AI EKG model raises annual sudden cardiac death risk detection from 4.6% to 7% in multi-national validation cohorts

The model analyzed raw waveform morphology rather than derived ejection fraction and was tested on independent US and Taiwan cohorts totaling thousands of records. It flagged patients who appeared low-risk by current ejection-fraction thresholds yet experienced higher rates of out-of-hospital arrest. This expands the candidate pool for implantable defibrillators while surfacing previously unrecognized electrical signatures linked to fatal arrhythmias.

Current ejection-fraction screening misses most sudden cardiac death events because many victims have preserved systolic function; observational data from registries show sensitivity below 30%. The AI approach improves discrimination but inherits the same downstream problem: deciding when to implant devices whose complications include infection and inappropriate shocks. No randomized data yet exist on whether earlier identification improves survival or merely increases procedures.

Two supporting analyses, a 2023 Circulation review of ICD utilization gaps and a Lancet Digital Health 2024 external-validation study of ECG-based risk models, underscore that algorithmic gains require prospective outcome trials and equity checks across ancestry groups before guideline adoption. Regulatory pathways will likely demand post-market surveillance of net clinical benefit.

Next steps include multi-center RCTs powered for mortality and cost-effectiveness endpoints, with FDA breakthrough-device designation already under discussion for the Berkeley algorithm.