Intratympanic vs Systemic Corticosteroids in First-line Treatment of Idiopathic Sudden Sensorineural Hearing Loss

Study Inclusion

Included studies were randomized studies that must have excluded identifiable causes of sensorineural hearing loss, such as vestibular schwannoma or Meniére disease. Corticosteroids must have been administered exclusively and not as combination therapy (such as with antiviral medicine or hyperbaric oxygen therapy). We excluded 1 study that did not apply the diagnostic criteria of a hearing loss of a minimum 30 dB within 72 hours.⁷

Search Strategy and Identification of Eligible Studies

PubMed, Embase, OvidSP, CINAHL, and Cochrane Library were searched from January 1, 1966, to July 1, 2018, as follows: “sudden deafness OR ISSNHL OR sudden sensorineural hearing loss OR acute hearing loss” AND “steroid OR corticosteroid OR dexamethasone OR methylprednisolone” AND “randomised OR randomized.” We restricted the search to randomized clinical trials (either placebo controlled or not), systematic reviews, and meta-analyses. We imposed no language restrictions.

The preliminary search identified 170 publications, which were screened by title or abstract by both authors (C.M. and T.O.). A total of 56 publications (32.9%) were eligible for full-text assessment. Subsequently, the abstract of these articles were jointly read by both authors: in total, 51 studies (91.1%) were excluded because they violated the inclusion criteria, including nonrandomized studies (eg, retrospective reviews and case reports) and studies investigating combination therapy with other drugs (eg, antiviral medicine). We screened the reference lists of systematic reviews and meta-analyses. Two studies were identified and added (eFigure in the Supplement).

In total, 7 eligible articles were identified and reviewed independently by both authors. Data of interest were subsequently extracted in consensus and compiled, of which the reported PTA gain in dB from baseline to follow-up was included first. The PTA was measured by taking the mean of 4 frequencies: 4 studies measured at 500, 1000, 2000, and 3000 Hz^8,9,10,11 and 3 studies measured at 500, 1000, 2000, and 4000 Hz.^12,13,14 The second type of data extracted was the severity of the mean hearing loss. Based on the baseline mean PTA, data from the individual studies were allocated accordingly to the Goodman criteria of hearing loss: mild (26 dB-40 dB), moderate (41 dB-55 dB), moderate to severe (56 dB-70 dB), severe (71 dB-90 dB), and profound (≥91 dB). Third, data were included on the proportion of patients achieving complete recovery as defined within each study and in the different treatment arms.

Data Management

Data were aggregated and allocated according to the different treatment arms: patients receiving the intratympanic corticosteroid (IT group), systemic treatment (ST group); and a combination of intratympanic and systemic treatment (CB group). The odds for achieving complete recovery were computed for the IT, ST, and CB groups. Odds ratios were subsequently calculated between the groups. An assessment of risk of bias was performed in accordance with previously established Cochrane Collaboration tools.¹⁵

Statistical Analysis

The fixed-effect model, assuming that 1 true effect size underlies all included studies in the analysis, was used for effect size calculation. Heterogeneity was quantified accordingly to Higgins et al,¹⁶ with low, moderate, and high corresponding to I² values of 25%, 50%, and 75%.¹⁶ We performed the statistical analyses in R-Studio by using the metafor package.¹⁷ Statistical significance was set at P < .05.

Risk of Bias Assessment

We assessed the risk of bias in each of the included studies. We present the detailed assessment in eTables 2 and 3 in the Supplement. The randomization and treatment allocation were known in all 7 studies (to the key staff and patient), yielding a high risk of bias in all 7 studies. Further associations with a high risk of bias were observed in the included studies: first, the allocation of patients was not concealed in 2 studies^9,10; second, 1 study allocated patients based on a principle of rotation and alternation⁸; and finally, 1 study did not report the length of follow-up.¹¹ Hence, insufficient concealment of randomization and nonmasked personnel and patients are predominant in the current literature, with the latter affecting all current publications. Similarly, there were numerous unknown risks of bias because of insufficient reporting (eTables 2 and 3 in the Supplement).

Intratympanic Treatment vs Systemic Treatment

Five studies investigated the IT group compared with the ST group,^{8,10,12,13,14} composed of a total of 464 patients, of whom 235 (51%) were allocated to the IT group and 229 (49%) to the ST group. According to the Goodman criteria, 3 studies had a baseline PTA equivalent to a severe hearing loss,^10,12,14 whereas the remaining 2 were classified as moderate to severe hearing loss^8,13 (Table 1^{8,9,10,11,12,13,14}). The mean days until treatment initiation ranged from 3.4 to 10.1 days in the IT group vs 3.1 to 7.1 days in the ST group. One study did not report days until treatment initiation.¹³ Follow-up ranged from 17 days to 3 months (eTable 1 in the Supplement).

Four studies were eligible for analysis of differences in mean PTA gain.^8,12,13,14 The fixed-effects model estimated no significant difference in mean PTA gain between the 2 groups (−2.01 dB; 95% CI, −5.61 dB to 1.59 dB; P = .96; I² = 0%; χ² P = .62). The model indicated no significant heterogeneity (Figure 1^8,12,13,14).

For the IT group, complete recovery was achieved in 56 of 235 patients (24%), and for the ST group, complete recovery was achieved in 58 of 229 patients (25%) (Table 2^{2,8,9,10,11,12,13,14,18}). Figure 2^{8,10,12,13,14} comprises odds for recovery stratified for baseline PTA; first, moderate to severe hearing loss, which indicated no significant difference, and second, severe hearing loss, which indicated no significant difference. The heterogeneity was moderate but not significant.

For the entire cohort comprising all 5 studies, the fixed-effects model estimate indicated no significant difference between the 2 groups, with an odds ratio of 0.94 (95% CI, 0.61 to 1.44). The heterogeneity was low and not significant (I² = 34.5%; χ² P = .19; Figure 2).

Combined Treatment vs Systemic Treatment

Four studies investigated the CB vs ST group.^8,9,11,12 In total, the studies comprised 301 patients, of whom 150 (49.8%) were randomly allocated to the CB group and the remaining 151 patients (50.2%) randomized to the ST group. The definition of complete recovery varied between studies and is listed in Table 2. All studies had a mean PTA at baseline corresponding to severe hearing loss according to Goodman criteria except for 1 study that corresponded to moderate to severe hearings loss⁸ (Table 1). The mean days until treatment initiation ranged between 4.0 to 9.6 days in the CB group and 3.1 and 7.1 days in the ST group. Follow-up ranged from 17 days to 3 months. One study did not report the duration of follow-up¹¹ (eTable 1 in the Supplement).

For the CB group, complete recovery was achieved in 49 of 150 patients (33%), and for the ST group, complete recovery was achieved in 46 of 151 patients (30%) (Table 2). The fixed-effects model estimated no significant difference. The odds ratio was 1.11 (95% CI, 0.68 to 1.82). The heterogeneity was low and not significant (I² = 0%; χ² P = 0.75) (Figure 3).

Combined Treatment vs Intratympanic Treatment

Two studies investigated the CB group compared with the IT group,^8,12 comprising a total of 107 eligible patients. The CB group consisted of 53 patients (50%), whereas the IT group consisted of 54 patients (50%). Neither of the 2 studies solitarily indicated any significant difference between the treatments; for severe hearing loss, the odds ratio was 2.7 (95% CI, 0.86 to 8.27), and for the moderate to severe hearing loss, the odds ratio was 3.78 (95% CI, 0.83 to 17.25) (eFigure in the Supplement).

Comparison With Existing Literature

In the investigation of the IT group vs the ST group, a meta-analysis applied a random-effects model to estimate a nonsignificant mean difference between the 2 groups.¹⁹ We were surprised to learn that specific data apparently used from 2 of the included studies did not appear in the original articles.^7,20 Further, the meta-analysis included Kosyakov et al,²¹ who report baseline PTA characteristics as a mean (SD) 41.0 (12.87) dB and 37.1(16.67) dB, respectively. This indicates that an unknown proportion of the included patients are not confined within the ISSNHL-definitory criteria of a minimum 30-dB hearing loss.²¹

Another meta-analysis applied a fixed-effects model to estimate the effect sizes of the IT and ST group.²² The mean difference in dB was reported significantly as 3.42 dB (95% CI, 0.17-6.67 dB; I² = 37%; P = .04) in favor for the IT group. However, we are concerned of the validity of these results. We observed that data from the Kosyakov et al¹⁹ were included and used equivalently to the beforementioned meta-analysis. Similarly, the meta-analysis included results from the Battaglia et al⁷ study; however, this specific study defined ISSNHL as a hearing loss of a minimum of 20 dB and not 30 dB. The meta-analysis stated to have included randomized trials exclusively. Therefore, we were surprised to learn that the included study of Filipo et al²³ was clearly stated as nonrandomized. These observations might highlight some perspectives as to why findings remain contradictory and evidence-based recommendations have not yet been established.

One randomized study concluded that patients in the CB group had a significantly higher likelihood of hearing recovery compared with the ST group.⁷ However, in addition to defining ISSNHL as a hearing loss of a minimum 20 dB, the mean days to treatment was 4 days in the CB group vs 11 days in in the ST group. Acknowledging time-to-treatment initiation as an important prognostic parameter, this baseline discrepancy is highly associated with treatment outcomes and could explain the results.²⁴

Five randomized trials met the inclusion criteria but were not included for analysis.^{7,20,25,26,27} First, the previously mentioned study from Battaglia et al⁷ defined ISSNHL as a hearing loss of a minimum 20 dB. This conflicts with the international consensus that ISSNHL is defined as a hearing loss of minimum 30 dB. Hence, this implies heterogeneity and we consequently excluded the article.

Further, none of the remaining 4 studies reported the PTA mean gain in dB ± the standard deviation. Three of the four studies defined complete recovery as a gain of a minimum 10 dB in PTA,^20,25,26 whereas the remaining defined complete recovery as a gain of a minimum 15 dB.²⁷ Thus, recovery rates were not graduated but dichotomous in association with whether complete recovery was achieved. Consequently, if we included the 4 studies in this meta-analysis, we would have had to define a gain of 10 dB as criteria for achieving complete recovery. Therefore, we included studies with graduated recovery exclusively.

Association With Future Clinical Practice and Research

The International Consensus (ICON) on treating sudden sensorineural hearing loss is a product of a panel discussion at the International Federation of Otorhinolaryngological Societies 2017 ENT World Congress. The ICON Study agreed that the main limitations present in the current literature are associated with the wide heterogeneity, which characterizes the initial hearing deficit and the amount of hearing recovery. Systemic steroids are considered as the current standard care; however, the evidence is debatable. The ICON Study proposes that the number of participants for future studies should be restricted to moderate to profound levels of hearing loss to improve the comparability.²⁸ This meta-analysis qualitatively adds to this consensus.

Strengths and Limitations

One major strength of meta-analyses of randomized studies is the reduced risk of bias. However, in general, the included studies did not provide a placebo: patients allocated to receive oral treatment should concomitantly be given intratympanic placebo (and vice versa). When a placebo was not administered, randomization was not masked to the personnel nor the patient. Thus, we associated all included studies with a high risk of bias on masking and it cannot be denied that the presented results were biased. We encountered other limitations too: first, the individual studies applied different guidelines in evaluating hearing recovery. Complete recovery was defined as a final PTA of at least 25 dB in 3 studies,^10,12,13 at least 30 dB in 1 study¹⁴ and, finally, as a PTA within 10 dB from the unaffected ear in 1 study.⁸ The definitions of complete recovery were fairly comparable and similar; however, we could not adjust for the heterogeneity implied within the use of different assessment schemes, nor the difference in baseline prognostic factors,²⁹ including (1) time to treatment onset, (2) age at baseline, and (3) PTA at baseline. We observed that the mean time to treatment onset varied no more than 7 days between the studies (ST group range, 3.1-7.1 days; IT group, 3.4-10.1 days; CB group, 4.0-9.6 days). We could not quantify the association of this difference with the results, but it has been estimated that the treatment window is between 2 to 4 weeks from the onset of symptoms.^24,30 As the maximum duration between the presentation of symptoms to treatment was 10.1 days, we do not consider this aspect a limitation of significance. We observed that the mean age range at the time of symptom presentation ranged between 45.9 to 56.9 years. Older age is often described as a prognostic factor,^24,31 although older age is not defined. Typically, the age described is presented as categories in the literature (eg, 20-29, 30-39) exclusively and is not treated as a continuous covariate. Thus, it is difficult to assess whether the presented mean ages of the included studies has inherent bias; however, we have observed that the older ages described in literature comprise the age categories 70 to 79 years and 80 to 89years.^24,31 Therefore, we do not consider the varying mean ages as a limitation of significance, as we do not consider the age interval of 45.9 to 56.9 years to be older ages nor is presbycusis a relevant parameter for patients in this age interval. Finally, we do not consider the different PTA at baseline a limitation of significance, as we allocated studies accordingly to mean PTA at baseline and performed our analyses based on the stratification according to Goodman criteria.³² We observed that follow-up in the included studies ranged from 17 days to 3 months. We cannot adjust for the inherent heterogeneity of varying follow-up and this must be considered a limitation for arriving at concise recommendations. Finally, there is the effect of applying different corticosteroids. There is also an emerging question on whether differences in hearing outcomes exist between dexamethasone vs methylprednisolone. The literature is sparse concerning this question, but it has been reported that no difference exists between the 2 drugs in association with intratympanic administration.³³ Based on this, we do not intuitively consider the application of different corticosteroids to be a significant limitation; however, future randomized or noninferiority trials should settle this question.