sign uplog in
Want to go ad-free? Find out how, here.

Dave Grimmond thinks a medical researcher's warning highlights a similar problem in economics; how to avoid the problem of self-selection. Your view?

Dave Grimmond thinks a medical researcher's warning highlights a similar problem in economics; how to avoid the problem of self-selection. Your view?
How do you avoid the problem of self selection when interpreting test results? Image sourced from

By Dave Grimmond*

It was with concern that I read a recent news article about the dangers of non-steroidal anti-inflammatory drugs (NSAID), in which a medical researcher is quoted as stating

"If you look at it internationally, diclofenac is the single most widely used NSAID,"

and that

"Clearly thousands of people die as a result of using [diclofenac]".

What the study reportedly found was that use of NSAIDs increased risks of heart attacks and strokes, and that diclofenac (also sold in New Zealand under the brand name of Voltarin) was identified as one of the higher risk NSAIDs. 

Now as an active, but ageing, cyclist and footballer I have often made use of diclofenac to nurse various soft-tissue joint injuries over the years, and so this news caused me some alarm that this use of medication might be imposing some unintended health risks for me.

However on a little reflection about the statistical issues underpinning the analysis I am instead beginning to have doubts about the strength of the conclusions and the authors recommendation that diclofenac should be banned.

The first thing that nagged at me is the growing evidence that inflammation of the arteries is being observed as being associated with the onset of heart attacks and strokes. This would suggest that the use of anti-inflammatory drugs should be associated with a reduction in the risk of cardiac events, not an increase. 

The next thing is that it is very difficult in fields like medicine and economics to isolate the true transmission path between cause and effect.

This is because it can be unethical to undertake fully controlled experiments and even when this can be done the complexities of human bodies (or economies) can mask the true transmission paths. A result of these analytical limitations is that while many studies can identify correlations between different observations, it is a far more difficult task to attribute causation.

Thus although the use of anti-inflammatory drugs may be associated with more cardiac events, this does not mean that the drugs cause the cardiac events. This is just the same as saying that as patients in hospitals have a higher probability of dying than people in their homes, then this means that hospitals cause people to die. Such an inference is of course nonsensical.

Thus returning to anti-inflammatory drugs, by definition people taking these drugs have a high probability of suffering from inflammation. Could these episodes of inflammation perhaps also heighten their risks of experiencing a cardiac event? If this is the case one might well observe a heightened correlation between people using anti-inflammatory drugs and cardiac events, yet there may be no causal relationship between the drug use and the cardiac event. Indeed it might even be the case that the use of the drugs might actually reduce the risk of cardiac events. Again this is analogous to patients in hospitals. By being a patient one potentially has an inherent heightened risk of dying, however the care one receives at the hospital will actually reduce this risk.

In economics we refer to situations like these described as “self-selection”.

At risk people are self-selecting their use of a drug or their visit to the hospital precisely because of their awareness of their heightened risk. The drug use or hospital visit is thus a proxy for this heightened risk and should not be confused with being a cause for the problem. 

Now it may be the case that the intervention exacerbates the risk, say by taking an inappropriate drug, but establishing that this is the case is usually a very onerous task and one that is unlikely to be achieved by a meta-analysis like the one underpinning the quoted news article.

This does not mean that the research undertaken is not useful, but rather that one should be circumspect with the conclusions drawn and with any resulting policy recommendations that you make.

In calling for a ban on specific types of drugs the authors here appear to have over played their hand.

The next step should be further investigation about the inter-linkage between inflammation, anti-inflammatory drug use and cardiac events.

It could be that these drugs are indeed responsible for heightened risk, but it is also possible that they mitigate inherent risk. In which case the banning of these drugs would have the perverse effect of both reducing consumer choice for anti-inflammatory drugs and increasing people’s risk of experiencing a life threatening cardiac event.


By Dave Grimmond a director and senior economist at Infometrics, an economic consultancy and forecasting service. You can contact him here »

We welcome your comments below. If you are not already registered, please register to comment.

Remember we welcome robust, respectful and insightful debate. We don't welcome abusive or defamatory comments and will de-register those repeatedly making such comments. Our current comment policy is here.


You might want to have gone to the actual original research article, rather than critiquing them based on the xth-hand media report. They were comparing diclofenac with safer NSAIDs such as naproxen. They way the conducted their study, your criticism has nothing to do with it, as it is a comparison of populations taking diclofenac and naproxen.  They were not saying "anti-inflamatorys are bad" they are saying "this inflamatory is worse than the other one". In both cases people may be taking inflamatorys because they have inflammation, but in the populations were it is diclofenac rather than a safer option they are observing an effect.
So, while this may be a valid criticism of what you imagine the article is talking about based on the media report, it doesn't have much to do with what they did for research.

Just so dh. A huge fail by the author, and as you point out caused by his inability to do the research on the original article.
Mind you should we be surprised? The arrogance of an economist attempting to critique a field which is actually evidence based (ie biomedical research) - talk about being out of your depth - but still charging on regardless.

From the research: "There are a number of limitations to this work. Most obviously, we do not have information on the risk profiles of patients taking NSAIDs."
I actually thought that the article was quite reasonable.

The link between NSAIDs and heightened risk of cardiovascular events is nothing new - Vioxx (Rofecoxib) was pulled off the market in 2004 after being associated with up to 27,000 deaths:
The study compared a range of NSAIDs and pretty conclusively demonstrated that diclofenac has a significantly worse profile than the others tested. There will no doubt be a series of follow up studies and it is not inconceivable that dicofenac will not share the same fate as Vioxx.
I for one am happy to let that process take its course and do not feel the interventions of an economist with a less than rudimentary knowledge of the issues involved remotely helpful.

agree, this is a lousy effort.
Can we get a bit more variety? Infometrics views seem to be getting a lot of airing here. 

These articles represent the view of the individual - not of Infometrics as an organisation.  If you have a problem with a specific article, it would be best to contact the author.
With regards to this article, the author seemed to be merely pointing out that we have to be careful making strong policy conclusions, especially when there is a potential issue of endgoeniety.  He was saying this as an economist - as economists face this issue constantly, and it is also prevelant in other disciplines using statistical analysis to go beyond description and move towards explanation and policy.
Again, if you felt the article was inaapropriate feel free to message the author.