2.5. Data analysis

DPOAE, ABR, and operant conditioning data sets were all analyzed separately. A comparison was then conducted between thresholds derived from ABRs and operant conditioning.

The ABR data set was sorted by stimulus type (clicks, 8, 16, 24, and 42 kHz tones) and blast-exposure groups (blast exposed vs. sham-blasted). Absolute thresholds were examined to establish changes in ABR thresholds over time. Threshold shifts were then extracted from the absolute threshold data to show effects of the blasts, to compare effects of the blasts across methodologies (ABRs, DPOAEs, and behavioral responses), and across studies (Newman et al., 2015; Smith et al., 2020). Threshold shifts were calculated for every subject for each frequency per observation using the following formula: dB shift = baseline dB – post-blast or post-sham-blast dB. Threshold shifts were used as the main dependent variable in the following analyses and were calculated relative to baseline such that a positive value reflects better sensitivity than baseline and a negative value reflects deficit from baseline.

Given the longitudinal design of this experiment and the unbalanced group composition, we utilized a linear mixed-effects model to examine whether blast exposure and day post-exposure predicted shifts in ABR thresholds across stimuli (LMM, lmer in the lme4 R package) (Bates et al., 2014; R Core Team, 2017). In this model, we examined whether shifts in thresholds (dB) could be predicted by fixed factors of day post-exposure (baseline, 3, 30, and 90), stimulus (click, 8, 16, 24, and 42 kHz tones), blast-exposure group (blast exposed and sham-blast exposed), sex (male and female) and by interactions between day post-exposure, stimulus, and group. The model was compared to the intercept only model (i.e., a model without predictors) for significance. To control for dependencies within our data from sampling each mouse repeatedly, we included a random intercept for mouse identity, sex, and the ear chosen for testing across day post exposure. Post hoc comparisons using Tukey’s method were performed to assess the relationship between day post exposure, stimulus type, and blast-exposure group (emmeans R package) with p values adjusted to the number of family-based comparisons to reduce type 1 error.

Similar to the ABR data set, threshold shifts were compared in these analyses. We utilized a linear mixed-effects model to examine whether blast-exposure group and day post-exposure predicted shifts in DPOAE thresholds across stimuli. The model was constructed using the same predictors as the ABR threshold shift data set. Post hoc comparisons using Tukey’s method were performed to assess the relationship between day post exposure, stimulus type, and blast-exposure group with p values adjusted to the number of family-based comparisons to reduce type 1 error.

Given the longitudinal design of this study, as well as the repeated sampling for each mouse, a linear mixed effects model was used to analyze the behavioral results. Threshold shifts (dB) were used as the dependent variable to examine whether the blast exposure, days after the blast, sex (male and female), and stimulus (8, 16, 24, and 42 kHz) predicted threshold shifts. In order to rule out the possibility that an uncharacteristic data point on any given day influenced our overall blast effect, a moving window average of the threshold shift across days was calculated. To compute this moving window average, the following formula was used: threshold shift today (dB) = average (today + (today +1) + (today +2) + (today + 3) + (today + 4) + (today + 5) + (today +6)). This smoothing technique has been used previously to reduce spurious day to day variability and show a more accurate reflection of the overall hearing sensitivity in animal subjects (Heinz et al., 2005). To limit the number of critical comparisons, to avoid an inflation of type I error, and to draw parallels between physiology and behavior, key days of baseline (henceforth −3), 3, 30, 60, and 90 days after the blasts were used in this model. The model was compared to the intercept only model for significance with mouse identity as a random intercept. Post hoc comparisons using Tukey’s method were performed to assess the relationships between days after blast exposure, stimulus, and sex with the p values adjusted to the number of family-based comparisons to reduce type I error.

A linear mixed effects model analysis on the false alarm rate before and after blast exposure was also constructed to see if incorrectly responding when no stimulus was present systematically varied with exposure. Mice had 3-7 data points of pre-blast baseline testing sessions and up to 90 sessions after blasts, leading to an extremely unbalanced design. As such, a linear model evaluating false alarm percentage across timepoints (before and after blast), and a post hoc comparison using Tukey’s method comparing these time points was conducted.

To examine the differences between the effects of a blast on behavioral and physiological assessments of hearing, a model was constructed using raw threshold and threshold shift data sets separately. The comparison of raw thresholds was utilized to determine if behavioral and physiological hearing assessments yield different threshold values. In this model, data collection technique (ABR and operant conditioning), stimulus (8, 16, 24, and 42 kHz tones), and days (−3, 3, 30, 60, and 90 days after the blast) were used as predictor variables for changes in thresholds (dB) after the blasts. Additionally, a separate model was constructed to examine whether threshold shifts (dB) could be predicted using sex (male and female), stimulus (8, 16, 24, and 42 kHz tones), days (−3, 3, 30, 60, and 90 days after the blast), and data collection technique (ABR and operant conditioning). Again, these models were compared to intercept models and Tukey tests were used for post hoc comparisons.