Efficacy and Safety of Mesenchymal Stem Cells in Knee Osteoarthritis: A Systematic Review and Meta-Analysis of Randomized Controlled Trials

Major Points and Findings:

Many meta-analyses of stem cells for knee osteoarthritis combine cells with PRP or concentrates, injections with surgery, and randomized trials with cohort studies. This one applies narrower criteria. The authors pooled only randomized controlled trials in which cultured mesenchymal stem cells (MSCs) were injected into the knee on their own, in people who had not had knee surgery. These criteria left 8 trials and 502 patients.

The pooled result favours MSCs, with better WOMAC, pain and KOOS scores than control at 6 and 12 months and no increase in adverse events. Three cautions apply. The benefit was concentrated in adipose-derived cells and in doses of 100 million cells or more. The 12-month function score was no better than control. Not one of the eight trials used umbilical cord cells, so the review provides no direct evidence on that cell source.

Aim:

To assess the efficacy and safety of MSC injection alone for unoperated knee osteoarthritis, and to explore whether cell source or dose changes the result.

Methods:

The authors performed a systematic review and meta-analysis following PRISMA, registered on PROSPERO (CRD42024571190).

They searched Web of Science, PubMed, EMBASE and Scopus to 1 August 2024, English language only, and checked reference lists. Two authors screened and extracted data independently.

Randomized controlled trials of MSC injection without surgery in knee osteoarthritis were included if the comparison was with “regular medication or placebo”.

The review excluded patients with knee replacement or other knee surgery, studies using bone marrow concentrate or PRP, and cohort studies, reviews and animal studies. The authors argue that earlier meta-analyses, which included trials adding microfracture, osteotomy or arthroscopic debridement, could not isolate the effect of the cells.

The primary outcome was WOMAC (total, and pain, stiffness and function subscores) at 6 and 12 months. Secondary outcomes were the 100 mm pain VAS, KOOS, adverse events and serious adverse events.

Fixed-effects models were used where heterogeneity was low (I² below 50%) and random-effects models where it was high. Risk of bias was assessed with the Cochrane tool. Where a trial had several dose arms, they were merged into one treatment group.

Results:

From 12,229 records, 52 were read in full and 8 included, published 2015 to 2023:

• Adipose-derived MSCs (5 trials): Kim 2023, Korea (125 treated, 127 control, 100 million cells); Chen 2021, China (49 and 8; 16, 32 or 64 million); Lu 2019, China (23 and 24; 50 million); Lee 2019, Korea (12 and 12; 100 million); Freitag 2019, Australia (19 and 10; 100 million, once or twice).
• Bone marrow MSCs (3 trials): Bastos 2019, Portugal (16 and 17; 40 million); Lamo-Espinosa 2016, Spain (20 and 10; 10 or 100 million); Vega 2015, Spain (15 and 15; 40 million).

Mean ages ran from about 52 to 70. All but one trial gave a single injection. The Korean trial by Kim supplies half of all patients. Follow-up was 6 to 12 months throughout.

Only 4 of the 8 trials described random sequence generation and only 2 described allocation concealment. Five blinded patients adequately and one was unblinded. Outcome assessment was blinded in four. The funnel plot did not suggest publication bias.

For WOMAC at 6 months (6 trials), MSCs were better than control:

• Pain subscore: mean difference 1.13 (95% CI 0.45 to 1.81, P = 0.001)
• Stiffness: 0.49 (0.11 to 0.86, P = 0.01)
• Function: 3.36 (0.98 to 5.73, P = 0.006)
• Total WOMAC: 7.44 (1.45 to 13.42, P = 0.01), with substantial heterogeneity (I² = 67%)

WOMAC at 12 months (5 trials):

• Pain: 1.03 (0.02 to 2.03, P = 0.04)
• Stiffness: 0.65 (0.02 to 1.27, P = 0.04)
• Function: 0.82 (-4.33 to 5.97, P = 0.76), no significant difference
• Total: 10.31 (0.96 to 19.67, P = 0.03), I² = 77%

The 12-month pain and stiffness intervals only just exclude zero. The paper does not state whether any of these differences reaches a clinically important size.

By dose: Low dose was defined as 10 to 64 million cells, high dose as 100 to 200 million.

• 6 months: high dose 8.14 (4.26 to 12.02, P = 0.002); low dose 3.20 (-1.29 to 7.68, P = 0.16), not significant
• 12 months: high dose 14.25 (7.44 to 21.07); low dose 4.84 (0.45 to 9.24, P = 0.03)

By cell source:

• Adipose: 7.53 (4.42 to 10.63) at 6 months and 8.76 (4.35 to 13.16) at 12 months, both significant
• Bone marrow: -6.30 (-17.45 to 4.85, P = 0.27) at 6 months and 6.00 (-3.32 to 15.31, P = 0.21) at 12 months, neither significant

Pain VAS (0 to 100): The difference was 19.39 points at 6 months (8.10 to 30.68, P = 0.0008; I² = 76%, 5 trials) and 16.21 at 12 months (7.38 to 25.04, P = 0.0003; 4 trials). All five KOOS subscales also favoured MSCs.

Safety: There was no significant difference between MSC and control groups for adverse events (pooled estimate 1.32, 95% CI 0.85 to 2.05, P = 0.21) or serious adverse events (0.78, 0.23 to 2.65, P = 0.69). Most events were pain and swelling after injection and resolved by themselves. The discussion notes that high doses “increased pain and swelling at the injection site”.

Conclusions:

The authors conclude that MSC injection alone was safe and significantly improved knee pain, function and quality of life, that subgroup results favour adipose-derived and high-dose cells, and that larger multicentre trials are needed to settle source and dose.

The limitations they list are that outcomes are self-reported and subjective, that many trials did not report adverse events in detail, that eligibility criteria varied between trials, and that publication bias can never be excluded.

Some further points apply. The comparator is defined only as “regular medication or placebo”. The paper does not list which control each trial used, and all are pooled together. The source and dose subgroups overlap heavily, because every 100 million cell arm except one used adipose cells, so the analysis cannot tell which factor drives the difference. No MRI or cartilage outcome was pooled. The methods mention endpoints at “6 and 12 weeks” where months are meant. Finally, the large MILES trial (Mautner and colleagues, 2023), in which cord tissue MSCs did no better than corticosteroid, is not among the included studies.

Background Information:

To explain why function may lag behind pain, the authors suggest MSCs may add cartilage volume without restoring cartilage quality, citing another meta-analysis, and that repaired tissue may wear again under load. The review did not measure cartilage, so this remains a hypothesis.

On dose, the discussion cites evidence that very high cell concentrations in a narrowed joint can lead to cell death and may push MSCs towards a pro-inflammatory state. A higher cell number is therefore not necessarily better.

The authors call umbilical cord MSCs (UC-MSCs) a promising future product because of low immunogenicity, fast growth and easy large-scale production. They present no data for this expectation. The evidence pooled here comes entirely from cells grown from abdominal fat or bone marrow.