Comparisons of Polarity Effect, Electrode Position, and Intracochlear Resistance
Figure 2 shows the 3D cochlear reconstructions of all subjects, arranged in order of electrode array type. Two subjects had a 1J-Helix array (S22 and S53), five subjects had a 1J array (S29, S40, S46, S50, and S52), and four subjects had a mid-scala array (S43, S47, S49, and S54). Each of the three types of electrode arrays is designed to achieve a different position within the cochlea (Dhanasingh and Jolly 2017). The standard 1J array has a lateral wall design. The 1J-Helix and the mid-scala arrays are both pre-curved. The 1J-Helix should achieve a more medial position relative to the 1J, whereas the mid-scala is designed for placement in the middle of the ST. Of the 176 total electrodes in the sample, 104 electrodes (59.1 %) were located in ST, 62 electrodes (35.2 %) were located in the intermediate position, and 8 electrodes (5.5 %) were located in SV. Two of subject S50’s electrodes were extracochlear (electrodes 15 and 16). Across electrodes, electrode-to-modiolus distance ranged from 0.18 to 2.2 mm (M = 1.14 mm, SD = 0.52; see Table 2).
Table 2 Means and standard deviations for sQP threshold, polarity effect, electrode-to-modiolus distance (EMD), and intracochlear resistance. Data are averaged across electrodes 2–15 Figure 3 shows individual site-specific thresholds in response to the ACA and CAC stimuli. Black squares indicate responses to the ACA stimulus, whereas red circles indicate responses to the CAC stimulus. Higher (i.e., worse) thresholds in response to the ACA stimulus compared to the CAC stimulus indicate a positive polarity effect. Conversely, higher (i.e., worse) thresholds in response to the CAC stimulus compared to the ACA stimulus indicate a negative polarity effect. The polarity effect could not be measured on electrodes 15 and 16 for subject S50, because those electrodes were extracochlear and stimulation did not result in auditory percepts.
Across electrodes, the polarity effect ranged from − 4.83 to 3.54 dB (M = 0.36 dB, SD = 1.42; see Table 2). The polarity effect varied both across subjects and across the electrode array within the same subject. For example, subject S40 shows consistently large, positive polarity effects at most electrode sites (Fig. 3). Conversely, subject S22 tends to have negative polarity effects in the apical portion of the electrode array, and positive polarity effects in the basal portion of the electrode array (Fig. 3).
Intracochlear resistance also varied across- and within-subjects, ranging from 91.71 to 2718.57 Ohms (M = 496.95 Ohms, SD = 518.17; see Table 2). Prior to data analysis, histograms were plotted to determine whether the data were normally distributed. This analysis revealed that Rlong values were highly skewed. To reduce skew, the Rlong data were log-transformed prior to data analysis and for visualization purposes. Of note, the size of the electrode contacts differs slightly between the three electrode arrays, which may influence electrode impedances (Hughes 2012). However, Rlong did not differ as a function of electrode array type in this study (β = 2.36, F(2,8) = 1.45, P > 0.05).
The first analysis assessed the relationship between the polarity effect and (1) electrode position and (2) intracochlear resistance. Figure 4 shows electrode-specific polarity effect plotted against (a) electrode-to-modiolus distance, and (c) intracochlear resistance. Individual subjects are distinguished by color and shape. The right-hand panels of Fig. 4 (b and d) show the best-fit lines for each individual. Results from a linear mixed-effects analysis showed that, across subjects, the polarity effect was not significantly predicted by electrode-to-modiolus distance (β = 0.11, F(1,103.23) = 0.15, P = 0.70), electrode scalar location (β = − 0.15 for intermediate, 0.45 for SV, F(2, 145.48) = 0.50, P = 0.61), or intracochlear resistance (β = − 0.39, F(1, 106.62) = 0.86, P = 0.36). As a complement to the multilevel model, repeated measures correlations (rrm) indicated very small effect sizes for each comparison: (1) polarity effect and electrode-to-modiolus distance (rrm(162) = − 0.02, 95 % CI [− 0.17, 0.14], P = 0.80) and (2) polarity effect and intracochlear resistance (rrm(151) = − 0.13, 95 % CI [− 0.28, 0.03], P = 0.11). These results suggest that the polarity effect varies independently of both CT-estimated electrode position and EFI-estimated longitudinal intracochlear resistance.
Predicting Focused Behavioral Thresholds
The second analysis assessed the relationship between the polarity effect and sQP thresholds and whether the polarity effect explains a significant portion of the variation in sQP thresholds after accounting for CT-estimated electrode position and EFI-estimated intracochlear resistance. Recall that sQP thresholds are believed to reflect the cumulative contributions of local neural health, electrode position relative to the auditory nerve, and intracochlear resistance. Figure 5 shows the relationships between electrode-specific sQP thresholds and (a) the polarity effect, (c) electrode-to-modiolus distance, and (e) intracochlear resistance. Again, individual data are distinguished by color and shape. The right-hand panels of Fig. 5 (b, d, and f) show the best-fit lines for each individual. Repeated measures correlations indicated significant relationships between focused thresholds and each of the three main predictors: (1) electrode-to-modiolus distance: rrm(140) = 0.51, 95 % CI (0.38, 0.63), P < 0.001; (2) intracochlear resistance: rrm(141) = − 0.37, 95 % CI (− 0.51, − 0.22), P < 0.001; and (3) the polarity effect: rrm(140) = 0.27, 95 % CI (0.11, 0.42), P = 0.001. Each relationship survived Bonferroni adjustment for multiple comparisons (adjusted α = 0.017).
A linear mixed-effects analysis was performed to assess whether the polarity effect remained a significant predictor of sQP thresholds after controlling for electrode position and intracochlear resistance. To determine whether each variable improved the overall model fit, an Akaike information criterion with a bias correction for small sample sizes (AICc) was used for model selection (Hurvich and Tsai 1989). Statistical models with lower AICc values are more parsimonious relative to those with higher AICc. First, we specified an empty model with only subjects as random effects to predict sQP thresholds (AICc = 877.62). Next, we consecutively added each of the following fixed effects: electrode-to-modiolus distance (AICc = 824.33), scalar location (AICc = 808.72), intracochlear resistance (AICc = 803.49), and the polarity effect (AICc = 792.99). The model fit improved with the addition of each predictor, and the lowest AICc value was associated with the model that included all four predictors. So, the final model specified electrode-to-modiolus distance, scalar location, intracochlear resistance, and the polarity effect as independent variables. The dependent variable was sQP threshold.
Traditional R2 values are not valid for linear mixed-effects models. However, Nakagawa and Schielzeth (2013) propose two types of pseudo-R2 values that can provide an indication of the variability explained by a multilevel model: (1) marginal R2 (R2marginal), which represents the proportion of the total variance explained by the fixed effects, and (2) conditional R2 (R2conditional), which represents the proportion of the variance explained by both the fixed and random effects. The difference between the R2marginal and R2conditional reflects the variability in the random effects; here, this represents across-subject variability.
Results from the linear mixed-effects analysis (R2marginal = 0.27, R2conditional = 0.80) showed that the polarity effect (β = 0.73, F(1, 138.54) = 14.83, P < 0.001), electrode-to-modiolus distance (β = 5.46, F(1,144.60) = 57.36, P < 0.001), scalar location (β = − 1.18 for intermediate, 3.52 for SV, F(2, 141.61) = 4.97, P = 0.01), and intracochlear resistance (β = − 2.42, F(1, 144.99) = 4.27, P = 0.04) were all significantly predictive of sQP thresholds. After adjustment for multiple comparisons (Tukey), results showed that thresholds were higher for electrodes located in SV relative to those located in the intermediate position (t(141.69) = 2.91, P = 0.01). There were no significant differences in sQP threshold between electrodes located in SV compared to ST (P = 0.08) or in ST compared to the intermediate position (P = 0.21).
Note that the group regression line in Fig. 5(e–f), which depicts the relationship between sQP thresholds and Rlong, appears to have a modest positive trajectory; however, the β coefficient for the Rlong predictor is − 2.42 and the repeated measures correlation coefficient (rrm) is − 0.37. This suggests that higher sQP thresholds are associated with lower Rlong values, demonstrating the importance of statistically accounting for clustered data.
Overall, the results indicate that, across subjects, relatively high sQP thresholds are associated with large, positive polarity effects, distant electrode position relative to the modiolus, and low intracochlear resistance. Electrodes with high sQP thresholds are also more likely to be translocated to the SV than electrodes with low sQP thresholds. Moreover, the polarity effect significantly predicts sQP thresholds, even after controlling for electrode-to-modiolus distance, electrode scalar location, and intracochlear resistance.
Despite these results, it is evident from Figs. 3, 4 and 5 that the within-subject relationships between variables are not the same for every participant. Table 3 shows within-subject correlations between sQP thresholds and electrode-to-modiolus distance (threshold-EMD), the polarity effect (threshold-PE), and Rlong (threshold-Rlong) for each individual subject. It was noted that five out of 11 subjects (S47, S43, S53, S49, and S50) had strong, positive correlations between focused thresholds and electrode-to-modiolus distance, with correlation coefficients (r) ranging from 0.71 to 0.94 (Cohen 1988). Each of those five threshold-EMD correlations was statistically significant (Ps < 0.05). This suggests that the variation in sQP thresholds for those five subjects is largely explained by variation in electrode position. The remaining six subjects had weak-to-moderate threshold-EMD correlations (r = 0.06 to 0.48) that were not statistically significant (Ps > 0.05). For that group, electrode position does not account for much of the variability in sQP thresholds.
Table 3 Individual correlation coefficients (r) and P values for correlations between sQP thresholds (dB rel. 1 μA) and each of the following variables: electrode-to-modiolus distance (EMD; in mm), polarity effect (PE; in dB), and intracochlear resistance (Rlong; in Ohms); data are arranged in descending order by strength of threshold-EMD correlation. Significant correlations (P < 0.05) are italicized Subjects were separated into two groups based on their threshold-EMD correlations for further analysis. Subjects with threshold-EMD correlation coefficients larger than 0.70 were in the “strong threshold-EMD” group (N = 5), whereas those below the 0.70 cutoff were in the “weak-to-moderate threshold-EMD” group (N = 6). Figure 6 shows the relationship between the polarity effect and sQP thresholds for (a) the five subjects with strong and statistically significant threshold-EMD correlations, and (b) the six subjects with weak-to-moderate, non-significant threshold-EMD correlations. Repeated measures correlations were performed to determine the strength of the relationship between sQP thresholds and the polarity effect within each group. Overall, the group with strong threshold-EMD correlations did not demonstrate a significant relationship between sQP thresholds and the polarity effect (rrm(63) = 0.03, 95 % CI [− 0.22, 0.28], P = 0.80). However, the relationship between the polarity effect and sQP thresholds was significant for the group with weak-to-moderate threshold-EMD correlations (rrm(76) = 0.48, 95 % CI [0.28, 0.63], P < 0.001). These results emphasize that sQP thresholds can reflect either electrode position or neural integrity (as estimated by the polarity effect). Relationships between sQP thresholds and neural integrity may be more likely to emerge when the variation in sQP thresholds cannot be explained by variation in electrode position.
Polarity Sensitivity and Duration of Deafness
Finally, we performed a preliminary analysis to quantify the relationship between the polarity effect and duration of deafness, a common implicit assessment of neural integrity. Figure 7 shows the relationship between duration of deafness and the across-site average polarity effect. For this analysis, the polarity effect was averaged across all available electrodes for each subject. Duration of deafness was defined as the time, in years, between diagnosis of severe-to-profound sensorineural hearing loss and CI activation. A linear regression analysis indicated that duration of deafness was significantly correlated with the across-site average polarity effect (R2 = 0.38, R2adjusted = 0.32, F(1,9) = 5.62, P = 0.04). Specifically, individuals that experienced relatively long durations of deafness prior to receiving a CI also tended to have relatively large, positive polarity effects.