Machine Learning Mortality Prediction Outperforms Other Models in COPD

Machine learning mortality prediction (MLMP-COPD) software outperformed other machine learning models and was similar to existing statistical methods for predicting all-cause mortality in patients with chronic obstructive pulmonary disease (COPD), according to study results published in CHEST.

A total of 30 clinical, spirometric, and imaging features of COPD in a subset of individuals with moderate to severe COPD from the COPDGene study were used to train a machine learning program called MLMP-COPD. The model was then tested on the remaining patients with moderate to severe COPD in the COPDGene and ECLIPSE studies. The primary outcome was prediction accuracy of all-cause mortality.

Of the 2632 CODPGene participants with moderate to severe COPD, 1974 were randomly assigned to the MLMP-COPD training group and 658 were randomly assigned to the MLMP-COPD testing group. Additionally, 1268 participants from the ECLIPSE study were included in the testing group. The top imaging predictor of all-cause mortality was pulmonary artery-to-aorta ratio. MLMP-COPD resulted in a C-index ≥0.7 in both COPDGene and ECLIPSE, significantly better than all tested mortality indices.

“The MLMP-COPD model demonstrated superior predictive performance to 4 prior mortality prediction indices in moderate to severe COPD subjects across two large cohorts,” the researchers wrote. “Further investigation across diverse populations and investigation of cause-specific mortality will help support the validity of this model.”

Reference

Moll M, Qiao D, Regan EA, et al. Machine learning and prediction of all-cause mortality in chronic obstructive pulmonary disease [published online April 27, 2020]. CHEST. doi: 10.1016/j.chest.2020.02.079