Smartwatch QT estimates show limited agreement with manual ECG readings
Source check: Source checked. Not independent scientific confirmation.
Finding: The QT-logs study applied a trained U-Net to smartwatch ECGs and compared QTc estimates with blinded manual12-lead measurements. Among66 day-ten comparisons, mean bias was about−1 millisecond, but agreement limits extended from−46 to45 milliseconds. The selected low-risk cohort contained very few critical QTc prolongations, leaving accuracy for dangerous prolongation uncertain.
Technology: U-Net convolutional neural network
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
- AI contribution
- A U-Net trained on6315 annotated resting/Holter recordings identifies waveform boundaries; beat-level QTc estimates are aggregated for received smartwatch tracings. No smartwatch ECGs were used in training.
- Test and result
- Newly transmitted smartwatch ECGs were automatically analyzed in a prospective paired-measurement study. Eighty-five people entered;76 received daily follow-up. Manual12-lead reference comparisons were available for71 baseline,53 day-six and66 day-ten visits, with blinded readers.
- Limitation / still unknown
- Recordings were not consistently simultaneous and agreement was only moderate; very few critical prolongations prevented validation of clinically dangerous thresholds. Young selected participants, technical exclusions and incomplete transmissions limit transfer. No drug-safety, arrhythmia-prevention or treatment-effect conclusion is adopted. Validation3 reflects the direct paired measurement check, not independent replication or proven model-guided clinical benefit.
Evidence sources and annotation context
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
Evidence sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC7845555/ · https://doi.org/10.1016/j.ijcard.2021.01.002
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review date: 21 Nov 2026. Review scope: Original clinical reference/readout Methods, CNN operation/training description, paired agreement Results/Table2 and limitations; drug efficacy not assessed.
No follow-up recorded; this is not a claim that no follow-up exists. Follow-up review scope and date are not recorded. No source-supported relationships recorded.
Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.