Audiometric threshold shift definitions: simulations and suggestions.

Abstract

OBJECTIVES: Determine (1) whether pure-tone average (PTA) or "any frequency" (AF) definitions of significant threshold shift (STS) are more accurate under different conditions; (2) whether STS definitions applied twice are more accurate than when applied once; (3) whether three surrogate analysis methods applicable to real-world audiometric data can appropriately measure the accuracy of STS definitions; and (4) whether evidence from previous studies supports a change in the current STS definition used by the Occupational Safety and Health Administration (OSHA); (5) make recommendations for future research. DESIGN: Computer simulation of periodic audiograms, with typical test-retest variability, from seven groups of individuals (N = 1000 each), with or without threshold shifts of different sizes at different frequencies; receiver operating characteristic area analysis to determine relative accuracy of PTA and AF decision variables, applied once or twice; testing of surrogate methods (confirmation, variability, and comparison) against receiver operating characteristic areas in estimation of accuracy of decision variables; review of previous studies, considering the limitations of the surrogate methods that were used. RESULTS: Either AF or PTA definitions may be preferable, depending on whether the genuine threshold shift has occurred at a single frequency or across a range of frequencies. Including too few or too many frequencies degrades the performance of both AF and PTA variables. An STS decision variable performs only slightly better when applied twice than when applied once. The surrogate methods can be useful, but all have important caveats: The comparison method is best, but only when an appropriate control group is available; the variability method is intermediate but requires that each decision variable be tested at multiple criterion values for reliable results. The confirmation method, which was the least accurate in assessing true performance, also requires testing at multiple levels. Even when this is done, it markedly exaggerates the benefit of applying an STS definition twice and makes STS definitions that identify large numbers of individuals falsely appear to be more accurate than definitions that identify smaller numbers of individuals. Taking these caveats into account, previous studies do not offer convincing evidence for a change in the current OSHA STS definition. CONCLUSIONS: Choice of an STS definition requires three judgments that cannot be made based on the type of analysis discussed in this report: The range of frequencies to be tested, whether to require that an STS definition be applied twice, and the acceptable magnitude of false-positive error. Once these judgments have been made, the techniques discussed here can assist in selecting the appropriate decision variable (typically PTA or AF) and a criterion value that yields an acceptable false-positive rate. Additional research using the techniques described in this report with either new or old audiometric databases could help determine whether any STS definition is significantly better than the current OSHA definition.