The neural response of the brainstem to brief sounds, known as the auditory brainstem response (ABR), is widely employed in the laboratory and the clinic to diagnose hearing loss. Current ABR approaches are limited to analyzing the ABR over hundreds of stimulus trials. Historically, trial-by-trial ABR analysis has not been possible owing to each trial's small signal-to-noise ratio. Here, we overcome this limitation and show how to score individual ABR trials. We use the distribution of individual-trial ABR scores to define thresholds, which we call auditory brainstem detection (ABD) thresholds, and to assess suprathreshold hearing. Scores based on individual-trial amplitude and similarity produce lower ABD thresholds than scores based on individual-trial amplitude alone. Our method enables us to rank the suprathreshold hearing of individual subjects. Despite having normal ABR thresholds, some subjects appear to have suprathreshold hearing deficits. Here we provide the experimental data and analysis code demonstrating our new method.

The Administrative Panel on Laboratory Animal Care at Stanford University (APLAC #14345) approved all animal procedures. All methods were performed in accordance with the relevant guidelines and regulations. Sprague Dawley rats of both sexes were obtained from Charles River Laboratories (Wilmington, MA). We measured ABRs in five rats at ages P21-P25 using Tucker Davis Technologies (TDT) System 3 RZ6. Rats were anesthetized by administering 3 % isoflurane for induction and 1.5-2 % isoflurane for maintenance, and their body temperatures were held constant at 37 degrees Celsius (FHC, DC-temperature controller and heating pad) until they fully recovered. Needle electrodes were placed in the vertex (non-inverting), below the left ear (inverting), and in the ipsilateral hind leg (ground). Subjects were placed in a sound-attenuating enclosure within a Faraday cage. Sound stimuli for eliciting ABRs were delivered using a high-frequency speaker (TDT, MF1) that was calibrated using a free-field prepolarized microphone (Model 377C01, PCB Piezotronics) prior to experiments. The rat was positioned 10 cm away from the speaker. The stimulus was a 5 ms sine wave tone pip with cosine-squared envelope rise and fall times of 0.2 ms. The repetition rate was 21 pips per second, the stimulus frequency was 16 kHz, and the intensity was increased in 5 dB SPL steps from 0 dB SPL to 90-100 dB SPL. The sampling rate was 200 kHz, the response was bandpass filtered between 500 Hz and 3 kHz, and artifact rejection was not employed. We recorded responses to 514 stimulus presentations (an equal number of alternating rarefaction and condensation stimuli) at each sound pressure level for offline analysis. Each response trial began at the onset of the stimulus, without correcting for the acoustic delay, and had a duration of 7.755 ms. We did not correct for the small acoustic delay of approximately 0.3 ms because our analysis methods do not require the correction.

Data analysis was performed in MATLAB (version R2018b, MathWorks). Error bars represent standard errors and, apart from the average scores, were estimated from studentized confidence intervals based on bootstrapping for each stimulus level and rat using 500 resamples for each observed set of scores. Doubling the number of bootstrapped samples changed the error estimates by less than 10 %. P-values, thresholds, cutoff values, false-positive rates, true-positive rates, and standard errors at points that were not directly sampled were estimated by linear interpolation between sampled points. We linearly interpolate between measurement points instead of fitting curves to the data, because fitting curves would introduce additional parameters and would make the quantities in question depend on stimulus levels very far from threshold (Shaheen et al., 2025). Defining thresholds or threshold differences using p-values does not necessitate multiple-comparisons corrections because we are requiring that many (instead of any) p-values are less than 0.05.