Optimizing DPOAEs for Serial Monitoring: From Nitty-gritty Solutions to Source Separation Dawn Konrad-Martin, PhD, CCC-A VA National Center for Rehabilitative Auditory Research (NCRAR) Portland, Oregon VA Medical Center One of 13 National VA RR&D Research Centers of Excellence Acknowledgements NCRAR Collaborators Garnett McMillan Marilyn Dille Kelly Reavis Wendy Helt Peter Jacobs Other Collaborators Lori Dreisbach at UCSD Lynne Marshall at Naval Submarine Medical Research Laboratory Research supported by grants from VA Rehabilitation Research & Disorders Service Cody Gohean LEARNING OBJECTIVES List benefits/limitations of various DPOAE recording paradigms Estimate your own test-retest variability Describe how to optimize test-retest reliability Discuss how changes in DPOAE fine structure might impact serial monitoring results PART 1 OUTLINE Define Screening Goals Why Screen Using OAEs? DPOAE Test Types Criteria for Valid DPOAEs Criteria for Significant DPOAE Changes 1
DEFINE GOALS DEFINE GOALS Screening and diagnosis have DIFFERENT goals Goal of screening is to determine which patients need follow-up diagnostic testing Screening for hearing changes (option A) and subclinical damage (option B) involve different analytic strategies PROTOCOL OPTION A PROTOCOL OPTION A Option A requires understanding the operating characteristics of the screening test i.e., Sensitivity and Specificity based on a comparison of DPOAE shifts to audiogram shifts in a sample of exposed subjects A shift of X db results correctly identifies hearing shift in X % of exposed ears PROTOCOL OPTION B PROTOCOL OPTION B Option B requires comparison to normative reference limits for screening test shifts i.e., norms show no more than X db shift in DPOAE level in 9% of unexposed subjects, so shifts larger than this are a screen fail DPOAE used as proxy for OHC function (not used as an estimate of hearing shift) Photos courtesy of Dr. Marc Lenoir, from Promenade around the cochlea EDU website: http://www.cochlea.org, by Remy Pujol et al., INSERM and University of Montpellier 2
DEFINE GOALS Audiogram is the current gold standard measure of ototoxicity in ototoxicity monitoring programs Audiogram is the current legal standard for noise damage in hearing conservation programs Follow-up for both screening options will involve audiometry NCRAR Ototoxicity Screening Protocol Pass Screening Audiogram Fail Responsive Screening Pass Audiogram Full <= 8 khz Audiogram (Fig. 4) DPOAE Fail Audiogram and/or DPOAE Assess Pre-Exposure Risk Exposure Pass DPIO Model Audiogram Shift Fail Option B Non- Responsive Screening Pass DPOAE Model of Audiogram <= 8 khz Shift The goal appropriate for MOST screening applications is Option B. Option A is generally reserved for patients who cannot take a behavioral screening test. DPOAE Fail WHY SCREEN WITH DPOAES? WHY SCREEN WITH DPOAES? Objective measure that tests outer hair cell (OHC) system function OHC system must be normal for hearing thresholds to be normal Initial damage from ototoxins and noise impacts OHCs DPOAE changes may signal hearing changes, or more subtle damage effects WHY SCREEN WITH DPOAES? DPOAEs were affected in more ears than conventional audiometry (CA) Noise: Lapsley Miller & Marshall, 27; Helleman & Dreschler, 212 Aminoglycosides: Katbamna et al., 1999; Stavroulaki et al., 22; Mulheran & Degg, 1997; Cisplatin: Stavroulaki et al., 21 DPOAEs were affected in more ears than CA, but in a similar number of ears compared with HFA Cisplatin: Ress et al., 1999 DPOAEs were affected in fewer ears and later than HFA Cisplatin, Carboplatin or both: Knight et al., 27 WHY SCREEN WITH DPOAES? DPOAE changes can occur earlier in DPs vs hearing threshold DPOAEs can be more reliable in certain populations Especially among the very sick or the very young Testing can be less time consuming Depending on DPOAE test type and number of frequencies tested 3
DPOAE TEST TYPES DPOAE TEST TYPES (1) Frequency sweep (DP-gram) Typical clinical measurement Correlated with hearing threshold Coarse- or Fine-frequency step sizes (2) Level sweep (I/O function) Slope estimates nonlinear BM growth (3) Fine frequency-ratio sweep (group delay or travel time) Delay is associated with tuning 1. DPOAE Frequency Sweep (DP-Gram) Lower levels give larger shifts f 2 Varied L 1 & L 2 Fixed f 2 /f 1 Fixed - Plots frequency on the x-axis & DPOAE level on y-axis - Course- or Fine-Frequency Step Size DPOAE level DATA FROM LORI DREISBACH Figure 1 from Marshall, Lapsley Miller, & Heller (21) Noise & Health, 3(12), 43-6 Also see review in Gorga et al. (27) 2. DPOAE Level Sweep (I/O function) Input-Output (I/O) Function f 2 Fixed L 1, L 2, or Both L 1 & L 2 Varied f 2 /f 1 Fixed -Plot L 1 (or L 2 ) on x-axis & DPOAE level on y-axis - Can determine threshold and slope of function Linear System Input Spectrum Amplitude Frequency Output Input Output Spectrum Amplitude Frequency threshold Nonlinear System Amplitude Frequency Output Input Amplitude Frequency Distortion products SLIDE FROM KIM SCHAIRER 4
3. DPOAE Frequency-Ratio Sweep (Group delay/travel Time) f 2 Fixed L 1 & L 2 Fixed f 2 /f 1 Varied (e.g. 1.1 to 1.3) Each measurement made has an accompanying amplitude and phase DPOAE freq on x-axis; DPOAE phase (time) on y-axis CRITERIA FOR SIGNIFICANT DPOAE CHANGE: ROC Curves (Protocol Option A) Reference Limits (Protocol Option B) Phase can be translated into time using the formula: Time (ms) = Δφ / 2π * Δf 2 Reproducibility (repeatability) of a measurement over time is paramount for serial monitoring Important to know and control for expected variance in performance to determine a real change 8 db PROTOCOL OPTION A And how well did you say this predicts the audiogram??? SCREENING TEST: Symbols CHANGED NOT CHANGED Some Definitions GOLD STANDARD: Hearing Shift CHANGED Hits (Sensitivity) False Negatives (Misses) NOT CHANGED False positives True Negatives (Specificity) % % 5
Receiver Operating Characteristic (ROC) curve FAKE DATA FOR ILLUSTRATIVE PURPOSES ONLY True Hearing Status Shift in Crashing Hearing No Hearing Symbols Threshold Change Change > 1 db 15 35 (hearing change) (TP = 1.) (FP = 1.) < 1 db (no hearing change) (FN =.) (TN =.) Total For any clinical test, there is always a trade-off between being sensitive and specific. Total 15 35 > 3 db (hearing change) 13 (TP =.87) 2 (FP =.57 ) < 3 db (no hearing change) 2 (FN =.13 ) 15 (TN =.43) Total 15 35 > 6 db (hearing change) (TP =.67) 3 (FP =.9) < 6 db (no hearing change) 5 (FN =.33) 32 (TN =.91) Total 15 35 > 12 db (hearing change) < 12 db (no hearing change) 5 1 (TP =.33) (FP =.3) (FN =.67) (TN =.97) Total 15 35 Reavis et al, Ear & Hear 31, 211 36 ears of 24 patients treated with cisplatin Mean age 58.5 years 3.4 monitoring visits on average Received ~ 4 mg of cisplatin over 42 days Half the subjects experienced ASHAsignificant hearing changes Reavis et al, Ear & Hear 31, 211 Measured hearing and DPOAEs before and during chemotherapy Used DPOAEs in a model to identify ( predict ) HL at SAME monitor visit Outcome variable was HL (yes,no) in the screening audiogram, called the SRO Predictors were DPOAE metrics alone or together w ototoxicity risk factors Reavis et al, Ear & Hear 31, 211 Reavis et al, Ear & Hear 31, 211 DP-gram search for highest DP freq f2= 1- khz; f2/f1=1.22; L1, L2=65, 59 DP I/O s at four highest frequencies tested in 1/3 rd octave steps f2=35-6 db SPL, L1 optimized using covaried paradigm (Kummer et al., 1998) Summary measures from I/O functions used as predictors 6
Better False Positive Rate When Risk Factors Are Included (blue) Pre-treatment Susceptibility Assessment (green) DPOAE-based Hearing Shift Assessment (black) Susceptibility + DPOAE-based Hearing Shift Assessment (red) Perfect accuracy compared with gold standard hearing test HOW WOULD DP-BASED PREDICTION MODEL WORK? At each monitoring visit, make DPOAE measurements Into the formula provided, enter DPOAE test-retest difference Enter relevant baseline patient factor information & exposure information Formula spits out likelihood that hearing has changed Formula is currently being validated PROTOCOL OPTION B PROTOCOL OPTION B Normative Reference Limits Use established norms for DPOAE changes (e.g., in level shifts) to determine if OHC function has changed or is stable Develop your own clinic norms to ensure your results are comparable to published data PROTOCOL OPTION B Standard error of measurement difference (SEM) Typically 2 X SEM = ~5 db for f 2 from 1-4 khz (Franklin et al. 1992; Beattie et al., 23) Average amplitude difference plus 2 SD 6 db for most frequencies from 1-6 khz (Roede et al., 1993) Cumulative distributions > 95% of ears had test-retest change of < 6 db for f 2 from 1 - khz (NCRAR; Reavis et al., 28) A Meta-Analysis of Reference Limits McMillan et al, in preparation 7
PROTOCOL OPTION B Reliability decreases with increasing time between measurements Weak emissions (i.e. near DPOAE threshold) less reliable than robust emissions Emissions at very low frequencies less reliable if averaging insufficient to achieve high SNR Emissions above 4 khz slightly less reliable than those from 2-4 khz Gorga et al., 212; Helleman & Dreschler, 212; our data in preparation HOW WOULD DP-BASED REFERNCE LIMITS WORK? DPOAE decrements or enhancements of >5-6 db indicates a screen failure DPOAE level differences between groups of noise-exposed and non-noise-exposed ears with normal hearing are often less than this, so Need to improve retest reliability! PART 2 OUTLINE BREAK TIME!!! Case Studies Improving Retest Reliability Determining Retest Reliability in YOUR Clinic DPOAE Source Generation OAE Source Generation Which One is Least Like The Others? DPOAE SFOAE Acoustic response measured in the ear canal Evoked using two-tone stimulation (f1 < f2) At least 2 sources (nonlinear distortion & coherent linear reflection) TEOAE SOAE 8
Noise floor Usually average amplitude in several frequency bins above and below 2f 1 -f 2 bin Noise is worse at low frequencies than high Signal-to-noise ratio (SNR) db difference between SPL at 2f 1 -f 2 and the estimated noise System distortion Some Definitions Distortion from the equipment that occurs at the 2f1-f2 frequency Greatest at high frequencies and high primary levels Determined the highest frequencies that provided a DPOAE level (Ldp) that was 6 db above the noise (Ndp). 2 2 3 4 DP Gram Ldp Ndp Monitored this octave range in fine resolution F2/F1 step sizes (1/48 th ) at baseline and at each treatment. Pure-tone Thresholds Subject 19 (LE) Fine Resolution Step Size 2 2 Ldp Ndp 3 4 cumulative dose (mg) 3223 34 3621 3832 455 431 4559 4828 DPOAE Levels Subject 14 (RE) Subject 38 (LE) Pure-tone Thresholds Pure-tone Thresholds dose (mg) dose (mg) DPOAE Levels DPOAE Levels 9
What region of the cochlea do ototoxic insults damage initially? frequency DPOAEs change first What region of the cochlea does industrial noise damage initially? khz DPOAEs change first IMPROVING RETEST RELIABILITY OAEs can change in regions with & without associated hearing changes Decrements & enhancements indicate damage Inside Your Probe Speaker 2 Plane of mic mic port Speaker 1 Drawing by S. Blatrix from "promenade around the cochlea" EDU website www.cochlea.org by Rémy Pujol et al., INSERM and University Montpellier 1 IMPROVING RETEST RELIABILITY Instrumentation-related sources of variability Noise & system distortion -Subject noise - Ambient noise - Electrical noise (wires are antennae) Probe fit - Affect both noise floor and system distortion In-the-Ear Calibration Measurement parameters - Level and frequency Tester(s) - Using these tips, practice makes perfect Instrumentation Reducing noise Keep test ear away from noise sources in test environment Maintain good probe tip seal Increase averaging time Average until noise floor is the level of your system distortion or artifact-free averaging time reaches 32 or even 64 seconds Ensure pre-amp batteries have good charge Microphone cable acts as an antenna and will transmit noise when rubbed or moved keep it away from other cables/equipment Instrumentation Improving fit of probe & probe tip Firm vs loose placement Ports facing tympanic membrane vs ports blocked by canal wall Wax issues Let foam tips expand fully before test Sound delivery tubes straight Anchor the probe so it is not pulling out Cable from microphone immobile, placed where it won t wiggle or vibrate
What is the problem with traditional calibration techniques for measuring OAEs? Errors in stimulus levels due to standing waves in the ear canal What will the result be? Exact primary levels generating the emission are unknown in certain frequency regions Practice to Improve Tester Reliability What can I do to address problem? For serial monitoring - Ensure probe placement is consistent across tests!!! 1. Establish a conservative estimate of your system distortion a) Measure with your own OAE equipment and paradigm in a coupler and, if you can, in ears with severe-profound HL b) Use this to revise criteria for a valid response (see rules for this) 2. Determine retest reliability in YOUR clinic and use it to establish criteria for a significant emission shift (SES) Determine reliability (SEM) for your group, tester, stimulus paradigm, equipment, setting. Compare to other people's. 3. Plot your test-retest reliability data look for trends (e.g., are certain frequencies/levels giving you outliers?) 4. Validity/Sanity check using +SES and SES Control group Exposed group Slide adapted from Lynne Marshall 1a) Establish conservative estimate of your system distortion Make DPOAE measurements in a coupler using test protocol Use standard coupler (e.g. 2cc coupler meeting IEC 711 specifications, such as the B&K ear simulator Type 4157) Make of these measurements; Mean plus 2 SDs to use as a conservative estimate of system distortion Perform frequent system distortion measurements to assess system performance over time 1b) Revise Criteria For Valid DPOAEs Favorable SNR (e.g., 6 db, or db in noisy environment) OAE amplitude is larger compared to conservative estimate of YOUR system distortion Middle ear function is stable DPOAE must be valid to be reliable 2. DETERMINING RELIABILITY IN YOUR CLINIC 11
2. DETERMINING RELIABILITY IN YOUR CLINIC Determine Clinician Retest Error Rate: 1. Measure DPOAE in 2 subjects at T= days for each frequency/level combination in your your test paradigm 2. Repeat (re-test) the test in the same 2 subjects at T=14 days 3. Calculate the average retest difference = Retest Amplitude - Test Amplitude = T amp T14 amp < 6 db 2. DETERMINING RELIABILITY IN YOUR CLINIC For a given stimulus condition, if no more than 1 / 2 subjects has a retest error > 6 db, then we are reasonably certain that for monitoring purposes, the retest error rate will be controlled below 5% If more than 1 error out of 2 subjects, then refer to tips and tricks for making reliable measures. Repeat with a different 2 subjects 3. Plot your retest differences to check for patterns & outliers 25 2 15 5 5 15 2 25 Delta for Condition A 5 15 2 25 2 15 5 5 15 2 25 Delta for Condition B 5 15 2 For each frequency/condition, calculate post-pre delta and plot by subject (arbitrary numbers used in example) Look for conditions with increased variability (B) And for conditions with extreme outliers (C) Examine data for measurement problems (high noise levels, bad stimulus levels, bad calibrations, transcription errors) Slide from Lynne Marshall 25 2 15 5 5 15 2 25 Delta for Condition C 5 15 2 Validity/Sanity Check Using Significant Emission Shifts Plot SES data for control group +SES vs SES should be few and randomly scattered across frequency Look for outliers & patterns Find sources of problems Were these associated with bad measurements? Plot SES data for noise-exposed group Indications of noise-induced change from Helleman & Dreschler, 212 SES rate higher than for control group More +SES than SES More SES- in 3-6 khz range than other frequencies THANK YOU FOR LISTENING Dawn.martin@va.gov 12