Paper: Eva A. S. Koster, Marije H. Sluiskes, Hein Putter, Frits R. Rosendaal, Astrid van Hylckama Vlieg, Mark G. J. de Boer, and Liesbeth C. de Wreede, "Dynamics of infection, vaccination and excess mortality during the COVID-19 pandemic among older individuals-a nationwide analysis," European Journal of Epidemiology, published online 8 June 2026. (traced - PDF supplied in-session and Springer/PubMed records fetched, accessed 2026-06-10)
Genre / field: Observational epidemiology; nationwide administrative-data cohort; relative survival and multi-state event-history modelling. (traced - paper methods and public code, accessed 2026-06-10)
Recommendation: Major revisions. (inference from traced design, methods, and claims)
Verdict: The paper is a valuable descriptive/state-model analysis of documented SARS-CoV-2 infection, registered vaccination, and all-cause excess mortality among older Dutch residents in 2020-2021, but its strongest vaccination-effect language should be downgraded from causal or benefit language to observational association unless additional confounding, misclassification, and survivorship analyses are added. (traced + inference)
H1: The paper is likely methodologically stronger than a simple vaccinated/unvaccinated comparison, because it uses time-varying multi-state methods rather than static baseline groups. (memory -> tested; traced to paper/code)
H2: The central risk is likely not immortal-time bias in the simple sense, but causal overstatement from residual confounding, healthy-vaccinee selection, undocumented infection, and the forced June 2020 infection-state origin. (intuition - unwarranted before source inspection; tested against paper/code)
H3: A methodological critique focused on unequal calendar periods, June 2020 testing start, and survivorship/landmark selection may identify real limitations, but may overstate the problem if it treats the model as a naive period comparison. (user-supplied - unverified; tested against paper/code)
The study analyzes Dutch residents aged over 63 on 1 January 2020, with a reported cohort size of 3,826,770 persons, using anonymized individual-level Statistics Netherlands microdata. (traced - paper abstract/methods, accessed 2026-06-10)
The primary endpoint is all-cause excess mortality in 2020-2021, estimated by splitting observed mortality into expected background mortality and excess mortality using relative survival. (traced - paper methods, accessed 2026-06-10)
The background mortality table is based on 2015-2019 observed mortality stratified by sex, age, and month of the year. (traced - paper methods and code, accessed 2026-06-10)
For infection/vaccination dynamics, the multi-state model begins on 1 June 2020 because large-scale testing in the Netherlands only became available from June 2020. (traced - paper methods/limitations, accessed 2026-06-10)
The multi-state model represents states including alive event-free, vaccination, recent documented infection, non-recent documented infection, infection after vaccination, vaccination after infection, and death states split into background and excess components. (traced - paper Fig. 1 and code transition matrix, accessed 2026-06-10)
Recent infection is defined as the first 28 days after a positive COVID-19 test; non-recent infection is the period after that until another infection, vaccination, or death. (traced - paper definitions, accessed 2026-06-10)
Vaccination status is based on first SARS-CoV-2 vaccination known to RIVM for which consent for data-sharing was given. (traced - paper methods, accessed 2026-06-10)
The paper reports total absolute excess mortality of 0.34% for 2020-2021, corresponding to about 13,120 excess deaths and 4.41% of observed mortality. (traced - paper results, accessed 2026-06-10)
The paper reports excess mortality was highest after recent documented infection, lower after non-recent documented infection, and lower after documented infection in vaccinated persons than after documented infection without prior vaccination. (traced - paper results, accessed 2026-06-10)
The paper's strongest interpretive claim is that the results show vaccination reduced excess mortality and gave a clear survival benefit after vaccination. (traced - paper abstract/conclusion, accessed 2026-06-10)
The paper avoids a naive baseline vaccinated/unvaccinated comparison by modelling vaccination and infection as time-varying intermediate states in a multi-state structure. (traced - paper methods and GitHub code, accessed 2026-06-10)
The paper uses calendar time rather than time-since-enrollment as the main timescale, which is appropriate for pandemic waves, testing policy changes, vaccination rollout, and seasonal mortality. (traced - paper methods, accessed 2026-06-10)
The paper recognizes that all-cause excess mortality cannot be assigned by death certificate alone and uses relative survival to split observed mortality into background and excess components. (traced - paper methods and Manevski et al. method paper, accessed 2026-06-10)
The paper explicitly acknowledges several important limitations: undocumented infections, missing vaccination consent data, healthy-vaccinee bias, socioeconomic testing/vaccine differences, and the June 2020 start of the infection-state model. (traced - paper limitations, accessed 2026-06-10)
The public analysis code improves transparency and makes the state structure, landmark dates, and modelling choices inspectable even though the underlying CBS microdata are restricted. (traced - GitHub repository, accessed/cloned 2026-06-10)
The authors run a sensitivity analysis shifting vaccination status by two weeks to allow time for immunity development, and report comparable landmark results. (traced - paper results, accessed 2026-06-10)
No fatal finding was identified that makes the paper unusable as a descriptive/state-model analysis of documented infection, registered vaccination, and excess mortality. (inference from traced sources)
The paper is not adequate as stand-alone causal proof that vaccination reduced all-cause excess mortality, but that is a Major overclaim rather than a Fatal flaw if the conclusion is rewritten as observational association and the limitations are foregrounded. (inference from traced sources)
Claim under review: The paper concludes that vaccination decreased excess mortality and produced a clear survival benefit after vaccination. (traced - paper conclusion, accessed 2026-06-10)
Finding: The design supports a strong association between registered vaccination state and lower subsequent excess mortality, particularly among documented infections, but it does not fully identify the causal effect of vaccination on all-cause excess mortality. (traced + inference)
Why this matters: Vaccination was voluntary and correlated with socioeconomic status, migration background, trust, adherence to preventive measures, testing behavior, and health status. The paper itself states that healthy-vaccinee bias likely exists and that the vaccinated population may be fitter than the average population. (traced - paper limitations, accessed 2026-06-10)
Required revision: Replace causal wording such as "vaccination decreased the risk" and "clear survival benefit" with language such as "registered vaccination was associated with lower excess mortality in this model, especially after documented infection." (inference from traced limitations)
Severity: Major. (inference)
Claim under review: The multi-state model assesses infection/vaccination/death dynamics in 2020-2021. (traced - paper methods, accessed 2026-06-10)
Finding: The infection/vaccination multi-state model begins on 1 June 2020, and all persons start that model in the alive event-free state. The first COVID-19 wave before June 2020 is therefore not represented as prior infection history in the multi-state infection states. (traced - paper limitations and code, accessed 2026-06-10)
Why this matters: Persons infected before June 2020 may have had long-term post-infection risk, immunity, altered testing behavior, and altered mortality risk, yet they enter the multi-state model as event-free. (traced - paper limitations, accessed 2026-06-10)
Interpretive effect: This does not invalidate the total 2020-2021 excess mortality estimate, which is estimated separately from 1 January 2020, but it weakens attribution of excess mortality to documented infection/vaccination states after 1 June 2020. (traced + inference)
Required revision: Add a clearer separation between total excess mortality for 2020-2021 and state-attributed excess mortality from 1 June 2020 onward; quantify plausible bias from first-wave seroprevalence and post-acute risk. (inference)
Severity: Major. (inference)
Claim under review: Landmark analyses compare excess mortality from different states at 1 January, 1 June, 1 September, and 1 December 2021. (traced - paper and code, accessed 2026-06-10)
Finding: Landmarking reduces non-Markov problems and avoids a crude immortal-time comparison, but each landmark comparison is conditional on being alive and in a classifiable state at that landmark. (traced + inference)
Why this matters: Later landmark states are selected by survival, vaccination eligibility, health status, infection history, testing access, and consent-recording status. A vaccinated group at a landmark is not exchangeable with an alive event-free or unvaccinated documented-infection group unless those selection processes are addressed. (traced + inference)
Required revision: Label landmark results as conditional state-specific prognosis rather than causal treatment effects. Add inverse-probability weighting, richer adjustment, negative-control outcomes, or stratification by long-term care/comorbidity if data access permits. (inference)
Severity: Major. (inference)
Claim under review: The model adjusts/stratifies by sex and age group and uses sex-age-month background mortality. (traced - paper methods/code, accessed 2026-06-10)
Finding: The main reported comparisons do not adequately adjust for comorbidity, frailty, long-term care status, prior health-care use, socioeconomic status, migration background, religious community, local epidemic intensity, exposure behavior, or testing propensity. (traced for included variables; inference for missing confounding burden)
Why this matters: The paper acknowledges lower vaccine uptake in groups with lower socioeconomic status, non-western migration background, low trust in government, and strong religious beliefs; it also acknowledges testing underreporting among lower-income groups. (traced - paper limitations, accessed 2026-06-10)
Required revision: Either add richer adjustment/sensitivity analyses or downgrade the paper's causal framing. (inference)
Severity: Major. (inference)
Claim under review: The paper estimates excess mortality after recent and non-recent documented infection. (traced - paper results, accessed 2026-06-10)
Finding: Documented infection substantially undercounts true infection. The paper reports cumulative documented infection around 9% by end-2021 while citing seroprevalence around 20% among older persons and 26% overall in November 2021. (traced - paper limitations, accessed 2026-06-10)
Why this matters: The documented-infection states are likely enriched for symptomatic, test-seeking, or administratively captured infections. This may overstate per-documented-infection excess mortality while understating population-level infection-attributable excess mortality. (traced + inference)
Required revision: Make clear in title/abstract/conclusion that infection-state claims are about documented infections, not all SARS-CoV-2 infections. (inference)
Severity: Major. (inference)
Claim under review: Whether the study rules out vaccines as a cause of excess mortality. (user-supplied query - tested against paper/code)
Finding: The paper does not explicitly exclude vaccination as a possible cause of any excess mortality. It models all-cause excess mortality by state, without using exact causes of death, and it contains "death after vaccination" states that mean death from a vaccination state, not death caused by vaccination. (traced - paper methods/code, accessed 2026-06-10)
Why this matters: The reported pattern is inconsistent with a large model-detectable vaccine-driven excess mortality signal among registered vaccinees, because excess mortality after vaccination without subsequent documented infection is reported below zero to near zero. But that does not rule out individual vaccine-caused deaths, small effects, short risk windows, or cause-specific adverse outcomes. (traced + inference)
Required revision: Add a clear statement that the model is not a cause-specific vaccine-safety analysis and cannot exclude rare or individual vaccine-attributable deaths. (inference)
Severity: Major if the paper is cited as vaccine-safety exoneration; Minor inside the paper if wording is confined to association. (inference)
Major 7: Event-date repair for infection/vaccination records after death is not externally auditable
Claim under review: The code handles registrations after death by moving some dates to the day before death. (traced - public code, accessed 2026-06-10)
Finding: The code shows that some vaccination dates and infection dates after death are interpreted as likely reporting delay and shifted to one day before death for modelling. (traced - GitHub code, accessed 2026-06-10)
Why this matters: This may be a defensible administrative-data repair, but without public counts and sensitivity analyses, the effect on acute infection/vaccination death-state attribution cannot be independently checked. (traced + inference)
Required revision: Report the number of records affected, stratify by event type/state, and rerun sensitivity analyses excluding those records or treating them as missing/late registrations. (inference)
Severity: Major for reproducibility of state attribution; Minor for the broad total excess mortality estimate. (inference)
The title and abstract should repeatedly use "documented infection" where relevant, because "infection" alone can be read as all SARS-CoV-2 infection. (traced + inference)
The paper should distinguish "lower excess mortality from the vaccination state" from "vaccine prevented excess death"; the former is a model result, the latter is causal interpretation. (traced + inference)
The paper should explicitly define whether "vaccination" means first recorded dose only in every public-facing summary, because booster timing and time since vaccination are not modelled in the main state definition. (traced - paper discussion, accessed 2026-06-10)
The supplement appears necessary for exact subgroup and landmark numeric verification; public HTML/PDF alone does not expose all supplemental table values in easily auditable form. (traced - paper references to supplemental tables; local supplement not available)
The code and paper should make simultaneous-event rules and administrative repairs easier to audit for non-author readers. (traced + inference)
Study design: Nationwide observational cohort using administrative microdata. (traced)
Appropriate core method: Relative survival is appropriate for estimating excess all-cause mortality when cause of death is not used or is uncertain. (traced - Manevski et al. and paper methods)
Appropriate event-history framework: Multi-state modelling is appropriate for sequences of infection, vaccination, and death, and is materially better than static group assignment. (traced + inference)
Main limitation: The method estimates state-specific excess mortality patterns, not a randomized vaccine effect. (traced + inference)
Confounding strategy: Sex/age stratification and sex-age-month background mortality are useful but insufficient for causal vaccination claims in a voluntary vaccination campaign. (traced + inference)
Temporality: Vaccination and documented infection are time-indexed, but pre-June 2020 infection history is absent from the multi-state infection model. (traced)
Reverse causation: Reverse causation is partly present as frailty/acute illness can delay vaccination, making vaccinated persons temporarily healthier. The authors explicitly identify healthy-vaccinee bias. (traced)
Outcome definition: All-cause excess mortality is useful for net pandemic mortality but cannot assign individual deaths to COVID-19, vaccination, delayed care, heat, influenza displacement, or other causes. (traced + inference)
The paper reports absolute excess mortality of 0.34% for 2020-2021, with a 95% confidence interval of 0.32-0.37. (traced)
The paper reports 0.19% absolute excess mortality in 2020 and 0.16% in 2021. (traced)
The paper reports excess mortality after recent documented infection as substantial and concentrated in the first four weeks after a positive test. (traced)
The paper reports lower excess mortality after documented infection in persons with prior vaccination than in persons without prior vaccination at several landmark dates. (traced)
The paper's confidence intervals assume fixed background hazards, which does not propagate uncertainty from the historical background mortality model. (traced - paper methods; inference about uncertainty)
The main causal weakness is non-exchangeability of vaccination states: vaccinated and unvaccinated persons differ by health, behavior, consent, socioeconomic status, and exposure/testing processes. (traced + inference)
The paper acknowledges the healthy-vaccinee bias and uses later landmark patterns to argue that the difference between vaccinated and unvaccinated documented-infection states persists, but this is suggestive rather than definitive causal identification. (traced + inference)
Hypothesis: The paper compares 2020 no-vaccine and 2021 vaccine periods like apples to oranges. Partially supported. The paper does not use a naive 2020-vs-2021 baseline comparison; it uses time-varying states and landmark comparisons. However, calendar-period changes, wave timing, vaccination rollout, testing availability, and selection into later landmarks remain important interpretive constraints. (user-supplied - unverified; tested: traced + inference)
Hypothesis: Starting infection measurement in June 2020 creates bias. Supported. The authors explicitly acknowledge that all persons start the multi-state model on 1 June 2020 as alive event-free and that first-wave infections may cause bias. (traced)
Hypothesis: Unequal periods and wave timing make long-term effects hard to measure. Partially supported. The paper can measure post-28-day mortality after documented infection until 31 December 2021, but the follow-up is truncated and varies by infection/vaccination date; later and longer-term pandemic dynamics cannot be assessed. (traced + inference)
Hypothesis: Survivorship bias affects the vaccination comparisons. Supported as a caution. Landmark analyses condition on being alive and in a state at the landmark; they are conditional prognostic comparisons, not randomized causal contrasts. (traced + inference)
Hypothesis: The model implies vaccination retroactively protects against old infection. Mostly not supported literally. The state labels and landmark comparisons can be confusing, but the model does not make vaccination retroactive; it compares persons occupying states at fixed landmark dates. However, state interpretation should be clearer. (traced + inference)
Hypothesis: The risk index is based on too few covariates. Supported for causal interpretation. Age, sex, and month-adjusted background mortality are not enough to remove confounding in voluntary vaccination and testing behavior. (traced + inference)
| Paper claim | Cited/source checked | Source actually supports | Verdict | Warrant |
|---|---|---|---|---|
| Dutch COVID data include testing and vaccination surveillance data | Geubbels et al., Scientific Data 2023 | The Dutch COVID database includes tests, individual-level positive cases, deaths, vaccinations, and other surveillance streams collected through RIVM/MHS systems | Supports | (traced - Nature article fetched 2026-06-10) |
| Relative survival can be integrated into multi-state models where exact cause of death is uncertain | Manevski et al. 2022 | Method paper proposes integrating population mortality tables into non-parametric multi-state models to split mortality into excess and population components | Supports | (traced - SAGE article fetched 2026-06-10) |
| Vaccination was associated with lower COVID-19 mortality in prior Dutch work | de Gier et al. 2023 | PubMed abstract reports vaccine effectiveness against COVID-19 mortality above 90% shortly after primary series and no increased non-COVID mortality signal | Supports, but separate design | (traced - NCBI EFetch 2026-06-10) |
| Consent-based vaccination registration biases VE estimates in older adults | van Werkhoven et al. 2024 | PubMed abstract reports nonconsent misclassified vaccinated persons as unvaccinated and biased VE estimates, largest in age 70+ | Supports | (traced - NCBI EFetch 2026-06-10) |
| Public code exists | GitHub repository | Repository contains RMarkdown scripts for dataset creation, rate tables, analyses, exports, and plots | Supports | (traced - GitHub cloned/inspected 2026-06-10) |
| Claim / framework | Requirement | Directly shown | Borrowed support | Status | Severity |
|---|---|---|---|---|---|
| "Nationwide analysis" | Cohort should cover national older population | Uses Statistics Netherlands microdata, cohort n=3,826,770 | Not all raw data public | Satisfied for population frame, restricted for audit | Minor |
| "Documented infection" | Infection status based on recorded positive test | GGD/RIVM data; starts June 2020 for multi-state model | Data descriptor supports surveillance architecture | Satisfied for documented infection only | Major if generalized to all infection |
| "Vaccination" | Vaccination status based on recorded first dose | RIVM data requiring consent; nonconsent misclassification acknowledged | van Werkhoven supports bias concern | Partial | Major |
| "Excess mortality" | Observed deaths minus expected background deaths | Relative survival using 2015-2019 sex-age-month background | Method paper supports approach | Satisfied for model-defined excess mortality | Minor |
| "Vaccination decreased risk" | Requires exchangeability or robust causal identification | Observational state comparison; healthy-vaccinee bias acknowledged | Prior VE studies support plausibility, not this causal estimate | Partial/overstated | Major |
| Claim | Construct | What was actually used | Adequate? | Severity |
|---|---|---|---|---|
| Infection increased excess mortality | Documented SARS-CoV-2 infection | Positive COVID-19 test known to GGD/RIVM systems | Adequate for documented infection; inadequate for all infection | Major if generalized |
| Vaccination reduced excess mortality | Registered first SARS-CoV-2 vaccination | First vaccination known by RIVM with consent for sharing | Adequate for registered vaccination state; inadequate for causal vaccine effect without more adjustment | Major |
| Excess mortality captures pandemic mortality | All-cause excess relative to expected background | Mortality difference versus 2015-2019 sex-age-month table | Useful net construct; not cause-specific | Minor |
| Death after vaccination | State after vaccination | Death while in vaccination-related state | Not evidence of vaccine-caused death | Important clarification |
The raw microdata are non-public Statistics Netherlands data, accessible under conditions for statistical/scientific research. (traced - paper data availability, accessed 2026-06-10)
The analysis syntax is public on GitHub. (traced - GitHub repository, accessed 2026-06-10)
The public code allows inspection of state definitions, landmark dates, background mortality table use, and event-date repairs, but it does not permit independent numerical reproduction without CBS data and exported intermediate RDS files. (traced + inference)
The paper reports ZonMw funding and states that the funder had no role in study design, data collection, analysis, interpretation, writing, or submission. (traced - paper acknowledgements/funding, accessed 2026-06-10)
The paper reports Mark de Boer as ESGAP secretary 2024-2026 and SWAB guideline committee chair, both non-financial interests; other authors report no relevant interests. (traced - paper declarations, accessed 2026-06-10)
Institutional network: authors are primarily affiliated with Leiden University Medical Center; RIVM/CBS data systems are government/public-health infrastructure; ZonMw is a Dutch health research funder. These are not independent nodes for all claims because the study, data infrastructure, and cited Dutch vaccination-bias literature are all embedded in Dutch public-health/statistical institutions. (traced + inference)
Would move toward Minor revisions if: conclusions are rewritten as observational association; supplement reports exact affected counts for date repairs; causal language is removed; additional sensitivity analyses are added for healthy-vaccinee bias, undocumented infections, consent misclassification, and first-wave infection history. (inference)
Would move toward Reject-resubmit if: the paper is presented as conclusive proof that vaccination causally reduced all-cause excess mortality or explicitly rules out vaccine-attributable excess mortality without cause-specific analysis. (inference)
Upward: Access to supplementary tables showing robust subgroup counts, sensitivity analyses excluding post-death-repaired records, richer adjustment for frailty/long-term care/SES/migration/comorbidity, and negative-control outcome checks. (inference)
Downward: Evidence that event-date repairs are numerous and materially alter death-state attribution; evidence that missing vaccination consent differentially tracks frailty or mortality more strongly than assumed; or evidence that documented-infection states are too selected to support the paper's infection/vaccination comparisons. (inference)
Major revisions. The paper should be accepted as a sophisticated descriptive analysis only if the interpretation is tightened. It should not be published or cited as if it explicitly excluded vaccines as a cause of excess mortality, nor as if it independently proved a causal vaccine survival benefit. Its strongest warranted contribution is narrower but still valuable: in registered Dutch older-adult data, documented infection was associated with excess mortality, and registered vaccination states were associated with lower excess mortality patterns, especially after documented infection, under a relative-survival multi-state model. (traced + inference)
| Source | URL / location | Access date | Publication date | Warrant | Funding / ownership / mandate / alignment |
|---|---|---|---|---|---|
| Koster et al. article PDF | C:\Users\Patrick\Downloads\s10654-026-01414-1.pdf and https://link.springer.com/article/10.1007/s10654-026-01414-1 |
2026-06-10 | 2026-06-08 | (traced) | Springer Nature journal; funded by ZonMw; Dutch academic authors |
| PubMed record for Koster et al. | https://pubmed.ncbi.nlm.nih.gov/42257779/ | 2026-06-10 | 2026-06-08 | (traced) | NCBI/NIH bibliographic database |
| Public analysis code | https://github.com/survival-lumc/COVID19_ExcessMortality | 2026-06-10 | Repository current as of clone | (traced) | survival-lumc GitHub repository; author-provided code |
| Geubbels et al. Dutch COVID database | https://www.nature.com/articles/s41597-023-02232-w | 2026-06-10 | 2023-07-20 | (traced) | RIVM/public health surveillance data descriptor |
| Manevski et al. relative survival multi-state method | https://doi.org/10.1177/09622802221074156 | 2026-06-10 | 2022-03-14 online | (traced) | Statistical Methods in Medical Research; some author overlap with reviewed paper |
| de Gier et al. Dutch vaccination mortality study | https://pubmed.ncbi.nlm.nih.gov/37328352/ | 2026-06-10 | 2023 | (traced) | RIVM-affiliated authors; public-health institutional source |
| van Werkhoven et al. consent-bias study | https://pubmed.ncbi.nlm.nih.gov/39032589/ | 2026-06-10 | 2024 | (traced) | RIVM/UMC Utrecht authors; directly relevant to Dutch vaccination-register bias |
| Methodological critique hypotheses | Conversation-supplied critique, not independently sourced | 2026-06-10 | n/a | (user-supplied - unverified) | Treated only as hypotheses to test, not as evidence |
Symmetry test: If the paper had concluded that vaccination increased excess mortality using the same observational design, the same review would downgrade causal language, demand richer confounding control, and object to causal claims beyond state-specific associations. (inference)
Standards applied: STROBE-style observational epidemiology standards, event-history/relative-survival methodology standards, and explicit causal-direction burden. (deferred to consensus)
Difference in approach vs fault: The use of multi-state relative survival is not the fault; the main fault is interpretive overreach from a sophisticated observational model into causal language. (inference)
Reviewer priors: COVID-19 vaccine effectiveness against severe outcomes has supportive literature, but this prior does not license this paper to overclaim causality. (deferred to consensus + traced cited studies)
The raw CBS microdata were not available for independent rerun. (traced)
Supplementary tables were referenced but not separately available as a local file during this review. (traced)
The exact counts of event-date repairs in the private data could not be verified. (traced)
This review did not perform a full reanalysis, only a document/code/citation audit. (traced)