Median regression spline modeling of longitudinal FEV(1) measurements in cystic fibrosis (CF) and chronic obstructive pulmonary disease (COPD) patients

RATIONALE: Clinical phenotyping, therapeutic investigations as well as genomic, airway secretion metabolomic and metagenomic investigations can benefit from robust, nonlinear modeling of FEV(1) in individual subjects. We demonstrate the utility of measuring FEV(1) dynamics in representative cystic fibrosis (CF) and chronic obstructive pulmonary disease (COPD) populations. METHODS: Individual FEV(1) data from CF and COPD subjects were modeled by estimating median regression splines and their predicted first and second derivatives. Classes were created from variables that capture the dynamics of these curves in both cohorts. RESULTS: Nine FEV(1) dynamic variables were identified from the splines and their predicted derivatives in individuals with CF (n = 177) and COPD (n = 374). Three FEV(1) dynamic classes (i.e. stable, intermediate and hypervariable) were generated and described using these variables from both cohorts. In the CF cohort, the FEV(1) hypervariable class (HV) was associated with a clinically unstable, female-dominated phenotypes while stable FEV(1) class (S) individuals were highly associated with the male-dominated milder clinical phenotype. In the COPD cohort, associations were found between the FEV(1) dynamic classes, the COPD GOLD grades, with exacerbation frequency and symptoms. CONCLUSION: Nonlinear modeling of FEV(1) with splines provides new insights and is useful in characterizing CF and COPD clinical phenotypes.