Distortion product otoacoustic emission test performance when both 2f1-f2 and 2f2-f1 are used to predict auditory status

The objective of this study was to determine whether distortion product otoacoustic emission (DPOAE) test performance, defined as its ability to distinguish normal-hearing ears from those with hearing loss, can be improved by examining response and noise amplitudes at 2 f(1) - f(2) and 2 f(2) - f(1) simultaneously. In addition, there was interest in knowing whether measurements at both DPs and for several primary frequency pairs can be used in a multivariate analysis to further optimize test performance. DPOAE and noise amplitudes were measured at 2 f(1) - f(2) and 2 f(2) - f(1) for 12 primary levels (L-2 from 10 to 65 dB SPL in 5-dB steps) and 9 pairs of primary frequencies (0.5 to 8 kHz in 1/2-octave steps). All data were collected in a sound-treated room from 70 subjects with normal hearing and 80 subjects with hearing loss. Subjects had normal middle-ear function at the time of the DPOAE test, based on standard tympanometric measurements. Measurement-based stopping rules were used such that the test terminated when the noise floor around the 2 f(1) - f(2) DP was less than or equal to-30 dB SPL or after 32 s of artifact-free averaging, whichever occurred first. Data were analyzed using clinical decision theory in which relative operating characteristics (ROC) curves were constructed and areas under the ROC curves were estimated. In addition, test performance was assessed by selecting the criterion value that resulted in a sensitivity of 90% and determining the specificity at that criterion value. Data were analyzed using traditional univariate comparisons, in which predictions about auditory status were based only on data obtained when f(2) = audiometric frequency. In addition, multivariate analysis techniques were used to determine whether test performance can be optimized by using many variables to predict auditory status. As expected, DPOAEs were larger for 2 f(1) - f(2) compared to 2 f(2) - f(1) in subjects with normal hearing. However, noise amplitudes were smaller for 2 f(2) - f(1), but this effect was restricted to the lowest f(2) frequencies. A comparison of signal-to-noise ratios (SNR) within nonnal-hearing cars showed that the 2 f(1) - f(2) DP was more frequently characterized by larger SNRs compared to 2 f(2) - f(1). However, there were several subjects in whom 2 f(2) - f(1) produced a larger SNR. ROC curve areas and specificities for a fixed sensitivity increased only slightly when data from both DPs were used to predict auditory status. Multivariate analyses, in which the inputs included both DPs for several primary frequency pairs surrounding each audiometric frequency, produced the highest areas and specificities. Thus, DPOAE test performance was improved slightly by examining data at two DP frequencies simultaneously. This improvement was achieved at no additional cost in terms of test time. When measurements at both DPs were combined with data obtained for several primary frequency pairs and then analyzed in a multivariate context, the best test performance was achieved. Excellent test performance (ROC) curve areas >0.95% and specificities >92% at all frequencies, including 500 Hz, were achieved for these conditions. Although the results described should be validated on an independent set of data, they suggest that the accuracy with which DPOAE measurements identify auditory status can be improved with multivariate analyses and measurements at multiple DPs. (C) 2000 Acoustical Society of America. [S0001-4966(00)04204-1].