Advanced lung function targeting small airway dysfunction (SAD) improves differentiation of mild COPD and smokers at risk from healthy controls – CAPTO-COPD

Abstract

Background: SAD is an early feature of COPD commonly overseen in routine diagnostics such as forced spirometry. Advanced tidal-breathing lung function testing showed potential to overcome this impediment. We aimed to assess their diagnostic value in differentiation of early disease using a machine learning-based approach.

Methods: We included 211 patients with mild COPD (post FEV1/VC<70% + FEV1≥70%pred), (ex)-smokers at risk (FEV₁/VC≥70%, ≥10py + CAT≥10 or long-acting bronchodilator) from CAPTO-COPD observational study and healthy controls. Conventional tests included spirometry, body plethysmography and transfer factor (TLCO). Oscillometry (OS) and SF₆-multiple breath washout (MBW) were additionally performed. We compared the accuracy of random forest class prediction including parameters of (1) spirometry only, (2) spirometry + body plethysmography + TLCO, (3) spirometry + OS + MBW and (4) all. We retained a test set for accuracy evaluation and used cross-validation for model tuning.

Results: Final analysis was performed in 90 COPD patients (mean age 65±9 years, mean FEV₁ 77±12%pred), 62 smokers at risk (60±10, 89±13) and 59 controls (45±19, 98±12, all p<0.001). Model accuracy was 63% (κ=42%) when using (1) spirometry only. Considerable improvement was achieved adding (2) conventional parameters (80%, κ=70%), (3) advanced parameters (78%, κ=66%) and (4) combining all approaches (85%, κ=77%, all p<0.01).

Conclusion: Machine learning-based differentiation of patients with only mild COPD and (ex)-smokers at risk from healthy controls can be considerably improved when including advanced lung function testing also targeting SAD.