Tuesday, February 26, 2008

Antidepression

I just looked at the guardian online, and saw the headline "Prozac does not work, say scientists". The article had a link to the original paper, which was very helpful, as this says something completely different.

First rule of reading scientific papers: just look at the plots, the rest is usually irrelevant. This plot from the paper shows how much patients improved when they were given either an antidepressive drug or a placebo.



Maybe there are too many dots and curvy lines to see what is happening. After squinting my eyes a bit, I have, using the power of MS Paint, reproduced the main features of this plot.



Note the baseline, which the authors felt was unnecessary. Mistake. First rule of drawing plots: add a baseline. Looking at the plot, and looking at the baseline, I drew the following conclusions:

1/ Taking antidepressive drugs improves, on average, the mood of people with all degrees of depression. Woohoo.

2/ For people with mild depression, the same improvement can be made with a placebo. For these people, it is not the active chemical of the antidepressive that is helping them, but the circumstances of being given and taking a "cure". This could lead to a long debate about the ethics and effectiveness of giving patients a placebo cure.

3/ For people with severe depression, the chemicals of the antidepressive work and improve their mood. Woohoo.

So antidepressive drugs have been shown to improve the mood of people with severe depression above the level achieved by a Placebo. Or, as summarised in the Guardian:

Prozac, the bestselling antidepressant taken by 40 million people worldwide, does not work and nor do similar drugs in the same class, according to a major review released today.


This is obviously bullshit, as a second look at those graphs will tell you. More subtle is the mistake made by the authors, who conclude:

The relationship between initial severity and antidepressant efficacy is attributable to decreased responsiveness to placebo among very severely depressed patients, rather than to increased responsiveness to medication.


This is confused thinking. Taking any drug gives potentially a medication effect and a placebo effect. If the total improvement caused by taking the drug stays the same, but the placebo effect diminishes, then the response to medication effect must have increased: the drug works.

Maybe there are other, better, options for some people, but antidepressive drugs do on average have a positive effect for people with severe depression.

Links:

This blog discusses the difference between statistical and clinical significance (the funny green area in the plot), so that I don't have to.

Don Jewett is an emeritus Prof. at the University of California and knows how to use his CAPSLOCK. He takes issue with the conclusions of the paper (link added 28. Feb)

The language log get into the graph-drawing spirit. I like their old-skool noughts and crosses marker styles. They also summarise the press coverage.(link added 1. Mar)

25 comments:

Anonymous said...

The clarity of the MS Paint graph is astonishing.

phil said...

Thankyou! Seeing as it is a knock-off of an open source graph, feel free to share and enjoy it!

Anonymous said...

Great post, and a great piece of art work.

Anonymous said...

A+ on that graph.

Anonymous said...

"Taking any drug gives potentially a medication effect and a placebo effect. If the total improvement caused by taking the drug stays the same, but the placebo effect diminishes, then the response to medication effect must have increased: the drug works."

That really depends on an assumption that treatment effect and placebo effects are additive. Do we know this definitely to be the case, or could it be that one effect *masks* the other. This second one seems t be the interpretation made by the authors. I am not very familiar ith research in this area, so don't know what evidence there might be for either assumption.

phil said...

Thankyou for the comments. To the second anonymous: You are right that I made an assumption of adding the effects. Maybe there is a more complex relation involving a combination of the two.

It is not necessary to know exactly how the placebo and chemical effects combine: if people taking the drug respond better than those taking a placebo, then the drug works. The graph shows that this is the case for some severely depressed patients.

Anonymous said...

"Note the baseline, which the authors felt was unnecessary."

Are you simply objecting to the jargon in the baseline given? The HRSD is used in the paper with some explanation, so it's not totally obscured.

Anonymous said...

from the paper:

"The curvilinear relation depended on only one trial of moderately depressed patients. When that outlier trial is excluded, there is no relation between baseline severity and antidepressant response."

Could this be the reason they concluded that "The relationship between initial severity and antidepressant efficacy is attributable to decreased responsiveness to placebo among very severely depressed patients, rather than to increased responsiveness to medication."?

pj said...

I think people need to be very careful what they mean by saying that the increased clinical effectiveness in severe depression is due to decreasing placebo response.

phil said...

To chai wallah: With "baseline", I meant the horizontal black line at the point of zero improvement. I drew one, while the original graph did not have one. I find it helpful for it to be drawn in so you can compare not just drug vs placebo, but drug or placebo vs "no treatment". As for your second post, I really don't know.

CCA said...

la exposición gráfica es excelente y gracias a ella he entendido el artículo mucho mejor. Muchas gracias y un saludo cordial.

Anonymous said...

if i understand the original graph correctly (if i dont please correct me) the dots and triangles are the specific data points represetnting individuale cases.
but between the initial severity levels of 16-22 there are only two datapoints. this does not seem to be enough infomation to base half of their curve graph on. while the it's fine to leave these points on the graph the relational line should surely not uses these points unless they are backed by more infomation. This would also make your graph better too.

Anonymous said...

Thiose two points are big shapes. That, I suspect, means lots and lots of points all in one place.

Anonymous said...

aahhh, i see, thanks Anonymous thats make a bit more sense.

pj said...

No that's one large study.

Anonymous said...

Unless that was one large case study there would still be many data points.

phil said...

Thanks again for the comments. All the above points about fitting a curve through the data seem fair to me. Areas of missing data or only small studies will lead to an uncertainty in the fitted curve. Note that nobody has attached any significance to the fitted placebo curve being better than the fitted drug curve for part of the graph.

pj said...

anonymous - true, but they wouldn't necessarily fall in or around the big shape since that is merely the average with its size due to the sample size.

phil - I guess no one attaches any significance to the cross-over of the lines because it is an extrapolation of the trend from the main body of the studies - looking at the only study in that region shows that the result is simply 'no difference' rather than placebo being better than drug.

Interestingly enough, I find that the graph flips over when you use the actual HRSD scores with drug effect sizes increasing with depression severity and placebo responses remaining fairly static.

Anonymous said...

Just to say, great post! Sorry I'm a bit late in posting a positive comment but I've been too depressed to get round to it (seriously!).

Anonymous said...

Who knows where to download XRumer 5.0 Palladium?
Help, please. All recommend this program to effectively advertise on the Internet, this is the best program!

customized term papers said...

Thanks for sharing this post. This is a very helpful and informative material. Good post and keep it up. Websites are always helpful in one way or the other, that’s cool stuff, anyways, a good way to get started to renovate your dreams into the world of reality.

auto insurance quotes said...

Doctors' use of drug for everything they can not explain could cause a big problem in the patients. One of my friend has been complaining about stomach pain. Doctor linked it to depression and gave him Prozac for years. Then he died of stomach cancer at the age of 28. Prozac should not be used like a painkiller of sort.

Anonymous said...

church is more and more crazy

Unknown said...

Nice post with awesome points! Can’t wait for the next one.

HP - EliteBook XT920UA 14" LED Notebook - Intel Core i7 i7-640M 2.80 GHz

Unknown said...

very nice and impressive one sharing you have done here i like it and would like to appreciate you on this so keep it up.
Apple Laptop