A new study published in the Archives of Pediatrics and Adolescent Medicine shows a positive correlation between counties in California, Oregon, and Washington with greater precipitation and a higher incidence of autism. While the results of this study are interesting, it needs to be put into proper context. Also of note, the authors had presented early results from this data previously.
Correlation is not Causation
This type of study is a correlational study, which means it asks whether or not there is a statistical correlation between two variables – in this case the rate of autism and the amount of precipitation. This type of data is extremely useful to medical science, but it has known limitations, which can be summarized by the statement that correlation is not causation.
I often see this principle used to dismiss correlation data entirely, but that is not the correct approach. Correlation, rather, needs to be considered in the proper context. When A correlates with B there are various possible interpretations: the correlation is a statistical fluke (coincidence); A causes B, B causes A, or both A and B correlate with another variable C, and there can be a variety of causal relationships among the three (or more) variables which would cause A to track with B.
Therefore, finding a correlation is a way to generate several hypotheses which can then be tested by further observations or research. That, in my opinion, is the best way to view correlational data – as a beginning step to help generate hypothesis. But they should not be used to reach firm conclusions.
It should also be noted that further correlations can be used to test various causal theories, and if multiple correlations all triangulate to a single causal hypothesis that can lead to a fairly confident conclusion – even in the absence of other evidence. For example, smoking correlates with certain types of lung cancer. There are no prospective studies in humans to establish that smoking causes lung cancer, but we can be confident that it does because the correlation holds up no matter how you choose to look at it. If smoking causes cancer (as opposed to other causal hypotheses stemming from the correlation) then we predict that increased duration of smoking increases risk of lung cancer, that stopping smoking decreases risk, that smoking unfiltered is more risky than filtered, etc. Each of the predictions turns out to be true, supporting the smoking causes lung cancer hypothesis.
Rain and Autism
I was immediately interested in how the authors of this study came by the notion that rain might correlate with autism rates. In their older publication they mention that they thought of it as a way to test the hypothesis that TV watching may be an environmental trigger for autism, and that increased precipitation correlates with greater TV watching. In their most recent publication they explore various other causal hypotheses.
For example, they discuss that increased rain leads to more time spent indoors. While this may lead to increased TV watching, it also leads to an increase in all indoor activities. It also leads to increased exposure to the indoor environment, and decreased exposure to the outdoor environment. Therefore there may be some indoor trigger or outdoor protective element.
The rain itself may be involved. Increased rain may increase exposure to atmospheric toxins.
The authors also raise the possibility that increased precipitation correlates with decrease sun exposure, which could lead to relative Vitamin D deficiency, which is the actual trigger for autism. I actually find this the most plausible scenario. Relative vitamin D deficiency has been linked recently to increased risk of a variety of neurological disorders, and has lead to an increase in the recommended daily allowance of Vitamin D for children.
But what the study does not discuss are those causal hypotheses that no one has yet considered (obviously because no one has considered them). This is always the primary weakness of pure correlation – there are so many variables that it is nearly impossible to anticipate them all.
Here is an example from an unrelated field. The research on the possible protective effects of alcohol and heart disease has been largely epidemiological (i.e. correlation). Many studies showed that those who drink a little every day are healthier than those who do not drink at all. It was not considered for many years after this data started to come out, however, that the group that does not drink at all contains ex-drinkers, who are unhealthy as a consequence of their prior alcohol abuse. That may seem like an obvious point now, but it was missed for a long time.
It therefore seems possible, even likely, that there can be a complex causal relationship hiding in this apparent correlation between rain and autism.
We won’t know until further studies are done. The authors of this study concluded:
These results are consistent with the existence of an environmental trigger for autism among genetically vulnerable children that is positively associated with precipitation. Further studies focused on establishing whether such a trigger exists and identifying the specific trigger are warranted.
This is the exactly correct way to word this conclusion – the results are consistent with an environment trigger (but do not prove that one exists), and the results warrant further research (but cannot be used to form reliable conclusions). We can now base predictions on various causal hypotheses stemming from this study as a guide to future research. Something interesting may ultimately come out of this.
Autism remains a poorly understood neurological entity, although it is an area of active research and we have a great deal of information upon which to base our thinking about autism. It is clear that there is a huge genetic component to autism, but it is also probably not a strictly genetic disorder (or, more likely, set of related disorders). While there is no clearly established environmental factor, most such disorders result from an interplay of genetics and environment.
I suspect, however, that the environment of the womb is the dominant environmental factor. Recent research is finding that subtle signs of autism are detectable earlier and earlier in an infants life. The environment of the infant may play a role in expression, but I am not convinced that any post-natal factors trigger autism in children that would otherwise not display any features. This latest study may be providing a clue (if the correlation itself holds up to replication) of a post-natal factor, but right now we are in the wild speculation phase of thinking about these results.
What is clear is that more research is warranted.