Validation of the Cepstral Spectral Index of Dysphonia (CSID) as a Screening Tool for Voice Disorders: Development of Clinical Cutoff Scores

详细信息    查看全文

• 作者：Shaheen N. Awan^&lowast ; ; ^{sawan@bloomu.edu" class="auth_mail" title="E-mail the corresponding author} ; Nelson Roy^&dagger ; ; Dong Zhang^&Dagger ; ; Seth M. Cohen^§ ;
• 关键词：Cepstrum ; Cepstral analysis ; Voice disorders ; Sensitivity ; Specificity ; Voice screening
• 刊名：Journal of Voice
• 出版年：2016
• 出版时间：March 2016
• 年：2016
• 卷：30
• 期：2
• 页码：130-144
• 全文大小：1070 K

文摘

The purposes of this study were to (1) evaluate the performance of the Cepstral Spectral Index of Dysphonia (CSID—a multivariate estimate of dysphonia severity) as a potential screening tool for voice disorder identification and (2) identify potential clinical cutoff scores to classify voice-disordered cases versus controls.

Methodology

Subjects were 332 men and women (116 men, 216 women) comprised of subjects who presented to a physician with a voice-related complaint and a group of non–voice-related control subjects. Voice-disordered cases versus controls were initially defined via three reference standards: (1) auditory-perceptual judgment (dysphonia +/−); (2) Voice Handicap Index (VHI) score (VHI +/−); and (3) laryngoscopic description (laryngoscopic +/−). Speech samples were analyzed using the Analysis of Dysphonia in Speech and Voice program. Cepstral and spectral measures were combined into a CSID multivariate formula which estimated dysphonia severity for Rainbow Passage samples (ie, the CSID_R). The ability of the CSID_R to accurately classify cases versus controls in relation to each reference standard was evaluated via a combination of logistic regression and receiver operating characteristic (ROC) analyses.

Results

The ability of the CSID_R to discriminate between cases and controls was represented by the “area under the ROC curve” (AUC). ROC classification of dysphonia-positive cases versus controls resulted in a strong AUC = 0.85. A CSID_R cutoff of ≈24 achieved the best balance between sensitivity and specificity, whereas a more liberal cutoff score of ≈19 resulted in higher sensitivity while maintaining respectable specificity which may be preferred for screening purposes. Weaker but adequate AUCs = 0.75 and 0.73 were observed for the classification of VHI-positive and laryngoscopic-positive cases versus controls, respectively. Logistic regression analyses indicated that subject age may be a significant covariate in the discrimination of dysphonia-positive and VHI-positive cases versus controls.

Conclusions

The CSID_R can provide a strong level of accuracy for the classification of voice-disordered cases versus controls, particularly when auditory-perceptual judgment is used as the reference standard. Although users often focus on a cutoff score that achieves a balance between sensitivity and specificity, more liberal cutoffs for screening purposes versus conservative cutoffs when cost or risk of further evaluation is deemed to be high should also be considered.