Performance of Ambulatory Electrocardiographic Data for Prediction of Stroke and Heart Failure Events

Background: Despite clear associations between arrhythmia burden and cardiovascular risk, clinical risk scores that predict cardiovascular events do not incorporate individual-level arrhythmia characteristics from long-term continuous monitoring (LTCM).

Objectives: This study evaluated the performance of risk models that use data from LTCM and patient claims for prediction of heart failure (HF) and ischemic stroke.

Methods: We retrospectively analyzed features extracted from up to 14 days of LTCM electrocardiogram (ECG) data linked to patient-level claims data for 320,974 Medicare beneficiaries who underwent ZioXT ambulatory monitoring. We created predictive models for HF hospitalization, stroke hospitalization, and new-onset HF within 1 year using LASSO Cox regression for variable selection among ambulatory ECG variables and components of the CHA₂DS₂-VASc score.

Results: A model that included components of the CHA₂DS₂-VASc and all ambulatory ECG variables had greater discrimination for HF hospitalization (C-statistic 0.85, 95% CI: 0.84-0.86) than the CHA₂DS₂-VASc (C-statistic 0.73, 95% CI: 0.72-0.74), but performed similarly to the CHA₂DS₂-VASc for prediction of stroke hospitalization (C-statistic 0.75 [95% CI: 0.74-0.77] vs 0.71 [95% CI: 0.70-0.72], respectively). Atrial fibrillation was associated with greater risk in the most predictive models (HF hospitalization, HR: 1.53 [95% CI: 1.35-1.72]; stroke hospitalization, HR: 1.58 [95% CI: 1.30-1.93]), and premature ventricular couplets were associated with greater risk of HF hospitalization (HR: 1.54, 95% CI: 1.43-1.65).

Conclusions: The CHA₂DS₂-VASc performed modestly for prediction of stroke and HF events; predictive ability improved significantly with addition of LTCM ECG covariates. The presence of atrial fibrillation and ventricular ectopy on 14-day LTCM were strongly associated with HF events.