Yesterday I wrote about the problem of using data mining to find chance correlations that are not “real.” I mentioned, but did not discuss further, the other problem with correlation data – the assumption of causation from correlation. By minor coincidence today many news outlets are reporting on a newly published study that correlates hypertension (high blood pressure) in men with use of aspirin, acetaminophen, and ibuprofen – common pain killers. But what does this correlation mean?

The assumption of causation from correlation is a common logical fallacy – A correlates with B therefore A causes B. A might cause B, but it is also possible that B causes A or that a third factor, C, causes both A and B. So how do we know what the real causal relationship is? The definitive way is to do a randomized controlled prospective study – which means randomly creating A and then observing to see if B follows. This type of protocol controls for B causing A or some other factor C causing both. But there are other ways of inferring causation, even to a high degree of confidence, without such a prospective study. First, you can consider mechanism and plausibility – does it make sense that A could cause B or B could cause A? Second you can look at multiple correlations that would be predicted from one causal relationship. For example, if A causes B then as A increases B should also increase, and if A goes away then B should diminish or disappear. You can also look at known (C) factors to see if they also correlate with A and B.

In this study over 16,000 men who did not have a history of hypertension were assessed for their use of either aspirin, acetaminophen, or ibuprofen – the three most common drugs used in the United States. Their use was reassessed after two years and their blood pressures were followed for four. Other factors, such as age, race, and weight, were also followed. The study found that use of all three drugs was independently associated with a 26%, 34%, and 38% increase in risk for hypertension respectively. This risk was greater as the number of pills per week increased and the number of days per week in which medication was taken increased.

This data sound pretty convincing, and it may in fact be true that use of these common analgesics can increase risk of hypertension. There are plausible mechanisms by which these drugs may cause increases in vascular resistance and therefore blood pressure. But let’s examine this type of data according to the principles outlined above. Analgesic use and blood pressure to correlate, and the correlation holds when you look at amount of analgesic use, as you would expect from a causal relationship. There is a plausible biological mechanism. Incidentally, since the study set out specifically to look for a correlation between analgesic use and blood pressure the results are not the result of data mining and they are statistically legitimate.

But we still need to consider the alternative causative relationships. Is it possible that hypertension causes analgesic use? It’s difficult to make a negative assertion – that there is no mechanism by which B could cause A, but we can say there is no known way. In this case, there is no plausible mechanism by which hypertension can cause analgesic use. Hypertension does not cause pain or any symptom that would lead someone to take these medications. It is possible (this is where researchers have to get creative) that hypertension raises concerns about heart health and therefore could lead some men to take aspirin for heart attack prevention, but then there would not be a dose correlation, and also this would not hold for acetaminophen and ibuprofen as these are not used for heart prophylaxis.

What about a third factor, C, that could cause both hypertension and analgesic use? This is trickier. The possibility of a mysterious third factor is always the specter haunting these types of studies. The researchers ruled out known factors, such as weight and race, but what about all the possible things they did not look at? Are there factors that cause both hypertension and analgesic use? The one thing that comes to mind is pain. Pain can certainly lead to painkiller use, and acute pain is known to increase blood pressure. So could subjects with chronic or intermittent pain causing frequent analgesic use have a higher rate of diagnosis of hypertension because of pain-induce hypertension? I don’t know, but this is the kind of question that correlations alone do not answer.

To their credit, the researchers appropriately acknowledged the nature of their study and did not make any claims that went beyond the evidence. They acknowledged the possibility of an unknown confounding factor – although they did not mention pain as a possibility and I am curious about that omission.

Science news reporting to the public, however, was not as careful. An article distributed by Reuters, for example, contains, “Popular painkillers such as aspirin, ibuprofen and acetaminophen can raise blood pressure and thus the risk of heart disease among men.” The headline read, “Common painkillers raise heart risk.” Although many reports included the conditional term “may” in their reporting, none I have read conveyed appropriately the nature of the study – that it established association only, not causation.

In medicine we often have to make decisions based upon imperfect information. In this case, what should we do with this interesting but preliminary data? The American Heart Association issued an advisory, essentially recommending avoiding unnecessary use of these medications and trying lifestyle changes to treat chronic pain prior to resorting to analgesic use. This is not unreasonable, but as a physician in the trenches I can say that telling a patient with pain they should lose weight and exercise rather than taking pain medication is often not a therapeutically successful strategy. Although I already do this, many patients are not ready to accept this advice. Ideally, we need follow up prospective data so that we can know with more confidence if analgesics really do cause increased blood pressure.