We evaluated the performance of risk models that incorporate ambulatory ECG data and clinical information for prediction of healthcare expenditures related to heart failure (HF) and stroke events in treated and untreated patients.

A retrospective cohort study of Medicare patients who underwent Zio XT ambulatory monitoring in the USA was conducted between 2014 and 2020.

14-day ambulatory ECG data and claims data were evaluated in the study sample which included 89 923 patients in the HF hospitalisation group, 75 870 in the new-onset HF group and 90 159 in the stroke hospitalisation group. Predictive models for new-onset HF, HF hospitalisation and stroke hospitalisation were generated using LASSO Cox regression with ambulatory ECG variables and components of the CHA₂DS₂-VASc. For each outcome, we scored patients using standardised linear predictors from three composite risk models, and we evaluated the association between risk score and total Medicare cost.

The following hazard ratios per one SD increase in the new risk score were observed for the model that included all CHA₂DS₂-VASc components and ECG variables: HF hospitalisation in treated 2.94, 95% CI 2.75 to 3.15; new-onset HF in treated 1.84, 95% CI 1.75 to 1.93; HF hospitalisation in untreated 3.51, 95% CI 3.23 to 3.82; and new-onset HF in untreated 1.92, 95% CI 1.85 to 2.00. Risk scores generated by the model were also predictive of Medicare cost in both treated and untreated patients, with patients in the high-risk category for all outcomes having the greatest Medicare costs during 1 year of follow-up.

Integrating arrhythmia data from ambulatory ECG monitoring into clinical risk models allows for better prediction of healthcare utilisation and cost in both treated and untreated patients at high risk for HF and stroke events.