This is a Physician in the Loop special topic.

The surgeon is the benchmark

Surgical robots are inching toward choosing where to clip, cut or sew. Any robot that does should be judged against a good surgeon doing the standard operation, not against other robots.

No robot has yet decided for itself where to clip, cut or sew a patient's soft tissue, a review published on Sept. 27 concluded. The Food and Drug Administration has nonetheless begun to consider how such machines will be judged. Its draft guidance for surgical robots, issued on Sept. 25, sets aside for case-by-case talks any autonomous machine that would perform significant parts of an operation independently of a surgeon. Comments on the draft are due by Nov. 24, and a public workshop on autonomous robots follows on Dec. 2 and 3.

The question before the agency is where the line falls between a robot that helps a surgeon and one that operates, and what the second kind should be measured against. For robots that surgeons steer, the draft prefers to compare clinical results with those of other robots, turning to laparoscopy only when robotic data are lacking. That is a fair test of whether a new machine works as well as the ones hospitals already own, but the wrong one for a feature that decides where to clip, cut or sew. Any such feature should have to prove itself against a good surgeon doing the standard operation, which for gallbladder removal is laparoscopy, on hard outcomes such as injury to the bile duct.

Intuitive, which had more than 11,700 da Vinci robots installed at midyear, has said where it is heading. In July it described its plans for artificial intelligence as five layers, from "Good data" at the base to "Supervised autonomy" at the top. That top layer, which the company calls its "longer-term vision," is artificial intelligence that "extends surgeon capability while preserving surgeon accountability and control." Iman Jeddi, who runs its da Vinci platforms, said the vision "is not about removing the surgeon" and spoke of technology that is "rigorously validated." The groundwork is in its newest robot, the da Vinci 5, cleared in 2024. By the company's count it has more than 10,000 times the computing power of the model before it, and optional instruments measure the force surgeons put on tissue and let them feel it. New features arrive through what its chief executive, Dave Rosa, calls "sequential software releases."

Rivals are arriving, with automation of a modest kind. Medtronic's Hugo, cleared for urology last December, works with the company's Touch Surgery software, which reviews cases after surgery using artificial intelligence and which, the company says, "is not intended to direct surgery." Johnson & Johnson's Ottava, authorized in July for operations including gallbladder removal, builds its four arms into the operating table, and its automation moves them into preset poses to simplify setting up and clearing away. Moon Surgical's robot goes a step further: its camera, the company says, can follow the tip of a chosen instrument, so the surgeon can steer the view with both hands still on the instruments.

The robots that have worked out for themselves where to clip, cut or sew soft tissue are research machines. In 2022 a Johns Hopkins robot called STAR stitched together pig intestine in keyhole operations on live animals, while an operator chose among its plans and approved the firing of its needle. In a synthetic model its stitching beat expert surgeons' for consistency and accuracy; in four pigs, compared with one sewn by hand, its builders wrote that it could match expert surgeons on leaks and on how open the bowel stayed. In 2025 a newer Hopkins system, trained on videos of surgeons, performed the clipping and cutting stage of gallbladder removal on eight organs outside the body with no human help. It found the ducts and arteries, placed the clips and cut, more slowly than a human surgeon but, its university says, with results comparable to an expert's.

The FDA's draft covers robots that surgeons steer, which may contain "some level of automated functions." It asks for clinical results such as complications and reoperations within 30 days, set against comparators that may come from published studies. The draft does not say how it would treat a feature that chose where to place a clip while a surgeon watched: its scope section mentions robots "capable of decision-making and independent execution of tasks" only to say that some of its advice may apply to them. Treated as an automated function of a steered robot, such a feature would be judged against other robots, and if it arrived as an update to a robot already sold, clinical data "may be requested," the draft says, in cases such as "different technology."

In gallbladder surgery, the operation the Hopkins team chose, other robots make a weak benchmark. A retrospective study of a decade of Medicare claims found robotic removal associated with about seven bile duct injuries needing surgical repair within a year for every 1,000 operations, against two for laparoscopy. A later study by largely the same University of Michigan team, running to 2021, found the excess among low-, medium- and high-risk patients alike, which, its authors write, suggests that patient selection may not be the cause.

That evidence is observational, and it is contested. A report on the same team's letter in May, covering 2020 to 2023, puts the robotic rate at about twice the laparoscopic rate and quotes the authors: outcomes "may be improving as robotic surgery matures," but "meaningful differences between approaches persist." In the original Medicare study, an analysis designed to account for hidden differences between patients narrowed the gap but did not close it. A study of commercial insurance claims in acute care found similar rates of bile duct injury but about half again as many major complications after robotic operations. And one academic hospital that compared closely matched patients found the reverse of the Medicare result, though its estimate rests on a dozen injuries. It also reported fewer conversions to open surgery with the robot.

Randomized comparisons with laparoscopy can be unflattering. In a blinded trial of ventral hernia repair in Houston, supported by a research grant from Intuitive, the robot added about an hour to the operation and a fifth to its cost, with no evidence that patients spent fewer days in hospital. A new robot measured only against older ones would inherit results like these unnoticed.

The best case against this standard is that it would slow a technology meant to make surgery safer. STAR's builders wrote that autonomous surgery could provide "efficacy, safety, and consistency independent of individual surgeon's skill and experience." Bile duct injury is rare enough that a trial built to detect a difference would need many thousands of patients, delaying a feature that might prevent it. And the draft says the FDA may require more data after a robot reaches the market when rare harms remain uncertain. None of this justifies the wrong benchmark. A trial need not be sized for bile duct injury alone: a feature can be shown to be no worse than laparoscopy on a composite of serious complications, with rare injuries counted afterward. Data gathered afterward mean little without a fair comparison, and for a machine that chooses where to clip, the fair comparison is a surgeon who chooses well.

The FDA should draw its line at decisions, not at independence from the surgeon alone, because a feature that picks the spot for a clip while a surgeon watches is still doing the surgeon's job. A robot that steadies a camera or sets out its own arms can sensibly be compared with other robots. One that chooses where to clip, cut or sew should first show that it is no worse than good surgeons doing the standard operation, and then be tracked in a registry that counts bile duct injuries and similar harms, by approach, for at least a year. Surgeons and their societies know what a good result looks like, and the comments due by Nov. 24 and the December workshop are their chance to say so. Hospitals that offer robotic gallbladder surgery owe the physicians who refer patients to them their own rates of bile duct injury and serious complications by approach, and those physicians should ask. A clear benchmark would also give courts a measure when they meet the question the September review raises: who is accountable when the model is wrong.

No robot has yet decided for itself where to clip, cut or sew a patient's soft tissue, which leaves time to set the standard before one does. The machine that chooses where to cut should have to match the surgeon who does that operation well, not the robot that did it before.

This special topic was read by an AI voice. Its sources are linked at physicianintheloop.org.