What difference was this trial powered to see?
The published protocol fixed the target before any patient was operated on. RACER-Knee set the Forgotten Joint Score at 12 months as its primary outcome, chose a 12-point target difference with an assumed standard deviation of 30 points — a moderate effect size of 0.4 — and calculated that 266 participants would give 90 per cent power at a two-sided error rate of 5 per cent. Allowing for up to 20 per cent loss to follow-up brought the recruitment target to 332. In the end 339 patients were randomised at 10 hospitals under 33 surgeons.[1]
At twelve months there was no meaningful difference on that score. Patients in both arms reported similar awareness of the joint, similar walking ability, similar pain in hospital and at three and twelve months, and a similar likelihood of further surgery. The robotic operations ran 10.5 minutes longer on average and cost about 950 pounds more, and under current NHS thresholds the system did not earn that cost back across the first year. It was not associated with more serious adverse events.[1]
What can a null result on this endpoint say?
A trial powered for a 12-point gap agreed in advance to treat anything smaller as uninformative. Robotic assistance, in RACER-Knee, did not produce the difference patients and payers were told to expect. That leaves the two operations unseparated on this measure rather than proven identical — the same distinction that separated a primary endpoint from a pooled headline in the digoxin trial I read on 17 August.[1], [2]
The likeliest reading is that the precision gain has no route to a patient-reported score in the first year, because within twelve months both groups have largely recovered and the Forgotten Joint Score compresses at the top. The alternative worth holding open is that the benefit is real but slow: alignment errors of a few degrees plausibly act through wear and loosening, which take years to declare themselves and would surface first in revision rates.[1]
Who carries the added cost while the question stays open?
Robotic assistance is already used in 6 per cent of UK knee replacements, 16 per cent in the United States and 42 per cent in Australia. The trial's question cannot be whether to start; it asks what the spread already under way is buying. At about 950 pounds a case, that is a cost a health system pays today against a benefit that has not yet been shown. The trial ran in partnership with Stryker, which makes the Mako system that was tested, and it still reported a null primary result — a device trial reporting against its own sponsor's product earns trust on this point.[1]
The measurable question now sits in the follow-up. The same patients are being followed to ten years, and revision and reoperation rates were prespecified as secondary outcomes, so the wear-and-loosening route to benefit has a scheduled test rather than an argument. If that follow-up reports on the randomised cohort with the same allocation, I expect the reoperation and revision difference between the arms to stay inside the range the twelve-month data already show, and a clear separation would be the finding that reopens the cost case. Until then the honest statement the evidence carries is that the robot makes the cut more precise and that no gain to the first year after surgery has been shown.[1]