[[alternative]]Analysis of factors and comparison of predictive models for outcome of sudden hearing loss treatment

[[abstract]]突發感音神經性耳聾 ( Sudden sensorineural hearing loss, SSNHL) 在聽力學研究以及臨床耳科中屬於十分棘手疾病，主要由於該病症的異質性和具有自發恢復的可能性，導致目前的研究中尚未建立相關的預測模型，所以對於分析預後的相關因子並建立預測模型是具有重大意義的。本研究使用羅吉斯回歸( Logistic Regression)建立預測治療後聽力恢復的預後模型。研究中回顧了888位在2011年11月至2021年4月間，在台灣北區醫學中心住院接受治療的SSNHL個案(男性431位，女性457位；平均年齡50.1歲；治療前聽力損失平均閾值73.7 dB )相關醫療數據，為回溯性的世代研究。SSNHL最常見的定義是：純音聽力檢查的結果為至少 3 個連續的頻率，出現 30 dB 以上的感音神經性聽力損失，並發生在 72 小時內，而聽力恢復的定義則根據Modified Siegel's criteria的標準，共407位屬於治療後聽力恢復。探究人口學特質、疾病特性以及身體徵狀與治療後是否屬於恢復進行統計分析，使用獨立樣本t檢定以及卡方檢定的結果建立研究的基準線(Baseline)，並將過往研究中發現到的預後因子以及基準線呈統計上顯著意義的因子納入羅吉斯回歸分析並比較其與神經網路多層感知器(MLP)以及分類樹狀結構CRT(Tree-CRT)的靈敏度與異度。在多變量羅吉斯回歸模型中，我們發現到：發病至就醫時間為14天內、治療前聽力1 kHz 閾值、治療前聽力4 kHz 閾值、轉院個案、耳垢嵌塞、耳咽管阻塞、耳鳴以及眩暈是重要的預後因子。當個案年齡小於30歲(含)以下時，在模型中預測能力最好，整體正確度為69.1%(Sensitivity：69%；Specificity：69.2%)。關鍵字：突發感音神經性耳聾、治療後聽力恢復、預後相關因子、多變量羅吉斯回歸預測模型[[abstract]]Sudden sensorineural hearing loss ( SSNHL ) is a troublesome symptom in audiology research and clinical otolaryngology. Due to the heterogeneity of the disease and the possibility of spontaneous recovery, no predictive models have been established in the current study. Therefore, it is of great significance to analyze the prognostic correlation factors and build predictive models. This study used Logistic regression to develop a prognostic model for predicting recovery of hearing after treatment. We retrospectively reviewed the medical data of 888 patients with SSNHL (Men, 431; Women, 457; mean age,50.1 years; mean of pretreat pure-tone audiometry, 73.7 dB ) who underwent treatment at the Northern Taiwan Medical Center between November 2011 and April 2021. Sudden Sensorineural Hearing Loss (SSNHL) refers to an abrupt onset of hearing loss at > 30 dB for at least three contiguous frequencies within 72 hours and the definition of hearing restoration is based on the Modified Siegel's criteria. Totally, 407 patients had their hearing restored after treatment. To investigate demographic characteristics, disease characteristics, and physical symptoms and recovery after treatment for statistical analysis. Baselines were established using independent sample t-testing and chi-squared results. Prognostic factors identified in previous studies and statistically significant factors in the baseline were included in Logistic regression analyses and compared to the sensitivity and Specificity of the Multilayer perceptron (MLP), and Classification Tree CRT(Classification, Regression Tree, Tree-CRT) .We found that onset of treatment within 14 days, pre-treatment hearing 1 kHz threshold, pre-treatment hearing 4 kHz threshold, Salvage cases, Impacted cerumen, Obstruction of Eustachian tube, unspecified, Tinnitus, unspecified, Dizziness and giddiness were significant prognostic factors in the multivariate Logistic regression model. Predictive power was best in the model when cases were younger than 31 years old, with an overall correctness of 69.1% (Sensitivity: 69%; Specificity: 69.2%).Keywords: sudden sensorineural hearing loss, post-treatment hearing recovery, prognostic correlates, multivariate Logistic regression prediction model