The main objective of our study was to investigate the appropriate disproportionality analysis to detect hearing impairment after mRNA COVID-19 vaccines as soon as possible during the COVID-19 primary vaccination campaign. Using a standard method, the ROR values showed a signal in July 2021, about 4 months after the beginning of the vaccination campaign in France, remaining constant during the second half of 2021 and 2022. Taking into account demographic factors such as age and sex, as well as the co-reporting of known ODs, the signal emerged 2 months earlier (May 2021). Finally, signal detection was brought forward 2 months (March 2021) when we added the exclusion of RRPT5 in the analysis. We assume that the exclusion of frequent and then well-known reported AEFIs enabled us to target rare AEFIs in the disproportionality assessments, thereby reducing event-competition bias. Salvo et al. suggested that event competition bias may be observed when other ADRs are reported with a much higher rate than the ADRs of interest: the removal of known and frequent ADRs for some drugs, such as statins and rhabdomyolysis/myopathy or oral anticoagulants and haemorrhages, revealed new signals of disproportionate reporting of which 12% appeared related to potential safety signals and were later confirmed [23, 24]. In fact, in 1 year, the RPVC Network collected, analysed and registered more than 120,000 AEFI reports associated with tozinameran and elasomeran vaccines alone. In contrast, over a 10-year period, from 2011 to 2022, the FPVDB recorded around 620,000 ADR and AEFI reports. Non-serious adverse reactions to COVID-19 vaccines were very frequently reported in 2021. This is most likely due the extensive media coverage of the vaccination campaign in France [27,28,29,30,31,32]. In particular, influenza-like illness emerged as the most frequently reported adverse event following COVID-19 mRNA vaccination, reported in 11% of cases.

According to our data, the choice of the threshold for excluding frequently and well-known reported AEFIs could have an impact on signal detection. Setting the threshold at 10% leads the exclusion of PT “influenza-like illness” from the reference reports, and a signal emerging in May, i.e. 1 month later than those detected with 5% threshold. Conversely, lowering the threshold to 2.5% leads to a signal emerging one month earlier, in March and excluded 15 PTs, corresponding to terms not known as AEFIs, such as “menstrual disorder”, “Herpes zoster”, and “arthralgia”. A 5% threshold appears consistent for mRNA COVID vaccines, considering terms like “influenza-like illness”, “asthenia”, “headache”, “vaccination site pain” and “lymphadenopathy”. The appropriate threshold depends on the specific drug studied and the database used and should be determined with careful consideration.

Finally, the most effective approach for early signal detection by ROR value estimation was to exclude RRPT5s combined to adjust for gender, age and OD. This approach led to an estimate ROR with a large confidence interval due to a reduction in the statistical power of the analysis. However, the high number of reports registered in the first few months of the vaccination campaign allows to have enough statistical power. The media and social attention given to the COVID-19 pandemic and the vaccination campaign has underlined the importance of pharmacovigilance. It can be assumed that this heightened awareness has encouraged the reporting of serious AEFIs that might not have been reported in another context.

This method of signal detection could be applied during public health crises similar to COVID-19. However, its application in routine pharmacovigilance practice requires further investigation. As suggested by Fusaroli et al., due to the lack of exposure data and the unquantified under and selective reporting, signals of disproportionate reporting cannot on their own be interpreted as conclusive scientific evidence of a causal relationship between a drug and an ADR [33]. We conducted this disproportionate reporting analysis following our previous study [15], which included an exhaustive, case-by-case assessment of all cases and contextualization within the knowledge already accrued from other sources, such as fundamental data, case series, literature and observational studies.

Concerning the previous and current cases review, despite oral or intra-tympanic corticosteroid therapy, deafness persisted for more than 3 months. The irreversible nature of the deafness reported is a cause for concern, since there is no treatment that can restore hearing apart from hearing aids. It is possible that the most severe SSNHL are over-represented in our case review. This may be due to reporting bias, with practitioners and patients tending to report the most severe cases more often. Also, the lack of medical confirmation of deafness led to the exclusion of the largest number of cases in our study. It is possible that patients with mild hearing loss did not consult an ear, nose and throat (ENT) specialist, explaining the higher rate of severe cases. It also implies that the reporting rate is underestimated for less severe cases. As expected, we gathered mainly cases occurring after booster doses, and most importantly, there was one positive rechallenge case. A patient developed SSNHL a few days after the first and second booster doses of tozinameran, suggesting the strong likelihood of a causal relationship [34]. Patients who had tolerated their primary COVID-19 vaccination well may still suffer from SSNHL following a booster dose. It seems important to take this risk into account during booster campaigns, particularly for patients who have already suffered from this AEFI.

Hearing disorders have also been observed in patients following COVID-19 infection. The effects of SARS-CoV-2 on neuronal tissue could be due to direct infection of the central nervous system, or linked to vascular damage caused by vasculitis or vasculopathy, in the same manner as the mechanism described for the varicella-zoster virus and human immunodeficiency virus [35]. The SARS-CoV-2 spike glycoprotein is known to exhibit tropism for human angiotensin-converting enzyme 2 [36], which is expressed not only in pulmonary respiratory epithelium and vascular endothelium, but also in olfactory neuroepithelium [37]. Indeed, the COVID-19 virus exhibits tropism for the central nervous system. Several studies have demonstrated the infection of neurons by COVID-19, causing characteristic metabolic changes, both in the animal model and in human neurons in vitro [38, 39], which explains the neurological symptoms observed in some patients, but does not explain the AEFIs related to hearing. The pathophysiological mechanisms that might explain post-vaccination SSNHL are those already suspected of causing idiopathic SSNHL. Vaccination could induce an immunological response leading to the release of antibodies and cytokines. The formation of immune complexes could trigger an autoimmune response directing antibodies to the cochlea. As a result, immunological and inflammatory responses lead to vasculitis and vascular ischaemia of the cochlea [40].

Recently, a study compared the SSNHL incidence rate in a university hospital in southern Israel over the COVID-19 outbreak and the COVID-19 vaccination campaign periods with pre-COVID-19 periods [41]. The authors found a low but significant increase of SSNHL incidence during the COVID-19 vaccination campaign, but the adjusted annual incidence rate of SSNHL for tozinameran-vaccinated patients was not significantly different compared with unvaccinated patients. This study is limited by the small sample size (single hospital, small number of patients), given the rarity of the event investigated. However, another study, through a very large retrospective cohort study of 2,602,557 patients in Israel, suggested an increase in the incidence of SSNHL within 21 days after the first or second dose of tozinameran compared with the incidence observed previously in 2018 and 2019 [42].

Damkier et al. carried out a study on the Danish national health register which failed to find an increased risk of SSNHL or vestibular neuritis following mRNA COVID-19 vaccination [43]. However, it did show that mRNA COVID-19 vaccination may be associated with a slight excess risk of a visit to an ENT specialist followed by a prescription for moderate to high doses of prednisolone, used as a proxy. This highlights the difficulty of identification of patients suffering from SSNHL in medico-administrative data. In practice, this event could be diagnosed and treated without requiring hospitalisation. Thus, SSNHL diagnostic codes are not sufficiently reliable to detect the event. Here, the authors attempted to measure the occurrence of the event indirectly by hypothesising that the initiation of high-dose corticosteroid therapy following an auditory procedure was most likely related to SSNHL. Recently, Shetty et al. adopted a multi-source data approach with, [44] on the one hand, a retrospective observational analysis of spontaneous reports of audio-vestibular events and, on the other, a self-controlled case series analysis using general practice data. Certain aetiologies such as intracranial lesions, neck pathologies, Meniere’s disease, hydrops or ischemic stroke were excluded, but case reports were not evaluated on the basis of a full medical analysis. The authors found a significant increase in vertigo and tinnitus following Vaxzevria® and tozinameran, but not in hearing loss. The difference in results could be due to the small number of cases of hearing loss selected (n = 76) compared with vertigo (n = 415) or tinnitus (n = 226), making it difficult to reach a threshold of significance, and to the subjectivity of the diagnosis of vertigo and tinnitus, whereas hearing loss is objectively diagnosable.

Several strengths could be highlighted in our study. First, the original approach to find a sensitive and reproducible method for early signal detection. Our data suggested the necessity of applying different methods of estimation in disproportionality analysis to identify those allowing an early detection of signals related to rare ADRs during pandemic or public health crisis. Secondly, to reduce the influence of notoriety bias, we included in the disproportionality analysis the cases of hearing impairment occurring before 1 February 2022, the date of WHO signals [5]. Consequently, all cases occurring prior to this date but reported between February 2022 and March 2023 were excluded. In fact, the impact of media is known on quantitative reporting of ADR [45,46,47]. Furthermore, the French pharmacovigilance system relies on a network of pharmacovigilance centres and ADR reports from healthcare professionals and patients are carefully reviewed by pharmacovigilance experts as well as clinicians, with this medical data making it possible to select a high-quality dataset. In contrast to epidemiological studies based on administrative databases, our case review is supported by application of strict diagnostic criteria, which reduces the risk of including false positives. In our first study, the reporting rate was similar for both mRNA vaccines, estimated at 1.5 cases per million injections [15]. The difference of the reporting rate in this study (0.83/1,000,000 doses for tozinameran and 4.3/1,000,000 for elasomeran) could be explained by the few cases selected (18 cases for tozinameran and 10 cases for elasomeran).

This study has some limitations, as disproportionality analysis can be affected by a variety of biases, since they are based on declarative data and under-reporting. However, regarding the context of intensive pharmacovigilance for COVID-19 vaccination campaign in France, the under-reporting rate could be considered as low. Moreover, few patient characteristics were taken into account by adjusting the analysis (age, gender and OD drugs). Other relevant criteria such as cardiovascular, autoimmune and otoneurological history were not considered.

An important step in the evaluation of vaccine-related ADRs is to compare their rates in vaccinated and unvaccinated populations, as was done for myocarditis [48]. As the adverse event studied is very rare, a case-control study appropriate to rare events is needed to better explore the association of mRNA COVID-19 vaccination with hearing impairment, as has been done for menstrual cycle [49]. We compared the rate of reporting of hearing impairment between mRNA vaccines and other vaccines, and not to the population not exposed to mRNA vaccines in the same period. Furthermore, we excluded non mRNA vaccines for COVID-19 considering the lower exposure of the population to these vaccines and in order to focus the analysis on mRNA vaccines.