Ultra-processed food

Ultra-processed food is a category, not an ingredient, which usually makes for weak evidence. This is the exception in the foods section: it is the dietary exposure that has been put against a quantified cartilage measurement, at scale, by people whose day job is reading knee MRIs.

The prospective cohort

163,987 UK Biobank participants, 1,461,447 person-years of follow-up, 11,540 incident cases of osteoarthritis. Compared with the lowest quartile of ultra-processed intake — classified by NOVA from 24-hour dietary recalls — the highest quartile had a 10% higher risk of knee osteoarthritis (HR 1.10, 95% CI 1.03–1.18).

Hip osteoarthritis showed no association at all.

Modelling a substitution — replacing 20% of the ultra-processed portion of the diet by weight with unprocessed or minimally processed food — gave a 6% lower knee OA risk (HR 0.94, 95% CI 0.89–0.98). That is a statistical estimate rather than something anyone did.

A hazard ratio of 1.10 is small, and ultra-processed intake travels with income, education, smoking, physical activity and total energy. Effects of that size are routinely produced by residual confounding in nutritional cohorts, and the knee and hip finding differently with no mechanism offered for the split is the shape that suggests it. The study is worth its place because of its size and its prospective design, not because a 10% relative risk settles anything.

The cartilage measurement

In 4,403 Osteoarthritis Initiative participants, ultra-processed proportion of the daily diet was set against WOMAC scores, chair-stand performance, gait speed and average cartilage thickness quantified from 3T MRI — adjusted for age, race, body mass index, daily calories, physical activity and insurance status.

In women, every one of those moved with more ultra-processed food: pain (β = 0.17, 95% CI 0.093–0.242), activities of daily living (β = 0.59, 0.365–0.832), total WOMAC (β = 0.81, 0.483–1.13), gait speed (β = −0.035, −0.042 to −0.027), and cartilage thickness (β = −0.013, −0.02 to −0.006), all p < 0.001. The sex interactions were significant. Chair-stand performance was worse with more ultra-processed food in both sexes; the stiffness association did not survive correction for multiple comparisons.

This is cross-sectional. Diet and cartilage were measured at the same moment, so neither can be shown to have come first, and a knee that already hurts changes what a person cooks. What the analysis has that almost nothing else in this section has is an endpoint measured in millimetres of cartilage rather than in answers to a questionnaire.

The route that isn't cartilage

The third study is the one that suggests a mechanism. In 615 OAI participants who were at risk of knee osteoarthritis but had neither radiographic disease (Kellgren-Lawrence ≤ 1) nor pain in either knee or hip, ultra-processed food made up an average of 41% of the diet — and the more of it someone ate, the more fatty infiltration their thigh muscles showed on T1 MRI. All thigh muscles β = 0.108 (95% CI 0.029–0.186, p = 0.007); flexors β = 0.111; adductors β = 0.122. The associations were stronger when abdominal circumference replaced body mass index in the model.

Thigh muscle is what unloads a knee, and fatty infiltration degrades its quality independently of its size. A diet-to-muscle-to-load route does not require food to act on cartilage at all — and this finding sits in people who have not yet developed the disease, which puts it upstream. It is also the one analysis here with no sex interaction, in a cohort where the cartilage analysis had a strong one.

What the adjustments are doing

Every one of these studies adjusts for body mass index, and that adjustment is doing uncertain work. If ultra-processed food acts on joints partly by adding body weight — which is the best-evidenced route to a worse knee anywhere on this site — then adjusting for weight removes part of the effect being measured. The associations reported above are therefore what is left after the most plausible mechanism has been statistically taken out, which cuts both ways: it makes them harder to explain, and it makes them harder to dismiss.

What would change this entry

The longitudinal analysis, in the cohort that already has the data. The Osteoarthritis Initiative has repeat MRI on these participants; running ultra-processed intake against change in cartilage thickness rather than against one cross-section would answer, in one paper, whether this is a finding or a correlation. And if the muscle-fat route holds up, this entry has as much to do with the strength section as with the food one.