Comparison of fixed percentage method and lower confidence limits for defining limits of normality for interpretation of spirometry.

BACKGROUND The use of the lower 90% confidence limit of the lower limit of normal (LLN(CI)), rather than a fixed percentage of the predicted value (LLN(%)), appears to be statistically more appropriate for interpretation of spirometry results. There has been no comparative assessment of these 2 definitions of the LLN in routine clinical practice. METHODS We studied results of spirometry interpretations made with these 2 approaches, and assessed various factors that influence discordant classification of spirometry results. Spirometry records from 18,112 consecutive adult patients referred for spirometry were interpreted as normal, obstructive, or restrictive, based on both LLN(CI) and LLN(%). Discordant results were analyzed using multiple logistic regression techniques to identify variables that significantly affected discordant classification of results. RESULTS Overall, 11.7% of the results were discordant between the 2 methods. Agreement between the 2 methods, calculated using the kappa estimate, was poorer with spirometry values from women and from patients at the extremes of height and age. Age, sex, and height independently influenced discordant classification. Limits of agreement between LLN(CI) and LLN(%) were wide for all the spirometric variables studied--more so in women and in shorter and older patients. CONCLUSIONS LLN(CI) and LLN(%) yielded different interpretations of spirometry data in several instances, and the 2 methods cannot be used interchangeably. When interpreting spirometry data in routine clinical practice, LLN(CI) should be preferred over LLN(%).