What's Actually Going On
Bacteria reach the joint in one of two ways: through the bloodstream, or by direct inoculation from surgery, injection, a penetrating wound or an overlying skin infection. The synovial membrane has no basement membrane, which makes it unusually permeable to blood-borne organisms and is the anatomical reason joints are seeded so readily from infections elsewhere.
The damage is not done by the bacteria. It is done by the response to them. Neutrophils flood the joint, release proteolytic enzymes and inflammatory cytokines, and those enzymes degrade the cartilage matrix. This is why chondral damage occurs even after prompt and correct treatment, and it is the mechanistic reason the field has spent a decade testing whether adjunctive corticosteroids protect the cartilage while antibiotics deal with the organism.
It is also why time matters. The inflammatory cascade does not pause while a diagnosis is arranged.
How to Identify It
The hallmark is global restriction of both active and passive movement, with the joint guarded against the examiner's hand. In almost every mechanical condition, passive range exceeds active range. Here it does not. The joint is held in the position of maximum capsular volume: hip in flexion, abduction and external rotation; knee in slight flexion.
Leukocyte esterase strip (native joint)Sn 86.5% · Sp 67.9%
Synovial PCRSn 49% · Sp 95.7%
Serum procalcitonin (children)Sn 0.72 · Sp 0.90
Two cautions that matter more than the numbers. Culture-negative infection is the second most common microbiological finding in adult hip series, so a negative culture is not a negative diagnosis. And synovial PCR, despite excellent specificity, misses roughly half of infections, so it can rule in and never rule out.
The Debate
Traditional: a synovial white cell count of 50,000/mm³ is the threshold for septic arthritis.
Recent: Mukerji 2024 (N=567) required 32,000/mm³ for 100% sensitivity. Salazar 2022 (31 studies, 7,434 native shoulders) found sepsis reported at counts as low as 30,000/mm³, attributed to short symptom duration before diagnosis.
Which to follow: treat 50,000 as rule-in only. Two independent cohorts, two different joints, both below the taught threshold, and the count is lowest exactly when the patient presents earliest, which is when the referral is most valuable.
Older: Mathews 2007 called synovial cell count "unhelpful" and ranked an experienced clinician's opinion above every investigation.
Recent: Mukerji 2024 found the count genuinely discriminative, AUROC 0.81 for septic arthritis and 0.87 for the combined septic-and-crystal presentation.
Which to follow: not actually a contradiction, a threshold question. Mathews rejected the count as a binary rule-out at 50,000. Mukerji shows it is informative as a continuous measure. Both reject 50,000-as-a-switch, which is how it is used.
Kennedy 2024: arthroscopic washout gave 20.18° better range of movement (95% CI 14.35-26.02, p<0.00001).
Nudelman 2025 (23 studies, 34,248 patients): no significant re-operation difference in 70% of studies.
Which to follow: both. They answer different questions. Arthroscopy probably does not change whether another operation is needed, and probably does change how the joint moves afterwards. Only one paper in the entire retrieved literature asked the second question.
Honest Limitations
Every diagnostic number here was measured in a population already selected for suspicion. Carpenter's 27% pre-test probability is emergency department patients with a single acutely painful joint. Mukerji's cohort had already had a joint aspirated. A musculoskeletal clinic sees an unselected caseload at a small fraction of that prevalence.
Clinical adjustment: this lowers the referral threshold rather than raising it. The job in a low-prevalence setting is not to estimate a probability but to avoid a miss, and the cost asymmetry between an unnecessary referral and a missed septic joint is enormous.
The literature is surgical and it answers surgical questions. Of 124 papers retrieved for this page, the overwhelming majority compare arthroscopy against arthrotomy or compare antibiotic regimens. Exactly one has range of movement as a pooled endpoint.
Clinical adjustment: what rehabilitation should look like after a septic joint is not answered anywhere in this evidence base. Not answered weakly. Not asked. Post-infection rehabilitation runs on clinical reasoning borrowed from post-surgical practice, and it must be labelled as that.
The pathogen mix and the health system are not transferable. Smith 2002, one of very few randomised trials here, studied 61 children in Malawi where 86% of positive cultures were non-typhoidal Salmonella. Ross 2017 reports MRSA as a major cause in the United States.
Clinical adjustment: use the direction of these findings, not their magnitudes. The direction is consistent everywhere: earlier treatment is better, and the joint is at risk regardless.
The Nuance
This is not a surgery-versus-conservative condition. Every patient receives antibiotics and joint drainage; the only choice is between drainage techniques, and the evidence is genuinely not good enough to settle it. What is not unsettled is the timing. Every source agrees drainage should happen urgently, and none suggests any technique compensates for delay. The physical therapist has no input into the technique and total input into the timing.
The hip in adults is materially worse than the other joints. After surgery for primary adult septic hip arthritis, 13.7% died within one year, 9.8% had a recurrent infection within two years, and 22% had a poor outcome overall. Among the 38 patients with an already-destructive hip, 18.4% died within one year. A single pooled prognosis across all joints would have hidden that inside a reassuring average.
Septic arthritis after ACL reconstruction is largely preventable. Vancomycin-soaking of the graft reduced incidence from 44/2,099 (2.1%) to 0/2,976, odds ratio 0.04 (0.01-0.16), with no difference in IKDC score, Tegner score, revision rate, biomechanical tendon properties or cartilage integrity.
On scans, because it frightens people. In the one randomised trial that followed this up, 70% of children's shoulders showed damage on X-ray at 6 months, and 23 of 24 of those joints had no deficit at all in movement or function at 1 year. That study is old, small, and in a very different setting, so it is not a promise. But a worrying-looking scan does not automatically mean a badly working joint, and that is worth knowing before you see your images.
