A cluster of birth defects

I’m not sure why the story got the media’s attentions this past week on what CNN refereed to as a “Mysterious cluster of birth defects” in Washington state. The story was also picked up by NBC News, Fox News, and others. The cluster was studied and reported on by the Centers for Disease Control (CDC)in July of 2013 and again in September of 2013. The CDC was unable to find a cause for the cluster of birth defects. The birth defect was anencephaly where the baby is born missing parts of their brain and skull.

Clusters such as this one are always hard to investigate. Rare events do happen. It only takes 23 people in room to have an 50 percent probability that two of those people share the same day of the year for their birthday. Statistics surprise and can be very counter intuitive. Many people have problems accepting that it only take 23 people to get to an equal probability of such an occurrence.

Computing probabilities can be very difficult in these types of situations. The investigation in this particular situation apparently started when a nurse notice two incidents with in the hospital where she was working and then discovered a third event in a nearby hospital while talking with a colleague.

Rare events happen. The CDC report tells us that the national rate for anencephaly is about 2.1 cases per 10,000 births. But there are millions of births per year. So is two cases in one hospital, or there cases in one area in a short period of time an anomaly?

The next step was when the state department of health reviewed the records for a three county area over a three year time span and found 23 cases. This translated into a rate of 8.1 defects per 10,000 births. Well above the national average. If we were to focus only on the 20 new cases that were seen in the state study the rate would still be 7.0 per 10,000 births which is still well above the national average of 2.1. At that point I cease being concerned that the nurse found a random outlier as the unusual number of cases was confirmed within the same geographic area.

The next step in the study was the selection of a random sample of 108 other pregnancies. The only constraints on the selected sample were “a pregnancy without an indication of a structural or genetic birth defect during routine prenatal care and prenatal residence in one of the three study counties.” I am not sure why that approach was used, and what other analysis was done that was not reported on by the state health department and the CDC.

Questions I would have liked to have seen addressed were things like: Did the 23 cases show any type of clustering based on the characteristics of the mother’s health, geography, medical care, and the like? Why was the random sample comprised of only 108 cases – I suspect money drove that decision – at least in part. And why was not a more targeted sample chosen that more closely represented the characteristics of the impacted mothers? Usually when trying to compare two groups reducing the variability within and between the two groups is crucial in reducing the statistical errors inherent in the data. Of course one must always deal with the realities of what data is available to select the sample. Perhaps there was not very much data on the other births in the three county area that would have made a more targeted sample possible. This makes me wonder if the inability to find a cause, if it was not just a random event, was due to the small sample size used and the inherent variability in the populations of pregnant mothers.