Multivariate analysis of objective vocal function

No standard and valid multidimensional index of objective voice function has been developed that integrates the information generated from the multiple objective parameters of voice function. The goals of this research were 1) to identify important objective voice parameters and 2) to create a multidimensional voice function index by combining relevant parameters. We evaluated 97 dysphonic patients and 35 normal volunteers on 14 objective voice parameters. Three multidimensional voice indices were created and evaluated: 1) nonweighted univariate index, 2) weighted odds ratio index, and 3) weighted multivariate regression index. The univariate index required all 14 parameters, while the odds ratio and logistic regression models required only 4 parameters (frequency range, airflow at lips, maximum phonation time, and subglottic pressure). The chi2 values for the 3 models were 37.8, 37.6, and 46.0, respectively. All 3 indices were able to satisfactorily classify voice function as normal or abnormal. However, the regression index performed best.