Methodological issues when comparing hearing thresholds of a group with population standards: the case of the ferry engineers.

OBJECTIVES: To discuss appropriate and inappropriate methods for comparing distributions of hearing thresholds of a study group with distributions in population standards and to determine whether the thresholds of Washington State Ferries engineers are different from those of men in the general population, using both frequency-by-frequency comparisons and analysis of audiometric shape.
DESIGN: The most recent hearing conservation program audiograms of 321 noise-exposed engineers, ages 35 to 64, were compared with the predictions of Annexes A, B, and C from ANSI S3.44. There was no screening by history or otoscopy; all audiograms were included. 95% confidence intervals (95% CIs) were calculated for the engineers' median thresholds for each ear, for the better ear (defined two ways), and for the binaural average. For Annex B, where 95% CIs are also available, it was possible to calculate z scores for the differences between Annex B and the engineers' better ears. Bulge depth, an audiometric shape statistic, measured curvature between 1 and 6 kHz.
RESULTS: Engineers' better-ear median thresholds were worse than those in Annex A but (except at 1 kHz) were as good as or better than those in Annexes B and C, which are more appropriate for comparison to an unscreened noise-exposed group like the engineers. Average bulge depth for the engineers was similar to that of the Annex B standard (no added occupational noise) and was much less than that of audiograms created by using the standard with added occupational noise between 90 and 100 dBA.
CONCLUSIONS: Audiograms from groups that have been selected for a particular exposure, but, without regard to severity, can appropriately be compared with population standards, if certain pitfalls are avoided. For unscreened study groups with large age-sex subgroups, a simple method to assess statistical significance, taking into consideration uncertainties in both the study group and the comparison standard, is the calculation of z scores for the proportion of better-ear thresholds above the Annex B median. A less powerful method combines small age-sex subgroups after age correction. Small threshold differences, even if statistically significant, may not be due to genuine differences in hearing sensitivity between study group and standard. Audiometric shape analysis offers an independent dimension of comparison between the study group and audiograms predicted from the ANSI S3.44 standard, with and without occupational noise exposure. Important pitfalls in comparison to population standards include nonrandom selection of study groups, inappropriate choice of population standard, use of the right and left ear thresholds instead of the better-ear threshold for comparison to Annex B, and comparing means with medians. The thresholds of the engineers in this study were similar to published standards for an unscreened population.