Reliability Study of GRBAS and CAPE-V Based on Large Samples in Chinese Context

To verify the intra-rater and inter-rater consistency of Grade, Roughness, Breathiness, Asthenia, Strain (GRBAS) and the Consensus Auditory-Perceptual Evaluation of Voice (CAPE-V) to clarify their reliability in the Chinese context.

A total of 662 voice samples were evaluated by five experts using GRBAS and CAPE-V in two rounds. Six hundred and sixty-two samples were used to analyze inter-rater agreement in the first round and 150 samples were randomly selected in the second round for intra-rater consistency analysis. Cohen's Kappa, Fleiss Kappa, and intra-class correlation coefficient (ICC) were used to analyze the consistency of scores, and Spearman correlation analysis was performed for features with equivalent meanings in the two scales.

For GRBAS, intra-rater consistency for the overall voice quality (G) feature ranged from 0.34 to 0.72, with fair to moderate consistency (Cohen's Kappa: 0.22-0.53) for other features. Inter-rater agreement was poor (Fleiss Kappa: 0.08-0.32). For CAPE-V, intra-rater consistency was moderate to excellent (ICC: 0.62-0.91), and inter-rater agreement was very high (ICC: 0.86-0.92). The four shared features (overall voice quality [G/OS], roughness [R], breathiness [B], strain [S]) showed highly consistent scores between the scales (r_s: 0.90-0.92, P < 0.05 for all).

In the Chinese context, CAPE-V demonstrates superior reliability, precise scoring, and enhanced suitability for research and efficacy tracking. While GRBAS exhibits poor inter- and intra-rater consistency, its practical value in rapid clinical assessments and historical data comparisons-rooted in extensive clinical application history-currently prevents CAPE-V from fully replacing it. This dependence, however, is transitional, with CAPE-V poised for expanded use as its adoption increases.