Are pet scans good enough to diagnose dementia?

I recently had the opportunity to speak at the national American Academy of Family Physicians annual meeting (called the AAFP Scientific Assembly).  I spoke on dementia from a primary care perspective. One question that came up was that a previous speaker stated the PET scans could be used to diagnose dementia.

Now I work at an academic setting and I have only ordered a PET scan once (and in retrospect I should have referred the patient to a memory center instead of ordering the PET scan).  And it was kind of interesting to me that there were primary care docs who order these things.  I see a lot more patients with dementia than the average primary care doc by far.  Probably half my patients have dementia.  It's one of the main reasons I get consulted and yet I never (ever) order PET scans.  So I decided to do some math to figure out why.

So some basic starting points-according to the Alzheimer's Association, PET scans to make the diagnosis of dementia are 95% sensitive and 75% specific.  Sensitivity means that if someone has dementia, the test will pick it up.  Specificity means that the test doesn't pick up other things like depression.  For more info you can look on google or wikipedia.

While the sensitivity seems great, the specificity is the achilles heel.

The next numbers are so key. It has to do with positive and negative predictive value.  Again, look on google.  Or you can use this handy dandy calculator.

Let me cut to the chase:

3 scenarios:
High likelihood.
If a clinician has a high suspicion that a person has dementia (let's say they are 75% sure), then if the test is positive (suggests dementia) then the likelihood increases to 92%.  If the test is negative then the test has an 83% chance that the person does not have dementia.  In other words, for patients for whom dementia is strongly suspected, then the test will be great and strengthening the diagnosis (but it still will be wrong nearly 1 in 10 times).  If the test is negative, it is good and making everyone rethink the tentative diagnosis.

So for patients for whom dementia is already strongly suspected, adding a test that agrees may add a shade of confidence which may not be terribly useful.  What is more useful in this setting is that if the test is normal, then there needs to be a re-evaluation of the cause of the cognitive deficits (memory loss)

Low likelihood
If the clinician thinks the pt does not have dementia (25% chance of having dementia), then a negative test essentially rules it out (98% chance that the person does not have dementia).  Again useful maybe but not terribly.  If the test is suggests dementia, it is only right 55% of the time which is a pretty horrible result for a test.  It means it's barely better than a coin flip.

So for patients for whom dementia is not suspected, it is good for confirming that (which is not terribly useful) and if it disagrees, it is no better than a coin flip (also not terribly useful).  In other words, for patients with a low suspicion of dementia, this test is not terribly useful.

For an intermediate likelihood.
Let's say the clinician has no idea.  The patient could have dementia but they might not (50% likelihood=coin flip).  If the test is positive, it means the person has dementia 80% of the time which may seem pretty good but it's wrong 1 in 5 times.  That's a lot of people to freak out.  If the test is negative, it is correct 94% of the time which is good for a rule out.

In other words, for a random probability, a PET scan can be used as a rule out but not so much to make the diagnosis.

Let me put this is in a chart if I can…

So summarizing, I would say that the test is useful as a rule out type of test for those who have an intermediate or low suspicion.  Not so much to make the diagnosis.  I wouldn't be comfortable telling someone they have a fatal neurodegenerative disease when I have a 20% chance of being wrong.  Or even 8% chance.

The second point is that the test is only useful in the setting of a clinical suspicion.  You can see how the characteristics of the test change depending on clinical suspicion.  However when people talk about PET scans, they imply that maybe it would be useful BEFORE a person has clinical symptoms.  It's not there yet. Maybe the new Amyloid PET scans but not the tagged glucose pet scans.

After going through the math, this is why I don't use PET scans.  I think it's more useful to hone my clinical skills than use a test to compensate for poor clinical skills.