Predicting audiometric status from distortion product otoacoustic emissions using multivariate analyses

Patricia A. Dorn, Pawel Piskorski, Michael P Gorga, Stephen T Neely, Douglas H Keefe

Research output: Contribution to journalArticle

64 Citations (Scopus)

Abstract

Objectives: 1) To determine whether multivariate statistical approaches improve the classification of normal and impaired ears based on distortion product otoacoustic emission (DPOAE) measurements, in comparison with the results obtained with more traditional single-variable applications of clinical decision theory. 2) To determine how well the multivariate predictors, derived from analysis on a training group, generalized to a validation group. 3) To provide a way to apply the multivariate approaches clinically. Design: Areas under the relative operating characteristic (ROC) curve and cumulative distributions derived from DPOAE, DPOAE/Noise, discriminant function (DF) scores and logit function (LF) scores were used to compare univariate and multivariate predictors of audiometric status. DPOAE and Noise amplitudes for 8 f2 frequencies were input to a discriminant analysis and to a logistic regression. These analyses generated a DF and LF, respectively, composed of a linear combination of selected variables. The DF and LF scores were the input variables to the decision theory analyses. For comparison purposes, DPOAE test performance was also evaluated using only one variable (DPOAE or DPOAE/Noise when f2 = audiometric frequency). Analyses were based on data from over 1200 ears of 806 subjects, ranging in age from 1.3 to 96 yr, with thresholds ranging from -5 to >120 dB HL. For statistical purposes, normal hearing was defined as thresholds of 20 dB HL or better. For the multivariate analyses, the database was randomly divided into two groups of equal size. One group served as the 'training' set, which was used to generate the DFs and LFs. The other group served as a 'validation' set to determine the robustness of the DF and LF solutions. Results: For all test frequencies, multivariate analyses yielded greater areas under the ROC curve than univariate analyses, and greater specificities at fixed sensitivities. Within the multivariate techniques, discriminant analysis and logistic regression yielded similar results and both yielded robust solutions that generalized well to the validation groups. The improvement in test performance with multivariate analyses was greatest for conditions in which the single predictor variable resulted in the poorest performance. Conclusions: A more accurate determination of auditory status at a specific frequency can be obtained by combining multiple predictor variables. Although the DF and LF multivariate approaches resulted in the greatest separation between normal and impaired distributions, overlap still exists, which suggests that there would be value in continued efforts to improve DPOAE test performance.

Original languageEnglish (US)
Pages (from-to)149-163
Number of pages15
JournalEar and hearing
Volume20
Issue number2
DOIs
StatePublished - Apr 1999

Fingerprint

Decision Theory
Noise
Multivariate Analysis
Discriminant Analysis
Ear
Logistic Models
Decision Support Techniques
Normal Distribution
Hearing
Databases

ASJC Scopus subject areas

  • Otorhinolaryngology
  • Speech and Hearing

Cite this

Predicting audiometric status from distortion product otoacoustic emissions using multivariate analyses. / Dorn, Patricia A.; Piskorski, Pawel; Gorga, Michael P; Neely, Stephen T; Keefe, Douglas H.

In: Ear and hearing, Vol. 20, No. 2, 04.1999, p. 149-163.

Research output: Contribution to journalArticle

@article{cab45ea421f64059bf9e5c3fbe4a0c6d,
title = "Predicting audiometric status from distortion product otoacoustic emissions using multivariate analyses",
abstract = "Objectives: 1) To determine whether multivariate statistical approaches improve the classification of normal and impaired ears based on distortion product otoacoustic emission (DPOAE) measurements, in comparison with the results obtained with more traditional single-variable applications of clinical decision theory. 2) To determine how well the multivariate predictors, derived from analysis on a training group, generalized to a validation group. 3) To provide a way to apply the multivariate approaches clinically. Design: Areas under the relative operating characteristic (ROC) curve and cumulative distributions derived from DPOAE, DPOAE/Noise, discriminant function (DF) scores and logit function (LF) scores were used to compare univariate and multivariate predictors of audiometric status. DPOAE and Noise amplitudes for 8 f2 frequencies were input to a discriminant analysis and to a logistic regression. These analyses generated a DF and LF, respectively, composed of a linear combination of selected variables. The DF and LF scores were the input variables to the decision theory analyses. For comparison purposes, DPOAE test performance was also evaluated using only one variable (DPOAE or DPOAE/Noise when f2 = audiometric frequency). Analyses were based on data from over 1200 ears of 806 subjects, ranging in age from 1.3 to 96 yr, with thresholds ranging from -5 to >120 dB HL. For statistical purposes, normal hearing was defined as thresholds of 20 dB HL or better. For the multivariate analyses, the database was randomly divided into two groups of equal size. One group served as the 'training' set, which was used to generate the DFs and LFs. The other group served as a 'validation' set to determine the robustness of the DF and LF solutions. Results: For all test frequencies, multivariate analyses yielded greater areas under the ROC curve than univariate analyses, and greater specificities at fixed sensitivities. Within the multivariate techniques, discriminant analysis and logistic regression yielded similar results and both yielded robust solutions that generalized well to the validation groups. The improvement in test performance with multivariate analyses was greatest for conditions in which the single predictor variable resulted in the poorest performance. Conclusions: A more accurate determination of auditory status at a specific frequency can be obtained by combining multiple predictor variables. Although the DF and LF multivariate approaches resulted in the greatest separation between normal and impaired distributions, overlap still exists, which suggests that there would be value in continued efforts to improve DPOAE test performance.",
author = "Dorn, {Patricia A.} and Pawel Piskorski and Gorga, {Michael P} and Neely, {Stephen T} and Keefe, {Douglas H}",
year = "1999",
month = "4",
doi = "10.1097/00003446-199904000-00006",
language = "English (US)",
volume = "20",
pages = "149--163",
journal = "Ear and Hearing",
issn = "0196-0202",
publisher = "Lippincott Williams and Wilkins",
number = "2",

}

TY - JOUR

T1 - Predicting audiometric status from distortion product otoacoustic emissions using multivariate analyses

AU - Dorn, Patricia A.

AU - Piskorski, Pawel

AU - Gorga, Michael P

AU - Neely, Stephen T

AU - Keefe, Douglas H

PY - 1999/4

Y1 - 1999/4

N2 - Objectives: 1) To determine whether multivariate statistical approaches improve the classification of normal and impaired ears based on distortion product otoacoustic emission (DPOAE) measurements, in comparison with the results obtained with more traditional single-variable applications of clinical decision theory. 2) To determine how well the multivariate predictors, derived from analysis on a training group, generalized to a validation group. 3) To provide a way to apply the multivariate approaches clinically. Design: Areas under the relative operating characteristic (ROC) curve and cumulative distributions derived from DPOAE, DPOAE/Noise, discriminant function (DF) scores and logit function (LF) scores were used to compare univariate and multivariate predictors of audiometric status. DPOAE and Noise amplitudes for 8 f2 frequencies were input to a discriminant analysis and to a logistic regression. These analyses generated a DF and LF, respectively, composed of a linear combination of selected variables. The DF and LF scores were the input variables to the decision theory analyses. For comparison purposes, DPOAE test performance was also evaluated using only one variable (DPOAE or DPOAE/Noise when f2 = audiometric frequency). Analyses were based on data from over 1200 ears of 806 subjects, ranging in age from 1.3 to 96 yr, with thresholds ranging from -5 to >120 dB HL. For statistical purposes, normal hearing was defined as thresholds of 20 dB HL or better. For the multivariate analyses, the database was randomly divided into two groups of equal size. One group served as the 'training' set, which was used to generate the DFs and LFs. The other group served as a 'validation' set to determine the robustness of the DF and LF solutions. Results: For all test frequencies, multivariate analyses yielded greater areas under the ROC curve than univariate analyses, and greater specificities at fixed sensitivities. Within the multivariate techniques, discriminant analysis and logistic regression yielded similar results and both yielded robust solutions that generalized well to the validation groups. The improvement in test performance with multivariate analyses was greatest for conditions in which the single predictor variable resulted in the poorest performance. Conclusions: A more accurate determination of auditory status at a specific frequency can be obtained by combining multiple predictor variables. Although the DF and LF multivariate approaches resulted in the greatest separation between normal and impaired distributions, overlap still exists, which suggests that there would be value in continued efforts to improve DPOAE test performance.

AB - Objectives: 1) To determine whether multivariate statistical approaches improve the classification of normal and impaired ears based on distortion product otoacoustic emission (DPOAE) measurements, in comparison with the results obtained with more traditional single-variable applications of clinical decision theory. 2) To determine how well the multivariate predictors, derived from analysis on a training group, generalized to a validation group. 3) To provide a way to apply the multivariate approaches clinically. Design: Areas under the relative operating characteristic (ROC) curve and cumulative distributions derived from DPOAE, DPOAE/Noise, discriminant function (DF) scores and logit function (LF) scores were used to compare univariate and multivariate predictors of audiometric status. DPOAE and Noise amplitudes for 8 f2 frequencies were input to a discriminant analysis and to a logistic regression. These analyses generated a DF and LF, respectively, composed of a linear combination of selected variables. The DF and LF scores were the input variables to the decision theory analyses. For comparison purposes, DPOAE test performance was also evaluated using only one variable (DPOAE or DPOAE/Noise when f2 = audiometric frequency). Analyses were based on data from over 1200 ears of 806 subjects, ranging in age from 1.3 to 96 yr, with thresholds ranging from -5 to >120 dB HL. For statistical purposes, normal hearing was defined as thresholds of 20 dB HL or better. For the multivariate analyses, the database was randomly divided into two groups of equal size. One group served as the 'training' set, which was used to generate the DFs and LFs. The other group served as a 'validation' set to determine the robustness of the DF and LF solutions. Results: For all test frequencies, multivariate analyses yielded greater areas under the ROC curve than univariate analyses, and greater specificities at fixed sensitivities. Within the multivariate techniques, discriminant analysis and logistic regression yielded similar results and both yielded robust solutions that generalized well to the validation groups. The improvement in test performance with multivariate analyses was greatest for conditions in which the single predictor variable resulted in the poorest performance. Conclusions: A more accurate determination of auditory status at a specific frequency can be obtained by combining multiple predictor variables. Although the DF and LF multivariate approaches resulted in the greatest separation between normal and impaired distributions, overlap still exists, which suggests that there would be value in continued efforts to improve DPOAE test performance.

UR - http://www.scopus.com/inward/record.url?scp=0032961355&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=0032961355&partnerID=8YFLogxK

U2 - 10.1097/00003446-199904000-00006

DO - 10.1097/00003446-199904000-00006

M3 - Article

C2 - 10229516

AN - SCOPUS:0032961355

VL - 20

SP - 149

EP - 163

JO - Ear and Hearing

JF - Ear and Hearing

SN - 0196-0202

IS - 2

ER -