South Africa’s first case of severe acute respiratory syndrome-coronavirus-2 (SARS-CoV-2) was reported on 5 March 2020, and by 12 October 2023, there were 4,072,522 laboratory-confirmed cases and 102,595 deaths reported [15]. In this retrospective pharmacovigilance study of AEFI and AESI reported to NDoH during the COVID-19 vaccine rollout, we observed several signals of disproportionate reporting for both Ad26.COV2.S and BNT162b2 vaccines. These signals largely aligned with AEFI observed globally. AEFI reporting rates varied across demographic features. A comparison of reporting rates with those of other countries revealed underreporting. Augmented database analyses revealed potential masking, highlighting data limitations.

Overall, 3846 AEFI were reported from 17 May 2021 to 31 December 2022, during which 37,537,009 doses of BNT162b2 and Ad26.COV2.S were administered. The overall AEFI reporting rate was 10.2 per 100,000 doses administered. A consolidated global report compiled by the Pan American Health Organization, reported AEFI rates ranging from 10.99/100,000 to 67.8/100,000 doses [16]. AEFI reporting rates for Argentina (67.8/100,000), and Colombia (58/100,000) [16], which have similar economic profiles and population sizes to South Africa, were 5–6 times higher than South Africa’s. AEFI surveillance relies on spontaneous reporting, and as such, underreporting is inevitable. Our low reporting rate may be indicative of underreporting in our setting.

Reporting rates by vaccine type differed substantially, with reporting rates of 18.1/100,000 and 7.9/100,000 for Ad26.COV2.S and BNT162b2, respectively. This observation aligned with findings in other countries. Canada reported rates of 243.24/100,000 and 47.32/100,000 for Ad26.COV2.S and BNT162b2, respectively [16]. A descriptive study on the frequency of AEFI with different SARS-CoV-2 vaccines in the Netherlands found that 82% of Ad26.COV2.S recipients reported at least one adverse event, compared with 45.2% of BNT162b2 recipients [17]. In addition, a cross-sectional study in Italy reported an adjusted odds ratio of 2.06 for AEFI reported in relation to the Ad26.COV2.S vaccine compared with that of the BNT162b2 vaccine [18]. Our reporting rates for serious/severe AEFI for each of the vaccine types were quite similar, with a slightly higher reporting rate for Ad26.COV2.S (4.7/100,000) compared with BNT162b2 (4.4/100,000). Overall, lower levels of reactogenicity have been reported for the BNT162b2 vaccine, compared with reactogenicity to other mRNA-based and adenovirus-vector vaccines [19, 20].

We observed differential administration rates of both vaccines across provinces in South Africa. However, AEFI reporting rates were not distributed in similar ratios to vaccine administration rates across provinces. The rural North West Province had the lowest AEFI reporting rate (1.6/100,000), despite having the third highest rate of Ad26.COV2.S doses administered per 100,000 persons (19,529/100,000). In contrast, the urban Western Cape Province had one of the lowest rates of Ad26.COV2.S doses administered per 100,000 persons (10,831/100,000), but the highest AEFI reporting rate (59.5/100,000). This pattern, however, is not consistent across other rural and urban provinces. For example, Gauteng Province, South Africa’s economic hub, had a relatively small Ad26.COV2.S AEFI reporting rate of 8.1/100,000. Variation of AEFI reporting rates across provinces can likely be attributed to differences in system functionality, provincial capacity and resources.

In dataset A, seven AEFI presented as signals of disproportionate reporting in relation to the Ad26.COV2.S vaccine (Table 9). Headache, nausea, myalgia, and injection site pain all appear on the vaccine PI, listed as ‘very common’ side effects, occurring at a frequency of ≥ 1/10 recipients [21]. Therefore, despite underreporting, our AEFI surveillance system detected signals that concur with AEFI frequencies from clinical trials and post-marketing surveillance. Pyrexia and chills are also listed in the Ad26.COV2.S PI as common side effects (≥ 1/100 to < 1/10), and dizziness as uncommon (≥ 1/1000 to < 1/100) [20]. Among the very common side effects listed on the PI, fatigue was not identified as a signal by the IC but was identified by the ROR in dataset A.

Regarding the BNT162b2 vaccine, we identified signals of disproportionate reporting for dyspnoea, chest pain, and rash (Table 10). Rash is listed in the PI as uncommon (≥ 1/1000 to < 1/100) whilst dyspnoea and chest pain are absent [22]. Myocarditis/pericarditis has been shown to be associated with BNT162b2. Chest pain is a commonly reported symptom of myocarditis/pericarditis [23] and presented as a signal in relation to the BNT162b2 vaccine. It is possible that some cases of chest pain could be undiagnosed mildly symptomatic cases of myocarditis/pericarditis; however, this cannot be assumed. Dataset A analyses did not detect signals for side effects that were listed as very common or common in the BNT162b2 PI. Disproportionality analysis is highly dependent on the data used as the background rate. Given that background rates were calculated from a vaccines database, AEs that are common for several vaccines will not be detected. This is a likely explanation as to why side effects common to both BNT162b2 and Ad26.COV2.S vaccines were not detected for both vaccines.

Dataset C was used to ascertain whether there was masking. In total, 7416 additional non-COVID AEFI were added which created a more stable background rate. If masking is present, we would expect signals to become stronger or new signals to emerge. For Ad26.COV2.S, three new signals emerged for fatigue, pain and injection site swelling. All side effects categorized as ‘very common’ on the PI were identified by dataset C, and all but one classified as ‘common’—injection site erythema—were identified. For BNT162b2, two signals emerged for fatigue and dizziness; both of which are listed in the vaccine PI as very common (≥ 1/10) and uncommon (≥ 1/1000 to < 1/100), respectively. For BNT162b2, rash appeared as a signal in dataset A and no longer presented as a signal in dataset C. The disproportionately large number of reports related to COVID-19 vaccines compared with other vaccines in dataset A has the potential to either mask existing signals or result in the generation of false associations. Stabilisation of the background rate is primarily driven by our need to elicit masking and to ensure that all safety signals are identified. However, this stabilisation of the background rate has the potential to identify false associations, which this signal of rash may be.

Of the nine AESI evaluated in this study, Guillain–Barré syndrome, thrombosis with thrombocytopenia, venous thromboembolism, and myocarditis/pericarditis are all included in the PI. Myocarditis/pericarditis presented as a signal of disproportionate reporting in relation to BNT162b2. In dataset C, signals were elicited for DVT and DIC. The significance of this in relation to the BNT162b2 vaccine is not well documented. We did not detect signals in our dataset for any coagulopathies in relation to Ad26.COV2.S, shown to be associated with adenovirus-vectored vaccines [24]. Given the rarity of such events, and limitations associated with a symptom-based database for AESI detection, not detecting signals for some AESI is not surprising.

In our study, we observed incomplete sharing of data between NDoH-EPI and SAHPRA databases (Fig. 1). Owing to incomplete data sharing and different data sources, not all data in the NDoH-EPI database contribute to VigiBase. Therefore, signals of disproportionate reporting regarding COVID-19 vaccines in the NDoH-EPI database may differ from those calculated from the SAHPRA database.

Other limitations regarding the use of SRS databases include the distribution of types of AEFI/AESI reported among different age, sex and ethnic groups. Prior to the pandemic, AEFI reported to NDoH-EPI were predominantly those that occurred among EPI program recipients. The nature and range of AEFI/AESI reported amongst persons 12 years and under differs from those seen in adults. The addition of non-COVID adult vaccination (non-EPI) AEFI to the augmented database was an attempt to reduce the impact of this issue.

A comprehensive evaluation of the South African AEFI surveillance system, involving both qualitative and quantitative analyses of numerous system attributes, identified broader areas in need of strengthening. The study identified that the system was heavily affected by under-resourcing, especially in more rural provinces. The study revealed that insufficient resourcing directly impacted reporting rates, investigation of suspected cases and causality assessment [25].