A Bayesian argument against rigid cut-offs in electrodiagnosis of median neuropathy at the wrist

Abstract

Background: Nerve conduction (NC) tests, using rigid cut-offs separating normal from abnormal test values, are commonly used to confirm median neuropathy at the wrist (MNW). The authors studied patients with clinically defined mild MNW and a normal median distal motor latency to determine 1) how much sensory or mixed NC test results increase (or decrease) the probability of MNW and 2) the NC test values required to confirm (or exclude) MNW for the range of pretest probabilities of MNW.Methods: Palmar, digit 4 (D4), and digit 2 (D2) median NC tests were reviewed in 125 hands with mild carpal tunnel syndrome (CTS) and 100 control hands with musculoskeletal pain. Receiver operating characteristic curves and interval likelihood ratios were plotted for the three tests. Using Bayes theorem, post-test probability of MNW was then determined for the range of pretest probabilities and NC test values.Results: Receiver operating characteristic curves showed that for a set specificity of 97%, palmar and D4 studies had higher electrodiagnostic utility than D2 studies with cut-off test values (sensitivities of 0.3 msec, 64.0%; 0.4 msec, 71.2%; and 50 m/sec, 44.8%). However, Bayesian analysis showed that to confirm MNW more conservative cut-off values (palmar 0.5 msec, D4 0.7 msec, D2 44 m/sec) were required for pretest probabilities ≤50%, whereas borderline abnormal values (palmar 0.4 msec, D4 0.5 msec, D2 48 m/sec) sufficed when pretest probabilities were ≥75%. Conversely, normal test values could exclude MNW only for pretest probabilities <25%.Conclusions: For a given NC test value, post-test probability of MNW can be determined from the estimated pretest probability (derived from clinical data), interval likelihood ratios, and Bayes theorem. Use of rigid cut-off values to confirm MNW is problematic, because more conservative cut-offs are required for low pretest probability. Conversely, NC tests with sensitivity <95% cannot exclude MNW when pretest probability is high.