Prediction of conductive hearing loss based on acoustic earcanal response using a multivariate clinical decision theory

Pawel Piskorski, Douglas H Keefe, Jeffrey L. Simmons, Michael P Gorga

Research output: Contribution to journalArticle

65 Citations (Scopus)

Abstract

This study evaluated the accuracy of acoustic response tests in predicting conductive hearing loss in 161 ears of subjects from the age of 2 to 10 yr, using as a 'gold standard' the air-bone gap to classify ears as normal or impaired. The acoustic tests included tympanometric peak- compensated static admittance magnitude (SA) and tympanometric gradient at 226 Hz, and admittance-reflectance (YR) measurements from 0.5 to 8 kHz. The performance of individual, frequency-specific, YR test variables as predictors was assessed. By applying logistic regression (LR) and discriminant analysis (DA) techniques to the multivariate YR response, two univariate functions were calculated as the linear combinations of YR variables across frequency that best separated normal and impaired ears. The tympanometric and YR tests were also combined in a multivariate manner to test whether predictive efficacy improved when 226-Hz tympanometry was added to the predictor set. Conductive hearing loss was predicted based on air- bone gap thresholds at 0.5 and 2 kHz, and on a maximum air-bone gap at any octave frequency from 0.5 to 4 kHz. Each air-bone gap threshold ranged from 5 to 30 dB in 5-dB steps. Areas under the relative operating characteristic curve for DA and LR were larger than for reflectance at 2 kHz, SA and Gr. For constant hit rates of 80% and 90%, both DA and LR scores had lower false- alarm rates than tympanometric tests-LR achieved a false-alarm rate of 6% for a sensitivity of 90%. In general, LR outperformed DA as the multivariate technique of choice. In predicting an impairment at 0.5 kHz, the reflectance scores at 0.5 kHz were less accurate predictors than reflectance at 2 and 4 kHz. This supports the hypothesis that the 2-4-kHz range is a particularly sensitive indicator of middle-ear status, in agreement with the spectral composition of the output predictor from the multivariate analyses. When tympanometric and YR tests were combined, the resulting predictor performed slightly better or the same as the predictor calculated from the use of the YR test alone. The main conclusion is that these multivariate acoustic tests of the middle ear, which are analyzed using a clinical decision theory, are effective predictors of conductive hearing loss.

Original languageEnglish (US)
Pages (from-to)1749-1764
Number of pages16
JournalJournal of the Acoustical Society of America
Volume105
Issue number3
DOIs
StatePublished - Mar 1 1999

Fingerprint

decision theory
auditory defects
logistics
acoustics
predictions
regression analysis
bones
ear
electrical impedance
middle ear
reflectance
air
false alarms
Prediction
Hearing Impairment
Decision Theory
Acoustics
thresholds
octaves
impairment

ASJC Scopus subject areas

  • Arts and Humanities (miscellaneous)
  • Acoustics and Ultrasonics

Cite this

Prediction of conductive hearing loss based on acoustic earcanal response using a multivariate clinical decision theory. / Piskorski, Pawel; Keefe, Douglas H; Simmons, Jeffrey L.; Gorga, Michael P.

In: Journal of the Acoustical Society of America, Vol. 105, No. 3, 01.03.1999, p. 1749-1764.

Research output: Contribution to journalArticle

@article{43eac8084455499aaaf80c7c48309bad,
title = "Prediction of conductive hearing loss based on acoustic earcanal response using a multivariate clinical decision theory",
abstract = "This study evaluated the accuracy of acoustic response tests in predicting conductive hearing loss in 161 ears of subjects from the age of 2 to 10 yr, using as a 'gold standard' the air-bone gap to classify ears as normal or impaired. The acoustic tests included tympanometric peak- compensated static admittance magnitude (SA) and tympanometric gradient at 226 Hz, and admittance-reflectance (YR) measurements from 0.5 to 8 kHz. The performance of individual, frequency-specific, YR test variables as predictors was assessed. By applying logistic regression (LR) and discriminant analysis (DA) techniques to the multivariate YR response, two univariate functions were calculated as the linear combinations of YR variables across frequency that best separated normal and impaired ears. The tympanometric and YR tests were also combined in a multivariate manner to test whether predictive efficacy improved when 226-Hz tympanometry was added to the predictor set. Conductive hearing loss was predicted based on air- bone gap thresholds at 0.5 and 2 kHz, and on a maximum air-bone gap at any octave frequency from 0.5 to 4 kHz. Each air-bone gap threshold ranged from 5 to 30 dB in 5-dB steps. Areas under the relative operating characteristic curve for DA and LR were larger than for reflectance at 2 kHz, SA and Gr. For constant hit rates of 80{\%} and 90{\%}, both DA and LR scores had lower false- alarm rates than tympanometric tests-LR achieved a false-alarm rate of 6{\%} for a sensitivity of 90{\%}. In general, LR outperformed DA as the multivariate technique of choice. In predicting an impairment at 0.5 kHz, the reflectance scores at 0.5 kHz were less accurate predictors than reflectance at 2 and 4 kHz. This supports the hypothesis that the 2-4-kHz range is a particularly sensitive indicator of middle-ear status, in agreement with the spectral composition of the output predictor from the multivariate analyses. When tympanometric and YR tests were combined, the resulting predictor performed slightly better or the same as the predictor calculated from the use of the YR test alone. The main conclusion is that these multivariate acoustic tests of the middle ear, which are analyzed using a clinical decision theory, are effective predictors of conductive hearing loss.",
author = "Pawel Piskorski and Keefe, {Douglas H} and Simmons, {Jeffrey L.} and Gorga, {Michael P}",
year = "1999",
month = "3",
day = "1",
doi = "10.1121/1.426713",
language = "English (US)",
volume = "105",
pages = "1749--1764",
journal = "Journal of the Acoustical Society of America",
issn = "0001-4966",
publisher = "Acoustical Society of America",
number = "3",

}

TY - JOUR

T1 - Prediction of conductive hearing loss based on acoustic earcanal response using a multivariate clinical decision theory

AU - Piskorski, Pawel

AU - Keefe, Douglas H

AU - Simmons, Jeffrey L.

AU - Gorga, Michael P

PY - 1999/3/1

Y1 - 1999/3/1

N2 - This study evaluated the accuracy of acoustic response tests in predicting conductive hearing loss in 161 ears of subjects from the age of 2 to 10 yr, using as a 'gold standard' the air-bone gap to classify ears as normal or impaired. The acoustic tests included tympanometric peak- compensated static admittance magnitude (SA) and tympanometric gradient at 226 Hz, and admittance-reflectance (YR) measurements from 0.5 to 8 kHz. The performance of individual, frequency-specific, YR test variables as predictors was assessed. By applying logistic regression (LR) and discriminant analysis (DA) techniques to the multivariate YR response, two univariate functions were calculated as the linear combinations of YR variables across frequency that best separated normal and impaired ears. The tympanometric and YR tests were also combined in a multivariate manner to test whether predictive efficacy improved when 226-Hz tympanometry was added to the predictor set. Conductive hearing loss was predicted based on air- bone gap thresholds at 0.5 and 2 kHz, and on a maximum air-bone gap at any octave frequency from 0.5 to 4 kHz. Each air-bone gap threshold ranged from 5 to 30 dB in 5-dB steps. Areas under the relative operating characteristic curve for DA and LR were larger than for reflectance at 2 kHz, SA and Gr. For constant hit rates of 80% and 90%, both DA and LR scores had lower false- alarm rates than tympanometric tests-LR achieved a false-alarm rate of 6% for a sensitivity of 90%. In general, LR outperformed DA as the multivariate technique of choice. In predicting an impairment at 0.5 kHz, the reflectance scores at 0.5 kHz were less accurate predictors than reflectance at 2 and 4 kHz. This supports the hypothesis that the 2-4-kHz range is a particularly sensitive indicator of middle-ear status, in agreement with the spectral composition of the output predictor from the multivariate analyses. When tympanometric and YR tests were combined, the resulting predictor performed slightly better or the same as the predictor calculated from the use of the YR test alone. The main conclusion is that these multivariate acoustic tests of the middle ear, which are analyzed using a clinical decision theory, are effective predictors of conductive hearing loss.

AB - This study evaluated the accuracy of acoustic response tests in predicting conductive hearing loss in 161 ears of subjects from the age of 2 to 10 yr, using as a 'gold standard' the air-bone gap to classify ears as normal or impaired. The acoustic tests included tympanometric peak- compensated static admittance magnitude (SA) and tympanometric gradient at 226 Hz, and admittance-reflectance (YR) measurements from 0.5 to 8 kHz. The performance of individual, frequency-specific, YR test variables as predictors was assessed. By applying logistic regression (LR) and discriminant analysis (DA) techniques to the multivariate YR response, two univariate functions were calculated as the linear combinations of YR variables across frequency that best separated normal and impaired ears. The tympanometric and YR tests were also combined in a multivariate manner to test whether predictive efficacy improved when 226-Hz tympanometry was added to the predictor set. Conductive hearing loss was predicted based on air- bone gap thresholds at 0.5 and 2 kHz, and on a maximum air-bone gap at any octave frequency from 0.5 to 4 kHz. Each air-bone gap threshold ranged from 5 to 30 dB in 5-dB steps. Areas under the relative operating characteristic curve for DA and LR were larger than for reflectance at 2 kHz, SA and Gr. For constant hit rates of 80% and 90%, both DA and LR scores had lower false- alarm rates than tympanometric tests-LR achieved a false-alarm rate of 6% for a sensitivity of 90%. In general, LR outperformed DA as the multivariate technique of choice. In predicting an impairment at 0.5 kHz, the reflectance scores at 0.5 kHz were less accurate predictors than reflectance at 2 and 4 kHz. This supports the hypothesis that the 2-4-kHz range is a particularly sensitive indicator of middle-ear status, in agreement with the spectral composition of the output predictor from the multivariate analyses. When tympanometric and YR tests were combined, the resulting predictor performed slightly better or the same as the predictor calculated from the use of the YR test alone. The main conclusion is that these multivariate acoustic tests of the middle ear, which are analyzed using a clinical decision theory, are effective predictors of conductive hearing loss.

UR - http://www.scopus.com/inward/record.url?scp=0033105996&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=0033105996&partnerID=8YFLogxK

U2 - 10.1121/1.426713

DO - 10.1121/1.426713

M3 - Article

C2 - 10089599

AN - SCOPUS:0033105996

VL - 105

SP - 1749

EP - 1764

JO - Journal of the Acoustical Society of America

JF - Journal of the Acoustical Society of America

SN - 0001-4966

IS - 3

ER -