Scaphoid Fracture

What Works

Read the framing first, because it is unusual. The graded treatments for this injury are diagnostic and surgical. There is no graded rehabilitation tier here, and that absence is a finding rather than an oversight.

Cinematic rendering of wrist immobilisation and fixation

Tier 1 — Strong Evidence

1. Get the right scan HIGH

An MRI catches 96 out of 100 of these breaks and correctly clears 99 out of 100. A CT catches 93 out of 100 and clears 99 out of 100. A plain X-ray, read by a person, catches about 71 out of 100. If the suspicion is there and the X-ray is clear, the scan is the treatment. (Yin 2010, 26 studies)

Timeline: immediate. Delay is the harm.

2. A cast, for a break that has not shifted HIGH

At one year, the wrist works the same as a surgically fixed one, and the cast group collects fewer and less serious complications. 93 out of 100 casted breaks heal. (Johnson 2022, 7 randomised trials; Singh 2012, 1,401 wrists)

Timeline: bone looks healed on a scan at around 79 days.

3. Surgery, when speed matters or the break has shifted HIGH

Fixation gets you back to work in about 46 days instead of 77, and the bone looks healed at about 44 days instead of 79. For a shifted break, 18 out of 100 fail to heal in a cast, so the calculation changes completely. (Alnaeem 2016; Singh 2012)

Timeline: bone looks healed on a scan at around 44 days.

Tier 2 and Tier 3 — the rest of the evidence

Tier 2 — Moderate Evidence

4. Any cast shape you like, for an unshifted break MODERATE

Thumb held still or thumb left free: 10.3% versus 10.1% failed to heal. Cast below the elbow or above it: 9.6% versus 10.5%. Four independent reviews spanning 2007 to 2023, adults and children, have failed to find any difference. (Siotos 2023; Yin 2007; Symes 2011; Shaterian 2019)

5. A plain bone graft, if it never healed MODERATE

The simpler graft heals 88.7 out of 100. The technically harder "living" graft heals 87.5. Taking bone from the wrist rather than the hip works just as well and hurts you less in the process. (Duncumb 2022, 78 studies, 7,671 patients)

6. A cast, in children MODERATE

96 out of 100 heal, 99 out of 100 get full movement back, and 97 out of 100 end up pain free without an operation. Children are genuinely a different population here. (Shaterian 2019, 812 fractures)

Tier 3 — Emerging, and weak

7. Keyhole-assisted grafting LOW

May shorten healing by about 8 weeks with no improvement in whether it heals at all. The authors themselves rate their certainty as very low. Direction only. (Noori 2024)

8. Cone-beam CT and AI X-ray reading LOW

Both look promising and neither is practice. The scans studies are 3-of-4 at high risk of bias. The AI studies are all backward-looking, 9 of 10 at high risk of bias, and none has been tested in a real clinic. Interesting, not actionable. (Yang 2021; Kraus 2024; Suen 2024)

Exercise Prescription

There isn't one, and that is the finding

Twenty-five systematic reviews of this injury. Zero rehabilitation trials. Not one study anywhere in this literature prescribes an exercise, a load, a set-and-rep scheme, or a progression rule for a broken scaphoid. This page could invent a convincing-looking list and it would look more complete. It would also be a fabrication dressed as evidence, for an injury whose signature harm is somebody being too confident about a wrist. Your exercises should come from the surgeon or hand therapist looking after your specific break, because which movements are safe depends entirely on where the break is and whether it has shifted, and neither of those is visible from the outside. What the evidence does strongly support is everything above this box: recognising it, scanning it, and not accepting a clear X-ray as an all-clear.

What Doesn't Work

Treating a clear X-ray as an all-clear. The single most common and most costly mistake in this injury, and the reason it has its reputation. The film is a screen, not a discharge.
"Come back in two weeks for another X-ray." It sounds careful and thrifty. It is the slowest and most expensive route there is: scanning immediately costs about $2,553 per break found, while waiting two weeks and then scanning runs to roughly $170,000 per break found. On top of that it leaves you either in a cast you never needed or walking around on a break you don't know about.
Defending the thumb-hole cast as if it were doing the healing. Four reviews, no difference. It persists because it is what everyone was taught, and because plaster wrapped around the thumb looks like it is doing something to the bone underneath.
Quoting "back to work faster" as if it meant "better wrist". Those are the numbers surgery wins, and they are unadjusted for your job, and for the simple fact that someone without a cast is allowed back sooner.
Applying "surgery is no better than a cast" to a break that has shifted. Every trial behind that sentence studied unshifted breaks only. Nobody has ever run the shifted trial. Using it there inverts the evidence.
Reaching for the fancier graft because it sounds more definitive. It heals no better. It just gets chosen for the worse breaks.

Return to Training

Honest labelling: the first two criteria are the ones the evidence actually tracks. The rest are clinical reasoning, because no study has validated a return threshold for this injury.

Healing confirmed on a scan by the team treating you. Not "it feels fine". This bone can be comfortable and still not healed, which is why it is first.
Explicit clearance for loaded use from the surgeon or hand therapist.
No tenderness in the hollow at the base of the thumb on firm pressing.
Full, pain-free wrist extension, the position you are about to load.
Grip within about 10% of the other side, measured with a gauge rather than guessed.
30 seconds holding a press-up position through the wrist, pain-free.
3 sets of 10 pressing at a submaximal load: no pain during, and no pain or swelling 24 hours later.
For anything with a fall risk (cycling, contact sport, gymnastics, skating): all of the above, plus explicit clearance for the fall risk itself.

While you wait, train everything else. A wrist does not deload a squat. Lower body, the other arm, and your trunk all stay in. What stops is pushing, gripping and taking body weight through that hand, and the temptation to "test it" to see how it feels. That is how a crack becomes a shift.

Red Flags — When to Get Seen

You fell onto an outstretched hand, the base of your thumb is tender, and the X-ray was called clear. That combination is the red flag, and the clear X-ray is part of it. Doctors catch this break on the first X-ray about 7 times out of 10. For the bigger bone right next door, 19 out of 20.
Thumb-side wrist pain that is still there weeks after an injury that was "cleared" on X-ray. This is the missed break arriving late.
A confirmed break that has shifted out of place, or sits at the end of the bone nearest your forearm. Those fail to heal 4 to 7 times more often, and neither can be judged from the outside.
Pain, stiffness, weak grip or loss of use that is not improving, or is getting worse, after treatment. That is what a bone that never healed feels like.
Visible deformity, numbness, tingling, weakness, or a hand you cannot use at all.

Where to go: Emergency Department or an urgent hand or orthopaedic assessment, for an MRI or CT scan. A hand surgeon for any confirmed shifted or proximal break, or any suspected non-healing. The target of the referral is the scan, not the appointment.

Cinematic anatomical rendering of the wrist highlighting the region of concern

Conviction

Moderate-High Overall Endpoint by endpoint, because the confidence is genuinely uneven.

Highest confidence HIGH

A plain X-ray cannot rule this out. MRI and CT are excellent. Thumb-side tenderness rules out but never rules in. For unshifted breaks, surgery and a cast give the same wrist at 12 months. Surgery is faster. Shifted and proximal breaks fail to heal far more often.

Moderate confidence MODERATE

Surgery causes more and more serious complications (three reviews say yes, one says no difference). Cast shape doesn't matter (no signal across four reviews, though the numbers are too loose to call it settled). The simpler graft is as good as the fancier one. Children do excellently without surgery.

Low, contested, or simply unknown LOW

Whether surgery reduces non-healing in unshifted breaks is contested and depends entirely on which studies you let in. Fixing versus casting a shifted break has never been tested in a randomised trial, despite that being the decision the whole thing turns on. And any specific rehabilitation exercise or dose: no trial exists, anywhere, in this entire literature.

What would change my mind on "surgery buys speed, not a better wrist"

This claim is bounded to breaks displaced by 2mm or less, because that is every trial's entry criterion. A multicentre randomised trial of 160 or more adults with CT-confirmed shifted breaks, comparing fixation against a cast, with patient-reported wrist function at 12 months and CT-confirmed healing as joint primary endpoints, would fill the largest hole here. If it came back null, "shifted means fix it" would lose its foundation. If it came back strongly positive, the equivalence finding gets formally quarantined to unshifted breaks, which is exactly where it already belongs and is routinely over-extended from.

What would change my mind on "there is no rehab evidence"

Any randomised trial at all. 100 or more adults after the bone has healed, randomised to a defined progressive loading programme (specified sets, reps, load, frequency and progression rules) against advice and self-directed use, with patient-reported wrist function at 6 and 12 months and grip strength alongside it. That single trial would create the first evidence-based rehabilitation claim in this condition. Right now there is nothing to upgrade or downgrade, which is precisely why this page refuses to give you an exercise table.

Go Deeper

Most injury advice online is guesswork with citations bolted on. The Verdict reads the actual evidence, tells you where it runs out, and refuses to invent the rest. One protocol a week, free.

Join The Verdict — free

The Full Picture — Anatomy, Diagnosis & Evidence

What's Actually Going On

The scaphoid is the most commonly broken bone in the wrist, making up 70% of all broken wrist-bone cases. Its name means "boat-shaped", and it sits diagonally across the two rows of small bones in your wrist, on the thumb side. It is the mechanical link between those rows, which is exactly why it takes the load when you fall onto an outstretched hand.

Cinematic anatomical rendering of the scaphoid within the carpus

It hides. Because the bone lies at an angle to the X-ray beam, a crack rarely presents itself cleanly on a standard picture. This has been measured rather than assumed: human readers caught scaphoid breaks on X-ray 71 times out of 100, while the same readers in the same review caught distal radius breaks 95 times out of 100. It is the bone that is the problem, not the reader.

Its blood runs backwards. The artery enters the bone at the far (distal) end and feeds it in reverse, so the half nearest your forearm is the last to be supplied and the first to be cut off. A crack interrupts the supply to the fragment furthest from the entry point. That is not a theory, it shows up in the healing numbers: breaks at that proximal end fail to heal 34% of the time without surgery, a risk 7.5 times higher than breaks further down the bone.

So location is not a detail, it is the prognosis. The same goes for displacement. Across 1,401 wrists treated in plaster, 93% healed. But of the 15% that were shifted by at least a millimetre, 18% never united, a risk 4.4 times higher.

And that is what everything here is trying to prevent: the fracture never bridges, the starved fragment can collapse, the wrist mechanics decompensate, and over years it drifts into a predictable arthritic pattern with pain, stiffness and lost grip.

How to Identify It

Cinematic rendering of clinical wrist examination

Three tests carry the clinical examination, and all three share the same personality: good at telling you the question is open, poor at closing it.

Anatomical snuffbox tenderness Sn: 93% | Sp: poor (0.03–0.98)
The most sensitive single finding. Its specificity varies so wildly between studies that the researchers refused to pool it at all, which tells you how unreliable it is at ruling anything in.
Thumb compression (axial loading) Sn: 48–100% | Sp: 66%
The most specific of the three, and still only 66%.
Scaphoid tubercle tenderness Sn: 82–100% | Sp: 17–57%
Same pattern again.

All three positive still leaves you at roughly 60%. Combining tests does raise specificity while keeping sensitivity high, which is why you do all three, and even then four in ten people with a fully positive examination do not have a break. This is a screen that earns a scan, not a diagnosis.

The imaging question, by contrast, is settled: MRI catches 96 and clears 99 out of 100; CT catches 93 and clears 99; a bone scan catches 97 but only clears 89.

The Debate

No clinical practice guideline was identified for this condition as of 17 July 2026, and none is cited on this page. The guideline layer here has not been verified, and rather than lean on an unverified summary of one, every threshold quoted above is taken straight from the primary research. So the debate is not guideline versus trial. It is the field arguing with itself, and one endpoint arguing with another.

Cinematic rendering contrasting treatment pathways

Does surgery give a better wrist?

Al-Ajmi 2018, 8 randomised trials

Surgery gave significantly better function, prevented delayed healing, and got people back to work sooner.

vs

Johnson 2022, 7 randomised trials, function at 12 months

No difference. p = 0.082. Complications were higher in the surgical group, and more serious.

Follow Johnson. The surgery-favouring result pools mixed, mostly early timepoints. And Al-Ajmi's own paper concedes, in the same paragraph, that superiority "could not be established". Only the first half of that abstract ever gets quoted.

What about the better grip and movement after surgery?

Chen 2023, 12 randomised trials

Significantly better grip strength and wrist movement after surgery (p < 0.01).

vs

Symes 2011, 8 trials

Grip and movement are "transiently" better. Johnson 2022: gone by 12 months.

A cast makes a wrist stiff. Taking the cast off fixes that. Pooling early timepoints with late ones manufactures a difference that does not survive the year, and Symes named the mechanism in a single word the field keeps ignoring.

Does surgery stop it failing to heal?

Li 2018, 10 trials + 4 observational studies

Surgery roughly halves the non-healing rate (RR 0.47, p = 0.023).

vs

Chen 2023, randomised trials only

No difference at all (p = 0.538).

The difference is entirely in what got let in. Li pools four observational studies alongside the trials, and in observational data the breaks that get operated on are not the breaks that get cast. When only randomised trials are counted, the benefit disappears. That makes it inclusion-criteria dependent, which means it is not established.

Is the thumb-spica cast doing anything?

Universal clinical tradition

The thumb must be immobilised, and the cast should run above the elbow, to hold the scaphoid still.

vs

Siotos 2023, 7 studies (+ 3 more reviews agreeing)

Thumb in vs free: 10.3% vs 10.1% failed to heal. Above vs below elbow: 9.6% vs 10.5%. Nothing.

Follow the evidence, but state it honestly: those numbers carry very wide error bars, wide enough to be compatible with both a real benefit and a real harm. What is supported is "no signal has emerged across four independent reviews", not "proven identical". That is enough to stop defending the longer cast as though evidence demanded it. It is not enough to call the question closed.

Honest Limitations

1. The whole treatment literature answers a question your physical therapist never asks

The research: every comparative study here compares surgery against a cast.

The real-world gap: not one of the 25 reviews tests a rehabilitation approach, prescribes an exercise, or defines a progression rule. This is not a gap in the search, it is a gap in the field. The person in front of you is out of a cast at six weeks asking what to do with their wrist, and the literature has nothing to say to them.

The adjustment: say so. Post-cast rehabilitation here is clinical reasoning, and it should be labelled as reasoning rather than dressed as evidence.

2. Back-to-work is not the same as better, and it is what surgery wins

The research: back to work at 46 days versus 77; bone looks healed at 44 days versus 79.

The real-world gap: those numbers are unadjusted for occupation, for expectation, and for the fact that someone without a cast is simply permitted back sooner. "Looks healed on a scan" is a picture of a bone, not a working wrist.

The adjustment: frame it as faster, not better, with more complications. A self-employed builder and a retired reader will rationally choose differently, and both are right.

3. Selection decides the outcome, and pooled percentages hide it

The research: "living" grafts look better (92% vs 88%) when series are pooled as raw percentages, and identical (87.5% vs 88.7%, p = 0.685) when compared head to head.

The real-world gap: the harder graft gets chosen for the worst breaks. The same review proves it: exclude the proximal and blood-starved cases and healing jumps to 96.5%, against 86.8% otherwise.

The adjustment: do not read a prognosis off which operation someone had. Read it off the fracture. A complex graft is a marker of a bad break, not a better outcome.

The Nuance

Cinematic rendering of the clinical decision pathway

Surgery buys speed, not a better wrist. The best evidence available is seven randomised trials pooled with a pre-planned patient-reported endpoint at twelve months, and it is null (p = 0.082), while the surgical group collected more complications and more serious ones. Everything surgery reliably wins is a process endpoint: the bone looks united sooner, the person is back at work sooner. Those things have genuine value to a real person with a mortgage, and someone who chooses fixation for that reason has read the evidence correctly. What is not supported is the belief that the screw produces a better wrist a year later.

And none of it crosses the displacement line. Every trial behind that paragraph studied breaks displaced by 2mm or less. For a shifted break, the risk of the bone never healing is 4 to 7 times higher, and nobody has ever run the trial. That makes it a conversation with a hand surgeon rather than something to settle from a meta-analysis of a different fracture.

Two success rates you will see quoted side by side, which must not be: 93% heal in a cast, and the twelve-month equivalence with surgery. They look like a head-to-head and they are not. The 93% comes from pooled observational series including shifted breaks; the equivalence comes from trials restricted to unshifted ones. Different populations. Nobody randomised them against each other.

The field cannot even compare its own results on its worst fracture. Three incompatible definitions of "proximal" are in circulation, and only half the papers using the word define it anatomically. The same injury carries a non-healing risk reported as 2.3, 3.4 or 7.5 times higher depending purely on where the author drew the line. The break with the worst prognosis has the least evidence and the shakiest definition. One author even published the sample size needed to settle it: 76 cases. Nobody has run it.

Sources

25 systematic reviews and meta-analyses retrieved. Every citation below was machine-verified against the source database. All are abstract-level; none declared funding.

Suen K, et al. 2024, Eur J Radiol (PMID 38981178). 20 studies. Human readers detect scaphoid fractures on X-ray at 71% sensitivity vs 95% for distal radius in the same review. Strong
Johnson NA, et al. 2022, Bone Joint J (PMID 35909381). Meta-analysis of 7 randomised trials. Function at 12 months: no difference (p = 0.082). Complications higher and more serious with surgery. Strong
Singh HP, Taub N, Dias JJ. 2012, Injury (PMID 22424701). 1,401 scaphoids. 93% healed in cast; displaced fractures 18% non-union, RR 4.4. No trial of fixation vs cast for displaced fractures exists.
Eastley N, et al. 2013, J Hand Surg Eur (PMID 22736744). 1,147 fractures. 34% of proximal fractures fail to unite without surgery, RR 7.5. Power calculation for the definitive trial: 76 cases.
Huynh KA, et al. 2021, Plast Reconstr Surg (PMID 33620933). 22 index tests, 1,940 patients. Snuffbox tenderness Sn 93%; all three tests positive gives ~60% probability.
Mallee WH, et al. 2014, J Hand Surg (PMID 25091335). 25 tests. Snuffbox Sn 0.87–1.00 / Sp 0.03–0.98; not pooled owing to heterogeneity.
Yin ZG, et al. 2010, Clin Orthop Relat Res (PMID 19756904). 26 studies. MRI Sn 96%/Sp 99%; CT Sn 93%/Sp 99%; bone scan Sn 97%/Sp 89%.
Carpenter CR, et al. 2014, Acad Emerg Med (PMID 24673666). 75 studies. Scaphoid = 70% of carpal fractures; plain X-rays insufficient to rule out.
Siotos C, et al. 2023, J Plast Surg Hand Surg (PMID 35034563). 7 studies. Cast type made no difference to non-union.
Duncumb JW, et al. 2022, Bone Joint J (PMID 35491585). 78 studies, 7,671 patients. Vascularized 87.5% vs non-vascularized 88.7% (p = 0.685). Excluding proximal pole and AVN: 96.5% vs 86.8%.
Yin ZG, et al. 2015, J Orthop Trauma (PMID 25756914). Decision model. Immediate CT $2,553/fracture detected; week-2 X-ray then MRI ~$170,000.
Alnaeem H, et al. 2016, J Hand Surg (PMID 27707564). 10 studies. Return to work 77 vs 46 days; union 79 vs 44 days.
Symes TH, Stothard J. 2011, J Hand Surg Eur (PMID 21700649). 8 trials. Grip and movement transiently better with surgery; more complications.
Chong HH, et al. 2022, J Plast Surg Hand Surg (PMID 34550858). 15 articles. Three incompatible definitions of "proximal"; RR 2.3 vs 3.4 depending on definition.
Shaterian A, et al. 2019, Hand (N Y) (PMID 29078712). 812 pediatric fractures. 96.2% union; 99% normal movement without surgery.
Chen S, et al. 2023, World J Surg (PMID 36484804). 12 randomised trials. Non-union p = 0.538 and complications p = 0.661 both null.
Li H, et al. 2018, Medicine (PMID 30508914). 10 trials + 4 cohorts. Non-union RR 0.47; no difference in satisfaction, pain or DASH.
Yin ZG, et al. 2007, Clin Orthop Relat Res (PMID 17310931). Operative offers no greater benefit for undisplaced fractures; no randomised evidence for the proximal pole.
Pinder RM, et al. 2015, J Hand Surg (PMID 26116095). 1,602 patients. Fixation of some kind beats none (88–91% vs 79%).
Noori A, et al. 2024, BMC Musculoskelet Disord (PMID 39090587). 822 patients. Arthroscopic grafting: very low certainty.

Your wrist X-ray was clear. For one bone at the base of your thumb, that isn't the all-clear.