Myers Joshua, Kei Joseph, Aithal Sreedevi, Aithal Venkatesh, Driscoll Carlie, Khan Asaduzzaman, Manuel Alehandrea, Joseph Anjali, Malicka Alicja N
Department of Audiology, Townsville Hospital and Health Service, Australia.
School of Health and Rehabilitation Sciences, The University of Queensland, Brisbane, Australia.
J Speech Lang Hear Res. 2018 Sep 19;61(9):2386-2404. doi: 10.1044/2018_JSLHR-H-18-0004.
The aim of this study was to develop a risk prediction model for detecting middle ear pathology in 6- to 9-month-old infants using wideband absorbance measures.
Two hundred forty-nine infants aged 23-39 weeks (Mdn = 28 weeks) participated in the study. Distortion product otoacoustic emissions and high-frequency tympanometry were tested in both ears of each infant to assess middle ear function. Wideband absorbance was measured at ambient pressure in each participant from 226 to 8000 Hz. Absorbance results from 1 ear of each infant were used to predict middle ear dysfunction, using logistic regression. To develop a model likely to generalize to new infants, the number of variables was reduced using principal component analysis, and a penalty was applied when fitting the model. The model was validated using the opposite ears and with bootstrap resampling. Model performance was evaluated through measures of discrimination and calibration. Discrimination was assessed with the area under the receiver operating characteristic curve (AUC); and calibration, with calibration curves, which plotted actual against predicted probabilities.
AUC of the fitted model was 0.887. The model validated adequately when applied to the opposite ears (AUC = 0.852) and with bootstrap resampling (AUC = 0.874). Calibration was satisfactory, with high agreement between predictions and observed results.
The risk prediction model had accurate discrimination and satisfactory calibration. Validation results indicate that it may generalize well to new infants. The model could potentially be used in diagnostic and screening settings. In the context of screening, probabilities provide an intuitive and flexible mechanism for setting the referral threshold that is sensitive to the costs associated with true and false-positive outcomes. In a diagnostic setting, predictions could be used to supplement visual inspection of absorbance for individualized diagnoses. Further research assessing the performance and impact of the model in these contexts is warranted.
本研究旨在开发一种风险预测模型，用于通过宽带吸光度测量来检测6至9个月大婴儿的中耳病变。
249名年龄在23至39周（中位数=28周）的婴儿参与了本研究。对每名婴儿的双耳进行畸变产物耳声发射和高频鼓室导抗测试，以评估中耳功能。在每个参与者的环境压力下，测量226至8000赫兹的宽带吸光度。使用逻辑回归，利用每名婴儿一只耳朵的吸光度结果来预测中耳功能障碍。为了开发一个可能适用于新婴儿的模型，使用主成分分析减少变量数量，并在拟合模型时施加惩罚。使用对侧耳朵和自助重采样对模型进行验证。通过判别和校准指标评估模型性能。通过接收者操作特征曲线（AUC）下的面积评估判别能力；通过校准曲线评估校准情况，校准曲线绘制实际概率与预测概率。
拟合模型的AUC为0.887。当应用于对侧耳朵时（AUC = 0.852）和进行自助重采样时（AUC = 0.874），模型得到了充分验证。校准情况令人满意，预测结果与观察结果高度一致。
风险预测模型具有准确的判别能力和令人满意的校准效果。验证结果表明，它可能很好地适用于新婴儿。该模型可能潜在地用于诊断和筛查环境。在筛查背景下，概率为设定转诊阈值提供了一种直观且灵活的机制，该阈值对与真阳性和假阳性结果相关的成本敏感。在诊断环境中，预测结果可用于补充吸光度的目视检查以进行个体化诊断。有必要进一步开展研究，评估该模型在这些背景下的性能和影响。
