The Fleet

← All digests

Digest — The Clinical, 2026-08-09

Four items crossed the escalation bar. Ordered by what you should do something about.


1. Amulet 360: VERITAS is published, launch is named for within 12 months — and Abbott's own press release understates device-related thrombus

Crosses the bar under the standing rule: Amulet 360, anything at all, until it launches.

Abbott's CEO, on the Q2 2026 earnings call (2026-07-16), verbatim: "Over the next 12 months … watch for U.S. launches of Libre® Duo, Amulet™ 360 left atrial appendage device, Coronary IVL System and TactiFlex Duo PFA catheter." Roughly mid-2027. The device is still investigational — Abbott states it is "not available for sale in any geography," and no PMA or PMA supplement for it exists.

The pivotal study is published and the numbers are strong: JACC: Clinical Electrophysiology 2026;12(6):1331–1341 (PMID 41701134). N=400, 34 global sites. Primary safety endpoint 0.0% (0/400). Implant success 99.8%. Complete closure at 0 mm leak, 93.9%.

The thing to act on is a conflict between the paper and Abbott's own PR.

JACC:EP paperAbbott press release
Device-related thrombus2.4%"one device-related thrombus event"
Complete closure definition0 mm leak"no leaks >3 mm"

Abbott does not reconcile either gap, and both run in the direction that flatters the device. Use 2.4%. It is still a good number in context and it is the one that survives an implanter who has read the paper. If you say "one DRT" to someone who has, you lose the room on a detail you could have had right.

Also: Abbott has published no IFU, no size matrix, no sheath spec for the 360. Anyone quoting a French size or recapture limit for this device is carrying over the current Amulet's numbers. Do not.

And a scope correction worth having: CATALYST tests the current Amulet, not the 360, and its primary completion is estimated January 2029. The readout is years away.

→ substrate/clinical--amulet-360.md


2. Abbott's ICM pause PPV is 0.01 — and the right response is a Clinical Specialist's argument

An independent, 25-center, four-vendor accuracy paper is out: Kamsani SH et al., JACC Clin Electrophysiol 2026;12(5):1113–1127 (PMID 41686115). 437,351 ECG-verified episodes, 6,756 patients, 25,826 adjudicated alerts.

Post-adjudication PPV by manufacturer: Boston Scientific 0.73, Medtronic 0.56, Abbott 0.51, BIOTRONIK 0.23.

Abbott pause PPV: 0.01, with 79.8% of those alerts attributable to undersensing. Boston Scientific's pause PPV is 0.70.

This is going to reach device clinic staff whether or not it reaches you first. The weak position is to be told. The strong one is to raise it and reframe it correctly: it is an undersensing problem, not an algorithm problem — and undersensing is fixable at implant and at follow-up. No cloud model recovers an R-wave the device never saw. That is your argument to make and nobody in a marketing function can make it. It turns the most awkward number about your own portfolio into a demonstration of what a Clinical Specialist is for.

The obvious counter — that the cohort's device-generation mix may predate Assert-IQ — is a fair question, not a deflection, and it is genuinely open. The full text was not retrieved; the generation breakdown would be in the methods.

→ substrate/clinical--icm-false-positive-evidence.md


3. The Implicity "76%" is a specificity, not an alert reduction — and it falls to 61.6% against modern devices

Closed the open question the corpus has carried since the seed run.

The 76% does not come from the 2026 papers Implicity cites. It comes from Crespin et al., Europace 2024;26(1):euad375 (PMID 38170474), and it is 215/283 — the share of already-adjudicated false-positive episodes the algorithm reclassified. That is a specificity. Against all 879 device-flagged episodes it is 24.5%.

Implicity's own EHRA 2026 abstract, in a cohort where the manufacturer's AI had already filtered, reports 61.6% specificity against AI-equipped devices versus 75.6% against non-AI ones. Neither 2026 source contains a 76% figure at all. Both are congress abstracts, not peer-reviewed papers, and the version tested was not the cleared one.

The defensible sentence: "The 76% is a specificity measured against adjudicated false positives on older, pre-AI devices. Their own 2026 data puts it at 61.6% once the device already has manufacturer AI — which is the only scenario the product is sold into."

Do not overcorrect either. 61.6% specificity on residual alerts with 98.3% sensitivity is a real reduction in review burden. The product works; the headline number is doing more work than it should.

Separately: the clearance date in the corpus was wrong by 26 days. FDA gives 2026-07-02 (K253463); July 28 was the press announcement. Fixed. Also: no Predetermined Change Control Plan was authorized, so Implicity needs new submissions to materially retrain — a real constraint on iteration speed.

→ substrate/clinical--implicity-ilr-ecg-analyzer.md


4. CMS is building a payment pathway for cardiac AI while narrowing who can bill for cardiac monitoring labour — and the comment window is open for 22 more days

Two corrections to what the corpus believed, both material.

The seed entry could not confirm any cardiac code in the proposed "Software as a Medical Service" set and warned against asserting one. The rule text has now been read. The set is cardiac-heavy. Coronary FFR-from-CT is in it as CPT 75580 (APC 1510, $850.50), alongside AI-QCT (75577, $950.50), algorithmic ECG risk assessment (0764T/0765T), computational arrhythmia simulation (0897T), perivascular fat cardiac risk (0992T/0993T), 3D cardiac procedure pre-planning (C9793), and more.

And the comment period has not closed. CMS-1850-P reads verbatim "comments must be received … by August 31, 2026" — 22 days from now. The corpus recorded that in the past tense. CMS explicitly asks whether additional codes should be included.

Meanwhile the PFS rule (CMS-1848-P, comments due Sept 14) leaves the CIED remote monitoring family entirely alone. A full-text search of the 2.8 MB rule returns zero occurrences of 93293–93299; they carry active status with ordinary RVUs and are not flagged as potentially misvalued. That is a stronger reassurance for your contract income than the seed entry could give — the exposure remains real but not immediate, and the live risk is extension in a later cycle, not this one.

The argument, now fully sourced: in one rulemaking cycle Medicare proposes to pay separately for algorithmic cardiac diagnostics while restricting who may bill for the human labour of cardiac monitoring. Both effective Jan 1 2027 if finalized. That is policy fluency almost nobody in a field role has, and you can name the codes.

One concrete downstream note: C9793 is the code family LAAO procedural planning software bills into. If SaMS finalizes, planning software for your procedure acquires a Medicare pathway in roughly the window Amulet 360 launches.

→ substrate/clinical--cms-software-as-medical-service.md, substrate/clinical--cms-cy2027-remote-monitoring.md


Below the bar, but you should know

FEops HEARTguide's randomized trial missed its primary endpoint. PREDICT-LAA (De Backer et al., JACC Cardiovasc Interv 2023;16(6):655–666, PMID 36990554) — run on Amulet, with Abbott as a registered collaborator — reported the composite primary endpoint at 41.8% vs 28.9%, RR 0.69 (95% CI 0.46–1.04), P = 0.08. The FEops press release announcing it is titled "FEops HEARTguide Proven to Significantly Enhance Efficiency and Outcomes." Efficiency findings are real and significant (fewer devices, fewer repositionings); the outcome claim is not. The widely quoted 21-vs-55-minute series is an unpublished company announcement from AF Symposium 2024. FEops was acquired by Materialise in 2024 — a 3D-printing company, not a device maker. → substrate/clinical--feops-heartguide.md

No FDA recall or safety communication touched Abbott's portfolio between July 1 and today. Searched, not assumed. Most recent Abbott CDRH record of any kind is 2026-06-11. Caveat: the database runs a ~5-week posting lag, so an action initiated after early July could exist unposted. Boston Scientific had three actions in the window; the FARASTAR one is an IFU labeling omission on 11 EU-only units. No WATCHMAN or LATITUDE actions. No Medtronic LINQ, Micra or CareLink actions.


What failed, and what I did not do

One thing for you to decide

The inbox handoff worked, but the CONSUMED- rename did not — the Drive connector has no rename, move, or delete operation. I left a receipt file CONSUMED-from-technologist-2026-08-07.md carrying the full original plus a routing table, but the un-prefixed original still sits in inbox/ and will appear in every future listing.

Proposed convention, yours to accept or reject: an inbox item is consumed if a file named CONSUMED-<original title> exists in inbox/. I have not amended AGENTS.md — that is your call, not the agent's. It resolves cleanly on the mini PC.