Predicting Heart Failure from ECGs

Not as sexy as predicting crime before it happens, but ....

A nice bit of work out of Yale (and a cast of thousands) using digitized images from ECGs to predict heart failure.

These authors used simple digital photographs of ECGs from persons not known to have heart failure, checked which ones ultimately received a hospital discharge diagnosis of heart failure, and then created a model to predict such. They used the Yale cohort for derivation, then validated it on cohorts from the United Kingdom and Brazil.

Here’s your take home message:

It “works” – but the AUCs and hazard ratios range from “awful” to “maybe”. There’s also a comparison against “PCP-HF”, which is a 10-year heart failure risk calculator based on a variety of clinical factors, and it performs favorably.

There’s certainly face validity in subtle changes in electrical conduction presaging clinical manifestations of heart failure, but we’re a few leaps away from: 1) validating this model with sufficient accuracy to be usable, and 2) determining what interventions, if any, in these “pre-HF” patients will preserve their cardiovascular health.