Osteomyelitis

If you have a wound or ulcer over a bone that is not healing, get it looked at this week. If you can see bone in it, go today.

A sore deep enough to reach bone is treated as a bone infection until a specialist says otherwise. In a diabetic foot, that finding raises the odds of amputation almost fourfold.

Takes one phone call. No equipment needed.

The Verdict

A sore that will not heal, or deep bone ache after surgery, needs a doctor, not exercise.

Antibiotics travel through your bloodstream, so they only reach living tissue. When an infection kills a patch of bone, that patch loses its blood supply, which means the drug arrives everywhere except the one place the bacteria are actually living. That is why this usually needs surgery to physically remove the dead piece, and why waiting almost never works on its own.

Here is what is really happening: an infection has got inside the bone itself, usually through a wound that will not close or an operation that left metal in a broken bone.
The one thing that makes it worse: waiting. The pain often does not get worse when you use the limb, so it never feels urgent, and that is exactly why people leave it for months.
What to watch for: a sore deep enough that a doctor can touch bone through it. That one finding is treated as a bone infection until proven otherwise.

Best for

Anyone with a non-healing wound over a bone, or deep unexplained bone pain after a fracture operation, who wants to know how urgent this is.

Skip if

Your pain is in your spine. That is a different condition with different tests. And nothing here replaces being seen, because this is not a self-treatment protocol.

Want the full evidence? Keep scrolling

What Works

Dark cinematic surgical anatomy of bone debridement

There is no physical therapy treatment for bone infection. Nothing in the physical therapy toolkit clears a biofilm on dead bone. This hierarchy is medical and surgical, recorded so you know what happens after the referral.

Surgical removal of all infected and dead bone STRONG

Source control is the treatment. Pooled cure 91 percent across 42 studies and 1,605 patients, with no significant difference between single-stage and two-stage surgery (Lari 2024).

Expected timeline: months, not weeks.

Culture-directed antibiotics, sampled before empiric treatment STRONG

Targeting requires a sample. Starting antibiotics first can make that sample uninformative (IWGDF/IDSA 2023, Senneville 2024).

Complete margin clearance in the diabetic foot MODERATE

Infected bone left at the resection margin raises amputation risk 4.3-fold (95% CI 2.4 to 7.6), re-infection 2.0-fold, and adds 16.3 days of antibiotics (Reyes 2024).

Non-removable knee-high offloading for a neuropathic plantar ulcer STRONG

First choice per the IWGDF 2023 guideline (Bus 2024). Healing risk ratio 1.24 (95% CI 1.09 to 1.41) versus removable devices. Carry the cost with it: device-related complications risk ratio 1.70, and wearers move measurably less.

Healing measured at 12 and 20 weeks.

Tier 2 and Tier 3 — reconstruction, local antibiotics, and the one study with physical therapy in it

Antibiotic-loaded calcium sulfate at debridement MODERATE

Overall eradication 92 percent across 16 studies and 917 patients (Shi 2022). Non-randomised series.

Staged reconstruction of segmental defects MODERATE

Induced-membrane (Masquelet) technique: union 89.7 percent and infection eradication 91.1 percent across 427 patients, but persistence of infection or non-union in 18 percent requiring further surgery, and patients operated for infection had significantly higher complication risk (Morelli 2016).

Physical therapy inside a multidisciplinary protocol EMERGING

The only retrieved study with physical therapy in the protocol is a Red Cross cohort in the Democratic Republic of Congo: 168 treated, 71 reviewed at mean 13.7 months, clinical cure 63.4 percent but good functional improvement in only 50.7 percent (Baldan 2014). One observational cohort in an austere setting. Its value is the gap it exposes, not the protocol it describes.

What Doesn't Work

Exercise, loading, or manual therapy as treatment for the infection. Loading a structurally weakened infected bone risks it breaking.
A trial of treatment while you see how it goes. Every week of delay is measured against an amputation odds ratio of 3.87 in the diabetic foot, and against a falling implant-retention success rate around metalwork.
Using a normal X-ray or a normal blood test as permission to keep treating conservatively. The X-ray misses roughly 4 in 10. Around metalwork, the blood tests miss more than half.
Borrowing an antibiotic duration from a headline. The pooled figures in this literature reverse inside their own subgroups.

Exercise Prescription

Nothing below treats the infection. It exists because the offloading device that heals ulcers best is also the one whose wearers move least, and because a systematic review that screened 3,332 records found no study has ever tested a rehabilitation programme for this condition (Jones 2023). Every item is a reasoned response to a measured problem, not a proven treatment. Your medical team's instructions override all of it, and nothing loads the affected limb until they say so.

Exercise	Sets × Reps	Frequency	Pain guide
Seated arm press Sitting tall, press the weight straight overhead and lower under control	3 × 10-12	3 days a week	Effort in shoulders and arms. Nothing at the affected limb
Seated row with a band Pull the band towards your ribs, squeezing the shoulder blades	3 × 10-12	3 days a week	Effort between the shoulder blades only
Unaffected-leg sit-to-stand Stand from a firm chair using only the good leg, lower slowly	3 × 8-10	Daily	Effort in the good thigh. Keep all weight off the affected limb
Seated arm cycling or brisk arm swings Any continuous arm movement that gets you slightly breathless	10-20 min	Most days	Able to talk but not sing

Red Flags

Dark cinematic anatomical study of infected bone tissue

Refer — do not treat

Same day · emergency department

Fever, shivering, confusion, or feeling systemically unwell. A rapidly spreading red limb. A foot that becomes cold, pale or numb in a way it was not before.

Same day · specialist foot service

An ulcer that a probe touches bone through. Visible bone at the base of a wound. A sinus discharging over bone or over metalwork.

This week · urgent orthopaedic

Deep, unremitting bone pain after fracture surgery, with or without fever. The clock matters: keeping the metal in succeeds in 86 to 100 percent of cases revised within three weeks of the infection starting, and 82 to 89 percent at three to ten weeks. Later is worse.

Absence does not reassure

Fever is commonly absent in chronic bone infection and in late infection around metalwork.
A normal X-ray does not exclude it. It misses roughly four cases in ten.
A normal blood test does not exclude it around metalwork. ESR catches fewer than half of these cases.

Refer to: diabetic foot with any suspicion of bone involvement, same-day multidisciplinary diabetic foot service or emergency department out of hours. Suspected infection around fracture metalwork, urgent referral to the treating orthopaedic team, naming infection in the referral. Systemic illness, emergency department now. Do not route any of these through a routine appointment in two weeks.

Return to Training

Every box is set by the treating team, not by how the limb feels. This is not a pain-guided progression.

Treating medical or surgical team has formally declared the infection treated
Full weight-bearing permitted in writing by the surgical or diabetic foot team
Wound fully healed, with no discharge and no sinus, confirmed at review
No deep, non-mechanical bone pain for 4 consecutive weeks
Inflammatory markers trending down or normalised, as interpreted by the medical team
Where reconstruction was performed: union confirmed on imaging by the surgical team
Able to walk your baseline daily distance without wound breakdown over 7 consecutive days
For lifters: 3 sets of 10 at bodyweight-only loading of the affected limb with no next-day pain or swelling, before any external load is added

Conviction

Overall MODERATE, split across nine claims. Actionability splits two ways and must not be averaged: MODERATE for recognition and referral, NOT ACTIONABLE for treatment.

Claim	Conviction
Bone infection substantially raises amputation risk in a diabetic foot ulcer (OR 3.87, 95% CI 2.28-6.57)	HIGH
Test accuracy differs by skeletal site and by population for the same test	HIGH
MRI is the most accurate widely available scan for diabetes-related foot bone infection	HIGH
A normal X-ray does not exclude it	HIGH
Blood tests are insufficient to diagnose late infection around fracture metalwork	MODERATE
Probe-to-bone is useful at the bedside, conditional on the setting	MODERATE
Earlier revision after fixation infection retains implants more often	MODERATE
Any specific antibiotic duration or route for peripheral disease in adults	LOW
Any rehabilitation protocol for this condition	NOT ACTIONABLE

What would change my mind — the probe-to-bone claim

A prospective study of probe-to-bone and blood markers run in a primary-contact musculoskeletal setting rather than a specialist service: consecutive adults with non-mechanical limb pain, N of at least 1,500 to accumulate enough cases at realistic rates, MRI or bone biopsy as the reference, reporting positive predictive value and the number needed to refer per case detected. That number does not currently exist anywhere in this literature, and it is the one a physical therapist actually needs. It would move this claim to HIGH or retire it.

What would change my mind — the "no rehabilitation evidence" claim

A randomised trial of a structured rehabilitation programme during and after offloading for diabetes-related foot ulceration, N of at least 300, non-removable knee-high device in both arms, with a defined upper-body and seated aerobic programme plus a staged post-device weight-bearing progression. Co-primary endpoints: ulcer healing at 12 weeks (non-inferiority) and objectively measured physical activity (superiority), with falls and new lesions as safety endpoints. This is the single trial that would create a genuine physical therapy role in this condition.

The Full Picture — Anatomy, Diagnosis & Evidence

What's Actually Going On

Dark cinematic cross-section of bone and marrow

Bacteria reach bone by three routes. Spreading from an adjacent soft-tissue infection is the dominant route in the diabetic foot: the ulcer comes first, the bone infection follows. Direct inoculation happens at the moment of an open fracture or during fixation surgery. Seeding through the bloodstream dominates in children and in spinal disease, both outside this page.

Once established, the organisms build a biofilm, a protective layer, on dead bone and on any implanted metal. This is the single fact that explains the whole condition. Biofilm survives antibiotic concentrations far above the level that kills the same organism free-floating, which is why physically removing the infected tissue does more of the work than choosing the right drug.

Dead bone, called a sequestrum, has no blood supply. Antibiotics travel in blood. The drug arrives everywhere except where the infection lives.

How to Identify It

Dark cinematic clinical study of a foot examination

How good each test is at catching this, and at ruling it out:

Probe-to-bone Sn: 0.60-0.87 | Sp: 0.83-0.91
The one bedside test that matters. Two published pooled sensitivities sit 27 points apart (Dinh 2008 at 0.60, Lam 2016 at 0.87), and Lam states outright that the test rules in where the disease is common and rules out where it is rare. Neither figure is the test's accuracy. Both are its accuracy in a setting.
MRI Sn: 96.4% | Sp: 83.8%
The most accurate widely available scan for the diabetic foot (Llewellyn 2020). Specificity falls to 60 percent in long-standing disease, and it cannot always separate infection from a Charcot foot.
Plain X-ray Sn: 54-62% | Sp: 68-78%
Misses roughly four cases in ten. A normal film excludes very little.
ESR blood test Foot: AUC 0.83 | Metalwork: Sn 45.1%
The most setting-dependent number on this page. Best of the blood markers in the diabetic foot (Ansert 2024, van Asten 2016), close to uninformative around fracture metalwork (van den Kieboom 2018).

The Debate

Oral antibiotics: non-inferior, or not?

Lima 2025, 9 RCTs, 1,723 patients

Oral versus intravenous, treatment failure RR 0.96 (95% CI 0.78-1.17). No difference.

vs

Lima 2025, same analysis, FRI excluded

RR 1.47 (95% CI 1.08-2.02, p=0.02), favouring intravenous.

These are the same paper. The headline non-inferiority is carried by the fracture-related subgroup. Read the subgroup before applying the headline.

Short antibiotic courses: safe, or not?

Huang 2019, 15 studies, 3,598 patients

Short versus long course, failure OR 1.50 (95% CI 0.97-2.34). Not significant.

vs

Huang 2019, spinal subgroup

OR 2.06 (95% CI 1.18-3.57). Short courses fail in spinal disease.

Again one paper. The pooled null is a mixture of a real spinal signal and a null childhood one, and neither of those populations is this page's.

Honest Limitations

The outcome literature does not agree what the disease is

The research finding: pooled cure rates of 91 percent and eradication rates of 92 percent look like precise, comparable numbers.

The real-world gap: Metsemakers 2018 reviewed 100 randomised trials on fracture fixation. Two cited a validated definition of infection. Seventy gave none at all in their methods while reporting infection as a result. Bezstarosti 2019 found the same across 93 treatment studies and 3,701 patients: one study used a standard definition, and nine different classification systems were in use.

Clinical adjustment: quote the direction, not the decimal. Most people do well, and the published number is less precise than it looks.

Every accuracy figure was measured where the disease is far more common

The research finding: probe-to-bone 0.87, MRI 96.4 percent, ESR AUC 0.83.

The real-world gap: these come almost entirely from specialist diabetic foot services and orthopaedic trauma units. A clinician seeing undifferentiated musculoskeletal pain works where this is orders of magnitude rarer, and no study reports accuracy in that setting. There is no published number needed to refer per case detected. That cost is unquantified, not acceptable.

Clinical adjustment: use these tests to raise or lower suspicion in someone who already has a wound or metalwork. Do not deploy them as a screen across a general caseload.

The physical therapy intervention has never been tested

The research finding: Jones 2023 searched four databases and trial registries to September 2022 and screened 3,332 records, including eight studies covering 441 participants.

The real-world gap: none of the eight delivered or tested a structured rehabilitation programme. The same review found the device with the best healing evidence is also the device whose wearers move least (SMD -0.45, 95% CI -0.87 to -0.04, p=0.03). The standard of care creates a measured fitness problem in people for whom inactivity is a reversible risk, and nobody has run the trial.

Clinical adjustment: not tested and failed. Not tested. Anything offered here is reasoned, and it has to be labelled that way to the patient.

The Nuance

Dark cinematic comparative anatomy of skeletal regions

The same test has a different accuracy in a different bone, and the disease has one name. Termaat 2005 pooled leukocyte scintigraphy, a scan that tracks labelled white blood cells to the site of infection, and reported 84 percent sensitivity in the arms and legs and 21 percent in the spine. One review, one pooling, one disease name, and a fourfold difference by location.

That is why this page covers the appendicular skeleton only and hands all spinal disease to a separate protocol. It is not a filing convenience. A clinician who borrows a number across that boundary is using a test that does not work where they are using it.

Curing the infection and restoring the limb are different outcomes. In the one retrieved cohort that measured both, the infection was cured in 63.4 percent and function meaningfully improved in 50.7 percent (Baldan 2014). Roughly one patient in eight got the first without the second. Curing the infection is the surgeon's job and it usually works. What happens to the limb afterwards is where the evidence stops.

Surgery versus conservative: in long-bone disease with dead bone, source control is the treatment, and pooled cure is 91 percent across 1,605 patients with at least 12 months of follow-up. Complications were reported in 26.6 percent of single-stage procedures. In the diabetic foot, medical management without surgery is a genuine option in selected cases under a multidisciplinary service.

Sources

Termaat MF, et al., 2005, J Bone Joint Surg Am — 23 studies. Leukocyte scintigraphy Sn 84% (72-91) peripheral versus 21% (11-38) axial; FDG-PET Sn 96% (88-99), Sp 91%. PMID 16264122.
Llewellyn A, et al., 2020, Eur J Radiol — 36 studies, 8 at high risk of bias. MRI Sn 96.4% (90.7-98.7), Sp 83.8%; X-ray Sn 61.9% (50.5-72.1). PMID 32862106.
Lam K, et al., 2016, Clin Infect Dis — 7 studies. Probe-to-bone Sn 0.87 (0.75-0.93), Sp 0.83 (0.65-0.93); accuracy stated by the authors to depend on how common the disease is in that setting. PMID 27369321.
Dinh MT, et al., 2008, Clin Infect Dis — probe-to-bone or exposed bone Sn 0.60, Sp 0.91; radiography Sn 0.54, Sp 0.68; MRI Sn 0.90, Sp 0.79. PMID 18611152.
van den Kieboom J, et al., 2018, Bone Joint J — 8,284 screened, only 6 suitable. Late fracture-related infection: CRP Sn 77.0%, ESR Sn 45.1%, leucocyte count Sn 51.7%. Authors conclude insufficiently accurate. PMID 30499325.
Ansert EA, et al., 2024, Wound Repair Regen — 19 papers, N=2,854. Diabetic foot: ESR AUC 0.83, CRP 0.77, procalcitonin 0.71; no marker rated outstanding. PMID 38566503.
Metsemakers WJ, et al., 2018, Injury — 100 RCTs. Two cited a validated infection definition; 70 gave none while reporting infection as an outcome. PMID 28245906.
Bezstarosti H, et al., 2019, Arch Orthop Trauma Surg — 93 studies, 3,701 patients. One study used a standard definition; nine classifications in use; tibia 64% of cases. PMID 30343322.
Morgenstern M, et al., 2021, Bone Joint J — 6 studies, 276 patients. Implant retention 86-100% under 3 weeks, 82-89% at 3-10 weeks; no time cut-off ever defined. PMID 33517743.
Lima JPC, et al., 2025, Bone — 9 RCTs, 1,723 patients. Oral versus IV failure RR 0.96 (0.78-1.17); RR 1.47 (1.08-2.02) favouring IV once fracture-related infections are excluded. PMID 40280255.
Huang CY, et al., 2019, Int J Antimicrob Agents — 15 studies, 3,598 patients. Short-course failure OR 1.50 (0.97-2.34) overall; spinal OR 2.06 (1.18-3.57). PMID 30639627.
Lari A, et al., 2024, J Orthop Surg Res — 42 studies, 1,605 patients. Pooled cure 91%, no difference single versus two-stage; complications 26.6% single-stage. PMID 38877562.
Reyes MC, et al., 2024, Wound Repair Regen — 530 patients. Residual disease at the margin: amputation OR 4.3 (2.4-7.6), re-infection OR 2.0 (1.1-3.4), antibiotics +16.3 days. PMID 39376015.
Zhang H, et al., 2023, Int Wound J — 32 studies, 9,934 subjects. Osteomyelitis OR 3.87 (2.28-6.57) for amputation; gangrene OR 14.45. PMID 37095728.
Senneville É, et al., 2024, Diabetes Metab Res Rev — IWGDF/IDSA 2023 guideline, GRADE, 149 studies. Current guideline for the diabetic foot. PMID 37779323.
Bus SA, et al., 2024, Diabetes Metab Res Rev — IWGDF 2023 offloading guideline. Non-removable knee-high device first choice. PMID 37226568.
Jones K, et al., 2023, J Foot Ankle Res — 3,332 records screened, 8 studies, 441 participants. None tested a structured rehabilitation programme. Non-removable cast wearers less active, SMD -0.45. PMID 36966316.
Baldan M, et al., 2014, Trop Med Int Health — 168 treated, 71 reviewed at mean 13.7 months. Clinical cure 63.4% versus functional improvement 50.7%. PMID 24702780.

Go Deeper

Most injury advice never tells you which problems are not injuries at all. The Verdict reviews one condition or supplement a week, with the evidence graded and the gaps named. Free.

Join The Verdict

A sore that will not heal, or deep bone ache after surgery, needs a doctor, not exercise.

Best for

Skip if

What Works

Surgical removal of all infected and dead bone STRONG

Culture-directed antibiotics, sampled before empiric treatment STRONG

Complete margin clearance in the diabetic foot MODERATE

Non-removable knee-high offloading for a neuropathic plantar ulcer STRONG

Antibiotic-loaded calcium sulfate at debridement MODERATE

Staged reconstruction of segmental defects MODERATE

Physical therapy inside a multidisciplinary protocol EMERGING

What Doesn't Work

Exercise Prescription

Red Flags

Absence does not reassure

Return to Training

Conviction

What's Actually Going On

How to Identify It

The Debate

Oral antibiotics: non-inferior, or not?

Short antibiotic courses: safe, or not?

Honest Limitations

The outcome literature does not agree what the disease is

Every accuracy figure was measured where the disease is far more common

The physical therapy intervention has never been tested

The Nuance

Sources

Dealing with something specific?

Get weekly evidence-based rehab verdicts

Want a coach, not just research?

Related free research

There are 500+ more inside