“Bisphenol A protects against Wheezing” screamed no headline ever, yet that is one of the findings of a paper just published in the Journal of Allergy and Clinical immunology.
I’m sorry, I really was going to talk about the WHO report, and I don’t want this blog to become all BPA all the time, but the press release that comes with the paper (and to some extent the paper itself) highlights significant issues with how science is communicated, and science by press release.
First, some background.
Asthma incidence has been rising in the past 30 years worldwide, except Australia, where the incidence of asthma has actually fallen by 25% in young people over the past 10 years. The reason for these changes is unclear, though environmental influences are likely to be involved. Prenatal exposure to compounds that mimic the sex hormone oestrogen are possible candidates for the increase (the fall in Australia is possibly related to our successful anti-smoking campaigns).
So how can we test this idea. One way is to measure the levels of BPA in pregnant women and their children after birth and see if they develop asthma as the children grow older. There are problems with this approach, but it can be useful and is fairly common.
This is what the authors of the new paper in the Journal of Allergy and Clinical immunology did. They examined perinatal and postnatal exposure to bisphenol A through urinary BPA excretion, and looked at wheezing and various indexes of asthma at 5, 6 and 7 years of age.
And what did they find?
One astonishing finding was that prenatal exposure to the oestrogen mimicking compound bisphenol A reduces the chance of children wheezing at age 5. It also had no effect on overall asthma and wheezing at other ages.
As one of the big worries people have is that prenatal exposure to oestrogen mimics will affect the developing foetus, this finding of no effect (and even a “protective” effect at 5 years of age) is of particular significance.
The “protective” effect of BPA at 5 years of age is quickly passed over in the press release and only briefly discussed in the paper. This should be kept in mind when trying to interpret the variety of results seen when the authors try to correlate post natal exposure to bisphenol A to indexes of asthma and respiratory problems.
When looking at postnatal exposure, the results are inconsistent and contradictory.
Levels of urinary BPA measured at 3 years of age are correlated with wheezing at 5 and 6 years but not at 7 years. But when we look at urinary levels of BPA measured at 5 years, there is only a marginal correlation with wheeze in 5 year old children and not at 6 or 7 years.
The size of the protective effect of prenatal exposure on wheezing, and its statistical significance, was as strong as the positive correlation of postnatal exposure with wheezing at 3 years and stronger than the correlation of postnatal exposure and wheezing at 5 years.
There was a correlation between clinician-diagnosed asthma and BPA exposure. But in stark contrast, there was no correlation between a breath test for asthma (which is equivalent to the standard lung function testing) and BPA at age 3 (which had the best correlation with wheeze) and age 5.
Issues with statistics and interpretation
Keep in mind the correlation does not imply causation (it may waggle its eyebrows suggestively while mouthing “look over there” though). A colleague notes that there is a strong correlation between ice cream sales and drownings.
But ice cream is not causing the drownings, hot weather makes people more likely to eat ice cream and go swimming. You always have to interpret your results with the knowledge that other factors may be involved.
Also keep in mind that to get these correlations, the authors have to apply mathematical calculations to correct for other factors which affect the incidence of asthma (mothers’ history of asthma, and tobacco smoke exposure amongst others).
This is all standard methodology, but again you have be very careful when interpreting the data. If you haven’t correctly identified all the important confounding factors, you can get spurious correlations.
Overall, given the number of tests done (which increases the chance of false positives), the sporadic and inconsistent correlations and the astonishing “protective effect” of prenatal BPA exposure, the results are more likely to be statistical artefacts rather than real associations.
Also, when you look at a number of the confidence intervals that were published alongside the significance values (eg in table IV), the confidence intervals include the no effect value, which is a red flag that the strength of the effects are overestimated.
As well, there are some inconsistencies in the data in the figures given for urinary bisphenol A. The urinary BPA levels in table IV are not the same as the BPA levels In table I. One table gives median values and the other mean values to be sure, but it is hard to interpret the values when they can’t be directly compared.
This is important as the table IV values means that these kids exposure to BPA is in the top 10% of concentrations seen in America according to the Fourth National Report on Human Exposure to Environmental Chemicals (2012), and the 3 year olds in the top 5% (ie 95% of the under-15 population have lower exposure to BPA than these kids do).
This means that these are not ordinary levels of exposure as claimed in the press release. It also suggests that these children have been exposed to unusually high levels of BPA (and possibly other environmental chemicals that were not controlled for).
The press release does not indicate any of these complexities, but is headlined “BPA raises risk for childhood asthma”. Since most people will read the press release and articles derived from it, rather than the actual study, once again we see that science by press release serves us poorly.