The Risks of Confounding

A study was just published titled “Short term effects of temperature on risk of myocardial infarction in England and Wales: time series regression analysis of the Myocardial Ischaemia National Audit Project (MINAP) registry” in the British Medical Journal. It’s kind of a mouthful but the authors are trying to draw a link between the risk of a heart attack with decreasing temperature.

There are probably a couple of factors in this that make this study somewhat flawed. The use of ambient temperature as an indicator for a heart attack in this case could be technically called a confounding variable. Here is a classic example to explain exactly what that is.

Assume a researcher wanted to study the causes of lung cancer but had no idea what factors could contribute to the development of disease. A case-control study would be the typical design used to look for the variables that contribute to disease. In essence, it takes a group of people and assigns them to two different groups – those who have disease and those who do not. The researcher would then look at variables that might explain the cause of the disease in the group that is ill.

Let’s say that our hypothetical researcher decides to study the presence of a lighter in a pocket or purse thinking that maybe some of the chemicals in the butane might cause lung cancer. Statistically, this variable is going to show up as a significant factor contributing to lung cancer. This is where we run into the problem of causality versus association. Just because a variable is associated with a particular outcome does not mean that it is the cause of the outcome. Today, we clearly know that carrying a lighter does not cause lung cancer. However, those who carry lighters are likely to smoke, which is a clear cause. Hence a lighter in this example is a confounding variable.

Back to the proposed link between temperature and heart attacks. There is a good chance that temperature is a confounding variable in this case. Colder temperatures also occur at the same times as snow and influenza. The risks of a heart attack increase after snowfall because of the extra exertion that is needed to shovel snow, especially when it is wet and heavy. Influenza causes inflammation, including in the coronary arteries, which increases the chance of a heart attack. It doesn’t look like the authors addressed these alternative hypotheses in their study.

So what are the two take-home messages? I’d argue that there is always more than meets the eye and you can’t believe everything you read, even if it comes from a reputable source.

Leave a Reply