Early Sentence Recognition in Adult Cochlear Implant Users

Normal-hearing subjects listening to acoustic simulations of cochlear implants (CI) can obtain sentence recognition scores near 100% in quiet and in 10 dB signal-to-noise ratio (SNR) noise with acute exposure. However, average sentence recognition scores for real CI listeners are generally lower, even after months of experience, and there is a high degree of heterogeneity. Our aim was to identify the relative importance and strength of factors that prevent CI listeners from achieving early, 1-mo scores as high as those for normal-hearing-listener acoustic simulations.

Sentence recognition scores (100 words/list, 65 dB SPL) using CI alone were collected for all adult unilateral CI listeners implanted in our center over a 5-yr period. Sentence recognition scores in quiet and in 10 dB SNR 8-talker babble, collected from 1 to 12 mo, were reduced to a single dependent variable, the "initial" score, via logarithmic regression. "Initial" scores equated to an improved estimate of 1-mo scores, and integrated the time to rise above zero score for poorer performing subjects. Demographic, device, and medical data were collected for 118 subjects who met standard CI candidacy criteria. Computed tomography of the electrode array allowing determination of the insertion depth as an angle, and the presence or absence of scala dislocation was available for 96 subjects. Predictive factors for initial scores were selected using stepwise multiple linear regression. The relative importance of predictive factors was estimated as partial r with a low bias method, and statistical significance tested with type II analysis of variance.

The etiologies chronic otitis and autoimmune disease were associated with lower, widely variable sentence recognition scores in the long-term. More than 60% of CI listeners scored >50/100 in quiet at 1 mo. Congenital hearing loss was associated with significantly lower initial scores in quiet (r 0.23, p < 0.001), as was longer duration of hearing loss (r 0.12, p < 0.001, -0.76 pts per year). Initial scores were negatively correlated with insertion depth (r 0.09, p < 0.001, -0.1 pts per degree), with the highest initial scores being obtained for insertion depths of 300° to 400°. A much greater proportion of scala dislocations was found for perimodiolar arrays compared with straight arrays. Scores were negatively correlated with the proportion of the active electrode array found in scala vestibuli for Nucleus perimodiolar devices (r 0.14, p < 0.01, coefficient -25). Similar overall results were obtained for sentence recognition scores in noise (+10 dB SNR). The intercept value for the obtained regression functions indicated that CI listeners with the least limiting factors generally scored ~95/100 in quiet and ~90/100 in noise. In addition, CI listeners with insertion angles as low as 315° to 360° could obtain sentence recognition scores >80/100 even at 1 day after activation. Insertion depths of 360° were estimated to produce frequency-place mismatches of about one octave upward shift.

Patient-related factors etiology and duration of deafness together explained ~40% of the variance in early sentence recognition scores, and electrode position factors ~20%. CI listeners with insertion depths of about one turn obtained the highest early sentence recognition scores in quiet and in noise, and these were comparable with those reported in the literature for normal-hearing subjects listening to 8 to 12 channel vocoder simulations. Differences between device brands were largely explained by differences in insertion depths. This indicates that physiological frequency-place mismatches of about one octave are rapidly accommodated by CI users for understanding sentences, between 1 day to 1 mo postactivation, and that channel efficiency may be significantly poorer for more deeply positioned electrode contacts.