When the noise exceeds the signal: day-to-day variability in spirometry dominates LLN discordance

Robert A. Rigby, Mikis D. Stasinopoulous, Achim Zeilis, Sanja Stanojevic, Gillian Heller, Fernanda de Bastiani, Thomas Kneib, Andreas Mayr, Reto Stauffer, and Nikolaus Umlauf contend that segmented linear regression (SLR) is inadequate for spirometry reference equations and over-/underdiagnoses abnormality compared with Generalized Additive Models of Location, Scale, and Shape (GAMLSS). Using a refined NHANES 2007-2012 dataset (n=16,596), I reassessed calibration and the lower limit of normal (LLN) discordance with additional analyses. Out-of-sample root mean square error (RMSE) and correlations were essentially identical across approaches, and overall, the prevalence below the LLN differed by <1 percentage point. Ages 5-80 were resampled to the 2020-2024 U.S. Census age structure, and observed below-LLN counts (z<-1.645) were compared with exact 95% binomial bands; most age-sex strata for FEV₁, FVC, and FEV₁/FVC fell within expected sampling variability, with modest departures mainly in older men. Age-stratified concordance showed 93.5-93.8% normal by both methods and 4.0-4.8% abnormal by both; discordance was uncommon (GAMLSS-only 0.20-0.78%, SLR-only 1.23-1.65%). Estimated week-to-week measurement variability produced LLN flip rates of ∼3% (FEV₁), ∼3.5% (FVC), and ∼6% (FEV₁/FVC), exceeding model-driven differences. GAMLSS remains useful, but simpler, transparent regressions can often yield comparable practical interpretation.