Utilizing Prediction Models to Classify Heart Failure

The current system for classifying patients with heart failure (HF) is dependent upon the confirmation of left ventricular ejection fraction (LVEF), a measurement of the heart’s ability to pump blood from the left ventricle. While patients with HF can be identified through the use of electronic medical records, the data on LVEF is oftentimes unavailable. In order to address this deficiency, a few previous studies have built and tested administrative data-based models in order to determine the likelihood of patients with HF having either HF with preserved ejection fraction (HFpEF) – when the heart muscle pumps blood normally but the ventricles are too rigid for it to adequately fill, or HF with reduced ejection fraction (HFrEF) – when the heart muscle is too weak to pump out blood effectively.

This study investigated the performance of these models in a population of over 25,000 patients with HF and known LVEF from Alberta, Canada. The data variables came from two models – the Desai model, which was built using Medicare data to predict ejection fraction (less than 45%, or greater than or equal to 45%) in patients with known LVEF, and the simplified Uijl model, which was built using data from the Swedish Heart Failure Registry.

The researchers found that these models had an acceptable performance in identifying different HF classes (HFrEF versus HFpEF) in the Alberta-based population that was relatively similar to what was shown for these models in their original studies. This suggests that when LVEF data are not available, these models are a viable option for classifying HF with moderate limitations.

“It would be great to have LVEF data for every patient with HF, but if we don’t, researchers can still use these models to find suitable patients for specific trials” said Dr. Nariman Sepehrvand, the study’s lead author and a Research Associate at the Canadian VIGOUR Centre. “This could help bring new therapies to these patients, and policymakers can also benefit from these models to understand the landscape and burden of these diseases and plan care more effectively.”

This research was co-authored by the CVC’s Drs. Nariman Sepehrvand, Justin Ezekowitz, Douglas Dover, Padma Kaul, Finlay McAslister, and Paul Armstrong, and Sunjidatul Islam.