As a result, drug treatment for depression generally involves starting with a cheap medication with relatively mild side-effects, and if that fails, moving onto a series of other drugs until one helps. But since it can take several weeks for any new drug to work, this can be a frustrating process for patients and doctors alike.
Some means of predicting the antidepressant response would thus be very useful. Many have been proposed, but none have entered widespread clinical use. Now, a pair of papers(1,2) from UCLA's Andrew Leuchter et al make the case for prediction using quantitative EEG (QEEG).EEG, electroencephalography, is a crude but effective way of recording electrical activity in the brain via electrodes attached to the head. "Quantitative" EEG just means using EEG to precisely measure the level of certain kinds of activity in the brain.
Leuchter et al's system is straightforward: it uses six electrodes on the front of the head. The patient simply relaxes with their eyes closed for a few minutes while neural activity is recorded.
This procedure is performed twice, once just before antidepressant treatment begins and then again a week later. The claim is that by examining the changes in the EEG signal after one week of drug treatment, the eventual benefit of the drug can be predicted. It's not an implausible idea, and if it did work, it would be rather helpful. But does it?Leuchter et al say: yes! The first paper reports that in 73 depressed patients who were given the antidepressant escitalopram 10mg/day, QEEG changes after one week predicted clinical improvement six weeks later. Specifically, people who got substantially better at seven weeks had a higher "Antidepressant Treatment Response Index" (ATR) at one week than people who didn't: 59.0 ± 10.2 vs 49.8 ± 7.8, which is highly significant (p less than 0.001).
In the companion paper, the authors examined patients who started on escitalopram and then either kept taking it or switched to a different antidepressant, bupropion. They found that patients who had a high ATR after a week of escitalopram tended to do well if they stayed on it, while patients who had a low ATR to escitalopram did better when they switched to the other drug.
These are interesting results, and they follow from ten years of previous work (mostly, but not exclusively, from the same group) on the topic. Because the current study didn't include a placebo group, we can't say that the QEEG predicts antidepressant response as such, only that it predicts improvement in depression symptoms. But even this is pretty exciting, if it really works.
In order to verify that it does, other researchers need to replicate this experiment. But they may find this a little difficult. What is the Antidepressant Treatment Response Index use in this study? It's derived from an analysis of the EEG signal, and we're told that you get it from this formula:
Some of the terms here are common parameters that any EEG expert will understand. But "A", "B", and "C" are not. They're constants, which are not given in the paper. They're secret numbers. Without knowing what those numbers are, no-one can calculate the "ATR" even if they have an EEG machine.Why keep them secret? Well...
"Financial support of this project was provided by Aspect Medical Systems. Aspect participated in the design and conduct of the study; collection, management, analysis, and interpretation of the data; and preparation and review of the manuscript."Aspect is a large medical electronics company who developed the system used here. Presumably, they want to patent it (or already have). We're told that
"To facilitate independent replication of the work reported here, Aspect intends to make available a limited number of investigational systems for academic researchers. Please contact Scott Greenwald, Ph.D... for further information."All very nice of them, but if they'd told us the three magic numbers, academics could start trying to independently replicate these results tomorrow. As it is, anyone who wants to do so will have to get Aspect's blessing, which, with the best will in the world, means they will not be entirely "independent".
[BPSDB]
Leuchter AF, Cook IA, Gilmer WS, Marangell LB, Burgoyne KS, Howland RH, Trivedi MH, Zisook S, Jain R, Fava M, Iosifescu D, & Greenwald S (2009). Effectiveness of a quantitative electroencephalographic biomarker for predicting differential response or remission with escitalopram and bupropion in major depressive disorder. Psychiatry research PMID: 19709754Leuchter AF, Cook IA, Marangell LB, Gilmer WS, Burgoyne KS, Howland RH, Trivedi MH, Zisook S, Jain R, McCracken JT, Fava M, Iosifescu D, & Greenwald S (2009). Comparative effectiveness of biomarkers and clinical indicators for predicting outcomes of SSRI treatment in Major Depressive Disorder: Results of the BRITE-MD study. Psychiatry research PMID: 19712979
16 comments:
Excuse my language, but what an utter load of crap. Thanks for this post though!
It's not necessarily crap, it could be though, time will tell...
Why would any decent journal carry a paper with secret parameter values?
IMHO, there a a lot more questions to rise.
#1 How could the the journal allow to split one project into 2 papers (an unfortunately well known but nevertheless incredible process).
#2 Everyone who is familiar with EEG recordings will know that recordings at this particular sites are very prone to ocular artifacts. Usually, one would use this sites to record ocular artifacts rather than the EEG. In our Lab we used to call these electrodes HEOG1 and HEOG2. However, the authors state nothing about rejection or correction algorithms.
#3 The analysis method used in the paper is very uncommon (using 2 different reference electrodes without averaging) since the brain is a volume conductor. Moreover, the formula (even if at least someone knows the magic numbers) is not at all based on a proper theory: which parts of the frontal cortex (which generators) are responsible for which frequency bands and why are the different bands summed up in this curious ATR formula? Does this formula really represent an suitable model for the interaction of the brain areas during depression and/or the treatment? Or is it a fitted curve used without any theory at all? Then, why they did not include the size of shoe to reduce the variance ;)
#4 The HAMD is a questionaire usually filled out by the clinician rather than by the patient. Altough it is widely used in studies one should be cautious in using it exclusively to assess the patients well-beeing.
Great points - I'm no EEG expert so I didn't pick up on those.
On #2, I suppose technically it doesn't matter where the predictive signal is coming from, so long as it predicts improvement, it could come from eye movements and it would still be clinically useful. But yes, that is troubling.
On #3, they say that they came up with it based on previous work, and to be fair there are a dozen or so previous papers (from Leuchter et al mostly) which might shed some light on how they came up with the formula. Or maybe they just made it up.
On #4, my reading is that the HAMD was clinician-administered (as it should be) - "Inventory of Depressive Symptomatology Clinician-rated and the Ham-D17 to measure core diagnostic and commonly associated symptoms of depression. The IDS-C and Ham-D17 were administered using a combined structured interview guide". The HAMD is far from perfect but everyone uses it so you can't really fault them on that...
dearieme: Well good question, but unfortunately journals rarely require authors to make their methods 100% transparent. Sometimes this is for reasons of space, in this case that's clearly not an issue because it would take about one line of text to print those three numbers. But journals, even very good ones, often do this kind of thing. The Lancet, an excellent journal, recently published a very influential meta-analysis of antidepressants with an opaque methodology.
People generally assume in these cases that if you wanted to replicate the results you could write to the authors and they would tell you exactly what they did, which is fair enough because usually academics do. But if someone wrote to these guys and they refused to divulge the numbers it would be a bit of a scandal... maybe someone should try.
It might be interesting to email the authors and ask for the parameter values. Not that anyone would replicate this in a zillion years, but I'm curious about their reaction.
Maybe A, B and C are whatever numbers you need to bring your p-value below 0.05?
On #2
"I suppose technically it doesn't matter where the predictive signal is coming from, so long as it predicts improvement, it could come from eye movements and it would still be clinically useful"
That's exactly the argument in favor of psychoanalysis, astrology, and many more 'treatments that work'. However, what we need IMHO is an experimentally derived scientific model of the relationship between the factors.
The electrode placement isn't that bad. they have their ground pretty close to their messurement electrode, so big things like eye blinks will be in both and therefore be removed ... at least in that version of the reference, I agree that having multiple is odd.
It should be quite easy for researchers to come along and figure out A B and C if they want to, as Yigal said, you perform the experiment, and see which values separate your two groups the most. A simple multiple regression would do that for you.
Kevin H: But if you do that you are guaranteed to find values that discriminate in your experimental group - the question then would be, do they work in a different group.
In other words, you might do that and end up with values of A,B and C which were completely different from the ones these authors used. In which case one (or both) of you is wrong.
Response to antidepressants is correlated with frontal asymmetry as measured by resting EEG. This paper is probably measuring this no?
There is, in fact, a patent held by Aspect Medical Systems, Inc for "System and method of assessment of neurological conditions using EEG" (no surprise there). The list of industry sponsors and affiliations in the disclosure section is longer than Appendix A.
"The electrode placement isn't that bad. they have their ground pretty close to their messurement electrode, so big things like eye blinks will be in both and therefore be removed ... "
In order to remove the ocular artifacts you would need to measure the ocular artifacts (independently) if you don't want to come up with an ICA. only with this data you would be able to calculate a regression. However, all they measured was the EOG.So they would only be able to reject trials with high amplitude ocular artifacts as eye blinks. Smaller eye movements (like saccades) can not be detected.
Neurocritic: I knew it! I just failed in my ability to find U.S. patents.
The method they have used here (ATR) is most commonly published as Cordance. This measure has been described in some papers and can be used for replication I guess. However, the Cordance algorithm itself has been patented by Leuchter & Cook long time ago, as I understand it Aspect is licensing that patent.
Altough these results are promising they still require patients to take antidepressants for 1 week. Furthermore, the false positive and false negative rates of these results are substantial as well, making the step to clinical use harder I think.
Post a Comment