The effect of spectrum bias on the utility of magnetic resonance imaging and evoked potentials in the diagnosis of suspected multiple sclerosis

Background: Spectrum bias is the tendency for the effectiveness of a test (or treatment) to vary as a function of clinical, pathologic, or comorbid variables including disease severity. Our null hypothesis was that within the population of suspected multiple sclerosis (MS) patients, there would be no difference in MRI and evoked potential (EP) sensitivity and specificity between those with mild MS versus those with more severe clinical signs of the disease, i.e., that spectrum bias is not a factor. Methods: Three hundred three patients with suspected MS were evaluated by a board certified neurologist, then scanned with MRI. Two hundred four patients also received EP testing. Results: The sensitivity of MRI in patients with suspected MS was 58 percent with a false-positive rate of 9%. The overall sensitivity was 64% in the probable and 45% in the possible group. In the low pretest probability group sensitivity was 20%, and it was 70% in the high pretest probability group. These differences in sensitivity are statistically significant (p < 0.03 In contrast, the specificity between groups did not differ significantly. EP sensitivity was 69% in the high probability subgroup and 5% in the low probability subgroup, (p < 0.01). Conclusions: In this study, both EP and MRI results demonstrated an association between disease frequency, disease severity, and test sensitivity with greater disease frequency and intensity suggesting more impressive diagnostic test performance. The distorting effect of the variable clinical severity on MRI and EP sensitivity in suspected MS underscores that diagnostic tests perform differently in different groups of patients.