Abstract
Background: Telehealth aims to detect imminent exacerbations in order to facilitate prompt action to prevent admissions. Current algorithms, however, are poor predictors. We aimed to develop a more advanced algorithm to estimate risk of COPD-related hospital admissions with better specificity/sensitivity.
Methods: We linked an existing telemonitoring dataset of 133 COPD patients monitored on average for 430 days, with their baseline data from the randomised controlled trial and data extracted from their electronic health record. Using this enhanced dataset we developed a probabilistic machine learning algorithm to predict next-day admissions due to COPD. We considered the complete-case and imputed scenarios. The quality of predictions was evaluated by 10-fold nested cross-validation.
Results: The standard score-counting method had a test area under the curve (AUC)=0.58. In the complete-case scenario, an approach based on patients' symptoms before the beginning of telemonitoring led to an aggregated test AUC of 0.54. In both scenarios our algorithm demonstrated significant improvements in the prediction of future admissions over the common symptom-counting methods; our machine learning algorithm resulted in a test AUC=0.71.
Conclusion: Our machine learning algorithm has significantly improved the ability of telemonitoring to predict COPD admissions. This offers the potential to improve the effectiveness both of telehealth and COPD self-management.
Funding: MRC Confidence in Concept.
- Copyright ©ERS 2015