Digest — The Clinical, 2026-08-23
Three items. One crosses the escalation bar on its own; two are clocks that run out before or just after the next clinical run on 2026-08-30.
1. The Merlin programmer recall is the SECOND time, not the first — and that changes what you should say about it
This is the escalation. The corpus already carried the 2026 recall. What it did not know is that the same operation on the same programmer line failed the same way in 2022.
| March 2022 | 2026 — Z-2405-2026 | |
|---|---|---|
| Programmer | Merlin / Merlin 2 PCS | Merlin 3650 PCS |
| Software | Model 3330 v25.0.x–v25.3.x; MER3400 v1.1.x–v1.3.x | Model 3330 v25.0.x–v28.7.x |
| Devices | Gallant, Neutrino NxT, Entrant ICDs / CRT-Ds | AVEIR leadless pacemakers |
| Operation | Pacing capture threshold test (Decrement Test Method) | Pacing capture threshold test |
| What went wrong | Test does not stop on button release; runs to 0.25 V or until telemetry breaks | Terminate command not transmitted; sub-threshold pacing persists until telemetry fully disconnects |
| Risk | Transient asystole in pacemaker-dependent patients | Transient asystole in a pacemaker-dependent patient (one, reported) |
| How it ends | Break telemetry | Telemetry disconnect ends it |
| Volume | 21 complaints / ~38,000 devices | 4 complaints / 11,182 programmers |
Two things to do with this.
Know how to break telemetry deliberately. Reposition the wand, disconnect the dongle, power the programmer down. That is the documented escape from both events, and it is the thing that matters in the ten seconds while it is happening in front of you. Abbott's 2022 letter also gave three preventatives that still read as sensible: position pacemaker-dependent patients supine, start testing well above the anticipated threshold, and hold the test button through the first decrement cycle (~2–3 seconds).
Get to "this is the second time" before someone else does. The natural description of the 2026 recall — a software bug, found, fixed, upgrade available — is accurate and incomplete. An experienced EP or device clinic lead who was around in 2022 may remember the first one. Volunteering "2022 was the Decrement Test on Gallant/Neutrino/Entrant, this one is Aveir" is a different position to be in than being told.
The rest of the recall detail, unchanged and still worth carrying. Affected: Merlin 3650 PCS software Model 3330, v25.0.x through v28.7.x, with AVEIR. Fix: v28.9.1 rev 1 or higher. Harm: four global complaints, three with no symptoms, one pacemaker-dependent patient with transient asystole, no permanent injuries, all resolved in-session. Saying "no injuries" is wrong — the trade coverage implies it and Abbott's own notification does not.
Status, checked today: the recall is still open, with no termination date, at the openFDA data cut of 2026-08-20. 11,182 units is a large denominator for a remedy that needs someone to physically accept an upgrade. A programmer in a clinic cabinet nobody touched is still in the affected range.
What is not known, and worth an internal question: what fraction of the 11,182 has actually been upgraded. Not public.
Also newly established, and it is the follow-up worth doing: Abbott's own advisories index lists at least three further Merlin programmer software actions nobody here has read — May 2020, June 2022 and November 2024. If any of those is also a threshold-test or telemetry-termination failure, "twice" becomes a much stronger word.
→ substrate/clinical--merlin-pcs-3650-recall.md
2. Clock: the CMS OPPS comment window closes 31 August — eight days
No new finding. This is a countdown, and the next clinical run is 2026-08-30 — one day before it shuts.
CMS-1850-P proposes Software as a Medical Service (SaMS), the first dedicated Medicare payment category for diagnostic AI, and the designated set is cardiac-heavy: 75580 (FFR from CCTA, APC 1510, ~$850), 75577 (AI-QCT), 0764T/0765T (algorithmic ECG risk assessment), 0897T (computational arrhythmia simulation), C9793 (3D predictive model generation for cardiac procedure pre-planning), 0932T, 0712T, 0992T/0993T, G0183, G0680, X658T, X668T, X669T.
The substantive work is done and sits in the entry — every code enumerated from the rule's own tables and CMS's Addendum B. Commenting is your decision, not an agent's. CMS explicitly invites additions to the list, and separately asks whether a "T"-style multiple-procedure discount should apply, which would cut the economics of every code in the set.
The second window — CY2027 PFS, comments close 14 September, 22 days — is the one that touches your contract income. Unchanged finding: the third-party RPM/RTM prohibition does not reach 93293–93299, confirmed by full-text search of the rule itself, with those codes carried at active status and ordinary RVUs. The exposure is real but not immediate; the live risk is extension in a later cycle, not this one.
→ substrate/clinical--cms-software-as-medical-service.md, substrate/clinical--cms-cy2027-remote-monitoring.md
3. There is now a second entrant in LAAO procedural planning, and it does not need a CT
Anumana "WatchMate Software," 510(k) K260300, cleared 2026-06-15. Indications, verbatim: "WatchMate Software enables visualization and measurement of structures of the heart and vessels for pre-procedural planning and sizing for the left atrial appendage closure (LAAC) procedure."
It builds a 3D reconstruction of the left atrial appendage from 2D TEE — not CT — through a five-stage deep-learning pipeline, with automated landing-zone measurement. Its reported comparison: mean absolute deviation versus CT of 3.1 ± 3.4 mm, against 5.2 ± 3.4 mm for physician-interpreted 2D TEE (P<0.001).
Three distinctions worth having ready, because each is a way this gets misused in a room:
- It is measurement, not simulation. FEops HEARTguide's digital twin predicts implant frame deformation and sits under a different FDA regulation. WatchMate sits in the same box as FEops's ALPACA line — anatomical measurement. Anyone calling it a competitor to HEARTguide's simulation is describing the wrong product.
- Its comparator is manual TEE, not FEops. 3.1 mm vs 5.2 mm is a claim about replacing physician-interpreted 2D TEE sizing. It is not a claim about beating CT-based planning, and the cleared file does not make one.
- Its IFU names no commercial occluder. It says "the LAAC procedure." It does not name Amulet and it does not name WATCHMAN as an indicated device.
Why it is worth your attention anyway: a planning tool that runs on imaging already being acquired in the lab is a different workflow and cost proposition from one that requires an additional CT, whatever the accuracy numbers say. And there is an open reimbursement question underneath it — C9793's descriptor reads "using data from cardiac CTA and/or MRI," so a TEE-derived planning model may fall outside the SaMS pathway entirely. That is a question, not a finding; no coding determination has been checked.
Unknown and not investigated: whether WatchMate is commercially available, and whether Anumana partners with anyone. That is what decides whether this is competitive intelligence or background.
→ substrate/clinical--feops-heartguide.md
Also this run, below the bar
- Amulet 360 — no movement, second consecutive negative walk. Zero PMA records mentioning Amulet or 360. Nothing new announced, no IFU, no size matrix, no CE statement. The DRT figure is still unreconciled — use 2.4%, not "one event." Meanwhile five PMA supplements landed on the TactiFlex/Ampere PFA line in five weeks. That contrast says the PFA programme is visibly moving through FDA and Amulet 360 is not yet visible; it is not evidence against the mid-2027 window, since PMA submissions are not public until decided.
- Portfolio safety sweep: clean, checked three ways including Abbott's own advisories index for the first time. No new Abbott device recall since 2026-05-12. No FDA safety communication for any cardiac device — the most recent posting of any kind is still 06/03/2026.
- The corpus has its first PFA entry, and its first finding is that there are two different 2026 multi-society PFA statements that are easy to merge by accident. The HRS/EHRA one names "device interactions" among PFA's new safety considerations; the EHRA-led one has a dedicated section on PFA in patients with CIEDs. Neither has been read in full — Heart Rhythm is Elsevier-blocked to every agent here. You can open the DOI in ten seconds; the section worth capturing is the CIED one. DOI 10.1016/j.hrthm.2026.02.006.
- FDA created a new Class II device category for cardiovascular outcome prediction (product code SIU, 21 CFR 870.2215, from the CaRi-Heart De Novo on 2026-07-28). Its own classification order never says "AI", restricts the whole class to adjunctive non-acute use, requires a standing postmarket performance-monitoring plan, and publishes no performance data at all.
- ESC Congress 2026 lands between runs — 28–31 August, Munich. Two late-breakers on your beat: VIRTUES ICD / VIRTUES PM (Sat 29 Aug, digital health platforms for ICDs and pacemakers) and ASPIRED (Mon 31 Aug, enhanced ambulatory ECG for unexplained syncope). Caveat carried from the Technologist and worth repeating: no late-breaker in the retrieved list is explicitly an AI-ECG or machine-learning trial. If someone claims there is a headline cardiac-AI late-breaker at ESC 2026, ask them to name it.
What failed this run
Stated as failures, not absences.
- **The EClinicalMedicine cardiac-AI meta-analysis could not be retrieved (PMID 42602970) — Europe PMC returned HTTP 429 on four attempts across two hosts, and NCBI E-utilities returned ROBOTS_DISALLOWED for this PMID specifically while succeeding for a different one minutes earlier. It is held in the ledger at
reportedand is deliberately not written up. It matters: 31 RCTs, n≈1.7M, and the headline is that AI-enhanced ECG did not significantly improve heart failure detection**. It is the strongest counterweight in the corpus to any vendor deck on cardiac AI, and it should not be quoted until someone has read it directly. - CorVista (K253576) and Abbott EnSite X (K262006) 510(k) summaries both returned HTTP 404 again. No performance figure for either exists in this corpus; none should be quoted.
- openFDA's clearance databases are frozen, not lagging —
device/510kanddevice/pmahave reported the same 2026-08-10 data cut for thirteen days. Every clearance negative above is ~16 days stale and getting staler daily. Recorded so it is not read as a fresh all-clear. - A wrong date was introduced into a substrate entry by this agent and corrected inside the same run. Recorded in the entry's changelog rather than fixed silently.