Emphysema Detection in Smokers: Diffusing Capacity for Nitric Oxide Beats Diffusing Capacity of Carbon Monoxide-Based...

Pulmonary diffusing capacity for nitric oxide (DLNO) remains underutilized despite potential advantages over pulmonary diffusing capacity for carbon monoxide (DLCO). We evaluated whether DLNO better detects emphysema than DLCO, spirometry, or lung volumes in smokers.

We performed an individual participant data meta-analysis of adult smokers (14-43 pack years) with and without computed tomography-defined emphysema using a standardized 10 ± 2-second breath-hold time double diffusion protocol. Variables were converted to z-scores. Prespecified models contrasted DLCO- versus DLNO-based approaches. Model selection used the Bayesian information criterion (BIC) and leave-one-out information criterion; discrimination used area under the receiver operating characteristic (AUROC) curve and Matthews correlation coefficient (MCC) with repeated cross-validation.

After harmonization and quality control, 408 participants (85 emphysema, 323 controls) were analyzed. The lowest BIC (164.6) occurred for the 3-predictor model with total lung capacity (TLC), forced expiratory volume in 1 second (FEV₁), and DLNO z-scores, with an 88% probability of being superior to the next-lowest BIC model (168.5). Discrimination (AUROC 0.97, 95% confidence interval [CI] 0.95-0.98) and classification (MCC 0.80, 95% CI 0.69-0.89) were high. Hierarchical partitioning showed unique contributions from FEV₁ z-scores (R²=0.35) > DLNO z-scores (R²=0.21) > TLC z-scores (R²=0.11), totaling McFadden's R²=0.663. Adding DLCO z-scores increased the total R² trivially (by 0.003) and contributed largely shared information with DLNO (variance inflation factors ≤ 4.5). Category-free reclassification and Youden-threshold analyses showed small but favorable gains; the case-control risk gap improved by up to ~5% when adding DLNO to a DLCO-based model.

When predicting the likelihood of emphysema in smokers, a parsimonious z-score model comprising TLC, FEV₁, and DLNO z-scores provides excellent performance and stable rank superiority.