Multivariate adaptive regression splines based prediction of peak expiratory flow with spirometric data

Abstract

Spirometry evaluates the integrated function of lung capacity and chest wall mechanics measuring the total volume of air forcefully exhaled from a fully inflated lung. This non-invasive, informative technique for characterizing pulmonary function has an important role in clinical trials to narrow the differential diagnosis of subjects being assessed for pulmonary disorders. The test however requires patient co-operation and sometimes sub maximal effort affects the results potentially thereby leading to incomplete test and misdiagnosis. The aim of this work is to develop a prediction model based on Multivariate adaptive regression splines (MARS) technique to estimate the spirometric parameter Peak Expiratory Flow (PEF) volume. In the present study, flow-volume data from N = 220 subjects are considered. Model performances are evaluated statistically with coefficient of determination (R2) and Root Mean Squared Error (RMSE). The significant spirometric features captured in the model were FEV1, FEF50, FEF25 and the demographic parameter weight. Bland-Altman plots for the estimated PEF values showed a minimal bias. The MARS model successfully adopted the important features for prediction of PEF parameter with overall good fit and these findings can assist clinicians with enhanced spirometric investigations on respiratory disorders.