Therapeutic ultrasound for the knee

A physiotherapist runs a small metal head over a knee through a layer of gel for five or ten minutes. That is therapeutic ultrasound, and it has been in clinical use for decades — long enough that most of the argument about it is not whether the sound reaches the joint, but whether anything follows from it.

The physics splits into two claims. A continuous beam deposits heat, raising local temperature, capillary permeability and tissue extensibility. A pulsed beam is meant to work without appreciable heating, through cavitation and acoustic streaming that change membrane permeability and protein synthesis. Low-intensity pulsed ultrasound is the pulsed form at very low power, and its credentials come from a different bone entirely: it has been cleared in the United States for fresh fracture and nonunion for more than twenty years. The knee argument borrows that mechanism and points it at cartilage.

Everything below is a symptom endpoint unless it says otherwise.

Two meta-analyses of one literature

The clearest way into this evidence is a disagreement, because the disagreement is not random.

The strict pooling allowed only trials in which ultrasound was the sole intervention against a sham. That left five trials and 264 people. Pain improved by a standardized mean difference of −0.33, which the authors convert to a 9.6% improvement on a 0–100 scale, and self-reported function by the same standardized amount, about 12.8%. Overall GRADE quality: very low. Their conclusion is that the evidence is not adequate in size or quality to make a clinical recommendation.

The permissive pooling allowed co-interventions so long as both arms received them, which gave 21 trials and 1,315 people, and larger effects — pain −0.64 and the WOMAC total −0.45, both statistically significant. Every one of its 21 trials carried an unclear risk of bias, and only three of them had more than 40 people per arm.

Neither is wrong. They are answering different questions: what the sound does alone, and what a course of physiotherapy containing it does.

Where it sits among the other devices

Two more recent syntheses ask harder questions of the same trials.

The first tests the way ultrasound is actually delivered — added to an exercise programme, against the identical exercise with a dummy device. Across 40 trials and 2,831 patients, ultrasound reached statistical significance on physical function and the reviewers judged it clinically unimportant, because the confidence interval excluded their six-point WOMAC threshold. Against exercise alone, no device in the review separated from exercise, ultrasound included. Certainty for every ultrasound comparison: low or very low.

The second ranks nine interventions in a Bayesian network of 32 trials. Ultrasound came eighth of nine on the visual analogue pain scale and last of nine on the WOMAC pain subscale, where it was the only intervention that did not reduce the score against exercise at all. Low-level light and shockwave were the two that beat exercise.

Low-intensity pulsed ultrasound

LIPUS is worth separating, because it is the version with a mechanism borrowed from bone healing and the version sold as a home device.

Thirteen trials and 807 people: pain fell by about one point on the visual analogue scale and the WOMAC by about four, with high heterogeneity on the pain estimate. The subgroup finding worth carrying is that the effect was larger when LIPUS was given on its own than when it was added to something else — the opposite of how it is used in a clinic, and a pattern the strict-versus-permissive split above should make a reader cautious about.

Something in that review is easy to miss and matters here. Its own table of outcomes shows that included trials measured Noyes cartilage defects, MRI T2 mapping and femoral articular cartilage thickness. None of those results is pooled or reported. The structural measurements exist, in the primary papers, and the synthesis of this literature has never carried them.

What was measured in cartilage

One human trial inside the pooled literature measured cartilage: central medial femoral cartilage volume at eight weeks, eleven people against twelve, and it did not move. Quality of life did not move either. That is one small trial at a short follow-up, which is a reason to hold the result loosely rather than a reason to call the question settled — but it is what exists.

The forty-trial adjunct review states the wider position for this whole shelf: radiographic joint structure was not reported in any of its trials.

What the animal work shows

In mice given a low dose of monoiodoacetate, four animals per group, four weeks of low-intensity pulsed ultrasound at 30 mW/cm² held the OARSI histological score where the untreated group's rose. Denatured type II collagen was widespread in untreated knees and sparse in treated ones; macrophage counts in the joint capsule fell; lubricin-positive cells rose. Aggrecan-positive cells did not differ. Both authors are employees of the company that funded the study and made the transducer, which is a reason to want the experiment repeated elsewhere rather than a reason to discard it.

The other frequently cited paper is thinner than its title. Its joint-level findings are an arthritis score, weight-bearing and synovial cytokines in mice; its chondrocyte findings are in two immortalized cell lines at 50 and 100 mW/cm². No cartilage histology appears in it at all.

Both used far less power than the human trials, which run up to 2.5 W/cm². Nobody has reconciled that gap.

Phonophoresis is a different question

Phonophoresis uses the same transducer to drive a drug through the skin. Pooled across nine trials and 423 patients, an NSAID gel beat a placebo gel on pain and a corticosteroid gel beat it on WOMAC function.

Both arms of those trials received ultrasound. What the comparison measures is the gel.

Safety

No treatment-related adverse event or discontinuation was reported in any trial in the strict pooling, the permissive pooling, the LIPUS pooling or the phonophoresis pooling. In the forty-trial adjunct review, one trial reported adverse events and none of them were ultrasound events.

What would change this

A trial powered for a structural endpoint, which nobody has run. The cheaper version already exists in fragments: the LIPUS trials that imaged cartilage have never had their structural results pooled, and that synthesis could be written from papers already published. Separately, no trial has randomized continuous against pulsed ultrasound at matched total energy, so the thermal and non-thermal accounts have never been told apart in a human knee.