Commentary and explanation

Meningitis can be a serious disease with devastating effects. This study set out to investigate one question about why some people develop hearing loss as a complication: is it something to do with eye colour?

The simplest way to look at this as shown in the Paper is to find a group of deafened people and compare the eye colours of those deafened by meningitis with those deafened by other causes. This is a type of case control study, in which cases of a condition (deafness caused by meningitis) are compared with controls (people who do not have the condition) to see whether they were exposed to the risk factor of interest (eye colour).

PAUL TAYLOR/PHOTONICA

It is clear from the results that the people deafened by meningitis were more likely to have light eyes than the people deafened by something else the proportions being 94% and 73% respectively. But is this difference big enough to get excited about?

The odds ratio is one way of expressing the size of the difference. Odds ratios are hard to get your head round, but because odds are important in both epidemiology and gambling you will find a use for them one way or the other. If 30 of the 32 patients deafened by meningitis had light eyes and two did not then the odds that any one of them had light eyes are 30 to 2, or in bookies' terms, 15 to 1. On the other hand, the odds that a member of the general population had light eyes are 1151 to 447, or 2.6 to 1. You get the odds ratio by dividing one set of odds by the other (see box).

The odds ratio reported in this study sounds like a large difference. But if we examined several other samples of patients we would probably get a different answer every time each one an estimate of the "true" underlying odds ratio in the entire population of deaf people. The 95% confidence interval is a way of expressing how precisely that "true" odds ratio has been estimated. In this case, it means that we can be 95% sure that people deafened by meningitis are between 1.4 and 24 times more likely to have light eyes than people deafened by other causes. An odds ratio of one would mean that the odds were the same in both groups. So it does seem likely that there is a real difference between the groups, but we do not have a precise estimate of how large that difference is.

How else might you investigate this question? You could identify every new case of meningitis, note the people's eye colour, and follow them up to see whether they became deaf. This type of study a cohort study would require more effort. For one thing, you would need to study many more patients to pick up the same number of deafened people. So a case control study is often a good pragmatic way to begin investigating a research question. Any epidemiology textbook will provide a list of the problems with case.control studies, and the author has identified some of them in the paper.

One problem is confounders. A confounder is a third factor which confounds the relationship between the two variables that we are actually interested in the exposure and the outcome. If genes which determine eye colour are linked to genes which determine the immune response then we might find an apparent association between deafness and eye colour which is actually explained by this third factor of immune response genes. If you identify possible confounders beforehand you can make allowances for them in the way that you choose the study participants or in the way that you analyse the data.

A second problem is bias. A bias is a feature of the study which makes a particular result more likely like a football pitch which slopes from one end to the other. If people with dark eyes are more likely to die of meningitis, then the group that we are studying people who survived meningitis, but became deaf will contain more people with light eyes, not because they were more susceptible to deafness, but because they stayed alive and were able to enter the study.

Two other questions you might ask are:
(1) Could the people in the study who all had cochlear implants be a biased (unrepresentative) sample of all deaf people?
(2) Could the classification of people's eye colour especially "borderline cases" have been biased?10

No one can design the perfect case control study, but you should be aware of the pitfalls when you read one. Case control studies are important tools in many public health investigations. However, sometimes they can raise more questions than they answer, acting as "research pointers" which lead to more work to get to the bottom of an issue.