Linda Johnson

Everyone with an interest in AI, ECGs, or cardiology should read our paper out in Nature Medicine today! The DRAI MARTINI study tested the performance of DeepRhythmAI from #MEDICALgorithmics for direct-to-physician reporting of ambulatory ECGs. The results overwhelmingly favour the AI, with 14 times fewer missed diagnoses of critical arrhytmias. Read it here: nature.com/articles/s4159… Here's the study in brief: 🏥 >200,000 days of ECGs from >14,000 patients 🩺 >50 experts provided 3-cardiologist beat-to-beat consensus panel annotations of >5,000 ECG events 🖥️ Innovative study design provides absolute rates of missed diagnoses of AF or SVT ≥30s, 3rd degree block, 3.5s asystole or >10s VT Results: 🫀Missed diagnoses in 3.2/1000 patients by DeepRhythm and 44.3/1000 patients by techs 🫀Superior sensitivity (98.6% vs 80.3%) 🫀NPV >99.9% for DeepRhythmAI. This study is the result of an awesome collaboration with true leaders in the field who not only gave their scientific expertise but also committed many hours of their own time doing beat-to-beat annotations of >5,300 ECG events, with consensus annotation on every single beat. This resulted in an impressive author list including @EmmaSvennberg @SZDiederichsen @DrJasonAndrade @WFMMD @AlexanderPBenz @EPjeff17 @PlatonovPyotr @StavrosStavrak1 @Cardiaficionado @Dominik_Linz @JuanBenezet @PhilippKrisai @SanjeevBhavnani @AlirezaOraii @ManningerMartin and many more!

I still think the whole workflow with an individualized model that suggests things to a physician should be tested! Maybe it’s better but maybe not. And certified for sure. Maybe in that case one need only certify the concept or the bare bones of it if the person vidence is robust enough, but for sure it needs to be certified as a decision aid.

Hey thanks - we mostly agree. Agree that models are good at summarizing but I think there a level of precision that they need to achieve to be freely useful in clinical care. They’ll get there soon/eventually. But right their imprecision is a limitation with respect to adoption in clinical contexts. I also fundamentally believe that the AI needs to “think the same way their human does” to be most useful; this of course is an opinion/belief only and I hold it lightly. As far as individual certification - I hadn’t implied that. And at least in the US I’d expect the individual physician to forever remain liable for any errors. So from a regulatory standpoint I don’t think it matters much whether products are certified at higher order levels or after tinkering by the individual.

I’ll die on this hill. AI models need to be trained at the individual clinician-level, not the population-level, in order to achieve widespread adoption. Physicians think about problems differently from one another, and do their work differently as well. A single AI model to be used “for all” will only be used by a minority of physicians and will be abandoned by the rest. Rather, AI systems need to be thought of as the individual clinician’s agent. And like a dog, I think that each clinician is going to have to spend substantial time working with and training the model to make it work in the way they need it to. Until I see this - a lot of this AI software is mostly hype to me.

Congrats @MToumpourleka! Very well deserved.

@leftbundle @DrJasonAndrade @drjohnm @Dominik_Linz @PrashSanders @Phiso_de Sure. So that would mean lower proportion cardioembolic strokes?

@lsjMD @DrJasonAndrade @drjohnm @Dominik_Linz @PrashSanders @Phiso_de More patients with AF today have lower burden and are lower chadsvasc.

#EPeeps - was discussing this today and interested if anyone has thoughts/a reference: AF is responsible for about 1/3 of strokes and 1/3 of ESUS But In a patients with AF - what proportion of strokes are due to AF? @drjohnm @Dominik_Linz @PrashSanders @Phiso_de

@DrJasonAndrade @drjohnm @Dominik_Linz @PrashSanders @Phiso_de I’m caught thinking an unoriginal thought apparently!

@lsjMD @drjohnm @Dominik_Linz @PrashSanders @Phiso_de That was the postulate of one of my mentors 15 years ago. The argument being OAC disproportionately reduces stroke beyond what would be expected for af stroke incidence

@leftbundle @DrJasonAndrade @drjohnm @Dominik_Linz @PrashSanders @Phiso_de And you think this changes the proportions? If so, how?

@DrJasonAndrade @lsjMD @drjohnm @Dominik_Linz @PrashSanders @Phiso_de SPAF had very sick patients.

I can’t believe the granddaddy of confounding gets 3.1 million views. Hey @adamcifu

New-onset postoperative atrial fibrillation management and outcomes: Read the results of the VISION Cardiac Surgery cohort study published in #EHJ 👉 ow.ly/gOk250YJ960 @RoccoMontone @ehj_ed #atrialfibrillation

@enricoferroMD @BIDMChealth Super interesting!

It was great to present our latest study #EHRA2026 in Paris for the Young Investigator Award. We ran one of the largest comparative safety analyses of #PFA vs radiofrequency #ablation for #AF, performed @BIDMChealth. We found a significantly higher stroke risk with PFA. Read on👇

Visionary Voices 2026 Goals | VISIONARY VOICES INCORPORATED donorbox.org/visionary-voic… I have a general skepticism for a lot of charity. Does it really do good? But this project is run by @sanjeevpb and seems just fantastic. Check it out and consider making a small donation

How would you choose to spend 31 million days of ECG? Smartwatches can easily generate this amount of monitoring need - check out our perspective out today: rdcu.be/eSeYc ⚡️Smartwatch sensitivity for AF is unknown ⚡️Downstream effects on healthcare systems are unknown @EmmaSvennberg @SZDiederichsen @WFMMD @EPjeff17 @SanjeevBhavnani @drjohnm

The sun is rising in New Orleans, on the first day of AHA 2025! Tomorrow we show data from the DRAI MARTINI study that answers whether the superior sensitivity of DeepRhythmAI for arrhythmia detection leads to relevant diagnoses or just a mess of incidental findings? Very proud on behalf of the whole study team that this abstract also won the Paul Dudley White Award for Norway. Clinical Science Zone 2, 9.55 in the session "Harnessing AI: Innovations in Arrhythmia Detection and Management". Hope to see you there! @SZDiederichsen @WFMMD @EmmaSvennberg @AlexanderPBenz #Medicalgorithmics #AHA2025

This is an important survey in a rapidly changing field! Please consider taking it!

Great analysis!

Our paper with @StavrosStavrak1 @MToumpourleka and others is now officially published!