Data Sheet 2_Predictive model for aminoglycoside induced ototoxicity.pdf

BackgroundIrreversible hearing loss is a well-known adverse effect of aminoglycosides, however, inability to accurately predict ototoxicity is a major limitation in clinical care. We addressed this limitation by developing a prediction model for aminoglycoside ototoxicity applicable to the general population.
MethodsWe employed a prospective non-drug-resistant tuberculosis (TB), non-HIV/AIDS cohort of 153 adults on Streptomycin based anti-TB therapy. High frequency pure-tone audiometry was done at regular intervals throughout the study. Clinical and audiological predictors of ototoxicity were collated and ototoxic threshold shift from the baseline audiogram computed. The prediction model was developed with logistic regression method by examining multiple predictors of ototoxicity. Series of models were fitted sequentially; the best model was identified using Akaike Information Criterion and likelihood ratio test. Key variables in the final model were used to develop a logit model for ototoxicity prediction.
ResultsOtotoxicity occurred in 35% of participants. Age, gender, weight, cumulative Streptomycin dosage, social class, baseline pure tone average (PTA) and prior hearing symptoms were explored as predictors. Multiple logistic regression showed that models with age, cumulative dosage and baseline PTA were best for predicting ototoxicity. Regression parameters for ototoxicity prediction showed that yearly age increment raised ototoxicity risk by 5% (AOR = 1.05; CI, 1.01–1.09), and a gram increase in cumulative dosage increased ototoxicity risk by 7% (AOR = 1.05; CI, 1.05–1.12) while a unit change in baseline log (PTA) was associated 254% higher risk of ototoxicity (AOR = 3.54, CI: 1.25, 10.01). Training and validation models had area under the receiver operating characteristic curve as 0.84 (CI, 0.76–0.92) and 0.79 (CI, 0.62–0.96) respectively, showing the model has discriminatory ability.
ConclusionThis model can predict aminoglycoside ototoxicity in the general population.

【背景】不可逆听力损失是氨基糖苷类（aminoglycosides）药物明确的不良反应，但目前无法精准预测耳毒性（ototoxicity）仍是临床诊疗中的一大局限。本研究针对该局限，开发了一款可应用于普通人群的氨基糖苷类耳毒性预测模型。【方法】本研究纳入153名接受链霉素（Streptomycin）抗结核治疗的成人非耐药结核病（TB）、非艾滋病病毒感染者队列，为前瞻性研究设计。研究全程定期对受试者进行高频纯音测听（pure-tone audiometry）。收集与耳毒性相关的临床及听力学预测指标，并计算相较于基线听力图的耳毒性阈值偏移量。本研究通过逻辑回归方法，对多种耳毒性预测指标进行分析以构建预测模型。依次拟合多组模型，采用赤池信息准则（Akaike Information Criterion）与似然比检验筛选最优模型。以最终模型中的关键变量构建耳毒性预测logit模型。【结果】35%的受试者出现了耳毒性。本研究将年龄、性别、体重、累计链霉素使用剂量、社会阶层、基线纯音听阈均值（pure tone average, PTA）以及既往听力症状作为候选预测指标。多因素逻辑回归分析显示，纳入年龄、累计剂量与基线PTA的模型为最优耳毒性预测模型。耳毒性预测的回归参数结果显示：年龄每增加1岁，耳毒性发病风险升高5%（调整后比值比（adjusted odds ratio, AOR）=1.05；置信区间（confidence interval, CI）：1.01~1.09）；累计剂量每增加1g，耳毒性发病风险升高7%（AOR=1.05；CI：1.05~1.12）；基线log转换后的PTA每变化1个单位，耳毒性发病风险升高254%（AOR=3.54，CI：1.25~10.01）。训练集与验证集模型的受试者工作特征曲线下面积（area under the receiver operating characteristic curve）分别为0.84（CI：0.76~0.92）与0.79（CI：0.62~0.96），表明该模型具有良好的区分能力。【结论】本研究所构建的模型可对普通人群的氨基糖苷类耳毒性进行有效预测。