Exploratory analyses were performed to assess worker demographics (gender, race, age at test, BMI, smoking status), test characteristics (maximum number of trials per subject, number of subjects at each test session, number of trials per test session), and distributions of spirometric variables (FVC, FEV1, FEV3, FEV6, EV, EV/FVC, and EV/FEV1). Graphic displays supported the assumption of normality of spirometric variables. The range and magnitude of Pearson correlations between spirometric measures on the same subject within the same session were calculated in preparation for the regression analyses. Three separate linear regression models that included data from all acceptable trials of 1,501 subjects (ie, EV/FVC < 5%) were analyzed to estimate the (statistical) predicted values of each dependent variable (EV/FEV1, EV/FEV3, and EV/FEV6) corresponding to an EV/FVC of approximately 5%. Because multiple acceptable trials for the same subject were included in the analyses, repeated-measures analyses were performed. The average correlation between pairs of spirometric measures on the same subject was constant for differing intervals between tests; therefore, an exchangeable correlation structure was assumed. The 95th percentile of the prediction interval of each dependent variable corresponding to EV/FVC = 5% was calculated and was used to identify new hesitating start cutoff values for EV/FEV1, EV/FEV3, and EV/FEV6. The newly calculated EV/FEV6 cutoff value was applied to trials rejected solely for not reaching a plateau and also to trials rejected solely for having a hesitating start as determined by OMI software. Secondary analyses were performed to investigate the effect of spirometry pattern on values of EV/FEV6 corresponding to values of EV/FVC approximating 5% for subjects classified into normal, obstructive, restrictive, and mixed spirometry patterns based on National Health and Nutrition Examination Study III reference equations.10 Again, the 95th percentile of the prediction interval of EV/FEV6 corresponding to EV/FVC = 5% was selected as the cutoff value. An additional analysis was performed in which the sensitivity and specificity of the new cutoff value at 6 s was calculated using the currently accepted hesitating start cutoff value based on EV/FVC as the gold standard. All data were analyzed using SAS statistical software, version 9.1 (SAS Institute; Cary, North Carolina).