Predicting wideband real-ear-to-coupler differences in children using wideband acoustic immittance

Abstract

Individual differences in ear-canal acoustics introduce variability into hearing aid output that can affect speech audibility. Measuring ear-canal acoustics in young children can be challenging, and relying on normative real-ear-to-coupler difference (RECD) transforms can lead to large fitting errors. Acoustic immittance measures characterize the impedance of the ear and are more easily measured than RECD. Using 226 Hz tympanometry to predict the RECD is more accurate than using age-based average RECD values. The current study sought to determine whether wideband acoustic immittance measurements could improve predictions of wideband real-ear-to-coupler difference (wRECD). 150 children ages 2-10 years with intact tympanic membranes underwent wRECD and wideband acoustic immittance measures in each ear. Three models were constructed to predict each child's measured wRECD: the age-based average wRECD, 226 Hz admittance wRECD, and wideband absorbance wRECD. The average age-based wRECD model predicted the child's measured wRECD within 3 dB in 62% of cases, but both the 226 Hz admittance and wideband absorbance wRECD were within 3 dB in 90% of cases. Using individual 226 Hz or wideband absorbance to predict wRECD improved the accuracy and precision of transforms used for pediatric hearing aid fitting.

FIG. 1.

Wideband absorbance wRECD vs measured wRECD by frequency, ear (R, right; L, left), and middle-ear status (normal vs abnormal). Perfect prediction is indicated by the diagonal dashed line. Uncertainty is represented by the width of the 89% highest-density CI around the prediction and is depicted as the size of the circles.

FIG. 2.

Violin plots of the difference between the average age-based wRECD model and measured wRECD by frequency and middle-ear status (upper panel, normal middle-ear status; lower panel, abnormal middle-ear status). Circles represent individual differences at each frequency between model predictions and measured wRECD. The violin plot provides a symmetrical representation of the distribution of values at each frequency with the vertical boundaries representing the range of differences.

FIG. 3.

Violin plots of the difference between the wideband absorbance wRECD model and measured wRECD by frequency and middle-ear status (upper panel, normal middle-ear status; lower panel, abnormal middle-ear status). Circles represent individual differences at each frequency between model predictions and measured wRECD. The violin plot provides a symmetrical representation of the distribution of values at each frequency with the vertical boundaries representing the range of differences.

FIG. 4.

Comparison of model RMSE by frequency (columns) and middle-ear status (upper panel, normal middle-ear status; lower panel, abnormal middle-ear status). The wideband absorbance, 226 Hz admittance, and average age-based wRECD models are shown within each panel. Points represent median estimates; horizonal bars indicate the 89% highest-density CI around the median estimates.

FIG. 5.

Bar plots indicating the percentage of cases for each model within 3 dB of the measured wRECD by frequency and middle-ear status (left panel, normal middle-ear status; right panel, abnormal middle-ear status). Dark gray bars represent the wideband absorbance model, light gray bars represent the 226 Hz admittance model, and white bars represent the average wRECD predicted model. The whiskers on each bar represent the 89% CI for model predictions at each frequency.

FIG. 6.

(Color online) Wideband absorbance wRECD (black line) and average age-based wRECD (red line) as a function of age in years by frequency (columns) and middle-ear status (upper row, normal middle-ear status; lower row, abnormal middle-ear status). The dashed lines in each panel represent the range of the 89% CI for the wideband immittance-predicted wRECD model. The average wRECD and wideband absorbance wRECD model had reasonable concordance of average values between 250 and 3000 Hz, but the average wRECD tended to be higher than the wideband absorbance wRECD from 4000 to 8000 Hz for both middle-ear status groups with a broader range of the 89% CI consistent with greater model uncertainty.

FIG. 7.

The conditional effects of age (left panel), equivalent ear-canal volume (middle panel), and absorbance (right panel) on wRECD from the wideband absorbance wRECD model. The solid line represents the average effect of age and equivalent ear-canal volume for all participants and the effect of absorbance for children with normal middle-ear function. The dashed line in the absorbance panel represents the conditional effect of absorbance on RECD for children with abnormal middle-ear function. The shaded area around each line represents the 89% CI around the average from the model.