If you haven’t had patients asking you to order them a PET scan to diagnose Alzheimer’s disease yet, brace yourself. It’s only a matter of time.
PET scanning for AD was recently endorsed by no less august a figure than Charlton Heston (of “Ben Hur,” “Ten Commandments,” and, less gloriously, “Bowling for Columbine” fame), who announced in 2002 that he had been diagnosed with AD. Heston had evidently been won over by the Academy of Molecular Imaging, a professional group that is boosting PET scanning with almost religious zeal. The actor’s endorsement was accompanied by a press release quoting UCLA’s Daniel Silverman’s claim that PET decreases false AD diagnosis “by almost half,” and that using PET would lead to a “62% decrease in avoidable months of nursing-home care and a 48% drop in unnecessary drug treatment.”
With numbers like these, who wouldn’t be sold on the technology?
While TCR hates to be a killjoy, the story on PET scanning for AD is quite a bit more complex than this. This article will guide you through this confusing landscape, and we will–sorry to say–be discussing such dreaded statistics as “sensitivity” and “specificity.”
Studies have shown that a psychiatrist’s conventional low-tech approach to diagnosing dementia is quite accurate. A clinical diagnosis of “possible or probable AD” has a sensitivity of about 90%, and a specificity of about 50% (for a review of these studies, see “Practice Parameter: Diagnosis of Dementia (An Evidence- Based Review)” Neurology 2001; 56:1143- 1153).
“Sensitivity” is defined as the proportion of true cases picked up by the diagnostic procedure. Thus, a 90% sensitivity means that you pick up most of the genuine AD patients you see, allowing very few patients to go undiagnosed. “Specificity” is a measure of the precision of your diagnosis. A specificity of only 50% means that half of the patients you diagnose as having AD don’t actually have it. You’re being overly inclusive, and you generate a lot of “false positive” diagnoses.
Is this such a bad thing? Not necessarily. By far the majority of the patients you’ve falsely diagnosed with AD have some other form of dementia, usually vascular dementia, and often some mixture of AD, vascular dementia, and Lewy Body Disease (J Am Geriatr Soc 1999; 47:564-569). The evidence is increasingly clear that all such patients benefit from immediate treatment with cholinesterase inhibitors, so the false AD label is unlikely to harm them in any way.
Now, the question we have to ask about PET is: Does a $1,500 PET scan add anything of value over and above what we can accomplish with the conventional workup?
The largest study reported to date relevant to this question involved 284 patients who were PET-scanned as part of the evaluation of cognitive deficits between 1984 and 1998 (JAMA 2001; 286:2120-2127). At least two years of clinical follow-up was obtained (via chart review) on about half of these patients (the UCLA portion of the study); the other half were treated at other sites, both in and out of the U.S., and researchers obtained autopsy data on these patients.
A nuclear medicine specialist read all the scans (he was, of course, blinded to the eventual outcome) and divided them into two categories: 1. “progressive scans” (a scan indicating dementia of some type) and 2. “nonprogressive scans” (no abnormal findings or nonspecific findings such as generalized cerebral atrophy).
How good were these two categories at predicting subsequent decline and diagnosis? Let’s look first at the 146 patients who were followed clinically at UCLA, and for whom we do not have autopsy data. Of these 146 patients, 86 of them subsequently experienced some type of dementia over an average of 3.2 years of follow-up. Researchers had correctly put 78 of those 86 PET scans in the “progressive” pile, for a diagnostic sensitivity of 91%. Thus, based on PET findings alone, 8 of 86 patients would have been falsely reassured that their memory would not get worse (i.e., false negative rate of 9%).
What about specificity? Of the 60 patients whose memory held steady over three years, PET correctly predicted 45 of them (specificity 45/60=75%). This means that 15 patients would have falsely been given the nasty news that the brain scan indicated that they had dementia (25% false positives). The autopsy portion of the study produced similar figures. At autopsy, 97 patients were diagnosed with AD. PET predicted 91/97 = 94% of these, and correctly predicted no AD in 30/41 = 73%.
So how do we interpret these figures? How likely is it that PET scan data added something of value to these patients’s subsequent management? Well, the patients who had true dementia would likely have been picked up by a standard clinical exam anyway (since the sensitivity is comparable to PET). Scan or no scan, they would have received a prescription for a cholinesterase inhibitor and been scheduled for follow-up in a month.
The patients whose scans accurately predicted lack of dementia could breathe a $1,500 sigh of relief. But we psychiatrists can produce that same sigh much more cheaply with a careful clinical exam.
Where PET scans caused problems was for those 15 “false positive” patients whose scans were read as indicating dementia but who turned out to be fine 3 years later. In clinical practice, patients whom we falsely label AD generally have another type of dementia, so the treatment they get is actually helpful, despite the misdiagnosis. But PET has a nasty habit of finding AD-like perfusion defects in cognitively normal people. Thus, PET risks causing a lot of normal patients to become extremely anxious, not to mention leading to unnecessary further workup and medication. So even though there are fewer false positive diagnoses with PET scan, they are clinically more significant.
It appears, then, that PET scans are not yet “ready for prime time,” an opinion seconded by the Agency for Health Research Quality (AHQR), which just concluded that PET does not add significant information to a competent clinical exam (http://www.cms.hhs.gov/coverage/ download/ADTAFinalReport-042904-2.pdf). This federal opinion constitutes a death knell for those seeking Medicare coverage for PET in evaluating AD.
It’s too bad. Those images sure are pretty.
TCR VERDICT: PET Scans: Too Much Bad News for the Absent-Minded