidence interval. Underlying numbers are shown in Supplementary Table S3-4.
Hazard ratio (HR), with 95% confidence interval, for risk of non-COVID-19 mortality after a second dose of COVID-19 vaccine, compared to person-time at least 2 weeks post-first dose, per stratum of long-term care use and birth cohort. Underlying numbers are shown in Supplementary Table S5.
Hazard ratio (HR), with 95% confidence interval, for risk of non-COVID-19 mortality after a booster dose of COVID-19 vaccine, compared to person-time at least 3 months post-second dose, per stratum of long-term care use and birth cohort. HR <0.1 are shown in grey without confidence interval. Underlying numbers are shown in Supplementary Table S6.
Results of the secondary analysis using all-cause mortality as outcome can be found in Supplementary Figures S2-4. There, HR estimates were comparable to the primary analysis, but somewhat lower for some analyses due to the protective effect of vaccination against COVID-19 death.
Discussion
Our analyses indicate that during the study period January 2021- January 2022, VE against COVID-19 mortality was high in the Netherlands. Given the high vaccination coverage in the population and the SARS-CoV-2 infection fatality rate, both higher among the elderly, the COVID-19 vaccination programme was therefore very successful in limiting the number of deaths due to COVID-19 in this period. We did not find indications for increased risk of death from other causes in the 5 to 8 weeks following immunization.
While VE started at a very high level in all age groups in the two months after completion of the primary series, effectiveness decreased gradually in the months thereafter. Booster vaccination restored VE, but with data up to 31 January 2022 we were not able to estimate VE three or more months after booster vaccination. The lower initial VE and faster waning among the oldest group and among elderly receiving a high level of LTC emphasizes the priority these groups should have within COVID-19 (re-)vaccination programmes.
Our analyses of risk of non-COVID-19 mortality in the 5-8 weeks following vaccination showed a lower to comparable risk, compared to the vaccination status before the respective vaccine dose. This was true for all ages, including LTC recipients. A true protective effect of COVID-19 vaccination on non-COVID-19 mortality is unexpected. Likely, healthy vaccinee bias has affected the results (Remschmidt et al, 2015). For example, fever was a contra-indication for vaccination, resulting in a selection of relatively healthy person-time shortly after vaccination. Also, people with a short remaining life expectancy may have opted to forego vaccination. Another possibility is that in deaths that were not attributed to COVID-19, SARS-CoV-2 infection actually had an unrecognized role in the causal pathway to death.
We did not stratify the study period corresponding to dominant SARS-CoV-2 variant. At the start of 2021, the SARS-CoV-2 wildtype still dominated, which was replaced by the Alpha variant in Spring 2021. During June 2021, Delta replaced Alpha, and around the turn of the year, Omicron BA.1 replaced Delta. Stratification into variant periods was explored, but did not result in informative estimates due to small numbers. Also, because the booster campaign coincided with the emergence of Omicron BA.1, the booster VE could not be stratified for different variants. We do not expect large differences in VE against COVID-19 mortality between Alpha and Delta, because the VE against hospitalization was similar for these variants (RIVM, 2021). However, it is possible that the lower VE estimates in months 8 to 10 post-primary series are associated with the emergence of Omicron BA.1 in the final months of the study period, consistent with a sharp decrease in the VE against hospitalization with Omicron compared to Delta for the primary series (RIVM, 2022; Tenforde et al, 2022).
The data we used has limitations. Because national registration of vaccination in CIMS is dependent on informed consent of the vaccinee, this register is incomplete by design. For the VE analysis this means that a part of the population defined as unvaccinated in the analysis was in fact vaccinated. Of people vaccinated at the Municipal Health Services (around 80% of all vaccinees in the Netherlands), around 7% of primary series recipients did not give consent for registration in CIMS, and around 5% of booster vaccination was not registered in CIMS. Older age groups consented to registration more often than younger people. The resulting misclassification will have led to an underestimation of VE, and likely more so in younger ages compared to elderly. On the other hand, some LTC homes may have refrained from centrally registering any of their vaccinations, contributing to a likely larger misclassification among LTC recipients, and thus a larger underestimation of VE.
Another important limitation is lack of data on previous SARS-CoV-2 infections in the study population. An increasing proportion of the population has experienced an infection, which will have contributed to protection against death from subsequent infections. Unvaccinated populations may have accrued these infections sooner, more often and more severe, leading to a faster build-up of infection-induced immunity in this group which may lead to lower VE estimates over time without actual waning of VE being present. Moreover, we were not able to adjust or stratify estimates according to comorbid conditions. It is likely that vaccination coverage is higher among people with a chronic condition that increases risk of severe COVID-19, which leads to an underestimation of VE if unadjusted for comorbidity (Remschmidt et al, 2015). Also, certain conditions or medications may impair VE. It is likely that the all-cause VE estimates are severely confounded by comorbidities: people with a medical condition putting them at high risk for severe COVID-19 were vaccinated earlier than their age cohort, therefore they are overrepresented in the later intervals post- vaccination (8+ months post primary series), whilst being at high risk for mortality in general. Because in the unvaccinated reference group there is likely not an overrepresentation of people with comorbidity, all-cause VE estimates become negative for these later intervals (Figure S1). This phenomenon is not visible in the LTC recipients, who are as a group likely more homogenous in terms of vulnerability. In future analyses, we hope to add adjustment or stratification for comorbid conditions.
The direction of bias from not adjusting for comorbid conditions is less clear in the analysis of risk of non-COVID mortality. People with a fragile health might be more inclined to get vaccinated, and also have a higher risk of death from other causes than COVID-19 (Malesza, M., & Bozym, M., 2021). However, frail health can be associated with lower socio-economic status, which is also associated with poor access to health care and lower vaccine uptake (Perry, M., 2021; Tessier et al, 2022).
Aside from the data limitations, the observational nature of this study warrants caution when interpreting its results. Characteristics associated with uptake of vaccination, and with consent for central registration, may also be associated with the chance of SARS-CoV-2 exposure or other health-related behaviors. The direction of such a bias is unclear: on the one hand, vaccinated people may exhibit more health-conscious behavior and lower their risk of SARS-CoV-2 exposure, on the other hand risk perception may be lowered after vaccination, leading to more exposure. Policies such as vaccination or test requirements for entry into establishments or events could have reinforced this bias. These exposure biases would mainly influence the VE point estimates, and to a lesser extent the comparison of VE between strata and over time.
Despite the abovementioned limitations, our results are in line with other studies of VE against COVID-19 mortality (Haas et al, 2021; Andrews et al, 2022b; Shrotri et al, 2022; Nunes et al, 2021), that also found a very high VE shortly after primary or booster vaccination, and faster VE waning among LTC recipients (Shrotri 2022). A Hungarian study also found a decreased risk of (all-cause) mortality after vaccination (Pálinkás, A., & Sándor, J., 2022). Two other studies, from the United Kingdom and the United States, did not find evidence for an increased risk of mortality after vaccination (Xu et al, 2021; Nafilyan et al, 2022). This is consistent with our results, where, on a population level, we found no indication of an increased mortality risk following vaccination. As COVID-19 vaccination is very effective against COVID-19 mortality, optimizing (re-)vaccination scheduling and uptake, especially among vulnerable populations, has major public health benefits.
This manuscript is an adaptation of two chapters of a report by Statistics Netherlands and the National Institute for Public Health and the Environment, Sterfte en oversterfte in 2020 en 2021 (cbs.nl) [in Dutch]
Data Availability
Under certain conditions, these microdata are accessible for statistical and scientific research. For further information: microdata -at- cbs.nl.
