Is image interpretation in drug-induced sleep endoscopy that reliable?

Previous studies have assessed drug-induced sleep endoscopy (DISE) as an interobserver reliable exam, with a learning curve effect. The objective was to check its assumed interobserver agreement and variability of between two groups of experienced and inexperienced French ENT physicians.

Prospective study. Seventy-six French ENT physicians (69 inexperienced in DISE and seven experienced) observed seven DISE videos. They were asked to determine the level(s), the configuration, and the degree of collapse, according to the VOTE classification. Specific and global agreements using the Fleiss' Kappa coefficient (k) were calculated.

The interobserver agreement varied from poor to good in determining the level; the best agreement being found for the oropharynx (global agreement = 0.82 and k = 0.6 in the experienced group, global agreement = 0.87 and k = 0.7 in the inexperienced group), followed by the soft palate and the larynx; the worst agreement being found for the tongue base (k = 0.29 in the experienced group and k = 0.38 in the inexperienced group). The agreement for the configuration and the degree of collapse was moderate except for the tongue base where the concordance was poor. In both groups, agreement was at best good without any statistically significant difference between the two levels of experience groups.

Even in a French ENT Sleep Experts group, DISE appears to be a technique with a limited interobserver agreement in the detection of obstructive sites, without any learning curve effect. In its current state, DISE interpretation may not be totally reliable.