Evidence-Based Electrodiagnostics

Operating characteristics and the meaning of a normal study

Every electrodiagnostic test has a finite sensitivity and specificity, and the numbers vary enormously by indication. Nerve conduction in suspected carpal tunnel syndrome performs well — sensitivity on the order of 85% and specificity around 95% when modern comparison techniques are used — so a positive study meaningfully confirms and a negative study meaningfully (though not completely) argues against the diagnosis. Needle electromyography in cervical or lumbar radiculopathy is a different instrument entirely, with a sensitivity commonly cited at only 50–70%: it detects motor-axon loss but is blind to the many radiculopathies that are purely sensory, purely irritative, or too acute to have produced denervation. The mechanistic consequence is the single most important inferential fact in the field: a normal study does not exclude diseasewhen the test's sensitivity is moderate. A normal needle examination in a patient with a clinically convincing radiculopathy refutes nothing; it merely places the patient in the substantial minority the test cannot detect.

Bayesian inference with likelihood ratios

Sensitivity and specificity describe the test in the abstract; the quantity that actually updates a diagnosis is the likelihood ratio, which converts a pretest probability into a post-test one. Expressed in odds, the relationship is exact and elegant: pretest odds × LR = post-test odds. The positive likelihood ratio is sensitivity divided by (1 − specificity); the negative likelihood ratio is (1 − sensitivity) divided by specificity. For carpal tunnel testing, an LR+ near 17(0.85 / 0.05) means a positive study multiplies the pretest odds roughly seventeen-fold, while an LR− near 0.16 (0.15 / 0.95) reduces them only about six-fold — quantifying directly why this test rules in more powerfully than it rules out. The conventional anchors are that LR+ >10 and LR− <0.1 constitute strong evidence that meaningfully shifts probability, ratios between 2 and 5 (or 0.2 and 0.5) provide modest movement, and ratios near 1 are diagnostically inert regardless of how statistically significant the underlying study was.

The practical force of this framework is that it makes the pretest probability mathematically inescapable. The same positive result moves a patient from a 10% to a 65% post-test probability, or from a 60% to a 96% one — the identical likelihood ratio acting on different priors. An abnormal response in a clinically implausible context may still leave the diagnosis improbable, and a normal response in a clinically compelling context may leave it likely; the result and the prior must be multiplied, never substituted for one another. Reporting a finding without reference to the clinical pretest probability is reporting half of a calculation.

Diagnostic-accuracy study design and its characteristic biases

The sensitivity and specificity values an electromyographer relies on are themselves estimates from diagnostic-accuracy studies, and those estimates are systematically inflated by recognisable design flaws. Spectrum bias arises when a test is validated on florid, advanced cases against obviously healthy controls; the apparent accuracy is real for that spectrum but collapses in the mild, early, and ambiguous patients who actually generate diagnostic uncertainty in clinic. Verification (work-up) biasoccurs when the reference standard is preferentially applied to patients who already tested positive — common when the “gold standard” is surgery or imaging ordered because the electrodiagnostic study was abnormal — which inflates sensitivity and deflates specificity. The reference-standard problem is more fundamental still: for many conditions there is no independent gold standard, and the electrodiagnostic study is itself part of the diagnostic criteria, so its accuracy cannot be cleanly measured against a truth it helps define. STARD reporting standards exist precisely to force the disclosure of patient spectrum, blinding, and verification so that a reader can judge whether a quoted operating characteristic applies to their patient. A sensitivity figure quoted without its study context is a number without a denominator.

Guideline-based, value-driven, question-led testing

Evidence-based electrodiagnosis is operationalised through the practice parameters of the American Academy of Neurology and the American Association of Neuromuscular & Electrodiagnostic Medicine, which synthesise the accuracy literature into indication-specific recommendations — which studies add information in carpal tunnel syndrome, which combinations of techniques raise yield, when needle examination is warranted, and where the evidence is too weak to support a test. These guidelines convert the Bayesian and operating-characteristic principles into concrete protocol decisions and protect against both under-testing and the more common failure of reflexive over-testing. The unifying ethic is value-based, question-drivenstudy design: order the components that can change the diagnosis or the management, omit those that cannot, and let the pretest probability and the test's operating characteristics — not habit or completeness — determine the extent of the study. A study that cannot alter a decision regardless of its result has no value, however technically immaculate; a study designed to resolve a genuine diagnostic uncertainty is the highest use of the laboratory.