Predicting New-onset Exertional and Resting Hypoxemia in Fibrotic Interstitial Lung Disease.

Abstract

Hypoxemia in fibrotic interstitial lung disease (ILD) indicates disease progression and is of prognostic significance. The onset of hypoxemia signifies disease progression and predicts mortality in fibrotic interstitial lung diseases (ILD). Accurately predicting new-onset exertional and resting hypoxemia prompts appropriate patient discussion and timely consideration of home oxygen. We derived and externally validated a risk prediction tool for both new-onset exertional and resting hypoxemia. This study used ILD registries from Canada for the derivation cohort and from Australia and United States for the validation cohort. New-onset exertional and resting hypoxemia were defined as nadir SpO2 <88% during 6-minute walk tests, resting SpO2 <88%, or the initiation of ambulatory or continuous oxygen. Candidate predictors included patient demographics, ILD subtypes, and pulmonary function. Time-varying Cox regression was used to identify the top performing prediction model according to Akaike information criterion and clinical usability. Model performance was assessed using Harrell's C-index and goodness-of-fit (GoF) likelihood ratio test. A categorized risk prediction tool was developed. The best-performing prediction model for both new-onset exertional and resting hypoxemia included age, body mass index, a diagnosis of idiopathic pulmonary fibrosis, and percent-predicted forced vital capacity and diffusing capacity of carbon monoxide. The risk prediction tool exhibited good performance for exertional hypoxemia (C-index=0.70, GoF=0.85) and resting hypoxemia (C-index=0.77, GoF=0.27) in the derivation cohort, with similar performance in the validation cohort except calibration for resting hypoxemia (GoF=0.001). This clinically applicable risk prediction tool predicted new-onset exertional and resting hypoxemia at six months in the derivation cohort and a diverse validation cohort. Suboptimal GoF in the validation cohort likely reflected overestimation of hypoxemia risk and indicated that the model is not flawed due to underestimation of hypoxemia.