Appendix D: Pass/Refer Criteria
Pass/Refer Criteria for DPOAE
The decision that a DPOAE exists is based on detecting a signal whose level is significantly above the
background noise level. This requires a statistical decision, since the random noise level in the DPOAE
filter channel can be expected to exceed the average of the random noise levels in the four adjacent
filter channels — used as the reference for comparison — roughly half the time.
Extended measurements of the noise distributions in both the DPOAE filter channel “DP level” and
the rms average of the 4 adjacent channels “N level” indicate that the signal-to-noise ratio (the
difference between DP and N) has a standard deviation of 5.5 dB. As shown in the figure below, this
implies a 10% probability of seeing a 7 dB SNR simply from the variability of the noise levels in the
2 filter sets.
Requiring an SNR of 6 dB in three out of four frequencies drops the probability of passing an ear with
significant hearing loss to 1% or less. Note: By the binomial distribution, two of three frequencies at
>8.4 dB or three of six frequencies at >7 dB should also ensure less than 1% probability of passing a
moderately-severe hearing-impaired infant.
Preliminary trials with infants indicate that the tester’s technique is the single most important variable
in the pass rate on normal-hearing infants. Some testers pick up the technique (see Operating
Instructions section, page 19) with only a couple of days’ practice, producing pass rates comparable
to those for other DPOAE equipment they have used for months; other testers take longer.
Occasional claims of extraordinarily low probabilities of missing an ear with hearing loss appear to be
based on poor statistics. As discussed by Gorga (Mayo Clinic Teleconference, 1998), since the
incidence of significant hearing loss is roughly 2 per 1,000, verifying a 99.7% accuracy would require
testing hundreds of thousands of babies with a given system. Thus to demonstrate that only 3 babies
out of 1,000 with hearing loss were missed would require follow-up testing on 500,000 babies. To our
knowledge,
no
one
has
performed
such
tests
to
date.
