select search filters
briefings
roundups & rapid reactions
before the headlines
Fiona fox's blog
A study, published in JAMA Network Open*, has looked at links between Bisphenol A (BPA) levels in humans and long-term risk of mortality.
Prof Kevin McConway, Emeritus Professor of Applied Statistics, The Open University, said:
“This research used data from quite a large sample of Americans, that is generally representative of the US adult population. But it’s an observational study, and there are always issues in deciding what the results of observational studies actually mean. I think that there are some specific features of this study that make the interpretation particularly awkward.
“The researchers used information collected from almost 4,000 people, on the level of bisphenol A (BPA) in a single sample of their urine, another urine measurement (creatinine) that improves estimation of BPA levels, and a number of other characteristics related to health, such as age, sex, ethnicity, smoking, alcohol use, physical activity, a measure of diet quality, and several others. Then the people were followed up for, about 10 years on average, and records were kept of when any of them died. But apart from this record keeping and data collection, the people just got on with their normal lives. As a result, there are large differences in many of the characteristics that were recorded, between people who had different levels of urinary BPA. As just some examples, the people with higher concentrations of BPA in their urine were more likely to be young, male, Black, have lower education and income, and lower quality diets. These other characteristics are related in various ways to health, and to the chance of death during the follow-up period. So, though the researchers found that (generally) the higher the urinary BPA level, the higher was the risk of death at a given time, this doesn’t mean that it is the BPA in the participants’ bodies that was causing the increased risk. It could be that the association between BPA level and death rate is one of cause and effect – but it’s also possible that what is causing the increase in risk is one of the other factors (called ‘confounders’) that are related to BPA levels and, separately, to the risk of earlier death. The researchers used appropriate statistical methods to adjust their estimates for several potential confounders on which they had data, and these adjustments made a difference. But there was still a statistically significant association, after the adjustments, between urinary BPA levels and the risk of death during the follow-up period. The big snag is that researchers cannot adjust for everything that might be a confounder – for one thing, they simply do not have data on all possible confounders. Possibly consuming certain specific foods, or working in certain occupations, or living in certain regions or places, might be independently associated with BPA levels and also with the risk of earlier death – the researchers did not adjust for factors like these, and in general they could not have done so because they did not have data on them. Because of this unavoidable possibility, this research (and any findings from similar studies, taken on their own) simply cannot establish whether having more BPA in one’s body has a causal effect on the risk of death in a certain time period. This does not rule out that such an effect might exist, but it cannot show that it does exist, and there are alternative explanations for the findings.
“The researchers do clearly acknowledge that this issue – called ‘residual confounding’ – might have arisen. They argue that residual confounding is unlikely to be important in their study, on the basis of their calculation of an E-value, which is a way of measuring how strong the relationship between, on one hand, confounders that were not adjusted for, and on the other, BPA levels and mortality would have to be, to make the apparent association between BPA and mortality disappear. However, I think they have considerably over-interpreted these E-values. This needs a little explanation, which cannot avoid being slightly technical. As recommended by the developers of E-values (VanderWeele and Ding), these researchers calculated two of them for each association, one based on their central estimate of the strength of the association between BPA and mortality, and the other on a smaller strength of association which is still consistent with the data. But in their interpretation of the E-values, they use only the one based on the central estimate, and ignore that the other one shows that an unmeasured confounder would not have to be so strongly associated to overturn the statistical significance of the relationship between BPA and mortality. Also, their explanation of the meaning of the E-values is in terms of a single confounder that was unaccounted for. So it apparently ignores the point, acknowledged very clearly by VanderWeele and Ding, that if there are several confounders that were not adjusted for, each of them individually would not have to have such strong associations with BA and mortality in order that they, taken together, could nullify the association between BPA and mortality risk. In this study, given the number of reasonably strong associations with confounders that were adjusted for, it seems credible to me that there could be several, not just one, that was not adjusted for.
“The research reports that the risk of death from any cause, in a given period, was 49% higher for people whose urinary BPA levels were in the top third, compared to those whose levels were in the lowest third. That sounds like quite a big increase in risk, but it must be considered alongside two other things. First, the risk of death during follow-up was not, overall, very high. On average, of every 1,000 people in the study followed up for a year, about 9 died. The increased risk at the higher BPA level, if it applied to everyone, would correspond to about an extra 5 deaths – but remember, we do not know whether it is the increased BPA that would be causing additional deaths. Second, partly because the death rate was not very high, the estimate of the increase risk is subject to a lot of statistical uncertainty. An increased risk of just 1% (almost no extra deaths per thousand people), or a risk that is over double for the higher BPA level (11 extra deaths per thousand people), are both consistent with the data.
“Another, separate, issue with this study is that the urinary BPA levels were determined from a single measurement at one point in time, and that is taken to stand for the participants’ BPA levels throughout the average 10 years’ follow-up. The researchers acknowledge in their report that this may be a problem, but quote another paper on BPA measurement (by Ye and colleagues) which, they report, says that using single samples to measure BPA might still produce an adequate measure of the average exposure of a population to BPA. The other paper does indeed say that, but it is in the context of pointing out that individual single measurements are not necessarily good measures of individual exposure, because urinary BPA levels vary considerably within a single day, and between different days in the same week. In this new research, the main aim was not to investigate the overall population levels of BPA, but to investigate, at an individual level, the association between BPA levels and risks of death – so I cannot see why the other paper that they quote is very relevant. Also, even if at the population level, the measurements could give information on BPA in the population at that time, I’m not sure how clearly relevant that is to relating BPA exposure at that time point to the risk of death during a 10-year follow-up.”
Prof Ieuan Hughes, Emeritus Professor of Paediatrics, University of Cambridge, said:
“This study used the longitudinal US National Health and Nutrition Examination Survey (NHANES) resource to examine the effect of BPA population exposure on long-term all-cause mortality. NHANES is a well-recognised and valuable epidemiological powerhouse. BPA exposure was determined by analysis of the chemical only in spot urine samples, a possible weakness in view of the rapid metabolism of BPA. The authors found an association with all-cause mortality which had essentially been sub-divided in to cardiovascular and cancer. However, these singly were not statistically associated with higher BPA levels. Much of the literature on the toxic effects of BPA, a chemical which is ubiquitously present in all ages including the fetus, has focused on endocrine-related effects in animals and in humans. For example, male reproductive disorders. But how can that be teased out in terms of mortality? So all-cause mortality is a blunt instrument to use and the authors recognise the challenge of using this observation to determine what may be the specific cause(s) of mortality. Cleansing the environment of BPA has started (witness BPA-free products on supermarket shelves) and the use of substitute analogues such as bisphenol S. The authors’ call for further studies are indeed warranted as humans will long remain exposed to this class of chemicals.”
Dr Fred Davis, Associate Professor in Organic Chemistry, University of Reading, said:
“This paper will add to a growing body of concern about the safety of polymers containing the monomer BPA. This material is certainly on the European Food Safety Authority’s radar; its use in bottles or cups intended for very young children is prohibited and in 2018 the levels allowed to migrate into food was reduced.
“Bisphenol A (BPA) also known as 4,4′-(propane-2,2-diyl)diphenol is a material used in the production of polycarbonate and epoxy resins and some other speciality polymers. As such is an extremely usefully material. The monomer is cheap, being available from the reaction between phenol and acetone and imparts rigidity and durability into polymers.
“Although BPA itself is known to be quite toxic, polymers containing this monomer have previously been thought to be safe since the monomer is chemically bound to the polymer and therefore in biologically accessible.
“Over the last few years, it has found that very small quantities of BPA are present in most people. This exposure is thought to be due to the use of BPA containing polymers in contact with food, in for example, plastic containers or coatings for cans. The exposure then occurs either due to the presence of unconverted monomer, or by decomposition (probably hydrolysis) of the polymer.
“The levels of exposure to this material is low and consequently this is challenging to investigate in terms of the hazards, particularly in view of the many overlapping factors that may influence mortality. Thus, the conclusions reached here rely on the models used, and despite a relatively large sample in research conducted over 10 years the authors concede more studies are needed.”
* ‘Association Between Bisphenol A Exposure and Risk of All-Cause and Cause-Specific Mortality in US Adults’ by Bao et al. was published in JAMA Network Open on Monday 17th August.
DOI: 10.1001/jamanetworkopen.2020.11620
Declared interests
Prof Kevin McConway: “I am a member of the SMC Advisory Committee, but my quote above is in my capacity as a professional statistician.”
Prof Ieuan Hughes: “No conflicts of interest to declare.”
None others received.