Dr Matthew BroadheadOrthopaedic Surgeon · Sydney & Coffs Harbour

Journal Club · September 2026 · updated 21 September 2026

Does robotic-assisted knee replacement improve your outcome?

Two 2026 randomised trials found robotic assistance placed the knee implant more precisely, but patients rated their new knee no better at one or two years.

Robotic-assisted knee replacement is heavily marketed, and patients often ask about it directly. If a robot positions the implant more precisely, will my knee replacement work better? When this entry first appeared in September it summarised a 2023 review of twelve randomised trials, and it closed by noting that larger trials using current robotic systems were still reporting. Two of them have now reported, in The Lancet and The Bone & Joint Journal, and this entry has been rewritten around them. One is British and one is Australian. The Australian trial also tests the other live argument in knee replacement, which is how the new knee should be lined up.

What the studies looked at

RACER-Knee is the largest masked trial of a robotic knee system yet published. Across ten NHS hospitals in England, Scotland and Wales, 33 surgeons randomised 339 patients with knee osteoarthritis to a knee replacement done either with the Mako robotic arm or with conventional instruments. The same implant was used in both groups. Patients did not know which they had received: everyone had the same draping, the same small skin incisions (real pin sites for the robot, sham cuts for the conventional group), and a masked operation note. The main outcome was the Forgotten Joint Score at 12 months, a 0 to 100 questionnaire that asks how often you are aware of your artificial knee in daily life, where a higher score means you forget it more. The trial was designed to detect a 12-point difference, roughly one point on each of its twelve questions, and it also costed both operations for the NHS.

RASKAL was run through the Australian joint replacement registry, at nine private and two public hospitals, by 14 surgeons. It randomised 303 patients twice over. First, to surgery with the same Mako robot or with computer navigation, the older, cheaper technology that guides the surgeon’s cuts without a robotic arm. Second, to one of two alignment philosophies. Mechanical alignment aims for a straight leg with the implant set square to the long axis of the bones, and has been the standard for decades. Functional alignment instead restores something close to the patient’s own natural alignment, within safe limits, and uses the technology’s measurements to balance the ligaments with as little cutting of soft tissue as possible. The main outcome was the change in the KOOS-12, a 0 to 100 knee pain-and-function questionnaire, over two years.

What they found

  • Patient outcome, RACER-Knee: no difference. At 12 months the Forgotten Joint Score averaged 49.2 with the robot and 50.2 without it. The adjusted difference was 1.5 points in favour of conventional surgery (95% CI −7.5 to 4.5), and the confidence interval excludes the 12-point difference the trial was looking for. Pain in the first three days, opioid use, time to discharge, the Oxford Knee Score and quality of life were all the same in both groups.
  • Patient outcome, RASKAL: no difference either. Over two years the KOOS-12 improved by a similar amount whether the surgeon used the robot or navigation (difference 2.8 points favouring navigation, 95% CI −6.4 to 0.9), and no secondary measure, including satisfaction, separated the groups at any time point.
  • Alignment philosophy: no difference in how the knee felt, but a real difference in how the operation went. Functional and mechanical alignment produced the same two-year scores (difference 0.3 points, 95% CI −3.4 to 4.0). Mechanically aligned knees, though, needed a soft-tissue release to get the ligaments balanced in 45% of cases, against 8% for functional alignment.
  • The robot did what it is designed to do. In RACER-Knee the implant ended up 0.8 degrees closer to the plan on average (95% CI 0.3 to 1.4), a precision gain that matches the 2023 review. In RASKAL the robot also shortened the operation by 11.5 minutes compared with navigation and caused less visible damage to the posterior cruciate ligament, a structure that the functional-alignment technique keeps. In RACER-Knee the robotic operation took 10.5 minutes longer than conventional instruments. The two findings are not in conflict: the robot is quicker than navigation and slower than a surgeon with jigs.
  • Cost. In the NHS, the robotic admission cost £995 more per patient. Over the first year the extra spend bought no measurable gain in quality of life, and the chance that robotic surgery was cost-effective at the usual NHS threshold was about 4%.
  • Safety. Serious adverse events were identical in RACER-Knee, 16 in each group. There were three deep infections with the robot and one without, numbers too small to mean anything on their own but worth watching in registry data. In RASKAL one patient in the robotic group fractured the bone through a pin site and needed it fixed, and the only two revisions were both in the navigated mechanical-alignment group.
  • One telling aside. RACER-Knee asked patients which operation they thought they had received. Those who believed they had the robot reported a much higher Forgotten Joint Score than those who believed they had conventional surgery, regardless of what had actually been done. That is an exploratory finding, not a result, but it shows why masking patients in a trial like this matters, and why unmasked comparisons of robotic surgery should be read with care.
  • What the surgeons thought. After each RASKAL operation, surgeons said they would have preferred the robot 82% of the time and functional alignment 73% of the time. They liked using them. The knees did not know the difference.

What it means for you

The picture has not changed since September, but it now rests on far stronger evidence. Robotic assistance places the implant closer to the plan, more consistently, than a surgeon working with jigs. What it has not been shown to do, in two well-run randomised trials on top of the earlier meta-analysis, is make the knee feel or work better for the patient in the first one to two years. That is exactly why this practice describes robotic assistance as a tool used where it adds value, rather than as something that by itself produces a better result.

On alignment, the honest position is that either philosophy, done well, gives the same result at two years. Functional alignment gets there with far less soft-tissue release, which is a reasonable thing for a surgeon to want. It is not, on this evidence, a reason for a patient to insist on one approach over the other.

None of this makes robotic surgery a bad idea. A shorter operation than navigation, less handling of the ligaments and tighter precision are real advantages in the operating theatre, and they may matter more in particular knees than they do on average. It does mean that if you are weighing up a robotic against a conventional knee replacement, “more accurate” and “better outcome” are still two different claims. The first is established. The second is not, and the evidence that it is not is now good.

A caveat worth knowing

Both trials report early results. RACER-Knee is following its patients for ten years and RASKAL sits inside the national registry, so the question that matters most over time, whether precision translates into fewer revisions, will be answered by these same patients, but not yet. Both trials studied one robot and one implant, so they say nothing about other systems. RASKAL compared the robot with navigation rather than with plain instruments, and its functional alignment was a restricted version that kept every knee within set limits, not the unrestricted kinematic technique some surgeons use. Both trials also excluded the knees where technology might plausibly earn its keep: severe deformity, previous fracture, complex implants. And each was designed to detect a moderate difference, so a small benefit, or a benefit in a particular group of patients, has not been ruled out. For now the finding is consistent across every randomised comparison available: more precise, yes; a better knee for the patient, not shown.

Assessing this studyA standardised appraisal for clinicians and registrars: study design, strength and validity.
Study design
Two randomised controlled trials of therapy, both published in 2026 and both using the Mako robotic arm with the same cemented implant. RACER-Knee (UK) randomised 339 patients to robotic-arm-assisted or conventional instrumented knee replacement, with patients and outcome assessors masked. RASKAL (Australia) randomised 303 patients in a 2 × 2 factorial design to robotic-assisted or computer-navigated surgery, and separately to functional or mechanical alignment.
Level of evidence
OCEBM Level 2 (therapy) for each trial, as individual randomised trials. The RACER-Knee authors also updated their systematic review to include their own trial and the smaller ROAM trial; that pooled analysis of 439 patients, Level 1 for this comparison, found no difference in patient-reported outcomes at 3 or 12 months.
Are the results valid?
RACER-Knee: patient and assessor masking was protected with sham pin-site incisions, extra draping and masked operation notes; randomisation by minimisation with a 70% random element; intention-to-treat analysis with 92% completion of the primary outcome; the statistical analysis plan was finalised before analysis. Funding came from the NIHR, although Stryker supplied robots, consumables and training and several authors report payments from Stryker. Masking was imperfect: 45% of patients thought they knew their allocation, and those patients were far more likely to guess robotic. RASKAL: nested in the Australian joint replacement registry, with patients, assessors, radiographers and statisticians blinded; randomisation stratified by surgeon; 91.7% follow-up at two years. The primary analysis was a modified intention-to-treat that excluded 31 randomised patients who did not proceed to surgery in the trial window, and no adjustment was made for multiple comparisons. It was funded by the Ramsay Hospital Research Foundation; the lead author reports unrelated fellowship funding from Stryker. Both trials studied one robot and one implant.
What are the results?
RACER-Knee: Forgotten Joint Score at 12 months 49.2 (robotic) against 50.2 (conventional), adjusted difference -1.5 (95% CI -7.5 to 4.5; p=0.62), which rules out the 12-point difference the trial was designed to detect. No clear difference in early pain, opioid use, time to discharge, Oxford Knee Score or EQ-5D-5L at any point. The robot was more precise (hip-knee-ankle angle 0.8 degrees closer to plan, 95% CI 0.3 to 1.4) and the operation took 10.5 minutes longer (95% CI 6.8 to 14.3). The initial admission cost £995 more (95% CI 795 to 1,194); over 12 months the robotic arm cost £958 more for -0.004 QALYs, a 4% probability of being cost-effective at £30,000 per QALY. Serious adverse events 16 against 16; deep infections 3 against 1, small numbers consistent with chance. RASKAL: change in KOOS-12 at two years, robotic against navigated -2.8 (95% CI -6.4 to 0.9; p=0.137); functional against mechanical alignment 0.3 (95% CI -3.4 to 4.0; p=0.867); no difference in Oxford Knee Score, Forgotten Joint Score, EQ-5D, pain or satisfaction at any time point. Robotic surgery was 11.5 minutes shorter than navigated surgery (95% CI 7.3 to 15.9) with less macroscopic injury to the posterior cruciate ligament (MASTI difference 0.7, 95% CI 0.3 to 1.1). Mechanical alignment needed a soft-tissue release in 44.8% of knees against 8.1% with functional alignment (OR 9.2, 95% CI 4.6 to 18.3). Two revisions, both in the navigated mechanical-alignment group.
Do they apply to our patients?
RASKAL is about as close to this practice's patients as a trial gets: Australian, 82% treated in private hospitals, mean age 67, mean BMI 32, operated on by 14 surgeons across nine private and two public hospitals. RACER-Knee ran in ten NHS hospitals in England, Scotland and Wales with 33 surgeons, median age 69. Both trials excluded the knees where technology might plausibly matter most: RASKAL excluded deformity beyond 15 degrees varus or 10 degrees valgus, flexion contracture over 15 degrees, and knees needing stems, augments or a posterior-stabilised design; RACER-Knee excluded inflammatory arthritis, previous fracture and complex implants. Both used the Mako system with a cemented Triathlon implant, so the findings do not transfer to other robots or implants, and RASKAL's comparator was computer navigation rather than conventional jigs.
Bottom line
Two well-run 2026 randomised trials, one British and one Australian, confirm the 2023 meta-analysis: robotic assistance delivers on precision, and in RASKAL on operating time and ligament protection, but neither it nor functional alignment made patients' knees feel or work better at one or two years, and in the NHS setting it cost more without being cost-effective.

Framework: study design graded with the Oxford Centre for Evidence-Based Medicine (OCEBM) Levels of Evidence; validity and applicability appraised in the CASP tradition.

This is a plain-language summary of published research, provided for general education. It is not medical advice and does not describe Dr Broadhead's own results. Whether any finding applies to you depends on your individual circumstances, so please discuss your care with your treating team.

← All Journal Club entries

CallBookEmail