Atrial Fibrillation Burden Signature and Near-Term Prediction of Stroke: A Machine Learning Analysis

Circ Cardiovasc Qual Outcomes. 2019 Oct;12(10):e005595. doi: 10.1161/CIRCOUTCOMES.118.005595. Epub 2019 Oct 15.


Background: Atrial fibrillation (AF) increases the risk of stroke 5-fold and there is rising interest to determine if AF severity or burden can further risk stratify these patients, particularly for near-term events. Using continuous remote monitoring data from cardiac implantable electronic devices, we sought to evaluate if machine learned signatures of AF burden could provide prognostic information on near-term risk of stroke when compared to conventional risk scores.

Methods and results: We retrospectively identified Veterans Health Administration serviced patients with cardiac implantable electronic device remote monitoring data and at least one day of device-registered AF. The first 30 days of remote monitoring in nonstroke controls were compared against the past 30 days of remote monitoring before stroke in cases. We trained 3 types of models on our data: (1) convolutional neural networks, (2) random forest, and (3) L1 regularized logistic regression (LASSO). We calculated the CHA2DS2-VASc score for each patient and compared its performance against machine learned indices based on AF burden in separate test cohorts. Finally, we investigated the effect of combining our AF burden models with CHA2DS2-VASc. We identified 3114 nonstroke controls and 71 stroke cases, with no significant differences in baseline characteristics. Random forest performed the best in the test data set (area under the curve [AUC]=0.662) and convolutional neural network in the validation dataset (AUC=0.702), whereas CHA2DS2-VASc had an AUC of 0.5 or less in both data sets. Combining CHA2DS2-VASc with random forest and convolutional neural network yielded a validation AUC of 0.696 and test AUC of 0.634, yielding the highest average AUC on nontraining data.

Conclusions: This proof-of-concept study found that machine learning and ensemble methods that incorporate daily AF burden signature provided incremental prognostic value for risk stratification beyond CHA2DS2-VASc for near-term risk of stroke.

Keywords: atrial fibrillation; machine learning; risk; stroke.

Publication types

  • Comparative Study
  • Research Support, N.I.H., Extramural

MeSH terms

  • Administrative Claims, Healthcare
  • Aged
  • Aged, 80 and over
  • Atrial Fibrillation / diagnosis*
  • Atrial Fibrillation / epidemiology
  • Atrial Fibrillation / physiopathology
  • Diagnosis, Computer-Assisted*
  • Electronic Health Records
  • Female
  • Humans
  • Logistic Models
  • Machine Learning*
  • Male
  • Middle Aged
  • Neural Networks, Computer*
  • Predictive Value of Tests
  • Prognosis
  • Proof of Concept Study
  • Retrospective Studies
  • Risk Assessment
  • Risk Factors
  • Signal Processing, Computer-Assisted
  • Stroke / diagnosis
  • Stroke / epidemiology*
  • Telemetry*
  • Time Factors
  • United States / epidemiology
  • Veterans Health Services