Comparability of meta-analyses of first-line steroids for sudden sensorineural hearing loss

Meta-analyses comparing first-line intratympanic and systemic corticosteroids for idiopathic sudden sensorineural hearing loss have reported differing conclusions despite drawing on many of the same randomized trials. We conducted a cross-sectional meta-research study to examine whether these differences were related to trial selection and to clinical and analytic specifications. Searches from database inception through July 18, 2026, identified 16 systematic reviews contributing 34 pooled estimates, 13 of which could be reconstructed against the Reference PICO. Published estimates were adjudicated at the estimate level, their contributing trials were mapped, and reconstructable estimates were harmonized for eligibility, follow-up, outcome definition, effect orientation, and statistical model. We assessed whether published pooled estimates matched the Reference PICO, how the evidence pool and historical shared core changed after harmonization, whether point directions changed, and whether alternative recovery definitions or statistical models affected the results. No published pooled estimate directly matched the complete Reference PICO in any of the six outcome-by-time strata, although eligible trial contributions allowed corrected-target reconstruction for selected outcomes. Across the 13 reconstructable estimates, 110 contributor slots were reduced to 28 corrected-target contributor slots, a 74.5% reduction at the review-estimate level. The 82 removed slots represented 49 trial-by-signature instances and 23 unique trials; 34 removals (41.5%) occurred within the historical shared core, representing 13 instances and seven unique trials. At the review-estimate level, 10 of 13 comparisons shifted from an IT-favoring to an ST-favoring point direction. These comparisons mapped to four corrected-target signatures with overlapping trials. Each signature included only two or three trials, and all modified Knapp–Hartung confidence intervals crossed the null. An illustrative single-trial analysis showed different point directions across recovery definitions, with all confidence intervals crossing the null. Model-only sensitivity analyses changed precision but not point direction. These findings show that substantial trial overlap did not ensure that meta-analyses addressed the same clinical question. The applicability of a pooled estimate depends on the patients represented, the treatment comparison, the timing of outcome assessment, and how hearing recovery was defined.