In this retrospective, observational study, 250,966 records of suspected SADRs following the administration of 733,837,251 vaccine doses against SARS-CoV-2 were explored. The main findings of our study are: (1) SADRs potentially associated with anti-SARS-CoV-2 vaccines seems to be rare and tend to manifest most commonly as neuropsychiatric, cardiovascular as well as musculoskeletal and connective tissue disorders; (2) regarding the administration of vaccine doses, the reporting of suspected SADRs following vaccination appears to be delayed and occurs over a longer time; (3) a monitoring platform based on a spontaneous ADR reporting system is an essential and effective data collection tool for long-term vaccine safety surveillance.

EMA recommends grouping the reported ADR frequency as follows: ≥ 1/10 administered doses as very common, ≥ 1/100 to < 1/10 as common, ≥ 1/1,000 to < 1/100 as uncommon, ≥ 1/10,000 to < 1/1,000 (≥ 100/1,000,000 to < 1000/1,000,000) as rare, and < 1/10,000 (< 100/1,000,000) as very rare. According to the results of our analysis, the summarized frequency of all SADRs potentially linked to anti-SARS-CoV-2 vaccination appears to be the lowest for Comirnaty (754 PMD) and Spikevax (785 PMD), followed by Jcovden (1,248 PMD) and Vaxzevria (2,301 PMD; p < 0.001; Fig. 2). Therefore, according to EMA’s classification, the frequency of suspected SADRs for Vaxzevria and Jcovden vaccines tends to be uncommon as well as rare for Comirnaty and Spievax. This data suggests a potentially superior safety profile of modern mRNA vaccines against SARS-CoV-2 (Comirnaty and Spikevax) in comparison to the vector vaccines (Jcovden and Vaxzevria). Moreover, the efficacy reported for adenovirus-vectored vaccines is about 65%^22,23. For mRNA vaccines, the claimed efficiency is over 90%^2,3,22,24. Thus, in addition to indicating a good safety profile that was also documented in other studies, mRNA vaccines appear to offer better protection against SARS-CoV-2 infections^2,3,22,24. Both Comirnaty and Spikevax seem to have very similar efficacy and safety profiles and there is no definitive “number one” vaccine^2,3,24.

Further on, based on our calculations, the most observed clinical manifestations of suspected SADRs linked to anti-SARS-CoV-2 vaccines encompassed neuropsychiatric, cardiovascular as well as musculoskeletal and connective tissue disorders (Tables 1, 2, Supplementary Materials 2 and 3). The 15 most frequently documented potential SADRs included headache, myalgia, nausea, dyspnoea, dizziness, arthralgia, pain in extremities, paraesthesia, arrhythmia, pulmonary embolism, vomiting, tachycardia, syncope, palpitations and hypoaesthesia. The majority of these SADRs are unspecific symptoms that can be reported by the patients themselves or can be diagnosed during a routine clinical examination. An exception to this are cases of arrhythmia and pulmonary embolism, which are more specific and usually require additional diagnostic modalities as the gold standard (electrocardiogram and computed tomography pulmonary angiogram, respectively)^25,26. Moreover, the occurrence of all described potential SADRs may be a pathophysiological molecular reaction to a postvaccination stimulation of the immune system²⁷. Following our data, except for rare headaches after Jcovden administration as well as headaches, myalgia and nausea after vaccination with Vaxzevria, all other associated SADRs can likely be classified with the use of EMA guidelines as “very rare” (< 100/1,000,000 doses), which can be considered a promising result.

Moreover, the EMA anti-SARS-CoV-2 vaccine product characteristics are documents that describe the use of medicine and provide the estimated frequency of the ADRs^{28,29,30,31,32}. They summarize all available data, incorporating the results of randomized controlled trials and phase III registration trials, considered a gold standard in evidence-based medicine and are the basis of information for healthcare professionals^{28,29,30,31,32,33}. According to these sources, the most common suspected SADRs associated with SARS-CoV-2 vaccination in our analysis are in-line with the EMA^{28,29,30,31,32}. Thus, our data align with the official product characteristics of the SARS-CoV-2 vaccines, suggesting the reliability and accuracy of our study.

However, EMA reports just the expected frequency of the ADRs, whereas the seriousness of the ADRs is not reported. Therefore, our study provides a more precise insight into the frequency of suspected SADRs linked to anti-SARS-CoV-2 vaccines and cannot be directly compared to the EMA’s product characteristics. In the future, the reporting of both the frequency and severity of the potential ADRs by EMA could provide a more detailed summary to clinicians as well as contain fine-grained research data on the expected complications of vaccinations.

Patients recruited to randomised controlled trials on anti-SARS-CoV-2 vaccine safety reported most commonly general pain, headaches, fatigue, injection-site pain, injection-site erythema, myalgia, arthralgia, nausea, vomiting and chills^2,3,23,24. Further on, the frequency of serious postvaccination ADRs was rare^2,3,23,24. It occurred in less than 1% of the vaccinated cohort^2,3,23,24. These prospective data are comparable to the insights from our retrospective study. Thus, the risk of SADR to the studied vaccine appears to be outweighed by the benefit of active immunisation against the SARS-CoV-2 infection and its complications like hospitalization, intensive care unit admission, severe long-term disability or death^2,3,23,24.

Myocarditis, pericarditis, anaphylaxis, Guillain-Barré syndrome and thrombosis with thrombocytopenia syndrome are reported by EMA to be the SADRs of “special interest”^5,8,20,21,34. According to the data collected in our study, myocarditis (n = 6,380), pericarditis (n = 4,804) and myopericarditis (n = 1,702) were ranked as the 19^th, 24^th and 76^th (Table 3). Thus, they were relatively frequently reported when compared with all studied cases of SADRs. Moreover, the total frequency of myo/pericarditis seems to be higher for mRNA vaccines (Comirnaty and Spikevax) when compared with adenovirus-based vaccines (Jcovden and Vaxzevria; p < 0.001).

Following the global study based on the World Health Organisation (WHO) pharmacovigilance database, the reported odds ratio (ROR, interpreted as “the higher, the stronger the association”) were also significantly higher for mRNA vaccines (ROR 37.77; 95% CI 37.00–38.56) compared to Ad5‐vectored vaccines (ROR 1.40; 95% CI 1.34–1.46) and inactivated whole‐virus anti-SARS-CoV-2 vaccines (ROR 0.22; 95% CI 0.17–0.29)³⁵. Gao et al. noted that the associated relative risk (RR) between anti-SARS-CoV-2 vaccination and the risk of myo/pericarditis was larger for mRNA vaccines: RR 4.15 (95% CI 1.87–9.22) for Spikevax and RR 2.19 (95% CI 1.46–3.29) for Comirnaty when compared with viral vector-based products (RR 1.11; 95% CI 0.81–1.53)³⁶. This data is in line with other published studies^5,35,37. Thus, there might be an association pointing to an increased risk of postvaccination inflammation of the heart tissue.

Further on, the total number of anaphylactic reaction/anaphylactic shock cases linked to vaccination against SARS-CoV-2 was 2,923 and was most often noted for Jcovden (total frequency of 7.0 PMD), followed by Vaxzevria (6.5 PMD), Comirnaty (3.7 PMD) and Spikevax (2.5 PMD; p < 0.001). According to the study by Boufidou et al., extremely rare fatalities—occurring at rates of 0.04 (95% CI: 0.03–0.06) per million doses for anaphylactic reactions and 0.02 (95% CI: 0.01–0.03) per million doses for anaphylactic shock—were more commonly linked to vector-based vaccines than to mRNA-based ones, both in the US and Europe³⁸. Conversely, other studies suggest that this association remains inconclusive^39,40,41. Nonetheless, healthcare professionals should remain vigilant, as anaphylaxis may occur following any vaccination, including those against SARS-CoV-2³⁹.

The Guillain-Barré syndrome was noted for 1,727 patients, and was most commonly linked to vaccination with Jcovden (10.8 PMD), next Vaxzevria (9.1 PMD), Comirnaty (1.5 PMD) and Spikevax (1.4 PMD; p < 0.001). This is consistent with other reports indicating a stronger association between post-vaccination Guillain-Barré syndrome and vector-based vaccines compared to mRNA-based vaccines^20,42,43.

In our analysis, TTS was identified in 198 cases and was most frequently associated with immunisation using Vaxzevria (2.8 PMD) and Jcovden (1.5 PMD), while rates for Comirnaty and Spikevax were markedly lower (< 0.1 PMD each; p < 0.001). However, a rare condition may sometimes go undiagnosed due to its low prevalence, making it essential to consider other related diseases within the same clinical group to ensure a more accurate and comprehensive diagnostic approach. Therefore, when considering all thrombosis-related events collectively (n = 22,507), the overall reporting rates were highest for Vaxzevria (128.6 PMD), followed by Jcovden (49.3 PMD), Comirnaty (21.1 PMD), and Spikevax (19.1 PMD; p < 0.001). These findings, indicating a higher likelihood of very rare thrombosis with thrombocytopenia syndrome following adenoviral vector-based vaccine administration compared to mRNA-based vaccines, are consistent with data reported by the EMA as well as with results from other large-scale studies conducted across diverse populations and geographic regions^44,45,46,47.

The timelines representing the weekly relative intensity of suspected SADRs reporting tends to be delayed compared to the curve representing the administration of vaccine doses. Moreover, the intensity of potential SADR reporting was more distributed in time and did not overlap with the administration of vaccine doses (p < 0.001; Fig. 3a–d and Supplementary Material 5). Nevertheless, the SADR reporting curves were visually similar to the vaccine administration timelines. Further on, the lag in SADR reporting and the smoother distribution in time were also observed for most clinical categories of the suspected SADRs. These trends were noted across all investigated vaccines, though further confirmation of these findings is required in prospective studies.

Based on the prospective studies, the greatest part of the SADRs tends to occur within a week after the vaccine injection^2,3,23,24. According to the EMA’s “Clinical Safety Data Management: Definitions and Standards for Expedited Reporting”, all serious and unexpected events should be reported no later than 15 calendar days after first knowledge, whereas a life-threatening or fatal ADR must be reported promptly and within 7 calendar days of initial awareness⁴⁸. Thus, the maximal delay from the SADR onset to its reporting should be maximally equal to three or four weeks. When the peak intensities of vaccine administration and SADR reporting are considered, our data show that this requirement was met for Vaxzevria and Spikevax. The most notable delays appeared to be associated with Comirnaty, suggesting a possible trend in this regard. However, for all vaccines, the SADR reporting tends to be more distributed in time than vaccine administration (p < 0.001). Therefore, the time from the event onset to spontaneous reporting should be shorter to meet the EMA’s guidelines⁴⁸. Moreover, this trend also underscores the importance of free-of-charge long-term spontaneous reporting systems used for global post-approval vaccine safety monitoring. Without the introduction of the EudraVigilance database, it would not be possible to investigate numerous potential postvaccination SADRs that were reported over an extended period. We endorse the maintenance of an open-source European pharmacovigilance database for future voluntary and spontaneous reporting of potential ADRs to authorized medicines. In addition to that, to improve adherence to the EMA’s ADR reporting guidelines, an educational intervention could be introduced in healthcare facilities that are responsible for vaccinations. Studies have proven that it can increase the reporting rate of the ADRs and pharmacovigilance knowledge. Workshops rather than telephone-based interventions are reported to be the most effective in improving spontaneous ADR reporting^49,50.

Hierarchical clustering is an unsupervised machine learning algorithm that groups data based on their relative similarity. The clustering was based on the timelines showing the weekly fluctuations in the reporting of the SADR clinical categories. We found four possible large clusters, each one potentially corresponding to one analysed vaccine (Fig. 4). Moreover, also for the smallest clusters the type of vaccine administered, not the clinical category of the reported SADRs, is likely to have been the primary clustering factor. The shapes of timelines depicting the changes in the intensity of clinical categories reporting were similar to each other within one administered vaccine type, which is also noticeable in the visual comparison of the Figures.

In June 2021, EMA officially raised awareness of “clinical care recommendations to manage TTS that might be potentially associated with vaccination with Vaxzevria or Jcovden”²¹. This warning rapidly spread out on social media platforms and in the news^51,52. Such events might have potentially shifted the focus of the medical professionals to one clinical category of suspected SADRs, in this case, cardiovascular disorders.

However, our hierarchical clustering suggests that across both large and small clusters the administered vaccine was the primary grouping variable. Therefore, the spontaneous reporting of the SADRs to anti-SARS-CoV-2 vaccines seemed to be independent of other possible shaping factors like ongoing discussions in the media. Based on the example of the EudraVigilance database, it also appears to provide confirmation of the robustness of a voluntary and spontaneous reporting system of potential ADRs for vaccine safety monitoring.

One of the limitations of this study is the difficulty in establishing a direct cause-effect relationship between the vaccination event and a suspected adverse drug reaction⁵³. Some of the most commonly reported SADRs, for example, “pain in extremities”, “arrhythmia” or “pulmonary embolism” might be ageing-related conditions and might have been reported due to the high populational risk of such diseases²⁵. Moreover, our study is a retrospective, observational study, which due to its study design cannot be used to confirm causal inference⁵⁴. To address these limitations, the causative relationship should be investigated in prospective studies⁵⁴. However, such studies are relatively rare due to their expensiveness and complexity. Moreover, the retrospective design of our study allowed us to comprehensively examine a very large sample of over 250,000 records of 897 types of SADRs, which is highly unlikely to be collected in any prospective study. Further on, the results of our study are consistent with the data gathered by the EMA. Thus, our research can be considered a cornerstone for future prospective research on SADRs to anti-SARS-CoV-2 vaccines.

The study is based on data from European databases, which may not fully capture the ethnic diversity present in global populations. As a result, the findings may not be generalizable to non-European or more heterogeneous populations. This limitation highlights the need for further research in more diverse cohorts to validate the study’s conclusions.

Due to the absence of randomization, unmeasured variables could serve as confounding factors, presenting a methodological limitation in our retrospective study. Additionally, the analyses are vulnerable to recall bias, as participants may have had to recall the potential link between the vaccination and the SADRs. Moreover, with limited clinical characteristics recorded for each entry in the EV database, it was not possible to provide a detailed clinical phenotype of patients who might be at the highest risk of experiencing SADRs.

The EudraVigilance database is based on a spontaneous reporting system. Such monitoring systems might be subject to under-reporting of the ADRs^55,56. It is estimated that on average, about 95% of all ADRs are under-reported. To reduce this bias, we decided to focus on SADRs only, which have a lower under-reporting rate of approximately 80%^55,56. Moreover, serious events tend to be more consistently reported because of their higher clinical significance⁵⁵.

In our study, we used the EudraVigilance-based system organ class (clinical category). However, some symptoms and diseases could be grouped into more than one clinical category. For example, “syncope” can be a manifestation of severe dehydration, a vasovagal reflex or arrhythmia⁵⁷. “Pulmonary embolism” can be considered a pulmonary or cardiovascular complication. Therefore, for future assessment of suspected ADRs, it would be beneficial to group symptoms and diseases into more than one organ class.