Deep Learning for Predicting Acute Exacerbation and Mortality of Interstitial Lung Disease

Area of Science:

• Pulmonary Medicine
• Artificial Intelligence in Healthcare
• Predictive Analytics

Background:

• Patients with interstitial lung disease (ILD) face high mortality and risk of acute exacerbation (AE-ILD).
• Accurate prediction of AE-ILD and mortality is crucial for effective treatment strategies.
• Longitudinal data analysis may improve prediction accuracy compared to static factors.

Purpose of the Study:

• To develop a deep learning (DL) model for predicting composite outcomes of AE-ILD and mortality.
• To utilize longitudinal clinical and environmental data for enhanced predictive capabilities.
• To validate the DL model against established methods like the ILD-GAP score.

Main Methods:

• Retrospective collection of longitudinal data from ILD patients (2008-2015).
• Development and internal/external validation of a DL model using 80% and 20% of data, respectively.
• Comparison of DL model performance with univariate/multivariate Cox proportional hazard models and the ILD-GAP score.

Main Results:

• The DL model demonstrated superior performance in predicting composite outcomes compared to CPH models.
• Concordance index values for the DL model reached 0.803 in external validation at 12 months.
• Key predictors identified include neutrophils, C-reactive protein, ILD-GAP score, and particulate matter exposure.

Conclusions:

• Deep learning models can effectively predict AE-ILD or mortality using longitudinal patient data.
• The developed DL model offers a promising tool for risk stratification in ILD patients.
• Integration of longitudinal data significantly enhances predictive accuracy for critical ILD events.