Reliability and validity of cervical auscultation: a controlled comparison using videofluoroscopy.

Cervical auscultation is experiencing a renaissance as an adjunct to the clinical swallowing assessment. It is a controversial technique with a small evidence base. We have aimed to establish whether cervical auscultation interpretation is based on the actual sounds heard or, in practice, influenced by information gleaned from other aspects of the clinical assessment, medical notes, or previous knowledge. We sought to determine (a) rater reliability and its impact on the clinical value of cervical auscultation and (b) how judgments compare with the "gold standard": videofluoroscopy. Swallow sounds were computer recorded via a Littmann stethoscope. Sounds were sampled from 10 healthy control swallows with no aspiration/penetration and 10 patient swallows with aspiration/penetration, all recorded during simultaneous videofluoroscopy. The system generated sound quality similar to "live" bedside listening, a feature rarely seen in cervical auscultation studies. The 20 sound clips were classified as "normal" or "abnormal" by 19 volunteer speech-language pathologists with experience in cervical auscultation. After at least four weeks, 11 of these judges rated the sounds rerandomized on a new CD. Intrarater reliability kappa ranged from -0.12 to 0.71. Individual reliability did not correlate with years of experience, practice pattern, or frequency of use. Interrater reliability kappa = 0.17. Comparison with radiologically defined aspiration/penetration yielded 66% specificity, 62% sensitivity, and majority consensus gave 90% specificity, 80% sensitivity. There was a significant relationship between individual reliability and true positive rate (r(s) = 0.623, p = 0.040). The reliability of individual judges varied widely and thus, inevitably, agreement between judges was poor. Validity is dependent upon reliability: Improving the poor raters would improve the overall accuracy of this technique in predicting abnormality in swallowing. The group consensus correctly identified 17 of the 20 clips so we may speculate that the swallow sound contains audible cues that should in principle permit reliable classification.