Thursday, April 28, 2005

Direct-to-Consumer Advertisement of Antidepressants

Direct-to-Consumer Advertisement of  Antidepressants; or, Why Doctors Should Remember Bayes' Theorem

In the latest (4/27/2005) issue of the Journal of the American Medical Association, there is a research paper entitled Influence of Patients’ Requests for Direct-to-Consumer Advertised Antidepressants.  The intent of the authors was to investigate the effect of patient's requests for antidepressants on the prescribing practices of primary care physicians.  The study was conducted by having actors go to doctor's offices and simulate a patient encounter.  They arrive at the unsurprising conclusion that direct-to-consumer (DTC) advertising of antidepressants can have an effect upon the prescribing practices of primary care physicians.

This is one of those studies, that, after reading it, one has the impression that the researchers have proven something...although it is not clear exactly what was proved.

The idea is this: simulated patients went to doctor's offices and mentioned a list of symptoms.  In some cases, the list was intended to mimic Adjustment Disorder; in others, Major Depression.  In both types of simulated encounter, actors presented one of three scenarios.  In some, they mentioned having seen an advertisement for a particular antidepressant; in others, they made a general request for an antidepressant, without mentioning a specific drug; in the third scenario, they did not mention a specific drug and did not request an antidepressant.  Thus, there were six types of simulated encounters.  The numbers in each cell indicate the percentage of times that an antidepressant prescription was written. 

Actor mentions specific drug Actor makes general request for drug Actor makes no request
Actors mimicking depression 53% 76% 31%
Actors mimicking adjustment disorder
55% 39% 10%

There are limitations to the study.  One major methodological problem is that the study was not a double-blind study.  The actors knew that they were not real patients, although the doctors did not know that the simulated patients were actors.  (The doctors did know that some of the patients that they would be seeing during the study period would be actors, but not which ones.)  Another limitation is the type of control.  The control in this study is provided by the encounters in which patients mimicked a condition similar, but not identical, to depression.  The intent was to show what would happen if the "patients" could not really benefit from an antidepressant, but asked for one anyway.  The fact is, some patients with adjustment disorder might benefit from a prescription. 

The patients who mimicked having adjustment disorder had been instructed to report symptoms of low back pain, feeling stressed, and having occasional insomnia.  Those are all nonspecific symptoms, but a patient with depression very well might present in the office with exactly those symptoms.  Those asked to mimic depression had been told to report wrist pain and feeling “down” for a month or so, loss of interest in activities, fatigue, low energy, poor appetite and poor sleep.  That is a more specific list of symptoms, but is not necessarily conclusive.  Thus, the doctors were were in the position of having to distinguish between two shades of gray, which is more difficult than distinguishing black from white. 

The most obvious conclusion from the numbers is that people who ask for a drug are much more likely to get one than those who do not.  What is mentioned in the report, but not shown in the table, is that most of time, when actors mimicked depression, they got an appropriate intervention: any combination of an antidepressant, mental health referral, or follow-up within 2 weeks.  Interestingly, the rate of appropriate intervention was a bit lower (90%) in the instances in which a specific request was made, compared to 98% in the instances in which a general request was made. Somewhat disappointing is the fact that only 56% of the instances in which depression was simulated, but no drug request was made, there was not appropriate intervention.  In my opinion, that is one of the major findings of the study, even though it was not one of the objectives.

The other really interesting finding is that the pseudo-depressed actors were more likely to get a prescription if they made a general request, rather than asking for a particular drug; the opposite was true for the pseudo-adjustment-disorder actors.

The authors' conclusions were as follows:
Conclusions  Patients’ requests have a profound effect on physician prescribing in major depression and adjustment disorder. Direct-to-consumer advertising may have competing effects on quality, potentially both averting underuse and promoting overuse.
I actually agree with these conclusions, although I am not sure that the study really proves the point.

The study has been picked up by several mainstream news organizations and some specialty services (1  2  3  4  5), and at least one blogger.  The LAT article mentions one point, in particular, that I would like to amplify.
"There's a whole lot of medicine that is practiced in the gray zone," where social influences matter as much as clinical findings, said Dr. Richard Kravitz, a professor of medicine at UC Davis and lead author of the study.

Depression can be difficult to diagnose, and many people resist the possibility that an illness may be mental. A openness to trying an antidepressant appeared to be an important cue to the physicians, Kravitz said.
I'm not sure what the author meant; the second paragraph contains a confusing mix of messages.  The last sentence, though, is the important one.  Here's why: When the doctor is trying to decide what to do, he or she first will formulate an hypothesis, then test the hypothesis.  Some of the testing is conscious; some not.  Once the doctor starts thinking 'maybe this patient is depressed,' she or he will then sift through the patient's presentation and look for clues.  On a conscious level, such clues would include the symptoms that the patient reported.  Perhaps, on an unconscious level, the fact that the patient asked for an antidepressant makes the diagnosis seem more likely.  I am not aware that this has been studied, specifically, but I suspect rather strongly that if it were tested, it would be found to be true.  That is, if you videotaped a zillion real patient encounters, and looked to see if the patients who asked for an antidepressant were more likely to be depressed, I think you would find that it is the case.  Unfortunately, the influence of DTC advertising of pharmaceuticals may degrade the usefulness of that as a diagnostic criterion.  Person who are really interested in this concept may be interested in an explanation of the mathematical basis for this phenomenon: An Intuitive Explanation of Bayesian Reasoning:
100 out of 10,000 women at age forty who participate in routine screening have breast cancer.  80 of every 100 women with breast cancer will get a positive mammography.  950 out of  9,900 women without breast cancer will also get a positive mammography.  If 10,000 women in this age group undergo a routine screening, about what fraction of women with positive mammographies will actually have breast cancer?

The correct answer is 7.8%, obtained as follows:  Out of 10,000 women, 100 have breast cancer; 80 of those 100 have positive mammographies.  From the same 10,000 women, 9,900 will not have breast cancer and of those 9,900 women, 950 will also get positive mammographies.  This makes the total number of women with positive mammographies 950+80 or 1,030.  Of those 1,030 women with positive mammographies, 80 will have cancer.  Expressed as a proportion, this is 80/1,030 or 0.07767 or 7.8%.
In my opinion, the authors may want to study this next:

Are the patients who come to the office and report symptoms of depression and ask for an antidepressant more likely to have depression than those who come to the office and report the symptoms, but don't ask for an antidepressant? 

If so, does the presence of DTC advertising affect the predictive value of the presence or absence of the request?

Why is this more interesting to me, than the study they actually did?  It is more interesting because it would be more useful.  It is very useful, clinically, to know which observations have predictive value, and to know what factors affect that predictive value.  It is less useful to know whether DTC advertising affects the prescribing habits of doctors.  (Of course it does; that is why the drug companies do it!)

How could the study that was actually done help a clinician?  If a patient comes in and asks for a specific drug, perhaps the doctor could take a few extra minutes to ask some questions that would facilitate a correct diagnosis.  Sure, but shouldn't that be done anyway?

I suspect that the authors of the study intended it to contribute to the public debate about the value and problems associated with DTC advertising.  It does that, but it also serves another purpose.  It reminds clinicians that Bayes' Theorem is a critical part of the diagnostic process, and it works best when it is considered on a conscious level.