The Researchers zone:
Sibling studies: Nature's own randomised trials?
COMMENT: Why do some people get sick while others remain healthy? We can learn more about that by comparing siblings to one another. But it is a delicate balancing act to perform such studies.
Why is it that some people who eat a lot of sweets do not get diabetes? Why do others who keep their hands out of the cookie jar become ill? Why do some children develop autism? And might it make a difference whether they came into the world via caesarean section?
The tireless questioner would be hard strung trying to mention all the questions each of us ask when attempting to make healthy choices – both for ourselves and our children.
It is even harder to come up with good answers – especially when it comes to the ordinary, everyday human life, which cannot be placed under the microscope in a laboratory.
But perhaps comparisons between siblings can bring us a few steps closer to the truth?
The answer is both yes and no. Although sibling studies are good at solving some challenges, they also create new problems.
At the Department of Public Health at the University of Copenhagen, my colleague Theis Lange and I have tried to take a closer look at the mathematical and statistical basis of sibling studies in a new study.
But before looking more closely at that, we have to take a few steps back. And, eventually, we will also get to learn a little bit about the connection between caesarean sections and autism.
The Holy Grail of Cause and Effect
Let us take our point of departure with the question of why some children develop autism and whether caesarean sections could be a cause.
The idea could stem from the fact that children born by caesarean section do not come into contact with their mother’s bacterial flora to the same degree, and we know from other research that the bacterial flora might affect the brain. So perhaps births by caesarean could be the cause of autism through an unfortunate development of the baby's bacterial flora.
How can such a question be examined?
In health research, the holy grail of knowledge about cause and effect is to carry out randomised trials. In such trials, you decide who gets a treatment by means of a draw (read more about this in the 'What is a randomised trial?' info box below this article).
This means that the effects of a treatment can be measured by comparing those who randomly received a treatment with those who randomly did not.
In our case, that would mean to randomly select those who were to give birth by caesarean section and then comparing how many children from the two groups would later develop autism.
But this idea will not get us very far. No doctor would like to randomly select pregnant women to give birth by caesarean section – and besides, probably not many pregnant women would do as the doctor told them!
So, we must come up with something different.
A glance at the statistics
Another idea could be to look at the statistics and see whether children born by caesarean section develop autism more frequently than other children.
If we consider children in Denmark born between 1997 and 2010, 1.2 percent of children born by vaginal delivery developed autism, while the corresponding figure is 1.3 percent of children born by caesarean section. That amounts to an eight percent increase, so, although autism is rare in both groups, the difference is actually quite large.
But we cannot use those numbers to say anything about whether births by caesarean section are the cause of the higher proportion of children with autism.
Perhaps the figures reflect the fact that women who are genetically disposed to develop autism themselves also, for some reason, will be more likely to have to give birth by caesarean section. Perhaps it is the case that a domestic environment with many quarrels increases the risk of both caesarean sections and autism.
It is possible to come up with lots of these hypothetical explanations. But we cannot say anything about whether or not some of them hold water just by looking at the population's distribution of autism and birth by caesarean section – simply finding a connection does not mean that we have come any closer to the cause.
Need to compare like for like
This brings us back to the sibling studies: rather than looking at an entire population at the same time, one can look at differences between siblings instead.
The idea is to make use of the fact that siblings are similar to one another in many ways; for example, they share much of their genetic material, they often grow up together in the same household, in the same local area and with the same adults in their lives.
As such, they will share a number of risk factors that could appear to make caesarean section look as if it were the cause of autism when that is really not the case.
For example, if we always observe that the children born by caesarean section are much more likely to develop autism than their siblings, then a possible causal link between caesarean section and autism is more likely.
By doing this, one has compared children that in other respects are similar. That is the idea behind sibling studies.
Nature's own randomised trials?
In health research, a whole lot of exciting sibling comparison studies have been carried out in recent years – especially in Scandinavia, where our national registers make it possible to connect siblings to each other and compare their health development.
As researchers, we have been very pleased with this possibility of approaching ‘questions of why’ that we otherwise would be unable to answer.
But we can also become a little too eager. Sometimes we end up thinking certain groups of siblings as 'nature's own randomised trials': we imagine that when, for example, there is a group of siblings where one was born by caesarean section while the others were not, then it is almost as if nature has randomly drawn who was to be born by caesarean section for us, because siblings are so similar in so many ways.
And is that, then, not almost the same as a real randomised trial?
Two steps forward and one step back
That is precisely the question that my colleague and I have delved into further. We have used mathematics and statistics to get a grip of how large the difference really is between sibling comparisons and randomised trials – and it is larger than we thought.
Although comparisons of siblings can help solve the problem of underlying factors that may trick us into believing that there is a causal link, one cannot actually directly interpret their results as if they proceeded from a randomised trial.
For it is not entirely ‘cost-free’ to use this method: in our usual statistical models, new complications arise from cramming siblings into the machinery. So, even though the sibling studies provide two steps forward, we also take a step back.
The models become more complicated, and, as researchers, we are sometimes not quite sure how to interpret the numbers they generate – albeit with one single, important exception.
A no is safer than a yes
When the statistical models in sibling comparisons suggest that there is no causal link, it is safer to trust that this is the case.
One might think of it as there being only one way to describe the fact that caesarean is not the cause of autism; namely, that there is approximately 0 percent difference in children born with or without caesarean section.
On the other hand, there are infinitely many ways in which there could be a difference: maybe there could be 8 percent greater (or lesser) probability, or 12 percent, or 25 percent, or...
Therefore, it is easier for statistical models to find 'negative answers' than 'positive answers' to questions of cause and effect.
Does caesarean section increase the risk of autism or not?
Are we now able to answer our initial question: does caesarean section increase your child's risk of developing autism?
In a new study from Denmark, we have examined that exact question based on information obtained from the national registers.
When we use common statistical models where we compare all children with each other at once, it seems that there is a greater risk of developing autism if one is born by caesarean section.
However, when we use statistical models with sibling comparisons, the difference between children born with and without caesarean section disappears.
This suggests that there is something we have not been able to take into account in our usual models that make it look as if caesarean sections affect autism. But by comparing siblings with each other, we can disprove that it is a matter of cause and effect.
This means that, based on our research, there is no reason to be concerned that children should be more at risk of developing autism just because, for one reason or another, they came into the world by caesarean section.
If we had not had the sibling study to help us, we would not have been able to reach the conclusion that there are more children born by caesarean that develop autism.
We would not be able to tell whether or not it was caused by the caesarean section.
Conversely, had the sibling study shown that there were a causal link between caesarean section and autism, then we would have had difficulties explaining how the connection should be interpreted more specifically. In that case, we would still not know how or to what extent caesarean actually affects the development of autism.
What is a randomised trial?
In medical science, randomised trials are considered the most valid way of examining whether there are causal links between a treatment and a consequence, such as a disease.
In randomised trials, lots are drawn among patients for who gets which treatment.
Typically, two treatments are compared; one of which is new and the other of which is either a placebo (i.e., no treatment) or the 'usual' treatment.
Since it is a matter of chance who receives which treatment, there will, on average, be no difference between the patients in the two groups before they receive their treatments.
This means that if there is a difference in how often the two groups fall ill, then that difference is related solely to the difference in treatment.
Ideally, randomised trials should be carried out as double-blinded trials, which means that neither the patient nor the healthcare professional knows what treatment the patient has received.
In that way, there is no risk that there may be a difference in how often the two groups fall ill, for example, due to the doctor lending more attention to those receiving the new treatment or to the patient being more apprehensive receiving placebo treatment.
Translated by Stine Zink Kaasgaard. Read the Danish version at Videnskab.dk's Forskerzonen.