Fig. 4: Discrimination and logistic regression of candidate audiometric factors.

A ROC curves depict the discriminative ability of individual audiometric factors for predicting SFHL in male workers from the training set (n = 4310 biologically independent participants). The AUC is indicated for each factor. B ROC analysis for female workers in the training set (n = 743 biologically independent participants). Forest plots display adjusted ORs with 95% CIs from multivariable logistic regression for the training set, analyzing the associations of (C) BH3kHz and (E) BH3_6kHz with SFHL, separately for males (n = 4310) and females (n = 743). All models were adjusted for age and CNE. The square represents the point estimate (OR), and the horizontal line represents the 95% CI. Exact two-sided P values are shown (all P values < 0.001); no adjustment for multiple comparisons was applied in this exploratory analysis. Corresponding forest plots from the external test set for (D) BH3kHz and (F) BH3_6kHz, for males (n = 1720) and females (n = 366 biologically independent participants). ORs and 95% CI were derived from multivariable logistic regression models. All P values are two-sided and exact values are provided. In this exploratory analysis comparing candidate audiometric factors, no statistical adjustment for multiple comparisons was applied. AUC area under the curve, BH binaural average hearing threshold, CI confidence interval, CNE cumulative noise exposure, FPR false positive rate, OR odds ratio, ROC receiver operating characteristic, TPR true positive rate.