Hearing preservation and restoration surgery in NF2-related schwannomatosis.

To analyze factors associated with hearing preservation in patients undergoing surgery with NF2-related schwannomatosis (NF2-SWN) vestibular schwannomas (VS), and to evaluate artificial intelligence/machine learning (AI/ML) predictors of hearing preservation. Retrospective analysis of NF2-SWN patients with a preoperative word recognition score (WRS) of greater than 50% who were undergoing surgical resection of VS with the goal of hearing preservation. Factors evaluated for influence on hearing preservation included preoperative tumor size and location, pure tone average and WRS, age of onset of NF2-SWN symptoms, auditory brainstem response (ABR), NF2 genetic severity score, inflammation in the cochlea and labyrinth, and surgical approach. The relationship between factors and hearing preservation was evaluated with univariate and multivariable analysis and a “SuperLearner” AI/ML model. When hearing was not preserved, outcomes of cochlear implants (CI) and auditory brainstem implants (ABI) are presented. Of 322 patients with NF2-SWN, 191 underwent surgical excision of VS. Fifty-two (27%) had WRS > 50% and tumor configuration allowing attempted hearing preservation. Twenty-three of the 52 patients had postoperative WRS > 50% (44%). The mean tumor size when hearing was preserved was 16.4 mm whereas the mean tumor size in those not preserved was 22.2 mm giving an odds ratio (OR) on univariate analysis of 0.90 (CI 0.85–0.95) suggesting larger tumors were less likely to have preserved hearing. For 12 patients with greater than 5-year follow up and preserved hearing, the mean time for their WRS to degrade below 50% was 11.6 years. Univariate regression also identified the presence of normal wave V on ABR having higher OR of hearing preservation 30.9 (95% CI, 3.78, 252.2) whereas tumors touching the brainstem lowered the odds ratio of hearing preservation of 0.07 (95%CI, 0.02, 0.24). However, none of the factors remained significant on multivariate regression after adjusting for other factors. In an AI/ML SuperLearner model, the most influential variables were normal ABR, lack of tumor contact with the brainstem, and preoperative WRS. The accuracy was 85% (79–91%); the sensitivity: 0.69 (0.50–0.86) and the specificity: 0.87 (0.80–0.95). This model of implies that collectively there is indeed useful information for predicting hearing preservation at surgery from these data. Cochlear implant and auditory brainstem implant rehabilitation is also reported. Seventy-three percent of NF2-SWN patients in this study had poor hearing at the time of surgery limiting hearing preservation. Current hearing preservation options are largely inadequate. Factors influencing postoperative hearing in NF2-SWN VS resection when combined with AI/ML may aid in predicting surgical outcomes. Further confirmational studies are needed. CI and ABI are shown to provide long-term hearing rehabilitation options for a modest proportion of NF2-SWN patients. Earlier intervention and better treatment options are needed.