Abstract Introduction Lung function evaluation commonly utilizes spirometry, but performing forced maneuvers can be challenging in young and developmentally impaired children. Oscillometry, performed during tidal breathing, measures respiratory impedance components of resistance (R) and reactance (X), with increasing magnitudes reflecting greater airway obstruction and lung tissue stiffness, respectively. Although prior studies have correlated individual spirometry and oscillometry metrics, less is understood how oscillometry metrics relate to standard spirometry definitions of airway obstruction and bronchodilator (BD) responsiveness. We aimed to construct supervised learning models of spirometry-based classifications using oscillometry metrics in school-age children. Methods Children (8-17 years) with (n = 113) and without (n = 29) asthma underwent sinusoidal oscillometry (5-37Hz; Tremoflo c100, Thorasys) followed by spirometry, pre- and post-albuterol, as part of the ENIGMA cohort study (NIH P01-HL132821). We used z-scores of oscillometry and spirometry metrics (FEV1, FVC) computed from race-neutral reference equations. Airway obstruction was defined as FEV1/FVC pre-BD z-score -1.645. Among children with obstruction (n = 41), BD-response was defined as 10% increase in FEV1 or FVC relative to the predicted value. Elastic net logistic regression identified the optimal sets of (A) pre-BD oscillometry metrics associated with spirometry-defined obstruction and (B) z-score change (pre-post) in oscillometry metrics associated with spirometry-defined BD-response. Five-fold cross-validation across a grid of alpha and lambda values was used to tune penalty strength and mixing ratio based on cross-validated deviance. Final models were refit with the selected values to estimate penalized coefficients and quantify discriminative ability (area under receiver operating characteristic curve AUC). Results Ten pre-BD oscillometry metrics were retained in the model of airway obstruction (Fig.A). Frequency dependence of resistance (R5-19), area of reactance (AX), and expiratory R5ex had the largest coefficients associated with the obstruction phenotype; increases in the magnitudes of these metrics are typically associated with small airways impairment. Notably, resistances at the highest frequencies (29, 31, 37 Hz) were also retained. The model achieved an AUC of 0.80 (95% CI 0.72-0.89), indicating good discriminatory ability for obstructive spirometry. All BD-change oscillometry metrics were retained in the model of BD-response among children with obstruction, with improvements in all metrics associated with BD responsiveness on spirometry (Fig.B). The model had limited discriminatory ability for BD-responsiveness (AUC of 0.63 95% CI 0.43-0.83). Conclusion Our results support oscillometry as an effort-independent method for detecting airway obstruction in children, which could lead to improvements in earlier detection and treatment. Future studies could further explore oscillometry as a predictor of BD-responsiveness. This abstract is funded by: NIH P01-HL132821, NIH/NCATS Colorado CTSA UM1 TR004399
Mcginn et al. (Fri,) studied this question.