Machine learning classifier is associated with mortality in interstitial lung disease: a retrospective validation...

Area of Science:

• Pulmonary Medicine
• Artificial Intelligence in Healthcare
• Radiology

Background:

• Interstitial lung disease (ILD) mortality prediction is challenging due to subtype heterogeneity.
• Current prognostication relies on forced vital capacity (FVC) and the Gender, Age, and Physiology (GAP) score.
• The machine learning system Fibresolve, initially for idiopathic pulmonary fibrosis (IPF) CT patterns, showed prior mortality associations across ILD subtypes.

Purpose of the Study:

• To retrospectively validate the association of Fibresolve with mortality in a large, external cohort of ILD patients.
• To assess Fibresolve's predictive performance beyond existing metrics like the modified GAP (mGAP) score.

Main Methods:

• A multi-center validation study applied Fibresolve to chest CT scans of ILD patients with follow-up data.
• Fibresolve scores, categorized by tertile, were analyzed using Cox regression, adjusted for tobacco use and mGAP score.

Main Results:

• In a cohort of 643 ILD patients, 69.3% died during follow-up.
• Fibresolve tertile 3 demonstrated a significant association with mortality (HR 3.12, p < 0.001) in multivariable analysis.
• Significant associations were found in non-IPF ILDs (Tertile 3 HR 4.66) and severe ILD (FVC ≤ 75%) (Tertile 3 HR 4.80).

Conclusions:

• Fibresolve is independently associated with mortality in ILD patients.
• The association is particularly strong in non-IPF ILDs and in patients with severe disease.
• Fibresolve offers a promising tool for ILD prognostication.