← Sample reviews

Review of COV009 vaccine follow-up

Independent demonstration review of a published article

Overall assessment and scientific impact

COV009 makes a useful contribution to the record of the UK COVID-19 vaccine rollout. Its strongest feature is the detailed follow-up of 4,468 former trial participants as they moved from controlled trial schedules into varied community vaccination histories. That transition is scientifically interesting in its own right. The cohort documents which schedules people actually received, the serious events captured during follow-up and how anti-spike IgG changed in the samples available for analysis. The long follow-up and detailed appendix give pharmacovigilance researchers a way to contextualise those observations and compare them with other surveillance designs. The likely impact is as a descriptive reference for mixed schedules and as a record of the extension cohort. The size of the cohort, its delayed entry, the concentration of withdrawals in one exit-defined vaccine group and the small numbers for individual outcomes limit its ability to resolve product-specific safety questions. The immunology results add longitudinal context, with interpretation depending on who provided paired samples and what vaccinations or infections occurred between visits. The paper would make a stronger contribution by explicitly defining the period and population for each result and by bringing the most consequential appendix findings into the main text. Article · Journal supplement

Editorial recommendation

Major revisions at pre-submission. This is a retrospective editorial assessment of the published article. Retain the cohort description and the event inventory. Correct the two prose percentage discrepancies, report follow-up completeness and observed person-time by group, and revise the abstract and Discussion around the descriptive estimand. Add uncertainty and case-definition details to the background comparisons. Reconcile protocol, registration and publication in a dated outcome crosswalk, and show the antibody analysis denominators by visit and paired contrast. If the authors wish to retain product-specific risk claims, those claims require a new dose-timed analysis with an appropriate comparator and defined surveillance windows.

Priorities

Findings

MRT-01 Separate observed follow-up from dose-specific safety

Evidence. The article reports 174 serious adverse events (SAEs) in 159 people and 71 adverse events of special interest (AESIs) in 68 people. Thirty-three events overlap across the categories in 32 people. This reconciles to 212 distinct events and 195 participants with at least one event, using the reported overlap. Median time from first COVID-19 dose to the first COV009 study visit was 574 days. The three groups describe final recorded histories. Of the 4,468 analysed participants, 3,069 received ChAdOx1 plus mRNA or a vaccine whose identity was unrecorded, 385 received ChAdOx1 only and 1,014 fell in the mRNA-only category as defined by the paper. The extension groups differ from randomised assignment in the parent trials. Article, §§2.1–2.4, 3.1, 3.4–3.5 and Tables 2–3

Why it matters. An event can occur before the dose that determines a participant’s final category, and earlier post-dose events may fall outside the extension window. The clinician’s event-level relationship assessment and an estimate of population-level excess risk answer different questions.

Action and closure. Make the time origin, ascertainment periods and censoring explicit. Present the descriptive event totals with observed person-time and distinguish them from the original randomised groups. Any product-specific follow-up analysis needs time-varying exposure, pre-defined windows, appropriate comparators and adjustment for confounding. Label the absence of a detected signal in terms of the observed cohort and surveillance window.

MRT-02 Put uncertainty and ascertainment alongside the expected-rate comparison

Evidence. Supplement Table C8 uses 48,116 participant-months, about 4,009.7 person-years, and an external annual Guillain–Barré syndrome rate of 0.0000139. That gives 0.0557 expected cases. With one observed case, the unrounded observed-to-expected ratio is 17.94, consistent with the reported 18. An exact Poisson interval based on the observed count and treating expected cases as fixed is approximately 0.45 to 100. It captures counting uncertainty alone. The reported case was in the mixed-history group. The table also includes anosmia despite stating that COV009 did not designate it as an AESI, and classifies unspecified cerebrovascular accidents as ischaemic in its comparison. Article, Discussion · Journal supplement, Tables C3/C5/C8

Action and closure. Show observed counts, person-time, expected counts and intervals. Document the outcome coding and exposure windows for the external comparison and standardise expected rates to available demographic and calendar strata. Exclude outcomes with incompatible ascertainment from comparative interpretation or clearly flag the mismatch. The aggregate check does not establish the cause of the observed case.

MRT-03 Reconcile selection, withdrawal and event observation

Evidence. The article identifies 11,889 eligible parent-trial participants, 4,470 who enrolled, 4,468 analysed and 3,845 analysed participants who completed follow-up. Among those analysed, 623 (13.9%) did not complete. The authors already report differences by age between completers and non-completers and provide withdrawal reasons and baseline summaries. These summaries describe selection but do not quantify how much follow-up or outcome ascertainment was lost. Article, §§3.1–3.2 and Discussion · Journal supplement, Tables A4–A6 · Final study report

Action and closure. Report start and end of observed person-time by vaccine history or original arm as appropriate, age, withdrawal reason and ascertainment source. Clarify whether outcomes after the last study visit were captured through routine records and what happened after consent withdrawal. Where feasible, compare estimates under plausible informative-censoring scenarios. Keep event counts and participants affected distinct.

MRT-04 Document changes between protocol, registry and publication

Evidence. The registry lists SAE and AESI ascertainment as the primary outcome and includes neutralising antibodies alongside anti-spike immunoglobulins among secondary measures. The article presents total anti-spike IgG for the main immunology analysis and refers separately to previous reporting of the fourth-dose subgroup. Protocol v3.0 §9 describes descriptive analyses by original randomised group and time since parent-trial vaccination, with medians and interquartile ranges for immune outcomes. The article’s main safety tables group participants by vaccination history at exit and its principal immune summaries use geometric means and ratios. The registry PDF lists 1,077 as final enrolment, compared with 4,470 in the article. ISRCTN registration · Protocol v3.0 · Article, Methods and Results · Journal supplement, Appendix D

Action and closure. Identify the analysis plan or dated amendments that account for changes, link each registered outcome to its result or reason for deferral, and reconcile the registration’s enrolment field. Specify whether neutralisation in the main extension cohort was measured, analysed or reported elsewhere. The published material can establish differences in presentation and scope. The chronology of analytical decisions needs dated documentation.

MRT-05 Distinguish available-case antibody summaries from longitudinal change

Evidence. In supplement Table D1, the geometric mean concentrations use 44, 40 and 38 specimens at the reference visit, SV1 and SV2. The two geometric mean ratios use 38 and 36. Dividing the displayed available-case means gives 8.94 and 29.82, compared with reported ratios of 9.34 and 32.90. Different paired subsets can explain this difference. The methods should say explicitly which participants entered each contrast and how its interval was calculated. The paper describes random selection for the immunology cohort and later describes the available immunology data as a small, non-randomised subset. These statements can coexist if selection for invitation and availability for analysis differed. Additional vaccination and infection also occurred during follow-up. Article, §§2.3–2.4, 3.6 and Discussion · Journal supplement, Appendix D

Action and closure. Show invitation, consent, sampling and paired-analysis counts by visit. Specify the estimator, within-person uncertainty and handling of assay limits and missing samples. Describe interim vaccination and infection histories where available. Frame the rising observed IgG as a result in the available cohort under mixed exposures.

MRT-06 Correct two percentages in the text

Evidence. The SAE paragraph assigns 86% and 14% to 146 and 28 of 174 events. The counts give 83.9% and 16.1%, consistent with Table 2’s 84% and 16%. The Discussion gives 86% for 7,771 mRNA community doses out of 9,254, which is 84.0%. Article, §3.4, Table 2 and Discussion

Action and closure. Verify both denominators against the analysis tables and generate prose percentages from the same underlying values. The reported counts themselves reconcile in these two checks.

Supplement interrogation

The audit followed participant counts through Appendix A, vaccination exposures through Appendix B, event definitions and comparison denominators through Appendix C, and paired antibody denominators through Appendix D. The calculations below were independently recomputed from the source tables. Journal supplement

Rare-event comparison rebuilt from Table C8

The table below uses the displayed annual rates and 48,116 participant-months. Expected count equals annual rate multiplied by participant-years. Exact two-sided Poisson count limits were divided by the expected count, treated as fixed. Values are rounded. The intervals quantify counting uncertainty and depend on comparable event definitions, ascertainment and risk periods. Journal supplement, Table C8

The comparison for ischaemic stroke depends heavily on the coding decision. Restricting the numerator to the one explicitly labelled ischaemic stroke changes the ratio from 1.17 to 0.19 against the same expected count. This is a coding sensitivity check. A defensible comparison would also harmonise the external outcome definition. The much lower observed death count motivates a check on population comparability and follow-up completeness before interpreting any background-rate comparison as a safety estimate. Journal supplement, Tables C1 and C8

Suggested revision to the Discussion

In this observational extension, we describe serious adverse events and events of special interest captured among 4,468 former trial participants. Most participants had mixed vaccine histories. Follow-up completeness differed across vaccination histories defined at exit, and the external comparisons used small counts for several outcomes. We therefore interpret the safety findings in the context of the extension cohort and its observed follow-up window.

Use the same scope in the abstract. State explicitly whether loss to scheduled follow-up also ended passive event ascertainment. Article · Journal supplement

Coverage and reproducibility

Sources

• Article DOI · Version of record PDF · Journal supplement
• ISRCTN16270186 registration · Protocol v3.0 · Final study report