A Non-Rhetorical Question: What’s the Accuracy of a Clinical Diagnosis of Alzheimer Disease?
The use of Avid/Lilly’s newly approved amyloid tracer, Amyvid, hinges on the accuracy (or inaccuracy) of a clinical diagnosis of Alzheimer disease (AD) during life. If the historic accuracy of a clinical diagnosis of AD is generally high—meaning the correlation between a clinical diagnosis of AD during life and postmortem evidence of the disease is good—then use of the very expensive radiolabeled tracer (and the consequent ordering of associated PET scanning) is less well justified, if at all.
In Avid/Lilly’s press release for the approval of Amyvid, Avid’s President and CEO, radiologist Daniel Skovronsky, implicitly promotes the use of amyloid imaging by claiming, “It’s estimated that one in five patients clinically diagnosed with
probable Alzheimer’s Disease during life do not end up having
Alzheimer’s Disease pathology upon autopsy.” The references cited to support his statement are the following:
- Lim A, Tsuang D, Kukull W, et al. Clinico-neuropathological
correlation of Alzheimer’s disease in a community-based case series. J Am Geriatr Soc. 1999;47(5):564-569.
- Petrovitch H, White LR, Ross GW, et al. Accuracy of clinical
criteria for AD in the Honolulu-Asia Aging Study, a population-based
study. Neurology. 2001;57(2):226-234.
Let’s examine these references to determine if they really do support Skovronsky’s claim of diagnostic accuracy and, moreover, if they’re representative of the literature at large.
The article by Lim et al, from 1999, is not available online. Therefore I’m currently limited to examining the PubMed abstract, which describes a “community-based case series” of 134 patients with “memory complaints” (who were presumably seen at the VA Puget Sound Health Care System in Seattle, Washington). The rationale for the study was to make a clinico-pathologic correlation for AD in non-research subjects. These community-based subjects, who were enrolled in the University of Washington/Group Health Cooperative Alzheimer’s Disease Patient Registry, met a diagnosis of “probable AD” on the basis of NINCDS-ADRDA criteria (1984 guidelines). Among the 134 individuals, postmortem criteria for AD (CERAD criteria) were met in 95 patients (71%). The authors calculated the sensitivity of a clinical diagnosis of AD (“probable AD” during life) at 83% and an overall diagnostic accuracy of 75%. Of the 94 cases with postmortem evidence of AD, most of these patients demonstrated coexisting pathology—like vascular lesions and evidence of Parkinson disease. Neuropathologic diagnoses in the patients without postmortem evidence of AD are, unfortunately, not provided in the abstract. In other words, the Lim abstract doesn’t reveal what was misdiagnosed as AD during life in this study.
The other cited article, by Petrovitch et al, describes a kind of a holy-smokes study, in which the clinico-pathologic correlation for a diagnosis of AD in Japanese-American men (who participated in the Honolulu-Asia Aging Study in the early-mid 90s) was a paltry 65%. (Again, the rationale for this study was to assess the diagnostic accuracy in a community-based sample [ie, outside of a referral center], as well as to examine a non-Caucasian population.) A closer look at the article reveals that, among the 20 men (ie, a small study) diagnosed with “pure AD” (as opposed to pure vascular dementia or AD plus something else) who came to autopsy, only 13 (65%) met CERAD criteria for “definite” or “probable” AD (which was partly based on the density of neuritic, amyloid-containing plaques*). Many of these patients also showed evidence of amyloid angiopathy and/or cerebrovascular disease. Among the patients with “possible” or no postmortem evidence of AD (ie, with minimal evidence of neuritic plaques), diagnoses for the most part were confined to vascular disease (which CT or MRI somehow missed). In one case, there was evidence of midbrain Lewy bodies (eg, Parkinson disease); in another, there was evidence of chronic subdural hematoma.
Petrovitch et al also cite several previous clinico-path studies (performed at referral centers in Western countries), in which the correlation of a diagnosis of “pure” AD during life and postmortem evidence varied between 80% and 90%. In cases of “probable” AD, the correlation rates were even higher: between 80% and 100%. The accuracy of the clinical diagnosis of “possible” AD, logically, has been historically lower: 60%-90%.
In a very recently published study, which draws on more recent NIA registry data (2005-2010), the sensitivity of a clinical diagnosis of “probable” or “possible” AD (when measured against neuropathologic data—which were based on amyloid-plaque density* and neurofibrillary tangles) was about
80%70%-90%. Meaning: Physicians correctly identified clinical AD as being neuropathologic AD about 85% 80% of the time (438/526 subjects). These data are consistent with the median values provided by the bulk of clinico-path studies in the literature. Conversely the specificity of clinical diagnosis was lower, from about 40%-70%. Meaning: Clinicians were not as good at correctly recognizing non-AD causes of dementia and often ascribed other causes of dementia to neuropathologic AD.
Among the most common primary neuropathologic diagnoses that were misidentified as AD during life were tangle-only dementia or
argyrophilic grain disease, frontotemporal lobar degeneration (aka Pick’s disease),
cerebrovascular disease, Lewy body disease, and hippocampal sclerosis. (Addendum: It’s also curious to note that among the 88 patients who were diagnosed with probable AD but who didn’t meet the minimum pathologic standards for AD at autopsy, 17 were still considered to have a “primary neuropathologic diagnosis of AD despite [the] low level of AD histopathology.” So the rate of diagnostic accuracy for clinical AD was a slightly higher, at ~85%. But the authors note that the criteria for the histopathologic diagnosis of AD has been a bit of a moving target over the last few decades.)
So circling back to Dr. Skovronsky’s quote, a
25% 20% rate of a misapplied diagnosis of AD during life is a bit of an overstatement reasonable—especially if the reference standard for clinical diagnosis is some type of AD referral center. (And I bet any primary care physician who might ever entertain an amyloid-imaging study would first refer his suspect patient to an AD referral center, or at least a neurologist.) A more accurate percentage of a misapplied diagnosis, on the basis of a totality of the literature, is closer to 15%. But despite the reasonable agreement with the literature at large, Skovronsky should have cited some other reference than the Honolulu-Asia Aging Study, which indicates a relatively high rate of misdiagnosis, at 35%.
Regardless, the crucial follow-up question is What are we missing when we mistakenly diagnose dementia as AD during life? And the answer is a number of related neurodegenerative, dementing illnesses and cerebrovascular disease—features of which may coexist. Moreover, evidence of some of these non-AD conditions (hippocampal sclerosis, vascular disease) should be detectable on standard MR imaging.
* What Amyvid is designed to detect.
NINCDS-ADRDA = National Institute of Neurological and Communicative Disorders and Stroke and the Alzheimer’s Disease and Related Disorders Association; CERAD = Consortium to Establish a Registry for Alzheimer’s Disease.