What's Actually Going On
A motor connection in the palm between the thumb branch of the median nerve and the deep branch of the ulnar nerve. In 80 dissected hands the ulnar side arose from the deep branch every single time, and it runs either outside the muscle or within it.
The textbook version of thumb-muscle supply is the exception, not the rule. In a dedicated 60-limb dissection, the superficial head of flexor pollicis brevis was median-supplied in only 70 percent, with shared supply in 30 percent, and the deep head was missing altogether in 14 percent. The authors' conclusion was that mixed supply is the normal pattern.
It carries motor fibres and leaves sensation alone. Sensory fibres to the fingers were normally distributed in the cases where anyone checked, which is the most useful mechanical fact on this page: it is exactly why the sensory test stays readable when the motor test has been destroyed.
Almost nobody measures how much of the muscle the crossover actually drives. One study quantified it, at an average of 27.6 percent with a standard deviation of 16.4, so the average tells you almost nothing about an individual hand. Everyone else records only whether the structure is there, in a variant whose real weight ranges from trivial to total.
How to Identify It
The finding that starts the whole process is a mismatch: thumb strength and thumb-pad bulk that do not agree with the severity on the report, in either direction. The comparison is the test.
- Ulnar stimulation with thumb-muscle recording Sn: NO DATA | Sp: NO DATA produces a large, easily obtained response at a lower stimulus than median stimulation, with a tell-tale onset.
- Median reading from the second lumbrical Sn: NO DATA | Sp: NO DATA stays interpretable when the routine comparison does not, because that muscle keeps its median supply.
- Needle EMG of the thumb muscles Sn: NO DATA | Sp: NO DATA confirms shared supply directly. Definitive, invasive, uncomfortable.
Those NO DATA labels were measured, not assumed. A search on this topic combined with "sensitivity" returns zero papers, and with "specificity" returns one that is about something else entirely. No test here carries a published accuracy figure, so any number you see quoted for one has been borrowed from elsewhere.
The Debate
Does the method explain why the numbers disagree?
The received story, repeated in reference summaries and in the 2026 meta-analysis itself
Dissection and nerve testing disagree because they detect different things. Dissection sees structures the electrical test cannot pick up.
vs
The primary studies, read together
Dissection holds both the highest and the lowest figure in the whole literature: 100 percent of 80 hands, and 16 percent of 50 hands. Nerve testing independently spans 2.48 percent to 83.3 percent.
The received story does not survive its own data. Method does not predict even the direction of the disagreement, let alone its size. What predicts the number is each author's threshold for calling a structure a connection, and almost nobody states theirs. This is a definitional failure, not a measurement dispute, so any prevalence figure is unusable rather than merely uncertain.
| Study | Method | Figure |
|---|
| Caetano 2019, 80 hands | Dissection | 100% |
| Kimura 1983, 150 limbs | Nerve testing | 83.3% |
| Ahadi 2016, 23 hands | Nerve testing | 82.6% |
| Budak 1999, 108 subjects | Nerve testing | 73.1% |
| Roy 2016, 10,562 limbs pooled | Meta-analysis | 55.5% (CI 30.6–79.1) |
| Falconer 1985, 10 specimens | Dissection | 30% |
| Ajmani 1996, 68 palms | Dissection | 19.1% |
| Lazo Velasquez 2018, 127 patients | Nerve testing | 16.5% |
| Loukas 2011, 50 hands | Dissection | 16% |
| Pastor Gomez 2001 | Nerve testing | 2.48% |
Is the "all-ulnar hand" a real entity?
The older case literature, and a paper title as recent as 2022
Some hands have thumb muscles supplied entirely by the ulnar nerve. The "all-ulnar hand" is a recognised pattern.
vs
Brown and Landau 2013, reviewing the dissection evidence
No dissection has ever shown the second lumbrical supplied by the crossover, which negates the concept of the all-ulnar hand.
Brown and Landau win, and the win pays for itself. The surviving median-supplied second lumbrical is not a technicality: it is the muscle that keeps the median nerve testable, and it is exactly what a 2024 case used to establish carpal tunnel syndrome after two coexisting variants had wrecked every routine measurement. The fact that refutes the entity is the fact that rescues the diagnosis.
Honest Limitations
Every figure comes from a clinic queue or a cadaver
What the studies show: prevalence anywhere from 2.48 to 100 percent.
The real-world gap: the only study of unselected, symptom-free adults is 43 years old. Cadaver figures depend entirely on how generously the dissector defines the structure, and clinic figures may be suppressed by the very nerve compression being investigated. There is no population estimate and no route to one from this literature.
What to do instead: assume it is possible in any hand you test. Do not scale your suspicion to a prevalence figure, because there isn't a usable one.
The literature reports presence; the clinic needs degree
What the studies show: mostly a binary present-or-absent.
The real-world gap: a hand where the crossover drives 5 percent of the muscle and one where it drives all of it both count as positive, and they behave nothing alike. One study, once, measured the amount.
What to do instead: judge the individual hand by the size of the mismatch between report and function, not by whether a label has been applied.
Twenty-two of the 30 papers are single cases
What the studies show: a collection of individually striking hands.
The real-world gap: that is enough to prove the failure mode is real and not enough to say how often you will meet it. All of it was read at abstract level.
What to do instead: use this as pattern recognition, not as a source of numbers to quote to a patient or a referrer.
The Nuance
The error runs four ways, not one. That is the part most summaries of this variant miss, and it is why it belongs on a clinician's radar rather than in a curiosities file.
- False severity. Nerve compression plus this variant produces an absent thumb-muscle response alongside a working, well-muscled hand. Read as end-stage disease. Documented consequence: an operation the patient did not need.
- False mildness. A severe or completely severed median nerve in one of these hands leaves the thumb working. A complete lesion gets recorded as partial, and in trauma that changes whether the nerve gets repaired.
- Wrong disease. A deep ulnar branch lesion in one of these hands wastes the thumb pad and the first web space with no numbness at all. Two published patients were diagnosed with, or worked up for, motor neuron disease before extended testing found a focal nerve lesion. In one, the cause was a compression point at the piso-hamate hiatus.
- Wrong location. Unrecognised connections produce apparent blocks and place the lesion at the wrong level along the nerve, which sends the investigation to the wrong part of the arm.
There is no surgery-versus-conservative comparison to present here, because there is no operation for the variant. What the record does show is that both documented harms are things done to patients rather than things the variant did: an unnecessary operation on one side, and a missed complete nerve injury on the other. All of the useful work sits upstream of any procedure, in reading the test properly. That is a job a physical therapist who examines the hand carefully is unusually well placed to do, because the mismatch shows up in what the hand can do long before anyone questions the report.
