Using audiometric thresholds and word recognition in a treatment study.

OBJECTIVES: First, to examine a possible limit on significant results imposed by a progressive floor effect for hearing threshold improvement in a treatment study. This floor effect for hearing recovery suggests that if inclusion criteria are not set sufficiently high, the superiority of a treatment group may not be detectable. Second, to examine the outcomes when using two different types of criteria for significant change in a subject's word recognition score.
METHODS: Several single-number criteria (e.g., 15 percentage points) are compared with the 95% (p=0.05) criteria from the binomial critical difference table for monosyllables. Critical differences for binomial variables change depending on whether the starting value lies in the middle (near 50% correct) or at either extreme of the range of scores (0 or 100%). Different judgments of significant word recognition improvement (or decrease) using binomial versus single-value criteria are presented. DATA SOURCE: A recent treatment study of sudden sensorineural hearing loss (n=318) is used to illustrate these effects.
CONCLUSION: First, there is a progressive floor effect of presenting severity that co-varies with the outcome measure hearing threshold recovery. In some designs, this may act to constrain the ability to detect a significant difference. Second, in the example data set, the use of single-value criteria for significant within-subject change in word recognition (e.g., 15 percentage points) introduced a miscategorization error rate of approximately 9% when compared with the result of the binomial 95% critical difference table.