Testing longitudinal data for prognostication in ambulatory heart failure patients with reduced ejection fraction. A...

Variability of parameter measurements in heart failure with reduced ejection fraction (HFrEF) may contribute to reducing the prediction accuracy of available prognostic models. We investigated whether the use of longitudinal versus cross-sectional measurements of established predictors of mortality in patients with HFrEF would increase the accuracy of prognostication.

We used longitudinal measurements of systolic blood pressure (SBP), heart rate, hemoglobin, creatinine and uric acid from HFrEF patients enrolled in the GISSI-HF trial. We performed linear mixed models to investigate the difference in first 6-month trajectories of these parameters between patients alive and dead at 4-year follow-up, and examined the change in prediction accuracy by comparing area under the curve (AUC) and net reclassification index (NRI) values obtained using a traditional cross-sectional survival model versus a longitudinal joint model using information up to 6-month follow-up.

We included 5469 patients with 32,206 repeated visits and measurements. We demonstrated a significant difference in the first 6-month change of each one of the selected parameters between those alive and dead at the end of follow-up (p-value for time∗mortality interaction ≤0.01). The comparison of prediction accuracy of the two models revealed a significant increase of about 2% in the AUCs when using longitudinal values of each parameter of interest up to 6 months, with significant concomitant increase in NRI. The greatest increase in accuracy was noticed when using longitudinal SBP measurements in patients with baseline SBP ≤ 110 mmHg.

Our findings support the use of longitudinal data to improve prognostication in patients with HFrEF, and warrant validation in external cohorts and creation of new prognostic tools.