The imaging, the evidence, and the honest comparison between injections, exercise and joint replacement.
What "Bone-on-Bone" Actually Means
When a surgeon points at an X-ray and says "bone-on-bone", they are describing Kellgren-Lawrence grade 3 or 4 — the radiographic end of a scale published in 1957 [1]:
Source: Kellgren and Lawrence, 1957 [1]
That grade describes structure. It does not describe how much your knee hurts.
The imaging–symptom disconnect is real and well documented
In the Framingham Osteoarthritis Study, among adults aged 63–94, approximately 33% had radiographic evidence of knee osteoarthritis while only about 9% reported symptomatic disease [3]. A systematic review of the discordance between clinical and radiographic knee OA found the same pattern across the literature [2]: many people with severe-looking films have manageable pain, and many with mild films are significantly limited.
What follows from this, and it matters more than anything else on the page: an X-ray grade alone does not determine what treatment you need. Your symptoms, your function, what you are trying to do, and what you have already tried matter more than the image. That applies in both directions — it is not a reason to avoid surgery any more than it is a reason to have it.
Before the Evidence: Why Placebo Matters Here
Knee osteoarthritis is strongly placebo-responsive. Injections in particular carry a large contextual effect — the needle, the clinic, the expectation, the attention, and the natural fluctuation of joint pain all contribute to what a patient feels afterwards.
This has been quantified. Yin and colleagues meta-analysed 8 randomised trials (467 patients) comparing intra-articular MSC injection with inert placebo, and calculated the proportion of the treatment effect attributable to contextual factors [7]:
Heterogeneity was very low (I [2]≤ 8% at 6 months, 0% at 12 months). GRADE certainty was rated low.
Their conclusion: "the majority of symptomatic improvement following intra-articular MSC injections is [attributable to contextual factors]."
Read every number below with that in mind. It does not mean the treatment does nothing — a real effect remains after the contextual fraction is removed. It means that roughly half of what someone reports feeling is not the cells.
What the Cell Therapy Evidence Shows — Both Sides
The evidence here genuinely disagrees, and a page that resolved the disagreement in either direction would be misleading.
The case in favour
Awad et al. 2026 — the most recent synthesis, 28 randomised controlled trials [6]:
The authors describe these as "modest improvements in pain outcomes… compared with control injections", and add two qualifications that belong with them: the role is "primarily symptom-modifying rather than structure-modifying", and "higher frequencies of local reactions must be weighed against symptomatic benefits."
Pas et al. 2017 reviewed the trials available at that point and found that all five RCTs reported superior efficacy on patient-reported outcomes at 24–48 months, with superior radiological outcomes and no serious adverse events [4].
The case against
Pas's own conclusion. Every trial in that review was at high risk of bias, producing level-3 evidence. The authors' verdict:
"Six trials with high risk of bias showed level-3 or level-4 evidence in favour of stem cell injections in KOA. In the absence of high-level evidence, we do not recommend stem cell therapy for KOA." [4]
Dai et al. 2021 — a meta-analysis restricted to 13 Level-I randomised trials, titled "Intra-Articular Mesenchymal Stromal Cell Injections Are No Different From Placebo in the Treatment of Knee Osteoarthritis" [5]:
- VAS pain: MD 1.62 (95% CI −0.60 to 3.85) — not significant
- WOMAC pain: MD 1.88 (−0.21 to 3.98) — not significant
- WOMAC function: MD −0.67 (−6.54 to 5.19) — not significant
- On every outcome, the smallest treatment effect failed to exceed the minimal clinically important difference
- Significant only against hyaluronic acid, not against placebo
*ADIPOA-2 (Pers et al. 2025, Annals of the Rheumatic Diseases)* — a phase 2b double-blind trial randomising 135 patients to low-dose cells, high-dose cells, or saline [8]:
At 6 months, OARSI/OMERACT strict responders: 26 patients (47.3%) with cells versus 23 (54.8%) with placebo — relative risk 0.86 (95% CI 0.58–1.28), P = .46. Overall secondary outcomes showed no significant between-group differences.
The placebo arm did numerically better.
How long does it last?
Jafar et al. 2026 — a randomised, double-blind, placebo-controlled trial in stage III knee OA, with blinding maintained for 12 months and the treated arm followed to 60 months [9]:
- Significant improvement versus baseline at 6, 12, 24 and 36 months (P < .0001)
- MRI structural improvement at 12 months (P < .02)
- "Clinical improvements declined steadily after 36 months, with scores nearing baseline at 60 months"
Their conclusion: benefit "for up to 3 years". Uncontrolled series followed for longer often report more durable benefit — which is precisely the difference a control group makes.
What this adds up to
Positive trials exist and their effect sizes are real. The largest Level-I synthesis found no difference from placebo. About half the observed benefit is contextual. The benefit that remains is symptomatic, modest, and fades over three to five years. All of those statements are true at once, and anyone presenting only one of them is not describing the evidence.
Other syntheses report favourably — a network meta-analysis found adipose MSCs and leukocyte-poor PRP associated with better outcomes than hyaluronic acid and saline [10], and smaller meta-analyses report sustained pain and function improvement [11][12]. The distinction between "better than hyaluronic acid" and "better than placebo" is the distinction Dai's analysis turns on, and it should not be collapsed.
Mechanism. MSCs are not understood to rebuild cartilage. They act through paracrine signalling — anti-inflammatory cytokines, growth factors and extracellular vesicles that influence the joint environment [13]. That is why the outcome measures that move are pain and function, and the ones that mostly do not are structural.
The Alternatives, Reported at Full Strength
Knee replacement
This is the treatment with the largest measured effect on this page, and it has been tested against non-surgical care in a randomised trial.
*Skou et al. 2015 (New England Journal of Medicine) randomised 100 patients with moderate-to-severe knee OA who were eligible for total knee replacement* to either TKR followed by 12 weeks of non-surgical treatment, or the same 12 weeks of non-surgical treatment alone [14]:
- KOOS4 improvement at 12 months: 32.5 versus 16.0 — adjusted mean difference 15.8 (95% CI 10.0–21.5)
- ⚠️ Serious adverse events: 24 in the replacement group versus 6 in the non-surgical group
- 13 of 50 patients (26%) in the non-surgical group went on to have TKR within the year
Skou et al. 2018, at two years across two parallel trials (n = 200) [15]:
- TKR versus non-surgical treatment alone: 18.3 points (95% CI 11.3–25.3)
- Non-surgical treatment versus written advice alone: 7.0 points (95% CI 0.4–13.5) — the supervised programme beats being told what to do
- Among TKR-eligible patients, 16 of 50 (32%) in the non-surgical group had surgery within two years — meaning two in three did not
Both halves of that matter. Knee replacement works better than anything else here, by a margin no injection approaches — and it carries four times the serious adverse events. Meanwhile, two thirds of people who qualified for it deferred or avoided it for at least two years after twelve weeks of structured care.
Exercise therapy
Schleimer et al. 2026 — an overview of 5 systematic reviews (n = 8,631) and 28 supplementary trials (n = 4,360) [16]:
- Knee OA pain, exercise versus placebo: MD −10.8 (95% CI −19.1 to −2.6); versus no treatment: −12.4 (−15.6 to −9.2)
- Certainty very low, and effects in larger and longer trials were smaller
- "Comparable outcomes to education, manual therapy, analgesics, injections and arthroscopy"
- "Exercise was less effective than knee osteotomy (12.4) and joint replacement (knee 17.1, 95% CI 10.4–23.8) at longer term"
Honest reading: exercise produces a small benefit on low-certainty evidence, roughly comparable to what injections achieve — and less than joint replacement over the long term. It is also cheap, available, and carries essentially no risk of serious harm.
Corticosteroid and hyaluronic acid injection
Corticosteroid injection provides genuine short-term relief but the benefit fades within months, and over two years of repeated injection triamcinolone produced significantly greater cartilage volume loss than saline with no pain advantage [17]. Hyaluronic acid is the comparator against which MSC therapy does show a significant difference [5]— which is a weaker benchmark than placebo.
Platelet-rich plasma
A point of calibration: the ESSKA-ORBIT European consensus issued grade-A recommendations — its highest evidence tier — supporting PRP in knee osteoarthritis for KL ≤ 3, and stating that PRP provides a longer effect than corticosteroid with a safer profile. There is no equivalent grade-A recommendation for MSC therapy. On current evidence PRP is the better-supported orthobiologic for this indication, not the lesser one.
Surgical options in brief
The Comparison Nobody Has Made
No trial has compared cell therapy with knee replacement, with osteotomy, or with any surgical procedure. No trial has compared it head-to-head with a structured exercise programme.
That means the question "should I have this instead of a knee replacement?" has no evidence-based answer. What can be said:
- Knee replacement has the largest randomised effect size and the highest serious adverse event rate
- Structured non-surgical care allowed two in three surgery-eligible patients to defer surgery for at least two years
- Cell therapy produces a modest symptomatic benefit, about half of it contextual, lasting around three years
- Cell therapy does not preclude later surgery
Anyone weighing these should be doing so with a surgeon who has seen their knee, their films and their function — not on the basis of a web page.
Realistic Expectations
Will it regrow cartilage? No. MSCs act by paracrine signalling, not by regenerating tissue [13]. The current 28-trial synthesis is explicit that the role is symptom-modifying rather than structure-modifying [6]. Structural improvements reported on MRI have been small and inconsistent.
How long will it last? Controlled follow-up shows benefit to about three years, declining toward baseline by five [9].
How much of it is the treatment? Roughly half at twelve months [7].
Who has the best evidence behind them? Trials have mostly enrolled Kellgren-Lawrence grades 2–3. Grade 4 is far less studied.
Who should not be considering it? Anyone with severe deformity requiring mechanical correction, active infection, inflammatory arthritis, or an expectation of cartilage regeneration.
Questions Worth Asking
- "What is my function, not just my grade?" — the X-ray is a poor guide to what you can do
- "Have I completed a proper course of supervised exercise?" — twelve weeks of structured care, not advice, is what the trials tested
- "How much of the expected benefit is contextual?" — for injections in this condition, roughly half
- "Is the evidence you're quoting against placebo, or against hyaluronic acid?" — the answers differ
- "How long does it last, and what happens then?"
- "If I have this, does it change my surgical options later?"
- "What are the serious adverse event rates for each option?"
The Bottom Line
"Bone-on-bone" is a description of a photograph. It is not a sentence, and the weak correlation between what an X-ray shows and what a knee feels like is one of the better-established facts in this field [2][3].
Cell therapy for knee osteoarthritis produces a modest, symptomatic, temporary benefit. The most recent 28-trial synthesis puts it at about 1.7 points on a VAS and 15 points on KOOS pain [6]; the largest Level-I synthesis finds no difference from placebo at all [5]; roughly half of what patients feel is contextual [7]; and in the one placebo-controlled trial followed for five years, the benefit was gone by the end [9]. The systematic review most often cited in its favour concluded "we do not recommend stem cell therapy for KOA" [4].
Knee replacement works better than anything else and harms more people doing it [14][15]. Twelve weeks of structured non-surgical care allowed two in three surgery-eligible patients to avoid surgery for two years [15]. Exercise helps modestly, costs little, and harms almost no one [16].
Those are the real options, at their real sizes. A patient with moderate knee OA who has genuinely exhausted conservative care, understands that about half the benefit is contextual and that it will likely fade within a few years, and who is not deferring a surgical decision they need to make, is making a defensible choice. Those conditions are the point.