A Duke orthopedic surgeon did a hip arthroscopy wearing an Apple Vision Pro — the FDA authorized the app on July 17, and it's the first one cleared to run inside a live case
September 6, 2026 · 5 items
A Duke orthopedic surgeon did a hip arthroscopy wearing an Apple Vision Pro — the FDA authorized the app on July 17, and it's the first one cleared to run inside a live case
The Rundown AI · Jennifer MossalguePractice operations
What actually happened. Stryker announced on Sept. 1 that the first surgical procedure using its SportSuite Vision app on Apple Vision Pro was completed at Duke Health. The surgeon was Chad Mather III, M.D., M.B.A., an orthopedic surgeon at Duke, and the case was a hip arthroscopy.
The regulatory fact is the load-bearing one. SportSuite Vision received FDA De Novo authorization on July 17 — the first application authorized for intraoperative use with Vision Pro. De Novo means there was no predicate device; FDA created the category. The Rundown adds that other surgical apps already run on the headset outside this pathway — ClearSurgery's ClearSphere has CE marking, and ScopeXR has been used in hundreds of cataract procedures.
What it puts in front of the surgeon's eyes. Arthroscopic video, HipCheck, HipMap and CT imaging in one customizable spatial field, replacing the several monitors positioned around an OR. Mather, verbatim: "Using spatial computing enabled me to customize the placement of key clinical information to fit my workflow and access it within the sterile field. This helped create a more comfortable, streamlined OR setup while keeping the information I needed in view."
Two things I'm flagging rather than smoothing over. The headset is $3,499 and this is one case, one surgeon, one center — no outcome data, no comparative study, nothing about complication rates or operative time. And Dr. Mather's name carries an asterisk in Stryker's own release, the standard marker for a disclosed financial relationship. This is a milestone announcement from the manufacturer, not evidence.
Why it matters for an independent practice: this is the closest thing to our clients' actual specialty that has come through this feed in weeks — hip arthroscopy, HipCheck/HipMap, sports medicine, the orthopedic end of the regen world. Three reads. One, the honest clinical one: nothing here shows a patient does better. What it shows is ergonomics and attention — the surgeon stops turning his head away from the field. That's a real workflow claim and a testable one, and it is not the same claim as "better outcomes." Don't let a rep collapse the two. Two, the timing read: the De Novo landed July 17 and the first case was announced Sept 1, which means the sales motion starts now. If a Stryker rep walks into a client's ASC this quarter with a headset, the physician should already know it's one authorized app on one procedure family and ask for the case volume. Three, the marketing read, and this is the one MMR can use: a practice that is early and honest about a technology like this — "here's what it does, here's what it doesn't prove yet" — builds more authority than one that either ignores it or oversells it. That posture is the P2 asset. The physicians who get burned are the ones who put "cutting-edge spatial computing surgery" on the website before they've done ten cases.
"There's no clinician time to bill for, no encounter code" — a health-policy institute says autonomous clinical AI breaks fee-for-service, and paying for it the old way will just raise costs
Healthcare IT News · Bill Siwicki, Managing EditorPractice operations
Who's saying it. A new report from the Peterson Health Technology Institute (PHTI), an independent non-profit that runs evidence-based evaluations of digital health. Executive director Caroline Pearson is the named voice throughout. The argument: today's reimbursement models were built for human-delivered care, and bolting clinical AI onto them "could unintentionally increase healthcare costs without improving outcomes."
The mechanism, in her words, and it's the sharpest sentence in the piece. "Today's healthcare spending is somewhat constrained by supply limits on clinician effort. A provider can only perform so many hours of clinical work per day." Autonomous AI removes that ceiling: "AI can generate far more clinical activity than any human workforce, without a proportional increase in clinician time, so tying payment to volume risks inflating spending at a scale fee-for-service was never built for." And there's no brake — "with low marginal costs for technology, fee-for-service has no way to ratchet spending back down."
The line PHTI draws is assistive vs. autonomous. Assistive AI fits current billing "because a clinician is still reviewing and approving every action." Autonomous AI doesn't. Her worked example is hypertension: protocolized care where an algorithm adjusts medication off remote BP readings with little human intervention — "there's no clinician time to bill for, no encounter code and often no mechanism to pay the technology developer directly."
Where their proposed fix admits its own limits. Outcomes-based payment "works best for chronic diseases with clear, measurable metrics — it begins to fall apart in areas like primary care or prevention, where we don't have near-term outcomes to tie payment to." And adoption stalls on money: health systems "must absorb substantial upfront implementation and licensing costs for AI deployment before realizing any returns," with returns "often uncertain and shared across multiple stakeholders who often did not bear the costs." Their answer is dynamic pricing — rates that start high to stimulate adoption, then step down as utilization scales and evidence accumulates.
Why it matters for an independent practice: read this one from the cash-pay side and it inverts. One — the structural point: every practice in our client base that runs on cash-pay regenerative procedures is already outside the mechanism PHTI says is broken. They don't need an encounter code to deploy an AI that handles follow-up, triage or monitoring, because nobody was going to reimburse the visit anyway. The payment problem that will slow hospital adoption for years is not their problem. That is a genuine, temporary structural advantage over the health system down the road, and it has a shelf life. Two — the discipline point: Pearson's "no way to ratchet spending back down" is the same logic pointed at a practice's own P&L. AI tools have low marginal cost, so they accumulate. Nobody cancels a $200/month tool. Count the subscriptions before adding the next one. Three — the caution: this is a policy report, not a dated action or a number you can act on this week, and that is the shape that has drawn :-1: here before. I ran it anyway because the assistive/autonomous distinction is a decision framework a practice can use on Monday — before you buy any clinical AI, ask which side of that line it sits on, because the answer determines who is liable when it's wrong, not just who pays.
Go check your saline. B. Braun recalled three lots of 0.9% Sodium Chloride Injection nationwide for iron oxide and plastic particulate — and it's the bag you dilute everything in
FDA · company announcement posted Sep 3, 2026 · September 3, 2026Pharmacy
What was recalled, exactly. B. Braun Medical Inc. is voluntarily recalling three lots of 0.9% Sodium Chloride Injection USP, 100 mL, in a 150 mL PAB® container, to the hospital / healthcare-facility / user level. Catalog S8004-5264. NDC 0264-1800-32. Lots J6B435, J6B444, J6B457. Expiration 30APR2027 — so this is stock that is nowhere near expiring and will be sitting on shelves. Distributed nationwide to hospitals and healthcare facilities from February 27 through August 3, 2026.
The contaminant list got longer after the fact. FDA's page carries the note "On September 3, 2026 firm updated their press release to include additional contaminants." The reason for the recall is now particulate matter identified as iron oxide, inorganic, cellulose or polystyrene particulate matter. It started narrower.
FDA's risk statement, verbatim, because it's blunt. Administration "has a reasonable probability of causing life-threatening adverse health consequences including pulmonary emboli … occlusions of other blood vessels (which can lead to tissue death and possible organ damage), and/or phlebitis." Systemically, infused foreign particles can cause "systemic activation of the immune system, organ dysfunction, and hemolysis." To date there are no reports of serious injury, death or other adverse events.
Why this specific product matters more than a normal recall. FDA's own description: this product "is designed for use as a diluent and delivery system for intermittent intravenous administration of compatible drug additives." It is not just hydration — it is the bag your compatible drugs go into. Returns and replacements are being arranged by B. Braun through distributors; questions go to <mailto:recalls@bbraunusa.com>, Mon–Fri 8am–5pm EST.
Why it matters for an independent practice: every IV suite, every med spa, every regen practice running infusions or diluting anything keeps 100 mL saline on the shelf, and almost nobody has a person whose job is reading FDA recall notices. Three concrete moves. One, today: someone physically reads the lot numbers on your saline stock against J6B435 / J6B444 / J6B457 — and while they're standing there, re-check the epinephrine lots from Friday's wrap (25128L1C0, 25280L1C0, 26108L1C0, NDC 0517-3030-01), because that's the same cabinet. Two, the systems point, which is the real one: two recalls in three weeks means the manual process failed twice. A standing FDA recall check is the single cheapest automation a practice can run — a scheduled pull of the recall feed filtered to the NDCs you actually stock, into a channel someone reads. That is a 30-minute build, and it is worth more than most of the AI tools that get pitched. Three, P2: the practice that can tell a patient "we checked, our lot isn't affected" within an hour of a recall is running a different operation than the one that finds out from a distributor email in October.
The car stopped, the driver assumed it saw the cyclist, and it hadn't. MIT built a readout that catches the AI being right for the wrong reason.
The Rundown AI · Jennifer Mossalgue · September 2, 2026Buildable AI
What it is. MIT and autonomous-driving company Motional built CW-Net, a system that makes part of a self-driving car's planning process visible while it drives — surfacing named concepts like "close to cyclist" and "approaching stopped vehicle" next to the trajectory it has chosen. Published in Nature on September 2; MIT announced it the same day; deployed in a Motional robotaxi on a private track generating explanations in real time. Crucially, the concept layer feeds the stage that scores trajectories — so the explanation isn't a narration bolted on afterward, it has a role in the decision.
The test case is the whole item. A safety driver watched the car stop and believed it had stopped because it accounted for a nearby cyclist. The concept display said otherwise: cyclist activation was low. Investigation found the perception system had detected the cyclist, but the experimental planner was not configured to consume cyclist inputs at all — and a backup braking system was overriding unsafe trajectories. The car did the right thing. The reason was broken. Without the readout, the correct-looking stop would have increased confidence in a planner with a serious blind spot.
A second one, same shape. A driver blamed a traffic cone for unexpected braking. The stopped-vehicle concept pointed elsewhere — and removing the cone did not eliminate the stop. The explanation turned a wrong guess into a testable hypothesis.
What was and wasn't shown, stated as the authors state it. Two studies: an online study with 9 Motional experts and 30 non-experts, and a situational-awareness study analyzing 99 participants after attention checks. Reported improvements in predicting vehicle behavior, including during surprising events. "Whether the system reduces crashes remains unestablished." The supplementary validation shows errors in concept classification — a readable label doesn't prove what's happening. As of Sep 6 the authors' own project page calls it a proof of concept, tested on limited scenarios and one planner type, and notes drivers needed time to interpret the probabilities.
Why it matters for an independent practice: zero medicine in this, and it is the most directly transferable item in today's digest. One — the failure mode has a name now, and every practice using clinical AI has it: right answer, wrong reasoning. Your scribe produces a clean note. Your intake bot routes the patient correctly. Your imaging tool flags the right knee. Each correct output raises your confidence in a system that may be ignoring an input entirely — and you will only find out on the case where the backup human isn't paying attention. The cyclist was detected and then discarded, which is exactly how a scribe drops the one sentence about prior failed injections. Two — the practical version: when you evaluate any clinical AI, stop scoring it on whether the output is right and start asking what it used to get there. "Show me which inputs the model actually consumed" is a vendor question almost nobody asks and very few can answer, and it's the same question as the audit-trail item from Sep 3 and the vendor-disclosure item from Sunday. Three — the honest limit: MIT is explicit that the explanation itself can be wrong. A confident dashboard about a confident model is two layers of unearned reassurance. The value isn't certainty, it's a reason to look closer — which is what a physician's skepticism is supposed to be for.
Video — "GPT-6 Astra Doesn't Need Your Instructions Anymore." (27:35)
Nate B Jones · uploaded Sep 6, 10:00 AM PDTBuildable AI
Not watched or transcribed — stated up front, as always. Verified: the channel (Trusted), the exact upload timestamp read from the page's own `uploadDate` metadata (2026-09-06T10:00:38-07:00), the 27:35 runtime, ~37K views at time of writing, and the specific watch URL. The claim in the title is his, not mine.
His framing, from his own description: Astra "changes what an AI agent can be trusted to do alone," and the video is about "what self-directed agents mean for your work, your skills, and the next six months." That is a workflow-and-skills argument, not a benchmark table.
Why this source and not the other one. Nate Herk's newest upload today is "I Tested GPT-6 Astra vs Fable 5.1 on 15 Real Use Cases" — a model comparison, which is the shape the standing preference de-emphasizes, and it would have been three consecutive days from the same channel. Nate B Jones is Trusted, sits at (0/0), and has supplied exactly one item in 46 days on the list. A vote here is genuinely informative.
The caveat I owe you. This is still Astra-anchored content, and Astra is the launch I've held twice as horse-race news. My read: "what can an agent be trusted to do without supervision" is a different question from "which model scored higher at what price." That's my read, not a rule you've set.
Why it matters for an independent practice: the whole premise — an agent that runs without step-by-step instruction — is the thing a small practice most wants and is least equipped to supervise. In a hospital there's an IT department that notices when an automation starts doing something odd. In an 8-person clinic, the automation is the process, and the only person who'd notice is the one it replaced. So take the video's claim seriously in both directions: the leverage is real (the front-desk lead describes the follow-up sequence and it exists), and so is the exposure — an unsupervised agent touching scheduling, intake or follow-up is touching patient data with nobody reading its log, which is the exact gap OpenAI admitted it has no standard for on Sunday. Concrete test this week: pick the lowest-stakes repeating task in the practice — appointment reminders, review requests, no-show follow-up — and let something run it unsupervised for five days with a human reading the output daily. You learn what "self-directed" actually means on a task where being wrong costs nothing.
