This analysis across five large cohorts covering distinct periods before, during, and after the COVID-19 pandemic showed that incident TTS events were very rare. The crude incident event rate of TTS was 0.42 per 100,000 person-years in the pre-COVID-19 cohort, 0.48 and 0.47 in the COVID-19 Period 1 and Period 2 cohorts, respectively, and 2.41 in the AZD1222-vaccinated cohort. There were notable differences in demographic and clinical characteristics between the individuals in the different cohorts, preventing direct comparisons of crude outcome rates between cohorts. No incident TTS events were seen in the confirmed COVID-19 cohort, but the total at-risk time of 3052 person-years limited the likelihood of observing a very rare event such as TTS. Collectively, the findings from this analysis show that in the AZD1222-vaccinated cohort the crude incident TTS event rate was very low and in line with expectations, with no new additional safety signals concerning thrombosis and TCP detected in this study with regard to AZD1222 vaccination.

This study provides important insights into the nature of TTS and the impact of its varying definitions, thanks to the large study population and the range of analyses performed, including the number and variety of sensitivity analyses. Other strengths of our analysis relate to the quality of the data available from the PCSC of the Oxford-RCGP RSC and linked immunization and COVID-19 laboratory test results databases, and how representative these data are to the general population in England. Furthermore, the use of SNOMED clinical terms provided more granular, high-quality data compared with using Read Codes terminology, or compared with using International Classification of Diseases codes, as discussed in prior reports [12, 23,24,25]. By utilizing a primary 28-day risk window rather than larger risk windows, which were ultimately less relevant, our study also benefited from a deeper understanding of TTS at the time of the analysis than in earlier studies attempting to evaluate the frequency and characteristics of the condition.

Based on this methodology, the background crude incident TTS event rates in the pre-COVID-19 and COVID-19 Period 1 and Period 2 cohorts in our analysis are similar to or lower than rates reported previously. In the ACCESS (vACCine covid-19 monitoring readinESS) cohort study of ten European safety databases, the incidence of TE events with TCP ranged from 0.06 to 4.53 per 100,000 person-years [26]. An analysis of pre-COVID-19 (from 2017 to 2019) electronic medical records for individuals in six European countries found background rates of 1.0–8.5, 0.5–20.8, 0.1–2.5, and 0.1 per 100,000 person-years for DVT, PE, SVT, and CVST, respectively, with TCP within 10 days [27]. A UK population analysis also showed low background rates within these ranges [16]. Furthermore, an analysis of the US Truven MarketScan Commercial Claims and Encounters Database from 1 January to 31 December, 2019 gave background incident TTS event rates of 5.6–6.4 per one million persons when standardized to a 21-day risk window [12]. The higher rates in these analyses compared with our findings may be due to differences in the populations analyzed, data source(s) used, and methodologies for identifying events. For example, the populations and data may have been derived from a mix of primary and secondary care; in contrast, our analyses used data from primary care only, meaning that our analysis may exclude in-hospital TTS cases not reported in primary care records.

To prevent misclassification of COVID-19 in the early pandemic period, our analyses used a very strict definition of medically attended, confirmed COVID-19 infection, requiring both a positive PCR test and a clinical diagnosis in the primary care record. Therefore, comparisons of the findings in the confirmed COVID-19 cohort with other studies using broader definitions of COVID-19 should be treated with caution. In addition, the confirmed COVID-19 cohort was limited to cases occurring from 1 July, 2020, which excluded the first wave of the COVID-19 pandemic. Nevertheless, multiple previous studies have demonstrated the impact of SARS-CoV-2 infection on rates of TE and TCP. At the start of the pandemic, it was determined that COVID-19 was associated with coagulopathies and an increased risk of thrombosis [28, 29]. Additionally, both SARS-CoV-2 infection [14, 16] and long COVID-19 [30, 31] have been shown to be associated with an elevated risk of TE and TCP events. Given the increased risk of TE and TCP associated with long COVID [30, 31], as the number of people infected with SARS-CoV-2 or had a COVID-19 diagnosis increased over the course of the pandemic, the ‘background risk’ of TE and TCP in the overall population may similarly have increased over time. This may have become evident with longer term follow-ups across cohorts during the pandemic.

Data from the AZD1222-vaccinated cohort confirmed that TTS is a very rare event and that there appears to be an extremely low risk of incident events following AZD1222 vaccination. Importantly, the crude incident TTS event rate in the AZD1222-vaccinated cohort must be interpreted in the context of the demographic and clinical characteristics and underlying comorbid conditions of the individuals in this cohort. Further, it should again be noted that, although <1% of the individuals in the AZD1222-vaccinated cohort had a confirmed prior COVID-19 diagnosis, this was based on the strict requirement for a positive PCR test and clinical diagnosis. It is possible that individuals may have been missed, such as those hospitalized without a test (during the early stages of the pandemic), those who tested positive but were not hospitalized and did not receive a clinical diagnosis, those who had previous asymptomatic SARS-CoV-2 infections, and those who were symptomatic but not tested or treated. Such groups may have an increased risk of subsequent TE/TCP events.

Potentially associated with these factors, our crude incident TTS rate of 2.41 per 100,000 person-years following AZD1222 was lower than some previous estimates. In a cohort analysis based on secondary care data for the whole population of England, a total of 116 cases of TTS were reported with a total at-risk time of 2,324,413 person-years, giving a TTS rate of 4.99 per 100,000 person-years [32]. An analysis of primary care data in the UK using the Clinical Practice Research Datalink AURUM database identified 16 cases in an at-risk time of 277,864 person-years, giving a TTS rate of 5.8 per 100,000 person-years [16]. The differences in rates between studies may be associated with several factors, including the criteria used for defining a TTS event. For example, an analysis where the TTS event definition used a platelet count of < 150 × 109/L for defining TCP, and included MI and stroke in the TE events, determined an even higher rate of TTS within 28 days following a first or second dose of AZD1222, of 85.9 per 100,000 person-years, based on 159 cases in 185,030 person-years [33].

Different definitions for TTS have been developed [3, 12, 17, 18, 34,35,36,37,38,39], and these have evolved over time [4]. TTS events occurring post-vaccination may have been incorrectly referred to as vaccine-induced immune thrombotic TCP, and it has been suggested that these events may differ from non-vaccination-related TTS [37]. Recent case definition criteria incorporate the following: thrombocytopenia, elevated D-dimer, confirmed thrombosis and/or severe persistent headache (5–30 days post-vaccination), onset of illness 4–30 days post-vaccination (or up to 42 days in isolated DVT/PE), and a positive functional anti-platelet factor 4-dependent antibody assay (with/without a positive enzyme-linked immunosorbent assay) [39]. However, some of these clinical and laboratory data are not available in large datasets such as the one used in the present analysis or in previous studies of the epidemiology of TTS. Other case definitions include varying criteria to define coincident TE and TCP events and it is therefore important to consider differences in TTS criteria when considering rates across analyses.

The sensitivity analyses can be examined in relation to other epidemiological analyses. Previous analyses using a TCP definition of a platelet count < 150 × 109/L reported crude incident TTS rates following AZD1222 vaccination up to 85.9 per 100,000 person-years [33], considerably higher than rates observed with the primary definition in our analyses (TCP definition of platelet count < 100 × 109/L). We showed similarly increased crude incident TTS rates across cohorts using a TCP definition comprising a clinical diagnosis or platelet count of 10–150 × 109/L. Furthermore, our primary analysis excluded events of MI and stroke from the TE event definition, whereas these are included in the current case definitions [39, 40]. We performed a number of sensitivity analyses to further elucidate the nature of events. Incorporating MI and stroke as TE events increased the TTS rate compared with the primary analysis in all evaluable cohorts, including a rate of 3.87 per 100,000 person-years in the AZD1222-vaccinated cohort. Other sensitivity analyses also showed expected changes in the TTS rate compared with the primary analysis, highlighting the impact of specific parameters on rates. Extending the coincident event window for TCP to + 42 days post-TE increased the TTS rate, whereas limiting the TCP definition to exclude those with TCP diagnostic codes without a platelet count of 10–100 × 109/L resulted in a lower TTS rate. Shortening the washout period for prior TCP/TE from 365 to 90 days had only limited effects on the incident TTS event rates, while increasing the post-vaccination follow-up from 28 to 42 days lowered the rate, indicating that events are most frequent within the 4 weeks post-dose, per the current criteria for time to onset [17].

Analyses stratified by sex showed a numerically higher, crude incident TTS event rate in male individuals compared with female individuals in the AZD1222-vaccinated cohort, although there was a substantial overlap in 95% CIs, limiting the interpretation of findings. This is in contrast to earlier smaller series of case reports of TTS [1, 4, 8] as well as analysis of the AZ Global Safety Database [1, 4, 8] showing that more cases occurred in female individuals. The crude incident TE event rate was also higher in male versus female individuals in the confirmed COVID-19 cohort (again with overlapping 95% CIs), while the crude incident TCP event rate was higher in male versus female individuals across all cohorts. Incident event rates of TTS, TE, and TCP all increased with age across all cohorts, notable among individuals aged 50–64 or ≥ 65 years. These findings highlight how differences in populations between cohorts prevent direct comparisons of rates—the AZD1222-vaccinated population included a high proportion of individuals in these two oldest age groups (as older more vulnerable groups were given priority for vaccination with AZ1222 in early 2021), which potentially resulted in an increased relative risk of TTS events. Our data are in line with prior analyses that also demonstrated an increasing incidence of TTS with age [12, 16]. Indeed, our multivariable analysis of covariates associated with TTS in the pre-COVID cohort and using the sensitivity analysis definition of a coincident event window of − 7 to + 42 days also showed that age 40–64 years, as well as the presence of various comorbidities per JCVI risk group definitions, was associated with an increased risk of TTS. However, these findings relate to TTS events in a pre-pandemic population (unlikely to be vaccine-induced immune thrombotic TCP), and it is possible that TTS events associated with a COVID-19 diagnosis, SARS-CoV-2 infection, or AZD1222 vaccination may have different characteristics and may be associated with different risk factors.

Epidemiological studies of TTS are inherently associated with limitations. Currently, there is no diagnostic code for TTS as a single entity within computerized medical records. Additionally, guidelines to define TTS have evolved over time, and physicians may also use their clinical acumen to establish a diagnosis outside of guidelines [41]. This makes accurate reporting of TTS challenging and highlights the need for a specific diagnostic code that can be used for research and to ensure a more comprehensive understanding of the incidence of TTS events. In this context, the use of our own case definition represents a limitation — we identified individuals with a co-occurrence of TE and TCP, which may not be specific enough to identify TTS from medical records. Another limitation is that SARS-CoV-2 infection appears to increase the risk of TTS, TE, and TCP events, and so prior asymptomatic infections or unreported cases of COVID-19 that are not captured in the datasets could impact rates of events in this study, particularly in the more contemporaneous AZD1222-vaccinated cohort.

Another limitation of these findings is that direct comparisons between cohorts in our analysis cannot be made owing to the observed differences in demographics and clinical characteristics. We were also unable to determine risk factors for TTS during the pandemic; the planned multivariable analyses of covariates associated with TTS were not feasible because of the low numbers of incident TTS events in all cohorts using the primary analysis definition and the large number of covariates investigated. Only by using the sensitivity analysis definition allowing a coincident event window of − 7 to + 42 days were sufficient TTS events available and only for the pre-COVID-19 cohort. Furthermore, we found differences in the recording of key covariates such as smoking status, BMI, and ethnicity between overall cohorts and the respective subgroups with TTS events.