Chronic obstructive pulmonary disease (COPD) is a leading cause of mortality. We hypothesized that applying machine learning to clinical and quantitative CT imaging features would improve mortality prediction in COPD.
We selected 30 clinical, spirometric, and imaging features as inputs for a random survival forest (RSF). We used top features in a Cox regression to create a machine learning mortality prediction (MLMP-COPD) model, and also assessed the performance of other statistical and machine learning models. We trained the models in moderate-to-severe COPD subjects from a subset of COPDGene, and tested prediction performance in the remainder of individuals with moderate-to-severe COPD in COPDGene and ECLIPSE. We compared our model to BODE; BODE modifications; and the age, dyspnea, obstruction (ADO) index.
We included 2,632 COPDGene and 1,268 ECLIPSE participants. The top predictors of mortality were 6-minute walk distance, FEV (% predicted) and age. The top imaging predictor was pulmonary artery-to-aorta ratio. MLMP-COPD resulted in a C-index of ≥ 0.7 in both COPDGene and ECLIPSE (6.4- and 7.2-year median follow up, respectively), significantly better than all tested mortality indices (p-value < 0.05). MLMP-COPD had fewer predictors, but similar performance to other models. The group with the highest BODE scores (7-10) had 64% mortality, while the highest mortality group defined by MLMP-COPD had 77% mortality (p = 0.012).
A MLMP-COPD model outperformed 4 existing models for predicting all-cause mortality across two COPD cohorts. Performance of machine learning was similar to traditional statistical methods. The model is available online at:

Copyright © 2020. Published by Elsevier Inc.