Non‐linear principal component analysis of maximum expiratory flow‐volume curves

The ability to classify maximum expiratory flow‐volume (MEFV) curves has been explored for several types of principal component analysis (PCA). The PCAs differed in their methods of transformation of the data. These included discrete non‐linear transformation with or without additional restrictions, or continuous linear (classical PCA) and piecewise linear transformation using first degree B‐splines. Two hundred and fifty curves of males between 21 and 30 years of age were involved in these analyses. Although no real pathology occurs among these males, still a clearly differentiated classification of curve types was found. The results were passively gauged with respect to smoking habits and respiratory symptoms. The analysis using piecewise linear transformations differentiated best between people with and without respiratory symptoms, and between smokers and non‐smokers. In general pathology is rare and to be found at the lower tail of a lung function measurement. As the piecewise linear transformations resulted in a more pronounced negative skewness of the relevant variables, the gain of this transformation with respect to the discrete analyses is to be found in the ability to use the information content within a category and with respect to the continuous linear analysis the ability to pronounce this information.