FormalPara Key Points

We assessed acute appendicitis risk after COVID-19 vaccines in Italy.

We used routinely collected data from linked health care registries, which permitted us to save time and costs.

There was no evidence of association of COVID-19 vaccines with acute appendicitis in the general population.

1 Introduction

Appendicitis was reported as serious sdverse event in the pivotal randomized phase 3 clinical trial on BNT162b2 mRNA vaccine (n = 43,448 enrolled participants) where more cases of appendicitis were experienced by the vaccinated volunteers (eight cases) compared with those unvaccinated (four cases) [1]. There have also been passive pharmacovigilance reports of appendicitis following COVID-19 vaccination, with both mRNA and adenoviral based vaccines [2]. The US Food and Drug Administration (FDA) added appendicitis to the list of adverse events of special interest to be monitored for COVID-19 vaccines [3].

A sequence of three cohort studies was therefore performed to assess this potential signal in mRNA vaccines providing inconsistent results. Barda et al. explored the association between the first dose of BNT162b2 and appendicitis in the Israelian population of 16 years and older, estimating an association with an adjusted relative risk (RR) equal to 1.40 (95% confidence interval [CI] 1.02–2.01) [4]. No association emerged from a US study that analysed the risk of the combined mRNA vaccines (RR of 0.82; 95% CI 0.73–0.93) [5]. Similarly, a Danish study did not find an increased risk of appendicitis after both doses of BNT162b2 (first dose: RR of 0.94, 95% CI 0.79–1.12; second dose: RR of 0.98, 95% CI 0.82–1.17) and mRNA 1273 (first dose: RR of 0.87, 95% CI 0.61–1.25; second dose: RR of 1.08, 95% CI 0.78–1.51) [6].

In a recent commentary conducted by Jambon et al., data of three cohort studies were meta-analysed. Findings were in favour of the null effect of mRNA vaccines on appendicitis. However, the authors highlighted the presence of relevant heterogeneity across these studies, which was attributed to differences in the choice of the comparators and matching/weighting procedures [7]. They recommended the choice of the self-controlled case series (SCCS) method [8] as the proper methodological solution to deal with the comparability issue between vaccinated and unvaccinated subjects in the evaluation of the relationship between COVID-19 vaccine and appendicitis [7].

To our knowledge, only two studies using SCCS methodology were conducted to examine the risk of appendicitis after the administration of COVID-19 vaccines, one in Japan [9] and the other one in Singapore [10], showing conflicting results. The Japanese study revealed no increased risks of appendicitis (first dose of mRNA vaccines: incidence rate ratio (IRR) of 0.23, 95% CI 0.02–2.71; second dose of mRNA vaccines: IRR of 0.68, 95% CI 0.12–3.83). However, this SCCS study was conducted on few events of appendicitis (cases), both overall (n = 51) and in the 21-day risk interval after vaccination (n = 1 and n = 3 cases of appendicitis after first and second dose, respectively), coming from a population of 184,491 subjects [9]. The study conducted in Singapore, which analysed 3314 cases of appendicites in a population of 5.5 million aged ≥ 12 years, reported an increased risk for the first dose in the risk intervals 1–7 days (relative incidence [RI] of 1.31, 95% CI 1.03–1.67) and 15–21 days (RI of 1.39, 95% CI 1.10–1.77); this increased risk was not observed in the second dose [10].

Therefore, in the attempt to further address the uncertainty about the risk of appendicitis in the COVID-19 vaccinated persons, we conducted a larger-scale SCCS study. We specifically evaluated the association of acute appendicitis with the first and the second dose of BNT162b2 and mRNA 1273, separately, in a population of about 16 million vaccinated aged 12 years or older in Italy. This study also expands upon previous research on COVID-19 vaccines and appendicitis by including in the analysis adenoviral based vaccines, ChAdOx1-S and Ad26.COV2-S.

2 Methods

2.1 Study Design

We used the SCCS method which was developed specifically for the vaccine safety assessment [11,12,13].

The method requires the identification of the events of interest and, since the exposed and non-exposed person-time is defined within each study subjects, it allows the implicit control for unknown and known confounders that remain fixed over the observation periods. It also permits the adjustment for potential time-varying confounder such as seasonality.

The SCCS method in its standard form requires some key assumptions to be applicable; one is that the occurrence of the event of interest does not alter the probability of subsequent exposures, and the other is that the occurrence of the event does not censor or alter the observation period (for example, if the occurrence of an event may delay or preclude vaccination or if the event has an associated higher mortality). In situations where these assumptions are violated, standard SCCS models lead to biased estimates; therefore, modified SCCS models, which deal with censored, perturbed or curtailed post-event exposures, must be used instead. The modified SSCS models were extensively described and discussed by Farrington CP et al. [14] and Ghebremichael-Weldeselassie et al. [15].

We used the modified SCCS method adapted for event-dependent exposures to estimate the relative incidence (RI) of appendicitis comparing the exposed time period after vaccination (risk interval) with the unexposed period (reference period).

2.2 Study Period and Population

In Italy, the COVID-19 vaccination programme started by the end of December 2020 and reached a very high population coverage in a few months [16].

Our study population comprised persons aged ≥ 12 years, who had received at least one COVID-19 vaccine and were admitted to emergency care department or hospital with acute appendicitis in the period from 27 December 2020 to 30 September 2021 in five Italian regions (Lombardia, Veneto, Friuli Venezia Giulia and Emilia Romagna and Lazio).

2.3 Data Sources

We used routinely collected data from health care information systems of five Italian Regions (Lombardia, Veneto, Friuli Venezia Giulia and Emilia Romagna and Lazio) representing 44% of the population aged ≥ 12 years in Italy. Data were linkable using a unique identifier. To obtain exposures data, we used a COVID-19 vaccination registry which includes vaccine type, date of administration and doses for all vaccinated persons. Data on the outcome of interest were extracted from hospitalization discharge and emergency care visit databases. Data on comorbidities were retrieved from hospitalization discharges, pharmacy claims and copayment exemptions databases preceding the vaccination of the study subjects. Age, sex and vital status were obtained from population registry without specification of cause of death. Data on SARS-CoV-2 infection was obtained from the COVID-19 surveillance system [16].

2.4 Data Collection, Linkage Procedures and Construction of the Analytical Dataset

This study used a distributed analysis approach based on a study-specific common data model (CDM) that facilitates analytical dataset production by collecting and organizing multiple regional data sources into the same database design using a common format (Supplementary Fig. 1) [17, 18]. The regional analytical datasets were locally created using TheShinISS, an R-based open-source statistical tool which was customized for the purpose of this study. TheShinISS locally processes data from regional health care databases that are linked at individual level according to an ad hoc, study-tailored CDM. TheShinISS was developed by researchers of the Istituto Superiore di Sanità (Italian National Institute of Health [NIH]) [19], and it is maintained by TheShinISS Network, which is a national initiative bringing together a group of researchers from Italian public health research organisations and academia: University of Verona, Department of Epidemiology of Lazio Region and NIH. It has already been used in various drug safety studies [17, 18, 20]. Local analytical datasets were produced by each region and sent pseudonymized to the NIH to conduct centralized analysis in compliance with the EU General Data Protection Regulation.

2.5 Definition of Exposures

The exposure variables were the first or the second dose of BNT162b2, mRNA-1273, ChAdOx1-S and the single dose of Ad26.COV2-S vaccines. The pre-defined risk interval after the first and second dose of COVID-19 vaccines (vaccination date), including day 0 (day of vaccination), was of 42 days, informed by the FDA guidance document on COVID-19 vaccine surveillance [3]. The risk interval was further subdivided into 14-day time periods. The reference period was defined as all days of observation period out of the risk intervals, i.e. before, between or after the risk intervals. According to the vaccination schedules of BNT162b2 (21-day interval between the first and second dose) and mRNA-1273 vaccines (28-day interval between the first and second dose), the risk intervals overlap and consequently, the risk intervals after first dose may end after the second dose. In the SCCS methodology precedence is given to the most recent exposure period, and the parameterization of the SCCS model is accordingly adjusted.

2.6 Definition of Outcome

The study outcome, acute appendicitis, was identified from discharge diagnoses of hospital admissions or emergency departments visits using the International Classification of Disease, 9th Revision, Clinical Modification (ICD-9-CM 540*) since in Italy ICD-10-CM has not yet been adopted. Incident cases were defined as those who experienced a first event of acute appendicitis in the study period, excluding subjects with a diagnosis of appendicitis in a look-back period of 5 years.

The observation period for each case ranged from 27 December 2020 to the last region-specific date of data availability (Lombardia up to 30 September 2021, Veneto up to 20 June 2021, Friuli Venezia Giulia up to 31 August 2021, Emilia Romagna up to 30 June 2021 and Lazio up to 16 June 2021).

2.7 Statistical Analysis

Vaccinated persons were described in terms of age, sex, comorbidities, comedications and region of residence. The number of first and second doses of vaccines and the number of the events, by calendar time, were plotted in a histogram. The frequency of deaths by days from the event of appendicitis and the occurrence of the event by time before or since the vaccination were also plotted in a histogram.

The RIs and their 95% CI were estimated for the 42-day risk interval and for each subrisk interval using the modified SCCS method for event-dependent exposures [14, 15] by unbiased estimating equations. To allow for any temporal change of the background rate of appendicitis, the time-dependent seasonal effect (six calendar periods of 45 days) was included in the model.

To handle event-dependent exposures, the SCCS model was properly modified considering a counterfactual exposure history for any exposure arising after the occurrence of an event [14, 15].

To handle individuals who received heterologous vaccination, the analyses were performed using an ‘initial treatment design’, without differentiating between types of vaccine administered as second dose after a first dose.

To handle mortality, as causes of deaths were unknown, we carried out the analyses following the approach of Ghebremichael-Weldeselassie [15] by estimating the duration (D) of short-term excess of mortality using a standard SCCS model where the event was death, and the exposure was the occurrence of the event. Deaths occurring less than D days after the event were attributed to the event, and deaths occurring D or more days after were attributed to other causes. Thus, when deaths were not attributed to the event, the end of observation period was set to the date of death, while when deaths were attributed to the event, the observation period was set equal to the planned end of observation period.

We carried out subgroup analyses by age groups (12–39, 40–59 and ≥ 60 years) and sex. The model, both in the main and in the subgroup analyses, was fitted only when the number of cases was greater than 10.

To assess the robustness of the results, we carried out four sensitivity analyses: (1) we explored the seasonal effect by removing the calendar time factor; (2) we investigated the effect of the SARS-CoV-2 infection on appendicitis by restricting the analyses to subjects without a positive SARS-CoV-2 test during the study period; (3) we handled heterologous vaccination using ‘on treatment follow-up’ analysis, wherein the second dose was considered only when of the same vaccine of the first dose and the observation time was censored otherwise; (4) we explored two extreme scenarios where all deaths were assumed to be caused by the event and all deaths were considered to be not caused by the event. Additionally, to support the choice of the modified SCCS model we performed the following three sensitivities analyses with the standard SCCS method: (1) beginning observation at the time 0; (2) beginning observation at exposure (starting the observation time at the first and second dose); (3) including a 28-day pre-risk interval [−28, 0).

The analyses were performed using R version 4.3.1 (R Core Team 2021) with SCCS package [21, 22] and STATA version 16.1.

3 Results

Between 27 December 2020 and 30 September 2021, 15,986,009 persons received at least one dose of COVID-19 vaccines (BNT162b2 n = 10,833,284; mRNA-1273 n = 1,706,979; ChAdOx1-S n = 2,863,950; Ad26.COV2-S n = 581,796). The median age of the entire vaccinated population was 56 years; vaccinees with mRNA-1273 were younger than persons vaccinated with ChAdOx1-S. Among males, Ad26.COV2-S was more represented, while ChAdOx1-S showed the lowest male/female ratio (Table 1).

Table 1 Characteristics of 15,986,009 vaccinated individuals by COVID-19 vaccine type (from December 2020 to September 2021)
Full size table

During the study period, there were 5351 cases of acute appendicitis; of these, 3869 occurred in BNT162b2 vaccinees, 669 in mRNA-1273 vaccinees and 647 and 166 cases in ChAdOx1-S and Ad26.COV2-S vaccinees, respectively (Fig. 1; Table 2). Within 42-day risk intervals 1285 cases occurred; of these, 880 cases occurred in BNT162b2 vaccinees, 177 in mRNA-1273 vaccinees, 199 and 29 cases in ChAdOx1-S and Ad26.COV2-S vaccinees, respectively (Table 3).

Fig. 1
figure 1

Flow chart of selection of the study population (27 December 2020–30 September 2021). n number

Full size image
Table 2 Characteristics of cases of acute appendicitis occurred during the study period by COVID-19 vaccine type
Full size table
Table 3 Relative Incidences estimated by self-controlled case series model for acute appendicitis by COVID-19 vaccine type and dose
Full size table

A total of 18 deaths were observed during the observation period; 5 deaths were among acute appendicitis events occurring in the 42-day risk interval and 13 deaths among acute appendicitis events occurring in the reference period (Table 2). The distribution of deaths by time from diagnosis of acute appendicitis is described in Supplementary Fig. 2; the median time of death from diagnosis of the acute appendicitis was 255 days. Our data sources did not allow to ascertain the causes of deaths.

The distribution of doses and appendicitis events are described by calendar time from the beginning of the study in Supplementary Fig. 3. Histograms of the interval between vaccination and the event, for each dose and by vaccine type, are also shown in Supplementary Fig. 4.

Table 3 presents the number of cases occurred in the reference period, in the risk and subrisk intervals, and the estimates of RIs with 95% CIs for the relationship between each COVID-19 vaccine and appendicitis. During the 42-day risk interval there were a total of 1285 cases of appendicitis: 727 cases after the first dose and 558 cases after the second dose of COVID-19 vaccination.

There was no evidence of an association of BNT162b and ChAdOx1-S vaccines with appendicitis in the main and in the subgroup analyses by sex and age (Table 3). There was no association of mRNA-1273 and Ad26.COV2-S vaccines in the main analyses, while the analyses by sex showed an increased RI estimates in the 14–27 day risk interval, in males after the first dose of mRNA-1273t, (RI of 1.71; 95% CI 1.08–2.70; p = 0.02) and in females after the single dose of Ad26.COV2-S (RI of 4.40; 95% CI 1.29–15.01; p = 0.02).

3.1 Sensitivity Analyses

Results of the sensitivity analyses agreed with the main analyses (Table 4). Sensitivity analyses were also conducted for ascertaining the robustness of the statistically significant associations between mRNA-1273 and Ad26.COV2-S vaccines with appendicitis in males and females, respectively. The sensitivity analyses confirmed these results (Supplementary Table 1). Notably, the sensitivity analyses obtained with the standard SCCS model starting the study period at the first dose did not change the findings, which was particularly reassuring: in males after first dose of mRNA-1273 vaccine in the 14–27 day risk interval (RI of 2.05, 95% CI 1.24–3.38) and in females after single dose of Ad26.COV2-S vaccine in the 14–27 risk interval (RI of 3.27, 95% CI 1.14–9.40). To note, these results were not influenced by heterologous vaccination, since this was confined to 66 cases with a first dose of ChAdOx1-S and 1 case of mRNA-1273 included in the female subgroup.

Table 4 Sensitivity analyses: relative incidences estimated by self-controlled case series model for acute appendicitis by vaccine type and dose
Full size table

4 Discussion

Our SCCS study, covering a population of about 16 million vaccinated in Italy, investigated the relationship of COVID-19 vaccines with acute appendicitis.

Results revealed no evidence of an association of BNT162b and ChAdOx1-S vaccination with acute appendicitis both in the main and the subgroup analyses by age and sex. In addition, the main analyses did not reveal any association with mRNA-1273 and Ad26.COV2-S vaccination. Conversely, we found for both vaccines increased risks in subgroup analyses by sex: an increased RI estimates in males after the first dose of for mRNA-1273 and in females after the single dose of Ad26.COV2-S. Comparisons with other studies are limited to the evaluation of the relationship with mRNA vaccines since our study is, to date, the first exploring the association of both mRNA (BNT162b2 and mRNA-1273) and viral vector (ChAdOx1-S and Ad26.COV2-S) COVID-19 vaccines with acute appendicitis. Our results are not in line with the first study evaluating the safety of BNT162b2 vaccine, which was conducted in Israel by using a cohort study design [4]. The authors estimated an adjusted RR of appendicitis equal to 1.40 (95% CI 1.02–2.01). This finding may indicate the presence of residual confounding as adjustment was made for a set of potential confounders selected for a broad range of adverse events, which our SCCS study inherently consider.

Conversely, our results generally confirm findings of the study conducted in the USA and Denmark [5, 6]. In the USA, Klein et al. did not detect any association analysing the effect of the first and second dose of mRNA vaccine combined [5]. This cohort study monitored 23 outcomes and compared the rate in the vaccinees of 12 years of age and older in their 1–21 day risk interval after vaccination (exposed group) with rate of appendicitis in vaccinees in the 22–42 day reference period after vaccination (unexposed group). In Denmark, Kildegaard et al. [6] conducted a registry-based cohort study in a population of approximately four million of residents of 12 years of age and older examining the risk of appendicitis at 21 days after first and second doses of mRNA vaccines separately. In line with our findings, the Danish study did not find an increased risk of appendicitis after doses of BNT162b2 (first dose: RR of 0.94, 95% CI 0.79–1.12; second dose: RR of 0.98, 95% CI 0.82–1.17) and doses of mRNA 1273 (first dose: RR of 0.87, 95% CI 0.61–1.25; second dose: RR of 1.08, 95% CI 0.78–1.51). The Singapore SCCS study [10], however, found in those ≥ 12 years, a 30% increased risk after the first dose in the analyses of mRNA combined, suggesting that a positive relationship might exist between appendicitis and mRNA vaccines. Our findings suggest an association of mRNA-1273 and Ad26.COV2-S with appendicitis in males and females respectively. However, such increased risks are difficult to interpret since the lack of prior knowledge on this field. Appendicitis has been associated with infection agents, greater variation of bacterial phylae and, even if more rare, luminal obstruction determined by lymphoid hyperplasia and lymphadenopathy [23]. Consequently, an immune mediated response has been offered to explain the relationship between COVID-19 vaccines and appendicitis [24]. However, the underlying causes of appendicitis remain multifactorial and challenging to ascertain definitively.

To our knowledge, the current study is the first exploring the risk of appendicitis by sex and age for each COVID-19 vaccines separately. The subgroup analyses, although pre-specified, were two of several hypotheses tested, introducing a possible multiplicity concern, as there were four vaccines, two doses, six groups/subgroups and four risk/subrisk intervals, corresponding to a total of 192 multiple tests performed. Moreover, the p value of the two statically significant results were close to 5% (p = 0.02); therefore, we cannot exclude the possible role of chance. Therefore, these results should be interpreted with caution and evaluated in the light of further studies of similar design to establish causality. Our study has several strengths. First of all, in that it includes a large population and different types of vaccines. The choice of the SCCS study design is another strength of the study. In this study design risk comparisons are made entirely within subjects thereby implicitly controlling for non-time varying covariates. This permitted to address key issues around vaccine safety design such as difficulties in the identification of the suitable control (marked differences between vaccinated and unvaccinated) and missing information on possible confounding variables. In addition, the inclusion of the seasonality effect in the model permitted to consider an important time-varying confounding factor. We also used the modified form of the SCCS model to handle event-dependent exposure to address also the issue of the effect of a postponed or cancelled vaccination leading to overestimation of the risk estimates. Furthermore, to provide greater confidence in our results we conducted sensitivity analyses that corroborated the choice of the model and the robustness of the results.

We used routinely collected data from linked health care registries, this permitted us to save time and costs. Moreover, in Italy, reporting health data to registries is mandatory within the national health care service. This ensures comprehensive data, near complete follow-up over time and minimisation of bias in case and exposure ascertainment.

Among the limitations of the study, the outcomes were not validated through review of clinical records. However, if outcome misclassification occurred, it was probably non-differential, leading to unchanged point estimates. Another limitation is that this study did not adjust for multiple comparisons potentially resulting in a higher family-wide type I error. Additionally, there is the possibility that some COVID-19 infections might have been missed due to asymptomatic or mild cases and lacking documented evidence of COVID-19 infections, despite the Italian tracking system’s high efficiency at the time of data collection.

5 Conclusions

In the main analyses our study did not observe an association between BNT162b, mRNA-1273, ChAdOx1-S and Ad26.COV2-S vaccination and acute appendicitis. Increased risks were observed in males after the first dose of mRNA 1273 vaccine and in females after the single dose of Ad26.COV2-S vaccine. Although potentially informative, the results of the subgroup analyses need to be interpretated cautiously. The key to demonstrating causality will be consistency with further large and well-designed SCCS studies carrying out similar subgroups analyses.