Curriculum/Pillar 8 · Diagnostic Reasoning in EMG

Diagnostic Pitfalls & Cognitive Bias

Module 34 of 36·28 min read

Learning objectives

1Catalogue the high-yield technical and physiologic pitfalls
2Name the cognitive biases that drive electrodiagnostic error
3Build habits that make the study falsifiable and reproducible

The most dangerous electrodiagnostic errors are not failures of technique but failures of inference: a normal physiologic variant promoted to disease, a stable old injury misread as an active one, an artifact accepted as signal because it confirmed what the referral already suggested. The instrument reports faithfully whatever the electrode senses; the diagnosis is constructed by the human interpreting it, and that interpretation is where the failure modes live. Defending against them requires two disciplines that are easy to state and hard to practise — explicit modelling of uncertainty, and a commitment to falsifiability over confirmation.

Physiologic pitfalls: normal variants masquerading as pathology

Several benign phenomena reliably impersonate serious disease. Benign fasciculations are common, frequently calf-predominant, and in isolation carry no prognostic weight; read as evidence of motor neuron disease, they convert a healthy person into a terrifying misdiagnosis. The discriminating principle is that fasciculations are pathological only in the company of their fellow travellers — active denervation (fibrillations, positive sharp waves), chronic reinnervation (large, polyphasic, unstable units), and reduced recruitment — distributed across multiple spinal levels and limbs. A fasciculation standing alone, in a muscle with normal insertional activity and normal motor units, is a normal variant until proven otherwise.

The mirror-image error is reading chronic, stable reinnervation as active disease. Large-amplitude, long- duration, polyphasic motor units are the fingerprint of a neurogenic process, but on their own they describe an injury that has already healed through collateral sprouting; they may date to a remote radiculopathy, an old poliomyelitis, or a long-resolved nerve injury. The variable that separates a remote, inactive lesion from an ongoing one is the presence and abundance of spontaneous activity and the stability of the units on firing. Reporting chronic reinnervation as if it were active denervation can launch an unnecessary search for a progressive disease that is not there.

The temperature artifact

Limb temperature is the single most underappreciated source of spurious abnormality in nerve conduction studies, and its effects are large, lawful, and bidirectional. As a cooled nerve's membrane slows, conduction velocity falls by roughly 1.5–2.4 m/s for every degree Celsius of cooling, distal latencies prolong, and — counterintuitively — sensory and motor amplitudes increase because slowed sodium-channel inactivation lengthens and synchronises the depolarisation, raising and broadening the response. The diagnostic hazard is twofold. Slowed velocities and prolonged latencies in a cold limb can fabricate a demyelinating picture in a normal nerve; conversely, the amplitude-boosting effect of cooling can resurrect a borderline response or mask a genuinely low amplitude, so that a warmed limb would have revealed an axonal loss the cold study concealed. The defence is procedural, not interpretive: measure and record surface temperature, maintain the hand at roughly 32°C and the foot at roughly 30°C, and warm a cold limb before drawing any conclusion. A latency abnormality in an unwarmed limb is uninterpretable.

Warm the limb before you trust the latency

Temperature corrections by formula are a last resort, not a substitute for physical warming; the relationship is approximately linear over the working range but varies by nerve and patient, and a corrected value inherits the uncertainty of the correction. The robust practice is to warm the limb to target, re-record, and interpret the real measurement. If a finding appears only in a cold limb and vanishes on warming, it was an artifact, and an artifact must not be reported as disease.

Technical pitfalls: when the signal is not what it seems

A second family of errors arises at the electrode. Voluntary co-activation during a motor study allows antagonist contraction to contaminate the response and can mimic an abnormality of recruitment or configuration. Sub-maximal stimulation understates amplitude and can manufacture a conduction block where none exists, while excessive current produces stimulus spread to adjacent nerves that fabricates an anomalously fast or large response. Electrode misplacement — an active electrode off the motor point, an incorrect inter-electrode distance — distorts latency, amplitude, and waveform morphology in ways that imitate genuine pathology. Environmental contamination compounds these: 60 Hz line noise obscures low-amplitude sensory responses and can be mistaken for spontaneous activity, and stimulus artifactbleeding into the recording can hide a short-latency response or be misread as the onset itself. None of these is exotic; each is a routine event whose only reliable countermeasure is the electromyographer's suspicion.

Cognitive biases: the failure modes of the interpreter

The errors that survive technical rigour are cognitive, and they operate below awareness. Anchoring fixes the interpretation on the first or most salient datum — often the referral diagnosis — and resists revision as contrary data accumulate. Confirmation bias selectively weights findings that support the expected answer and explains away those that contradict it, so the study is mined for agreement rather than tested for refutation. Premature closure ends the study the moment a plausible answer appears, before the comparison that could have overturned it is performed, and is especially seductive in electrodiagnosis because the protocol can always be stopped early. Base-rate neglect reads a positive finding without reference to the pretest probability, inflating the post-test confidence in a diagnosis that was improbable to begin with. Framingfrom the referral — the way the question was posed — quietly constrains the hypotheses considered, so that a study ordered to “confirm” a diagnosis is unconsciously designed and read to do exactly that. These biases are not signs of carelessness; they are the default operating mode of expert pattern-recognition, and they must be actively counteracted.

Clinical Pearl

Make uncertainty explicit on the screen, not implicit in your head. State the competing hypothesis aloud, name the single finding that would refute your leading interpretation, and then go acquire it. A diagnosis you have genuinely tried and failed to disprove is far stronger than one you have only assembled supporting evidence for — falsification, not accumulation, is what separates inference from rationalisation.

The reproducibility principle

The operational antidote to both artifact and bias is a single rule: a finding that cannot be reproduced should not be reported. Genuine physiology is stable and lawful; artifacts are fragile and conditional. The discipline is to change one variable and re-test— reposition the stimulator and re-stimulate, warm the limb and re-record, move the recording electrode, increase the current to supramaximal, sample the adjacent muscle, ask the patient to relax fully — and observe whether the finding persists. A conduction block that disappears when stimulation is made truly supramaximal was sub-maximal stimulation; a fibrillation potential that vanishes when 60 Hz noise is suppressed was line interference; a slowed velocity that normalises on warming was temperature. The reproducibility test is the practical form of falsifiability: it asks of every abnormality whether it can survive an honest attempt to make it disappear, and only findings that survive belong in the report.

Common Pitfall

Reporting an unreproduced abnormality — a single low-amplitude response, an isolated run of spontaneous activity, a borderline latency — recorded once and never challenged. The borderline finding is precisely where artifact and pathology are indistinguishable on a single observation, and precisely where confirmation bias is most likely to resolve the ambiguity in favour of the expected diagnosis. If you cannot reproduce it after changing a variable, it does not go in the report.

Key points

Benign fasciculations are pathological only alongside denervation, chronic reinnervation, and reduced recruitment across multiple levels — in isolation they are a normal variant.
Chronic, stable, polyphasic units describe an already-healed injury; only spontaneous activity and unit instability mark an active, ongoing process.
Cooling slows conduction ~1.5–2.4 m/s/°C and enlarges amplitudes, fabricating demyelination or masking axonal loss — warm the hand to ~32°C and foot to ~30°C before interpreting.
Co-activation, sub-maximal or spreading stimulation, electrode misplacement, 60 Hz noise, and stimulus artifact each imitate genuine pathology and demand active suspicion.
Anchoring, confirmation bias, premature closure, base-rate neglect, and referral framing are the default mode of expert pattern-recognition and must be deliberately counteracted.
A finding that cannot be reproduced after changing a variable should not be reported — reproducibility is the practical test of falsifiability.