First-dose ChAdOx1 and BNT162b2 COVID-19 vaccines and thrombocytopenic, thromboembolic and hemorrhagic events in Scotland

Reports of ChAdOx1 vaccine–associated thrombocytopenia and vascular adverse events have led to some countries restricting its use. Using a national prospective cohort, we estimated associations between exposure to first-dose ChAdOx1 or BNT162b2 vaccination and hematological and vascular adverse events using a nested incident-matched case-control study and a confirmatory self-controlled case series (SCCS) analysis. An association was found between ChAdOx1 vaccination and idiopathic thrombocytopenic purpura (ITP) (0–27 d after vaccination; adjusted rate ratio (aRR) = 5.77, 95% confidence interval (CI), 2.41–13.83), with an estimated incidence of 1.13 (0.62–1.63) cases per 100,000 doses. An SCCS analysis confirmed that this was unlikely due to bias (RR = 1.98 (1.29–3.02)). There was also an increased risk for arterial thromboembolic events (aRR = 1.22, 1.12–1.34) 0–27 d after vaccination, with an SCCS RR of 0.97 (0.93–1.02). For hemorrhagic events 0–27 d after vaccination, the aRR was 1.48 (1.12–1.96), with an SCCS RR of 0.95 (0.82–1.11). A first dose of ChAdOx1 was found to be associated with small increased risks of ITP, with suggestive evidence of an increased risk of arterial thromboembolic and hemorrhagic events. The attenuation of effect found in the SCCS analysis means that there is the potential for overestimation of the reported results, which might indicate the presence of some residual confounding or confounding by indication. Public health authorities should inform their jurisdictions of these relatively small increased risks associated with ChAdOx1. No positive associations were seen between BNT162b2 and thrombocytopenic, thromboembolic and hemorrhagic events.

T he Coronavirus Disease 2019 (COVID-19) immunization program in the United Kingdom (UK) recommends COVID-19 vaccination for adults aged 18 and over. An independent UK-wide body, the Joint Committee on Vaccination and Immunisation (JCVI), has recommended vaccination of all adults beginning with those at highest risk of serious COVID-19 outcomes-in particular, hospitalizations and deaths (Supplementary Table 1) 1 . The three vaccines currently being administered in the UK-ChAdOx1 nCoV-19 (Oxford-AstraZeneca, hereafter ChAdOx1), BNT162b2 mRNA (Pfizer-BioNTech, hereafter BNT162b2) and mRNA-1273 (Moderna)-have been shown to reduce COVID-19 infections, hospitalizations and deaths [2][3][4][5] . Given that the first dose of mRNA-1273 was given in Scotland only on 7 April 2021, this article will focus on ChAdOx1 and BNT162b2.
The risk of adverse events after vaccine administration has been assessed in clinical trials. These found that the ChAdOx1 and BNT162b2 vaccines have been generally well tolerated 6,7 . The most commonly reported adverse events included injection site reactions, such as pain, redness, tenderness and swelling, and non-serious systemic reactions, such as myalgia, headache, nausea and fever [6][7][8] .
Reports of serious adverse events have been rare 9 . However, concerns have been raised over the safety of the ChAdOx1 vaccine, and this has resulted in several countries initially temporarily suspending and then restricting use of ChAdOx1 to certain age groups [10][11][12][13] . By 4 April 2021, the European Medicines Agency (EMA) had received 169 reports of central venous thromboembolic events 14,15 , and an EMA signal assessment report on 8 April 2021 concluded that a signal of disproportionality was noted for rare events, such as disseminated intravascular coagulation, cerebral venous sinus thrombosis (CVST), as well as arterial thromboembolic and hemorrhagic stroke, which warranted further investigation 16 . The UK's Medicines & Healthcare products and Regulatory Agency (MHRA) had received, as of 21 April 2021, 209 reports of thrombocytopenic and thromboembolic cases, after 22 million first doses and 6.8 million second doses of the ChAdOx1 vaccine 17 . Three separate case series have described patients who developed thrombocytopenic and thrombotic events after ChAdOx1 vaccination, which clinically mimics heparin-induced thrombocytopenia [11][12][13] . ChAdOx1's summary of product characteristics has now been updated accordingly. The JCVI has recommended that 18-39-year-old individuals who do not have an underlying health condition should be offered an alternative to ChAdOx1, if available 18,19 . There have also been reports of post-vaccination exacerbation of chronic idiopathic or immune thrombocytopenic purpura among individuals receiving mRNA vaccines (including BNT162b2) [20][21][22] .
Given ongoing public, professional and regulatory concerns and the limited body of population-based evidence, there is an urgent need for evidence on the safety of all COVID-19 vaccines and, in particular, any association between COVID-19 vaccines and ITP and venous thromboembolic (including CVST) and arterial thromboembolic and hemorrhagic events. To investigate this, we used a national prospective COVID-19 surveillance cohort in Scotland, which consisted of linked databases containing individual patient-level data relating to vaccination status, virological reverse transcription polymerase chain reaction (RT-PCR) COVID-19, laboratory tests and clinical and mortality records covering 5.4 million people (~99% of the Scottish population).

Results
Between 8 December 2020 and 14 April 2021, 2.53 million people (57.5% of the adult population aged ≥18 years) received first doses of COVID-19 vaccines in Scotland. Of these, 1.71 milion people were vaccinated with the ChAdOx1 vaccine, and 0.82 million people were vaccinated with the BNT162b2 vaccine (Extended Data Fig. 1), with fewer than 10,000 people receiving the mRNA-1273 vaccine.
No increased risk of thrombocytopenia events was found for the BNT162b2 vaccine ( Table 2) during any of the post-vaccination time periods. There was no association found for the BNT162b2 vaccine in the SCCS analysis (Supplementary Table 3 (Table 3). In the adult population studied, the difference in expected versus observed events was 0.46 events (95% CI, −0.44-1.33) per 100,000 doses.
Patients who had ITP after ChAdOx1 vaccination compared to those who were unvaccinated at the time of the event tended to be older (median age, 69 years versus 54 years, P = 0.01), to be more likely to have at least one clinical risk condition (85% compared to 40%, P = 0.001) and to have been in hospital at the time of the event (52% versus 28%, P = 0.05). The gender distribution was the same for ITP after ChAdOx1 vaccination compared to those who were unvaccinated. For the 22 patients with post-vaccination ITP and for whom platelet counts were available after vaccination, all but two had counts below 100,000 per µl. In addition, 48% of patients with post-ChAdOx1 ITP had prior prescriptions that could induce ITP, compared to 35% of those who were unvaccinated at the time of their ITP event. Five or fewer (≤10%) patients with ITP were prescribed ITP therapies by general practitioners in the community after vaccination with ChAdOx1.
No positive association was found between the BNT162b2 vaccine and ITP (Table 2): aRR at 0-27 d after vaccination was 0.54 (95% CI, 0.10-3.02). There was also no clear evidence of association found for the BNT162b2 vaccine in the SCCS analysis (Supplementary Table 3).
In total, three deaths were reported after ITP. These deaths occurred in both vaccinated and unvaccinated individuals all aged over 70 and for reasons not associated with ITP.
The number of events per day for ITP since September 2019 is available in Extended Data Fig. 2, showing stable rates until January 2021 when the ChAdOx1 vaccine was introduced in the UK, followed by an increase in the number of events per day. Dates of vaccination and type of vaccine for individuals with an ITP event during the study period are available in Extended Data Fig. 3.

Venous thromboembolic events.
We found no association between prior ChAdOx1 vaccination and venous thromboembolic events (including CVST) at 0-27 d after vaccination (aRR = 1.03, 95% CI, 0.89-1.21) or for the BNT162b2 vaccination 0-27 d after vaccination (aRR = 0.50, 95% CI, 0.40-0.62). No increase in the odds of venous thromboembolic events was found during any of the post-vaccination time periods analyzed for ChAdOx1 (Table 1) or BNT162b2 (Table 2) vaccines. The SCCS analysis for ChAdOx1 and venous thromboembolic events found an RR = 0.94 (95% CI, 0.87-1.02) (Supplementary Table 3). No association was found in the SCCS analysis for the BNT162b2 vaccine (Supplementary Table 3) The difference between observed versus expected cases for younger age groups after ChAdOx1 vaccination was 10.41 (95% CI, 4.64-16.65) per 100,000 doses for 16-39-year-old individuals and 9.60 (95% CI, 5.57-13.69) per 100,000 doses for 40-59-year-old individuals (Table 4). In the adult population included in the study, the difference in expected versus observed events was 4. CVST. When focusing on CVST events, 19 total incident events were recorded among all people (vaccinated and unvaccinated). Among those vaccinated before the incident event, there were insufficient events to adequately power an analysis (n = 6). Post-vaccination incident events were recorded for both vaccines. Two of those individuals with a post-vaccination CVST event died. Platelet count results were available for 17 of 19 individuals identified as having CVST. There was no evidence of platelet counts <150,000 per µl at any time point in any of the 17 individuals with a post-vaccination CVST event.
There was no increased risk of arterial thromboembolic events associated with BNT162b2 vaccination at 0-27 d after vaccination (aRR = 0.92, 95% CI, 0.81-1.04) or during any of the post-vaccination time periods analyzed (Table 2 and Supplementary Table 3).
Multiple outcomes and predictors. Overlap in our outcomes with individuals presenting with multiple events of interest was rare. Among those vaccinated with ChAdOx1, there were six occurrences of ITP with hemorrhagic, venous thromboembolic or thrombocytopenia events, with the most common combination being ITP and thrombocytopenia.
We found increased risk for post-vaccination ITP and arterial thromboembolic and hemorrhagic events combined associated with increasing age (especially over 60), male sex, having certain comorbidities (such as heart failure, coronary heart disease, peripheral vascular disease, severe mental illness, sickle cell disease, prior stroke, type 1 and 2 diabetes and chronic kidney disease (stage 5)), very high blood pressure and smoking (Supplementary Table 5 Table 2).

Discussion
This Scottish national population-based analysis among 2.53 million people who received their first doses of SARS-CoV-2 vaccines reveals a potential association between receiving a first-dose ChAdOx1 vaccination and occurrence of ITP, with an incidence of 1.13 cases per 100,000 vaccinations. For ChAdOx1 vaccination, there was suggestive evidence of an association with arterial thromboembolic and hemorrhagic events. For these outcomes, an attenuation of effect was found in the SCCS analysis, which might indicate the presence of residual confounding or confounding by indication. For any venous thromboembolic event, there were more observed than expected events for younger age groups (16-59 years old) associated with ChAdOx1, but this was not seen in our primary incident case-control analysis. Because of our limited ability to match in the observed versus expected analysis, this finding should be treated with considerable caution. There were 19 incident CVST events seen in our study population: 6 of these occurred after vaccination, with these events being seen after both ChAdOx1 and BNT162b2 vaccines. Two individuals with post-vaccination CVST events died. For CVST (and other rare conditions), there were insufficient numbers to draw any reliable conclusions other than, if there is any association, it is likely to represent an extremely rare outcome. For the BNT162b2 vaccine, our analysis found no evidence of increased adverse events for the thrombocytopenic, venous thromboembolic or hemorrhagic outcomes of interest.
To our knowledge, this is the one of the first real-world contemporaneous studies identifying all vaccinated individuals within a national population and assessing COVID-19 vaccine-related thrombocytopenic, venous or arterial thromboembolic and hemorrhagic adverse events. One published study of people aged 18-65 years who received the ChAdOx1 vaccine in Denmark and Norway observed increased rates of venous thromboembolic events, including cerebral venous thrombosis (standardized morbidity ratio of 1.97 and 95% CI, 1.50-2.54) and intracerebral hemorrhage (standardized morbidity ratio of 2.33 and 95% CI, 1.01-4.59) 23 . In that study, Pottegård et al. found a standardized morbidity ratio for any thrombocytopenia/coagulation disorders of 1.52 (0.97-2.25) and for any bleeding of 1.23 (0.97-1.55).
Our study has several strengths, including our ability to rapidly access and analyze data on vaccination status and medical and death records from linked national databases [24][25][26] . This study is, therefore, less susceptible to recall or misclassification bias than studies of samples of the population. A large population aided study power to facilitate the analysis of rare events such as ITP. We think that our findings have generalizability across countries using these vaccines as part of national vaccination programs that have prioritized vaccination of high-risk populations.
Our study has several limitations. As few individuals had received two vaccine doses at the time of analysis, this (second-dose) The observed events were incident cases after vaccination counted for the observed duration of the post-vaccination period. b The expected events are the number of events per day in the pre-vaccination period divided by the population and multiplied by the days at risk in the post-vaccination period for all vaccinated and then summed. c The difference between expected (the number of events per day in the pre-vaccination period divided by the population and multiplied by the days at risk in the post-vaccination period for all vaccinated and then summed) and observed events during the post-vaccination period, with CIs obtained from a parametric bootstrap, was based on the Poisson distribution using 10,000 samples. There was limited opportunity for matching in this analysis, and the findings, therefore, need to be interpreted with caution. d n ≤ 5 denotes minimum allowable reported value.
subgroup was not investigated separately. A further analysis on second doses will be conducted in due course. Furthermore, our study included few young vaccinated people (<40 years), especially for the ChAdOx1 vaccine, because the vaccination program has been predominantly targeted by age and underlying comorbidities so far. Although electronic general practice records of hospitalization and deaths were accessible, and linked with hospitalization and mortality records, lags in final coded hospital discharge data and postmortem changes to death certification might have resulted in over-riding of initial recorded causes of hospitalizations and deaths in some instances. However, our sensitivity analysis restricting to an earlier date of follow-up is less subject to such potential biases and found similar results. Additionally, ITP is a diagnosis of exclusion. Given that we based this analysis on clinician-recorded data, we had to assume that clinicians had appropriately investigated patients for their thrombocytopenia before recording this diagnosis. Discussions with Scottish hematologists indicated that this was a reasonable assumption, as the diagnosis of ITP is made only by specialists in a Scottish context. However, there can be uncertainty about the diagnosis of ITP, and published experience indicates that the diagnosis of ITP is often changed when patients are followed by skilled hematologists 27,28 . There is also the possibility that some of these cases of ITP could have represented reactivation of disease that had been in remission for more than 1 year. We did, however, carry out a post hoc analysis of all post-vaccination ITP events for those with prior available platelet count (tested in the primary care setting) and relevant prescriptions that could cause thrombocytopenia. We also carried out an analysis of ITP-directed therapy after vaccination (including oral corticosteroids). We were unable to access blood smear information as this is not routinely captured in the record systems that we had access to. Furthermore, 48% of patients with post-ChAdOx1 ITP events had prior prescriptions that could induce ITP, compared to 35% of those who were unvaccinated at the time of their ITP event. ITP-directed therapy prescribed by general practitioners in the community to patients with post-vaccination ITP was uncommon (≤10%). The overwhelming majority of ITP-directed therapy (for example, pulsed dexamathasone, prednisolone with or without intravenous immunoglobulin, rituximab and immunosuppressants) is, however, is likely to have been initiated in the hospital setting by hematologists, but these data were not accessible to us, as hospital prescribing in Scotland remains predominantly paper based 29 . These patients with community prescribing of oral corticosteroids are likely to have had persistent ITP that was managed in primary care. Although we used a nested case-control study design matched by age, sex and geography, and adjusted for several confounders, unmeasured confounders could still have influenced our estimates (Supplementary Table 2). To mitigate this risk, where associations between ChAdOx1 and any adverse event were seen (that is, ITP and arterial thromboembolic and hemorrhagic events), we conducted a confirmatory post hoc SCCS analysis. SCCS designs can account for time-invariant confounding but are less suitable where recurrent events are not independent. Although the pattern of findings was largely similar across different analytical approaches, it is worth noting that the magnitude of associations did differ. Estimates tended to be greater in the case-control analysis, which could arise from potential residual confounding by indication that would most likely result in an overestimate of the real effect sizes. By contrast, the SCCS analysis tended to estimate smaller effect sizes, but the potential correlation of outcomes within an individual over time could bias estimates toward the null. The two approaches, therefore, provide reasonable bounds for the true effect, with our primary results potentially overestimating the risk of vaccine-associated harm and, therefore, being the most conservative for decision-making. Owing The observed events were incident cases after vaccination counted for the observed duration of the post-vaccination period. b The expected events are the number of events per day in the pre-vaccination period divided by the population and multiplied by the days at risk in the post-vaccination period for all vaccinated and then summed. c The difference between expected (the number of events per day in the pre-vaccination period divided by the population and multiplied by the days at risk in the post-vaccination period for all vaccinated and then summed) and observed events during the post-vaccination period, with CIs obtained from a parametric bootstrap, was based on the Poisson distribution using 10,000 samples. There was limited opportunity for matching in this analysis, and the findings, therefore, need to be interpreted with caution.
to the small number of adverse events, to identify predictors among vaccinated individuals we combined the outcomes of interest: ITP and arterial thromboembolic and hemorrhagic events. This as an area for future work, for instance through a meta-analyses of vaccine safety studies. Finally, the EAVE II platform is a national public health surveillance platform that was established at the request of the Scottish Government to help inform the public health response to the pandemic. It brought together a range of national whole-population healthcare datasets for the first time into Public Health Scotland. Ethical permission for this study was granted, and the Public Benefit and Privacy Panel Committee of Public Health Scotland approved the linkage and analysis of the de-identified datasets. As the policy aim was for national coverage, it was not feasible to obtain individual patient consent. This, therefore, restricted our ability to interrogate and report on certain individual record data in detail. For CVST, for instance, there were very few events, and, in keeping with our permissions, we suppressed the actual number of events found to minimize the risk of inadvertent disclosure of identity 30 . Also, centralized adjudication of our outcomes through case record review by an independent group of experts was not possible because access to data was limited to a small number of approved researchers. The Centers for Disease Control and Prevention estimated that 60,000-100,000 Americans die annually due to venous thromboembolism (United States of America: 2.8 million deaths annually) 31 . Venous thromboembolic events are common in patients with COVID-19. Approximately 10% of patients with COVID-19 in hospitals (non-intensive care unit (ICU)) are diagnosed with venous thromboembolism and 28% of those in ICU 11,32,33,34 . The vaccine-induced adverse events after administration of the adenovirus-based SARS-CoV-2 vaccines (including the ChAdOx1 vaccine) have been described as vaccine-induced immune thrombotic thrombocytopenia (VITT) syndrome or thrombosis with thrombocytopenia syndrome resulting in a venous or arterial thrombosis, including CVST and thrombocytopenia 35 . The syndrome has been characterized as being similar to heparin-induced thrombocytopenia, a pro-thrombotic adverse drug reaction caused by the transient production of platelet-activating antibodies of IgG class that recognize multi-molecular complexes of (cationic) platelet factor 4 bound to (polyanionic) heparin 36 . We were insufficiently powered to provide estimates of the rarer VITT CVST and splanchnic vein thrombosis. This is an area for further work likely best pursued through larger datasets and meta-analyses.
ITP has also emerged as an important complication of COVID-19, with early epidemiological evidence suggesting a rate of 0.34% among hospitalized patients. There have also been reports of post-vaccination ITP in patients who received mRNA vaccines (including BNT162b2), and it has been postulated that some individuals might have had mild 'compensated' thrombocytopenia of diverse causes, and severe thrombocytopenia might have been induced by enhancement of macrophage-mediated clearance or impaired platelet production as part of a systemic inflammatory response to vaccination 21 . ITP, however, as an adverse event after vaccine administration, is very rare. Our study suggests that there might be an increase in the risk of this very rare outcome for ChAdOx1 that is similar to other vaccines, including hepatitis B; measles, mumps and rubella; and influenza 37,38 . This very small risk is important but needs to be seen within the context of the very clear benefits of the ChAdOx1 vaccine.
As a result of findings from UK pharmacovigilance and surveillance data (including from EAVE II investigators), advice was issued in April 2021 regarding age group limits for the ChAdOx1 vaccine for individuals younger than 30 years of age 18 and then, in May 2021, for individuals younger than 40 years of age 19 . Replication of our study in other countries is needed to confirm our results. We plan to update our analysis as the vaccine program is extended to younger, healthier individuals and as new vaccines become available. We also plan to extend our pharmacovigilance efforts to cover the second doses of these and other vaccines.
In conclusion, we did not identify any overall increased risk in the adverse events of interest in individuals receiving BNT162b2. First dose of ChAdOx1 was found to be associated with small increased risks of ITP, with suggestive evidence of an increased risk of arterial thromboembolic and hemorrhagic events. Given these small increased risks for ChAdOx1, alternative vaccines for individuals at low COVID-19 risk might be warranted when supply allows.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/ s41591-021-01408-4.

Methods
Ethics and permissions. Ethical permission for this study was granted by the South East Scotland Research Ethics Committee 02 (12/SS/0201). The Public Benefit and Privacy Panel Committee of Public Health Scotland approved the linkage and analysis of the de-identified datasets for this project (1920-0279).
Study setting and population. The National Health Service in Scotland (NHS Scotland) provides comprehensive health services that are free at the point of care for all residents. Our base population for this study was 5.4 million residents (~99% of the population) registered with a general medical practice in Scotland.
Study design. Following a pre-specified analysis plan, we started with a matched case-control study nested within the EAVE II prospective cohort. A case was defined as anyone with a recorded incident event of thrombocytopenia, venous thromboembolism, arterial thromboembolism or hemorrhage after the start of the COVID-19 vaccination program in Scotland. The general practice clinical data contained records of the specified events from 1 September 2019 to 14 April 2021. An incident case was defined as the first event in the period from when vaccination started on 8 December 2020 with no prior thrombocytopenic, venous thromboembolic or hemorrhagic clinical events since 1 September 2019. We matched cases who had experienced the outcomes of interest during the period from 8 December 2020 to 14 April 2021 based on recording of demographic and pre-existing comorbidities (Supplementary Table 2) to controls who had not yet experienced the outcome in an incident-matched nested case-control design. Incident cases were matched to controls based on age (in exact years up to 80 years old and then 2-year age groups up to 90 years old and 5-year age bands above that to allow for sparse data), sex and area of residence using Intermediate Zone classification. There are 1,271 such zones in Scotland (comprising between 2,500 and 6,000 households, with an average population of 4,200), which are geographically aligned with general practices. Ten controls were selected per case. Diagnosis dates of the cases were taken as the index dates for the controls.
We investigated confounding by indication and carried out a post hoc SCSS analysis to estimate the risk of COVID-19 vaccination and ITP events using conditional logistic regression with an offset for the length of the risk period 39 .
Data sources. Almost all residents in Scotland are registered with a general practice and have a unique Community Health Index (CHI) number used by NHS Scotland. We used the CHI number to deterministically link all datasets with vaccination records in Public Health Scotland (Extended Data Fig. 4). Vaccination information was extracted from the general practice records and the Turas Vaccination Management Tool system; together, these captured all vaccination records, including those vaccinated in general practices, community vaccination hubs and other settings, such as care homes and hospitals in Scotland 40 . Further details on the data sources used in this study are available in a published project protocol 24 .
Exposure definition. We studied the first doses of the BNT162b2 (ref. 41 ) and ChAdOx1 (ref. 42 ) vaccines. An individual was defined as exposed to each vaccine if they received their first dose of vaccine between 8 December 2020 and 14 April 2021. Given the limited number of people who had received their second vaccine dose by this time, we have reported only on first-dose-associated events.
Outcomes. The outcomes analyzed in this study were one or more (1) thrombocytopenic, (2) arterial or venous thromboembolic or (3) hemorrhagic (excluding traumatic, gastrointestinal and genitourinary bleeding) events. We undertook additional a priori subgroup analyses focused on ITP and CVST and post hoc subgroup analyses for DVT and PE 43 . Read codes (version 2) were used to determine adverse incident events recorded in the primary care electronic health record (Supplementary Table 6), which were then followed up in the linked RAPID dataset and National Records Scotland for hospitalization and mortality outcomes, respectively (Extended Data Fig. 4).
Covariates. Data on the number of comorbidities recorded in the general practice record were derived from the QCovid risk groups on 1 December 2020 (ref. 22 ). Socioeconomic status was measured by quintiles of the Scottish Index of Multiple Deprivation 18 . The number of pre-vaccination program RT-PCR tests was extracted from the records as a marker of being in a population at high risk of exposure (for example, health and social care workers in whom regular testing was recommended).

Statistical analysis.
Conditional logistic regression analysis was used with case and control matching groups as the strata. The unadjusted model included no covariates other than the strata and the exposure. Separately, a simple adjustment was made for the number of clinical risk groups per patient 44 . Then, a full adjustment included socioeconomic status 24 and number of RT-PCR tests an individual had before 8 December 2020, because we considered a priori these variables to affect vaccination receipt (all three) and risk of outcomes (risk groups and deprivation). Odds ratios for being a case among vaccinated versus unvaccinated individuals were estimated from the logistic regression, which, given the incidence-density sampling design, are mathematically equivalent to RRs. Unadjusted and adjusted RRs, along with 95% CIs, were calculated. Additionally, we investigated the individual QCovid risk groups as potential confounding variables (through statistical adjustment) or effect-modifying variables (through interaction) 44 . We also identified predictors (QCovid risk groups and age) for our combined outcomes events of interest, namely ITP and hemorrhage and arterial thromboembolic events.
We carried out an analysis of observed incident cases in the post-vaccination period compared to those expected from a pre-vaccination period. The pre-vaccination period used was the 28-d period after 1 October 2020. For the event to be incident, there were no thrombocytopenic, venous thromboembolic or hemorrhagic events in the same individual in the period from 1 September 2019. In the post-vaccination period, clinical events were linked to the vaccination records, and incident cases after vaccination were counted for the observed duration of the post-vaccination period. The expected events were the number of events per day in the pre-vaccination period divided by the population and multiplied by the days at risk in the post-vaccination period for all vaccinated and then summed. The difference between expected and observed events during the post-vaccination period was used to calculate expected additional cases per 100,000 vaccine doses. CIs were obtained from a parametric bootstrap based on the Poisson distribution using 10,000 samples.
Analyses were carried out by one statistician (C.R.) and independently checked by additional statisticians (E.M., U.A. and R.M.). All analyses were carried out with R software version 3.6.1 (ref. 45 ).
Post-hoc analyses. The nested case-control method was used in our primary analysis, as opposed to an SCCS method, because recurrent adverse events of interest were considered unlikely to be independent. Nonetheless, we investigated confounding by indication and carried out a post hoc SCSS analysis to estimate the risk of COVID-19 vaccination and the adverse events of interest using conditional logistic regression with an offset for the length of the risk period 39 . This analysis compared risk for the same individuals in the time periods before and after vaccination. We considered the time period from the date of first vaccination dose to 28 d after as the risk (exposed) period. The control (pre-risk) period was the 90-d period before 14 d before vaccination (that is, 15-104 d before vaccine receipt), allowing for a 14-d clearance period. The main comparisons were the rate of adverse events between (1) the risk period and the pre-risk period and (2) the clearance period and the pre-risk period (Fig. 1).
As an additional post hoc analysis to understand whether cases of ITP could have represented reactivation of disease and whether relevant prescriptions that could cause thrombocytopenia were prescribed and ITP-directed therapy after vaccination was administered, platelet counts for individuals with the outcome were extracted from the general practice electronic health record (Extended Data Fig. 4), and prescriptions were extracted that could cause thrombocytopenia (including amiodarone cephalosporins, ciprofloxacin, ethambutol, furosemide, glycoprotein IIb/IIIa inhibitors (tirofiban, abciximab and eptifibatide), haloperidol, heparin, ibuprofen, irinotecan, linezolid, mirtazapine, naproxen, oxaliplatin, paracetamol, penicillins, phenytoin, quinidine, quinine, rifampicin, salmeterol, sodium valproate, sulphonamides, tacrolimus thiazides, trimethoprim and vancomycin). The following ITP-related prescriptions were extracted: azathioprine, ciclosporin, cyclophosphamide, danazol, dapsone, intravenous immunoglobulin, mycophenolate, oral corticosteroids, rituximab, vinca alkaloids, eltrombopag and romiplostim therapy.

Sensitivity analyses.
To reduce the potential for ascertainment bias, we carried out a sensitivity analysis with a censoring date before the first media reports of possible thrombotic events associated with ChAdOx1 (that is, 21 February 2021) 46 . An analysis to explore any influence on testing (RT-PCR) positive before an event (restricting our analysis to those who never tested positive previously) was undertaken. Fig. 1 | Schematic presentation of the self-controlled case series study design. Blue, the control (pre-risk) period: the 90-day period prior to 14 d before vaccination (i.e., 15-104 d before vaccine receipt). Green, a 14-day clearance period. Orange, the time period from the date of first vaccination dose to 28 d after, as the risk (exposed) period. . This is derived from the cohort set up on 1 March 2020 by removing individuals who died. Individuals in the population who died before vaccination will count in the denominator but cannot count in the numerator and this explains the drop off in uptake percentage in the elderly.