Modelling the reproducibility of acoustic rhinometry

This paper presents an application of a linear model with correlated errors to the assessment of the precision of acoustic rhinometry in clinical medicine. The design of the reproducibility study involved 20 healthy subjects measured by five investigators with three replicate measurements of four variates taken from the right and left nasal cavities, altogether 2400 measurements. The parametric model may be structured to account for random variation between subjects, serial correlation within subjects and measurement error. The model-based reproducibility correlation coefficient was estimated to be fair both for the nasal volume measurements and for the nasal minimal cross-sectional area measurements with an overall average of 0.65. An alternative measure, the within-subject coefficient of variation, averaged 15 per cent. The choice between the measures of reproducibility depends on the degree of heterogeneity of the studied populations, the presence of time trend and the scale of measurement of the rhinometric variates. In clinical practice, repeated measurements should always be made to attain a reasonable level of reproducibility in acoustic rhinometry.