Review Reports - Association Between COVID-19 Vaccination and Long COVID Symptoms in Hospitalised Survivors: Distinguishing Prevention from Reverse Causality

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The manuscript addresses a highly relevant and timely topic—the relationship between COVID-19 vaccination and the risk of long COVID—using a retrospective dataset and advanced propensity score methods. The study is clearly structured, and the analytical approach is detailed and transparent. However, several aspects of the manuscript would benefit from clarification and refinement to strengthen methodological rigor, improve interpretability, and ensure that the conclusions remain aligned with the evidence presented. The suggestions below aim to support the authors in enhancing the clarity, scientific soundness, and overall impact of the work.

Clarify the definition of long COVID.
Please provide the operational criteria used to classify participants as having long COVID (symptom duration, specific ICD codes, clinical descriptors). This is essential for reproducibility and for interpreting comparisons with other studies.
Improve transparency in describing the vaccination variables.
The manuscript would benefit from clearer descriptions of:
- Which vaccines/platforms composed the “mixed-dose” vs. “same-dose” regimens.
- The temporal interval between vaccination and infection in both “vaccinated before infection” and “vaccinated after infection” groups.
- The rationale for treating the 0–4 dose variable as categorical rather than ordinal.
Address potential confounding more thoroughly.
Although propensity scoring and IPTW are used, several relevant confounders were not incorporated (e.g., comorbidities, socioeconomic status, occupational exposure, reinfections). These factors may substantially influence both vaccine uptake and long COVID risk. Please discuss how this may affect effect estimates.
Provide additional diagnostics for IPTW and matching quality.
Including tables or plots showing standardized mean differences before and after weighting/matching (similar to Figure 3, but presented comprehensively) would strengthen confidence in covariate balance. Clarify whether weight truncation or stabilization was applied to mitigate extreme IPTW values.
Interpret the findings with more caution.
Several conclusions in the Discussion imply or suggest biological mechanisms (immune modulation, viral persistence, antigen retention, autoimmune processes). Because the current study does not include biological sampling, immunological assays, or temporal viral-spike measurements, these mechanistic explanations should be clearly labeled as hypotheses, not interpretations derived from your data.
Reorganize and streamline the Introduction.
While informative, the first half of the Introduction provides extensive historical background on vaccine development. Consider condensing these details and focusing more directly on:
- existing knowledge gaps in vaccine effects on long COVID, and
- the conceptual rationale for using propensity-weighted models.
Strengthen the coherence of the Results and Discussion.
Some interpretations extend beyond the statistical findings. For example, the conclusion that two or more vaccines “increase the risk of long COVID” should be softened to reflect associations rather than causal relationships. Please emphasize alternative explanations such as reverse causation, risk-based vaccination prioritization, and survivor bias.
Clarify long COVID prevalence estimation.
Only 40% of participants were classified as having long COVID, but the manuscript does not describe whether symptom onset predated vaccination in individuals vaccinated after infection. This temporal ambiguity should be resolved or acknowledged as a limitation.
Consider discussing dose-group imbalance as a limitation.
The one-dose group is very small and shows extreme IPTW variability. This influences the stability of effect estimates. A brief comment acknowledging this limitation would increase methodological transparency.
Improve language clarity and reduce redundancy.
The manuscript is generally readable but contains long paragraphs, repeated concepts, and occasional colloquial phrasing. A language refinement focusing on conciseness and precision would improve readability.
Update the Conclusion to align with the evidence strength.
Conclusions should more clearly indicate the exploratory nature of the findings and avoid implying direct causation. Highlighting the need for prospective, biomarker-based studies would strengthen the final section.

Comments on the Quality of English Language

The English language used in the manuscript is generally understandable; however, several sections would benefit from careful editing to improve clarity, conciseness, and flow. Some sentences are overly long or contain redundant information, particularly in the Introduction and Discussion, which can obscure key points. Occasional grammatical inconsistencies, informal expressions, and repetitive phrasing should be revised for greater precision and academic tone. A thorough language review—focusing on sentence structure, transitions between ideas, and reduction of speculative or conversational wording—will enhance readability and strengthen the overall presentation of the manuscript.

Author Response

Please see the response in the attached file.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

I want to congratulate the authors regarding this study. I appreciate the retrospective observational study that includes 627 adults with confirmed COVID-19 in London and the assessment whether COVID-19 vaccination—including number of vaccine doses, timing of vaccination, and heterologous versus homologous regimens—affects the risk of developing long COVID.

I have a few recommendations in order to improve the manuscript:

-definition and measurement of “Long COVID” must adequately to be described

-provide exact clinical criteria used for long COVID classification and if symptoms are ≥12 weeks duration, as defined by WHO

-clarify if the symptoms were clinician-diagnosed or self-reported and how variability in patient charts was handled

-please provide more details regarding the timing between infection and vaccination which now is not quantified, the study results relies on whether vaccination occurred “before” or “after” infection, but the timing is not described numerically.

-the vaccination status and timing are often not random but influenced by a lot of factors, such as health-seeking behaviour, presence of comorbidities, age, prior COVID severity. Without controlling for these factors, the conclusions may reflect pre-existing risk differences rather than true vaccine effects. Please clarify this aspect.

-I appreciate the figures

-since the cohort is from a single centre in London with unique demographics aspects, findings may not extend to other populations. The study includes infections from early 2020–2022 but does not account for different variants. Variant of virus periods and availability of specific vaccines (AstraZeneca, Pfizer, Moderna, J&J) should be described if possible.

-in the discussion section, as the article raises the possibility of immunological burden, post-viral complications, and altered host responses after SARS-CoV-2 infection and vaccination, we believe that including this paper https://doi.org/10.3390/medicina59101858 could add valuable context on bacterial co-infections and comorbidities that might modulate post-COVID and long-COVID risk. This could help strengthen the manuscript’s discussion of non-viral co-factors and their potential role in symptom persistence.

Author Response

Please see the response in the attached file.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

I congratulate with the authors, the work is well structured, interesting and scientifically valid. I have only a few minor considerations.

Minor questions:

In the introduction section (lines 80–92), the authors emphasize that COVID vaccine studies have shown a significant reduction in symptoms, hospitalizations, and deaths. They also emphasize that few studies exist on the ability of vaccination to prevent long COVID. I would suggest the authors include a brief description (two lines are sufficient) to define the condition of “long COVID”.
In the Materials and Methods section (lines 102–111), the authors state that they analyzed the records of 627 individuals diagnosed with COVID-19 between April 2020 and December 2022 who met the study criteria. I would ask the authors to specify whether the COVID diagnosis included (as I assume) laboratory tests such as molecular biology diagnostic methods (PCR) for the diagnosis of COVID-19 and/or rapid methods such as rapid antigen tests (immunochromatographic tests); a half-line description is sufficient.
In the Results section (lines 191–194), the authors describe the mean age of the 627 subjects included in the study. I would suggest the authors include the maximum and minimum ages in addition to the mean age and standard deviation.
In the Results section (lines 194-203), the authors describe the average BMI, the percentage of subjects divided by ethnic origin, the percentage of smokers, etc., as well as the percentages of vaccinated and unvaccinated, and the percentages of those who received one dose, two doses, or three doses of the vaccine, etc. I would ask the authors (if possible) to also include a couple of lines describing the type of vaccines used (for example, Pfizer, AstraZeneca, Moderna, etc.).
In the Results section (lines 253–256), the authors state that there is a strong correlation between long COVID and advanced age or increased BMI. If the authors also had data on immunological status, this would be very interesting. For example, knowing the quantitative level of T lymphocytes specific for the SARS-CoV-2 S protein after vaccination, measured in blood samples and after appropriate stimulation with SARS-CoV-2-specific peptides and flow cytometric analysis, would be very important. I realize how difficult such a test is, and it would certainly have been performed on only a small portion of the subjects studied, but if such data were available, it would greatly strengthen the study.
In the Discussion section (lines 323-327), the authors describe how some studies have detected the persistence of the S protein in the blood or tissues of subjects vaccinated with mRNA for a very long time after vaccination, or the persistence of fragments of the S1 subunit of the spike protein in CD16-positive monocytes, even 245 days after vaccination. I would suggest that the authors also mention the persistence of specific T lymphocytes against COVID-19, for long time in infected and vaccinated subjects (for example: Guihai Liu, et al. Long-persisting SARS-CoV-2 spike-specific CD4+ T cells associated with mild disease and increased cytotoxicity post COVID-19. Nature Communications volume 16, Article number: 8743 (2025)).

Author Response

Please see the response in the attached file.

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

This manuscript from Daodu et al describes a retrospective observational study evaluating whether COVID-19 vaccination influences the risk of developing long COVID, focussing on number of doses, heterologous vs. homologous schedules, and timing relative to infection. Using electronic health records from 627 adults in London and applying multinomial propensity score weighting and matching, the authors report that two or more vaccine doses were associated with higher odds ratios of long COVID, while single doses or a mixed-dose regimen showed a borderline protective effect.

The topic is clinically relevant and has implications for future vaccine strategies, but the data does disagree with a large body of existing literature. The implications would in fact be quite significant for vaccination and public health and therefore would need to be substantiated with very strong and convincing data.

The manuscript unfortunately has serious methodological and interpretative problems (detailed below) that significantly undermine the validity of the findings.

No Definition or diagnostic method for Long COVID

The manuscript does not provide any definition of long COVID used in this study, nor does it specify:

the time threshold used after acute infection,
which symptoms were included,
the diagnostic criteria applied (WHO/NICE/CDC/clinician-documented diagnosis),
how cases were identified in the electronic patient records (as long COVID diagnoses are difficult to obtain, 40% Long COVID in the EPRs seems beyond credibility)
whether symptoms continued for the duration of the study (or if patients recovered)
how reinfections were handled if at all

The Methods vaguely state that “we reviewed the symptoms associated with long COVID” but provide no definition and could be subjective. This omission makes the primary outcome non-reproducible, non-interpretable, and unpublishable.

Long COVID has over 200 listed symptoms which overlap with many other conditions. The definition of long covid used greatly affects any interpretation of any scientific study. For example, if vaccinations are causing a major increase in severe fatigue among a population where Long COVID rates are averaging 40%, this is a public health emergency. Alternatively, if COVID vaccines increase the risk of a persistent cough 3 months after infection, but prevent patients from dying, this is a wonderful trade off.

As there is no stated Long COVID definition, the findings cannot currently be evaluated.

“All individuals lacking a definitive vaccination status were classified as unvaccinated”

Missing data should be removed from the dataset and absolutely cannot be classified with any confidence as an unvaccinated individual and to do so is a catastrophic error. It is noticeable in Figure 1 and 2 that there is no difference between 2-4 doses of vaccine, that 1 dose is a small cohort and therefore unreliable (see comment 4). As around 90% of the UK population were vaccinated, one could estimate that anywhere from 50%-80% of the unvaccinated cohort are likely vaccinated but without a note in their electronic health record. If the zero-dose category has such a number of patients with a missing data value, the entire study has to be called into question.

Misinterpretation of Causality Amid Strong Confounding and Reverse Causation

The manuscript concludes that two, three, or four vaccine doses increase the odds of long COVID. However, the analysis cannot support causal statements due to:

Reverse causation where individuals who were older, sicker, or had persistent symptoms may have been more likely to receive additional doses.
Indication bias where high-risk patients are systematically more vaccinated and also more prone to long COVID.
Reporting bias: patients who get vaccinated more visit their doctors more and so have more extensive information in their electronic health records.

Critically, the authors find that vaccination after infection increases long-COVID odds nearly sixfold. Because vaccine dose number and vaccination timing are correlated in this dataset, the apparent “dose effect” may simply reflect infection timing and illness trajectory.

Propensity Score Weighting Likely Invalid Due to Sparse Treatment Groups and Unstable Weights

The multinomial IPTW model produces highly unstable weights, especially for the one-dose group (n=18). Figure 1 shows extreme values (median of 14, outliers approaching 50) indicating poor covariate balance and unreliable estimates.

The manuscript does not present post-weighting balance diagnostics for all dose levels, nor does it justify causal inference under these conditions. IPTW is known to perform poorly when treatment groups are highly imbalanced, which is the case here.

As a result, the estimated odds ratios for dose levels cannot be considered robust.

Missing Key Covariates: Acute Disease Severity, Comorbidities, Reinfections, Socioeconomic Factors

The analysis lacks several established predictors of long COVID:

comorbidities (diabetes, cardiovascular disease, lung disease)
severity of initial COVID-19 (hospitalization, oxygen requirement)
reinfection history
occupation, deprivation measures

Without these variables, key confounding pathways remain uncontrolled, and causal interpretation is not possible.

Overemphasis on Speculative Biological Mechanisms Not Supported by the Dataset

The Discussion introduces hypotheses such as persistent spike protein after vaccination; vaccine-induced autoimmune activation; and immune system “modulation”.

These topics, while appearing in the literature, are not measured in this study, and many cited sources are preliminary or reports rather than peer-reviewed primary data. This are of study is well known to suffer from publication bias. The authors inclusion distracts from more plausible explanations rooted in confounding, reverse causation, and mistakes in data handling.

Contradictions with the Broader Evidence Base Are Not Adequately Addressed

Most high-quality studies find that vaccination reduces long-COVID risk when administered before infection. The manuscript acknowledges these findings but does not reconcile them with its own results, and simply tries to provide a mechanistic explanation for its own findings.

Rather than suggesting biological mechanisms that imply harm from vaccination, the authors should more thoroughly explore their data set.

Minor Issues

Table numbering errors: Table 4 is referred to as “Table 15” in the text.
Figures need improved labelling: Figure 2 seems to have the doses out of order
Almost 10% of all citations are news articles and not primary literature. In some cases the primary literature does not agree with the news article, or is at least more cautious in their findings.

Author Response

Please see the response in the attached file.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

After reviewing the revised version, I find that the manuscript has clearly improved and now presents a much more coherent and convincing analysis. From my perspective, the greatest strength of the revised work is the explicit separation between vaccination before and after infection. This single change substantially clarifies the results and resolves much of the confusion seen in previous observational studies. The explanation of reverse causality is clear, logical, and well supported by the data, and it strengthens the credibility of the conclusions. I also find the statistical strategy appropriate for the dataset. The use of Bayesian models is justified, and the authors do a good job explaining why this approach was necessary given the sample size and subgroup structure. Importantly, the conclusions are now more cautious and better aligned with what the data can realistically support. Clinically, I think the emphasis on comorbidities and acute disease severity as the main predictors of long COVID is one of the most valuable contributions of the study. This shifts attention away from misleading vaccine-related associations and toward factors that are directly relevant to patient management and follow-up. That said, a few aspects could still be improved from a reader’s point of view. Personally, I found parts of the Results section overly detailed, which makes it harder to immediately grasp the key messages. A more direct statement of the main findings at the beginning of each Results subsection would improve readability. In addition, the findings related to vaccine dose number and mixed versus same regimens should be consistently framed as exploratory, to avoid overinterpretation. Overall, my impression is that the manuscript is now methodologically solid, balanced in tone, and much clearer in its message. With minor refinements focused on clarity and emphasis, it is well positioned for publication.

Comments on the Quality of English Language

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

The manuscript has been improved and can be published!

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 4 Report

Comments and Suggestions for Authors

The manuscript now clearly defines the study population as adults hospitalised with acute COVID-19, and have removed the unclear population from the dataset, which has made the findings much more sensible. However, the implications of restricting the analysis to a high-severity cohort could be more explicitly integrated into the interpretation of the vaccination findings (this is a minor correction).

In particular, the lack of statistical significance for the protective effect of pre-infection vaccination against long COVID may plausibly reflect effect modification by baseline disease severity, rather than an absence of vaccine benefit per se. In a population already sick enough to require hospital admission, the dominant drivers of long COVID risk such as comorbidity burden, prolonged length of stay, and older age; may overwhelm or attenuate the marginal downstream protective effect of vaccination. This is supported by the authors’ own findings, which show comorbidities and acute severity as the strongest independent predictors of long COVID. I think this should be mentioned explicitly in the discussion.

Clarifying this point explicitly would strengthen the manuscript and help prevent misinterpretation of the null result as evidence of vaccine ineffectiveness. Emphasising that vaccination may exert a larger protective effect in lower-severity or community-based cohorts, as has been seen in other studied, would also aid contextualisation and alignment with existing literature. I recommend that the authors explicitly discuss effect modification by acute disease severity when interpreting the non-significant association observed in the prevention cohort.

Finally, there is a lot of literature on this question which the authors should acknowledge in the discussion section of their work:

https://pmc.ncbi.nlm.nih.gov/articles/PMC9978692/

https://pmc.ncbi.nlm.nih.gov/articles/PMC12644529/

https://www.thelancet.com/journals/lanres/article/PIIS2213-2600%2824%2900082-1/fulltext?

https://academic.oup.com/cid/advance-article/doi/10.1093/cid/ciaf549/8268019?

https://www.sciencedirect.com/science/article/pii/S0163445324002937

https://pmc.ncbi.nlm.nih.gov/articles/PMC10849259/

Author Response

Please see the attachment

Author Response File: Author Response.docx