Spinal Epidural Abscess

The Verdict

Back pain that ignores everything you do, plus one infection risk, is worth a blood test today.

Think of the space around your spinal cord as the packing foam inside a shipping box. If an infection collects in that foam it swells, and the cord inside gets squeezed and has its blood supply pinched off at the same time. That second part is why someone can go from sore to weak faster than the size of the collection would suggest, and it is why the clock matters more than the scan does.

What this actually is: an infection collects in the space around the spinal cord, and it starts out feeling like an ordinary bad back.
The myth that won't die: clinicians are taught to look for back pain plus fever plus nerve symptoms together, and only about one patient in ten ever has all three.
What to watch for: new weakness, numbness between the legs, or trouble passing urine means the emergency department today, not next week.

Best for

Anyone with back pain that does not ease with rest or position changes, especially alongside diabetes, a recent infection anywhere in the body, a recent spinal injection or operation, or injected drug use.

Skip if

Your back pain is clearly load-related, eases when you rest or change position, and is settling as expected. That is ordinary back pain, and this is not your condition.

Want the full evidence? Keep scrolling

What Works

Cinematic anatomical rendering of the lumbar spine and epidural space

There is no physical therapy treatment for this condition, and that is the finding rather than a gap. Every tier below is delivered by a medical team. The physical therapist's entire contribution is recognising this is not mechanical and moving the patient into the medical pathway before a deficit arrives.

Urgent contrast MRI and blood cultures Strong

The diagnostic step, and the one a physical therapist actually influences by referring. MRI is described as the most sensitive and specific modality available. (Tetsuka 2020, PMID 33324773; Reihsaus 2000, PMID 11153548)

Targeted intravenous antibiotics after a sample Strong

Organism-directed and prolonged, measured in weeks. The guideline underpinning it (IDSA 2015, PMID 26229122) is written for vertebral osteomyelitis rather than this condition specifically, and is 11 years old.

Surgical decompression and drainage Strong indicationWeak timing

For neurological deficit, progressive deterioration, instability, sepsis, or failed medical management. 69% of pooled patients underwent surgery (Budiman 2026, PMID 41479378). Tuchman 2014 states outright that the point at which neurological injury becomes irreversible is unknown, which is exactly why no defensible waiting window exists (PMID 25081968).

Tier 2 and Tier 3 — non-operative management and later rehabilitation

Non-operative management in selected patients Moderate

Antibiotics with or without image-guided aspiration, in patients without deficit, with a known organism, and without failure risk factors. Pooled failure of medical management was 29.3% (95% CI 21.4 to 37.2) (Stratton 2017, PMID 27636865) and 29.40% (Gardner 2021, PMID 32684428), two independent reviews landing within 0.1 of each other. Stratton warns the estimates are highly heterogeneous and that no consensus definition of failure exists.

Reconditioning after the infection is controlled Emerging

General post-illness and neurological rehabilitation reasoning, labelled as such. No trial of physical therapy in this condition was retrieved, and it is not condition-specific.

Exercise Prescription

There are no exercises for this condition, and the absence is the instruction. No exercise, stretch, or hands-on treatment helps a spinal infection, and doing the wrong thing here wastes the only resource that matters, which is time. Spinal manipulation is contraindicated the moment infection is suspected. Loaded spinal training stops until infection is excluded. Reconditioning is planned by the medical team afterwards.

What Doesn't Work

Screening on the classic triad. It identifies roughly one case in eight, and by the time all three components are present the deficit that determines the outcome has usually already happened.
Being reassured by a normal neurological examination. 40% present with no deficit at all (Budiman 2026, PMID 41479378).
Being reassured by the absence of fever. Absent in about 50%, and recorded in only 19.7% of cases in a modern emergency cohort (Long 2022, PMID 35063888; Davis WT 2020, PMID 31128933).
Being reassured by the absence of risk factors. Long 2022 states explicitly that their absence cannot exclude this condition.
Treating night pain as a specific infection flag. Without an infection history or fever it is a false positive for infection more than 96% of the time (Premkumar 2018, PMID 29509613).
Spinal manipulation or aggressive loading on suspicion. No trial exists and none ever will. Contraindicated.
Lumbar puncture. Not needed for diagnosis, and it risks causing meningitis (Reihsaus 2000, PMID 11153548).

Red Flags

Dark cinematic rendering of spinal cord compression

Any new or progressive motor weakness, sensory loss, or gait disturbance. Diagnostic delay was associated with residual motor weakness in 45% of patients versus 13% without delay (odds ratio 5.65, 95% CI 1.15 to 27.71) (Davis 2004, PMID 15028325).
Bladder or bowel disturbance, or saddle numbness. Cord or cauda equina compression.
Fever with spinal pain. But fever is absent in about half of cases, so its absence is not a downgrade.
Non-mechanical spinal pain plus any named infection risk factor, even with a completely normal neurological examination, because 40% present with no deficit.
A patient who is systemically unwell with spinal pain. Sepsis.
Spinal pain with known bacteraemia, endocarditis, or a recent spinal injection or operation.

Refer to the emergency department the same day for any neurological deficit, bladder or bowel change, saddle numbness, or a febrile unwell patient. Telephone ahead and say you are concerned about a spinal epidural abscess. Otherwise make an urgent medical referral explicitly requesting ESR, CRP, blood cultures and MRI, and name suspected spinal infection on the referral. Do not treat and review.

Return to Training

These apply only after the infection is treated and the medical team has cleared return to activity. No published return-to-activity protocol for this condition exists, so every criterion below is clinical reasoning rather than trial-tested.

Medical team has confirmed the infection is treated and cleared return to loaded activity
Inflammatory markers (ESR and CRP) normalised, or the medical team is satisfied with the trend
Off intravenous antibiotics, or explicitly cleared to train while on oral therapy
No neurological deficit, or a stable documented deficit with a rehabilitation plan in place
Pain-free full active spinal range of motion
Can perform bodyweight movement patterns without pain or disproportionate fatigue
Two consecutive weeks of light activity with no return of night pain, rest pain, or systemic symptoms
Progressive reloading from the lightest available load, with a documented plan

Conviction

Moderate

Split by claim, because the underlying evidence is not uniform and averaging it would be misleading.

High The classic triad appears in roughly one in ten cases. Five sources converge.
High Diagnostic delay is common and is associated with worse neurological outcome.
High Risk-factor screening substantially outperforms the triad: 98% against 13%, in the same study, the same patients, and the same matched controls.
Moderate ESR is the right screening blood test and beats CRP. One centre, one substudy, never replicated here.
Moderate Non-operative management fails in roughly 29% of selected patients.
Low Any specific operative timing threshold is correct.
Not actionable Physical therapy treatment of the condition. There is none.

What would change my mind on the screening claim

A prospective diagnostic-accuracy study in first-contact physical therapy or primary care rather than an emergency department, enrolling consecutive adults with spinal pain (tens of thousands, given a 0.01% prevalence), applying a pre-specified risk-factor checklist and ESR/CRP at intake with MRI-confirmed diagnosis as the reference standard, and reporting sensitivity, specificity, likelihood ratios and above all the number needed to refer per case detected. That last number is the single most important missing quantity in this entire literature, and it is what decides whether a 98%-sensitive screen is deployable at primary-care prevalence.

What would change my mind on ESR over CRP

A multi-centre replication of the ESR-versus-CRP head-to-head in at least 300 risk-factor-positive spinal-pain patients with MRI as the reference standard, reporting both areas under the curve with confidence intervals. The current preference rests on a single centre reporting 0.96 against 0.81, and has not been replicated.

The Full Picture — Anatomy, Diagnosis & Evidence

What's Actually Going On

Cinematic cross-section of the epidural space

The epidural space is a fat-filled sleeve between the dura, the sac holding the spinal cord and nerve roots, and the bony wall of the spinal canal. Bacteria reach it three ways: through the bloodstream from an infection somewhere else in the body, by direct spread from an infected vertebra or disc next door, or by direct inoculation during a spinal injection or operation.

Staphylococcus aureus dominates, reported in 27% of spinal infection cases (Yusuf 2019, PMID 31836000) and as the most common isolate in a 101-case series where MRSA accounted for 25% of those (Vakili 2017, PMID 28797646).

Once pus collects, the cord is injured two ways at once: direct mechanical compression, and compromise of the cord's own blood supply. That second mechanism is why a patient can deteriorate faster than the size of the collection on the scan appears to justify, and it is why "the collection is small" is not reassurance.

How to Identify It

Cinematic anatomical study of the lumbar spine

There is no orthopaedic special test for this condition. The tests that carry diagnostic weight are a history question set and a blood test.

Risk-factor assessment, at least one factor present Sn 98%Sp ~79% — diabetes, recent infection anywhere, IV drug use, recent spinal injection or surgery, indwelling line, immunosuppression, malignancy (Davis 2004, PMID 15028325)
Classic triad (spinal pain + fever + neurological deficit) Sn 13%Sp ~99% — rules in, cannot rule out, useless as a screen (Davis 2004, PMID 15028325)
ESR in risk-factor-positive patients Sn 100%Sp 67% — area under the curve 0.96, against 0.81 for CRP (Davis 2011, PMID 21417700)
Gadolinium-enhanced MRI — the definitive test, ordered by the medical team
Lumbar puncture — contraindicated, risks causing meningitis

The first two rows read together are the whole point. The test the profession is taught to apply has a sensitivity of 13%. The test almost nobody calls a test, asking about risk factors, has a sensitivity of 98%, in the same patients, in the same study, against the same matched controls.

The Debate

No condition-specific clinical practice guideline exists as of August 2026. No NICE, APTA, BOA, EULAR or ACR guideline covers it directly. The nearest guideline-class document, IDSA 2015 (PMID 26229122), is 11 years old and written for vertebral osteomyelitis rather than this condition. Both are flags, and it must not be presented as a guideline for this condition.

Triad screening versus risk-factor screening. Emergency medicine made this switch years ago. Alerhand 2017 calls risk-factor-based evaluation "the current optimal strategy" (PMID 28779448), and Schwab 2020 found risk-factor assessment to determine the need for MRI reduces diagnostic delays compared with relying on clinical or laboratory findings alone (PMID 32694325).

The finding that matters most to this profession. Yusuf 2019 (PMID 31836000) is the only physiotherapy-authored systematic review here, commissioned and funded by the Chartered Society of Physiotherapy. Its abstract conclusion and its conclusion section both state that the classic triad was "the most frequently reported clinical feature." Its own discussion says the opposite, and is correct: it reports the triad at 8% sensitivity, reports risk-factor determinants at 98%, calls them "a better predictor," and warns that relying on the triad "is likely to result in missed cases."

The mechanism is worth understanding because it is easy to repeat. The review counted how often each of 23 clinical features was reported across papers, finding spinal pain in 72% and fever in 55%. That is component frequency, not how often all three appear together in one patient. Two features at 72% and 55% are entirely compatible with a triad under 15%, which is what everyone else measured.

Nothing new had to be discovered for this profession to update. The corrective was already inside its own commissioned document, four paragraphs above the conclusion.

Honest Limitations

Every accuracy figure here comes from an emergency department, and a physical therapy caseload is not one. Infections were 0.01% of 1,568,704 primary-care physical therapy referrals (Budtz 2021, PMID 34034209), roughly one in ten thousand. A 98%-sensitive, ~79%-specific screen at that prevalence refers many well patients, and nobody has published the number needed to refer per case detected, so that cost is unquantified rather than acceptable. It is still the right screen, because the alternative finds one case in eight.

The screening rule has never been tested in a physical therapy setting, and its only implementation study is uncontrolled. Davis 2011 compares a 9-year control period against a 5-year study period at one centre. Over those fourteen years MRI availability, referral culture and awareness all changed, and none of it was controlled. The effect is far too large for secular trend alone to explain, but the design cannot separate the rule from the era.

The entire treatment literature is Class III. Tuchman 2014 reviewed 28 case series of at least 30 patients each and graded every one as Class III evidence, all but two retrospective (PMID 25081968). No randomised trial of operative versus non-operative management exists, and given the rarity and severity, none is likely.

The Nuance

Cinematic anatomical rendering of spinal structures

Surgery versus conservative management, and why the obvious reading of the statistics is wrong. Wang 2019 found surgery itself was not associated with neurological outcome (odds ratio 1.01, 95% CI 0.40 to 2.59) (PMID 30589647). That sounds like an argument against operating and is not one: in non-randomised cohorts the sicker patients get operated on, so the surgical arm carries the worse cases. This is confounding by indication, and it must not be used to justify waiting.

A printed error worth knowing about. The same paper gives delayed surgery an odds ratio of 0.01 with a 95% CI of 0.02 to 0.62. The point estimate lies outside its own confidence interval, which is arithmetically impossible. The direction of the finding is consistent with the rest of the literature, so quote the direction and never the number.

What the evidence does say clearly. Roughly three in ten carefully selected non-operative patients fail and cross over to surgery: 29.3% (Stratton 2017, PMID 27636865) and 29.40% (Gardner 2021, PMID 32684428) across two independent reviews. Delay is associated with worse neurological outcome. And nobody knows the hour at which a deficit becomes permanent. For a physical therapist none of this changes the action: the referral is the same either way, and it is urgent either way.

Sources

Davis DP, 2004, J Emerg Med. PMID 15028325. 63 cases hand-matched to 126 emergency department controls. Triad present in 13% of cases versus 1% of controls; one or more risk factors in 98% versus 21%; diagnostic delay in 75%; residual motor weakness 45% with delay versus 13% without (OR 5.65, 95% CI 1.15 to 27.71). The foundational comparison.
Davis DP, 2011, J Neurosurg Spine. PMID 21417700. Implementing a risk-factor-then-ESR/CRP guideline cut diagnostic delays from 83.6% to 9.7% and motor deficit at diagnosis from 81.8% to 19.4%, both p<0.001. ESR sensitivity 100%, specificity 67%; AUC 0.96 against CRP 0.81.
Yusuf M, Finucane L, Selfe J, 2019, BMC Musculoskelet Disord. PMID 31836000. CSP-commissioned systematic review, 40 papers, 2,224 spinal infection cases. Spinal pain 72%, fever 55%.
Budtz CR, 2021, Physiotherapy. PMID 34034209. 1,568,704 primary-care physical therapy courses. Serious pathology 2.30% overall; infections 0.01%.
Long B, 2022, Am J Emerg Med. PMID 35063888. Classic triad in under 8% of cases; up to 90% misdiagnosed on first emergency visit; fever absent in 50%.
Budiman L, 2026, Emerg Med Australas. PMID 41479378. Systematic review and meta-analysis, 79 manuscripts. Mortality 11%; 40% presented with no neurological deficit.
Reihsaus E, 2000, Neurosurg Rev. PMID 11153548. Meta-analysis of 915 published cases. Incidence 0.2 to 2 per 10,000 admissions; back pain the initial symptom in 71%, fever 66%, paralysis 34%.
Premkumar A, 2018, J Bone Joint Surg Am. PMID 29509613. 9,940 low back pain patients. Night pain a false positive for infection more than 96% of the time absent infection history and fever.
Stratton A, 2017, J Neurosurg Spine. PMID 27636865. 12 studies, 489 medically treated patients. Pooled failure of medical management 29.3% (95% CI 21.4 to 37.2).
Tuchman A, 2014, Neurosurg Focus. PMID 25081968. 28 case series, all graded Class III evidence. The point at which neurological injury becomes irreversible is unknown.
Berbari EF, 2015, Clin Infect Dis. PMID 26229122. IDSA guideline for native vertebral osteomyelitis. Not specific to this condition, and 11 years old.