ABSTRACT
OBJECTIVE: Receiver operating characteristic (ROC) methodology was used to compare the performance of different statistics in the automatic detection of multiple auditory steady-state responses (MSSR) and click auditory brainstem responses (cABR). STUDY SAMPLE: Thirty-five healthy newborns tested within the first two weeks of birth. DESIGN: In each case cABR and MSSR (0.5 and 2 kHz) were recorded with and without acoustic stimulation, using AUDIX equipment. With this test sample of recordings, ROC curves were evaluated separately for each statistic evaluated: (1) the standard deviation ratio (SDR) and the correlation coefficient ratio (CCR) for the cABR; (2) the Hotelling T2 (HT2) and circular T2 (CT2) for the MSSR. RESULTS: All objective detection methods performed well (areas under ROC (AUC) > 0.9). The MSSR statistics showed significantly larger AUCs at both frequencies (HT2: 0.98 and 1; CT2: 0.96 and 0.99) than the cABR measures (SDR: 0.91 and CCR: 0.92). The HT2 hits rate was the highest (97-100% at 0.5 and 2 kHz) for fixed false alarms rates of both 10 and 20%. CONCLUSIONS: This superiority of performance of T2-like statistics, reflecting inherent advantages of MSSR analysis for automation, warrants serious consideration for further development of newborn screening technology.
