I read with dismay the article by Frisoni et al regarding the radial width of the temporal horn in Alzheimer disease (1). The authors claim that, armed with one CT scan of the brain and a ruler, they can make a single measurement that will distinguish patients with early Alzheimer disease from age-matched control subjects with a sensitivity of 93% and specificity of 97%, up to the age of 90 years; this was better in fact than they could apparently achieve with MR imaging and far better than other measurements they tried, about which others have made similar extravagant claims. This, then, is another article with the “magic measurement,” like the thickness of the substantia innominata, also in an article appearing in the same issue (2).
A moment’s reflection surely makes clear the implausibility of a result like this. One suspects that the answer lies in the control subjects, selected because of no clinical or CT evidence of a neurodegenerative disease. It seems clear that subjects with CT findings of excessive atrophy were excluded as controls and that the results reflect only how efficient this exclusion process was. I, and many others, will take a lot of convincing by the authors that the situation is otherwise. Indeed, so many articles in related fields seem to make the same mistake that I think an editorial or commentary should be dedicated to it—not the sort of laudatory commentary as appear on page 33 of this issue (3), but a critical appraisal. The “past glory” is not all that glorious, and the “future promise” is most uncertain, if it is the truth about the real world we are seeking as opposed to the pursuit of producing nice studies that have value only as an art form.
References
Reply
Basically, Dr Stevens claims that the positive results of our study (1) were obtained by excluding unwelcome data (ie, control subjects with CT findings of excessive atrophy). We strongly disagree with Dr. Stevens’ assertions.
Our controls were persons with no cognitive symptoms who underwent CT mostly for headache or dizziness (83% of our 29 controls) and whose CT findings were normal. Thus, although we did not take measures of physical comorbidity, it is likely that these persons were in reasonably good physical health. Because age-associated temporal atrophy in physically healthy elderly persons is absent or minimal (2), a serene mind would not find it surprising that our controls had very little age-associated medial temporal lobe atrophy and that their atrophy measures separated them well from Alzheimer disease patients.
What might have been contended with more support is rather that these controls are not representative of the clinical world, where physicians are challenged with patients who do report cognitive symptoms. Indeed, our own is a phase I study of a diagnostic tool, aiming to answer the question, “Do test results in patients with the target disorder differ from those in normal people?” (3). Final evidence of clinical usefulness would require demonstration of high positive and negative predictive values (phase II), high sensitivity and specificity in clinically meaningful groups (phase III), and good test efficacy on ultimate health outcomes (phase IV). Radiologic, as well as nonradiologic, diagnostic tools are usually supported by phase I, seldom by phase II, and very rarely by phase III and IV studies (3).
Indeed, for our own diagnostic tool—as well as for most others—this evidence still needs to be provided (4). However, the high frequency of use of CT in the diagnosis of cognitive impairment and the often limited human and technological resources in diagnostic facilities make a feasible CT-based marker of Alzheimer disease a potentially significant incremental diagnostic value. This is not magic but simply good clinical practice.
References
- Copyright © American Society of Neuroradiology