From the aBiostatistics Branch, National Institute of Environmental Health Sciences, Research Triangle Park, NC; and bInstitute of Public Health, University of Aarhus, Aarhus, Denmark.
The author reported no financial interests related to this research.
Correspondence: Chunyuan Fei, National Institute of Environmental Health Sciences (NIEHS), Epidemiology Branch, PO Box 12233, Mail Drop A3-03, Research Triangle Park, NC 27709. E-mail: email@example.com.
In 2008, we reported that high maternal levels of perfluorooctanoate (PFOA) and perfluorooctane sulfonate (PFOS) were associated with longer time to pregnancy (TTP) in the Danish National Birth Cohort.1 Reverse causality is a possible explanation for the association,2 as has been pointed out by Olsen and colleagues. Even with age adjustment, past pregnancies and deliveries may serve to lower stored levels of PFOA and PFOS. On average, women with longer TTP will have had more time to reaccumulate perfluorinated chemicals (PFCs).
In a study reported in this issue of EPIDEMIOLOGY, Whitworth and colleagues attempted to replicate our findings in data from the Norwegian Mother and Child Cohort Study.3 They found a positive association between maternal PFC exposure and subfecundity (defined as TTP >12 months) in parous women, but found no association in nulliparous women. These observations appear to support the theory of reversal causality.
In fact, our unpublished analyses had shown a strong association in both nulliparous as well as parous women between maternal PFC exposure and both TTP and infertility (defined as TTP >12 months). Parity is the strongest predictor of PFC levels in blood and is also a function of fecundity. Stratification by parity would therefore be controversial, so we reported the results without stratification, but adjusted for parity in the multivariate models.
A directed acyclic graph (DAG) representing the relationships among these factors is shown in the Figure. Present and past fecundability share common determinants, and those determinants confound the relationship between PFOA/PFOS and present fecundability. Adjusting for parity should serve to block that pathway and hence control confounding. However, a subtlety not captured by the DAG is that PFOA/PFOS were not measured at the beginning of the attempt at conception (which would have been ideal), but at the end, after a pregnancy had been achieved. Thus, in the available data, the measurement of PFOA/PFOS can potentially be influenced by TTP for parous women through reaccumulation of the chemicals. Such influence produces a cycle in the graph through the arrow from TTP to the measured PFOA/PFOS. However, for nulliparous women, that arrow does not exist in a model that adjusts for age.
To make a comparison with the results from the Norwegian study, we now present analyses of the Danish data stratified by parity, using multiple imputation techniques to deal with missing data on covariates (n = 94). The associations between PFOS and infertility remained strong in nulliparous women (Table). The associations between PFOA and infertility were attenuated in nulliparous women, and an increased risk was observed only in the highest quartile of PFOA exposure. Similar associations were observed for the discrete measurement of fecundability (TTP). In parous women, the magnitude of the association between PFOS and infertility or TTP did not differ from that for all women, whereas the association between PFOA and infertility became stronger (Table).
Stratification by gravidity showed stronger associations in nulligravid women. The adjusted fecundability odds ratios were lower (less fertile): 0.55 (95% CI = 0.36–0.85) and 0.51 (0.32–0.79) for the top 2 quartiles of PFOS (vs. the lowest quartile), and 0.51 (0.27–0.98) and 0.36 (0.19–0.68) for the top 2 quartiles of PFOA (data not shown).
The strong associations we see in nulliparous or nullgravid women suggest that reverse causality is not the explanation for the association in our data, although the association observed in parous women may partly be explained by this. After stratification by parity, the proportion of infertility was reduced by one-half in the lowest quartile of PFOS and by two-third in the lowest quartile of PFOA in parous women (Table). The number of pregnancies in the lowest quartile in nulliparous was much fewer, which gave less precise estimates. Still compared with the proportion of infertility in general population (usually 15%), the proportions in parous women were lower. In addition, our data previously showed that maternal PFOA concentrations were close to concentrations in the cord blood of their newborns, suggesting that PFOA may be more easily transferred to the fetus than PFOS and, thus, more substantially reduced in parous women, who are also generally more fertile than the general population. If the toxicokinetics of PFCs explain the lower PFC concentrations in these fertile women, the results from parous women are less informative.
Maternal PFC concentrations in the Danish National Birth Cohort were much higher than those in the Norwegian study; for example, in the Danish study, PFOS and PFOA concentrations in the lowest quartile were higher than the MoBa exposure levels in the fourth quartiles. If there is a threshold level below which the exposure presents no risk to human health, we would expect no association in a population with lower exposure levels, except what is produced by reverse causality. In conclusion, we found limited evidence for reverse causation as an explanation of our results, and we welcome additional studies in populations with elevated and variable levels of PFC exposure.
1. Fei C, McLaughlin JK, Tarone RE, Olsen J. Perfluorinated chemicals and fetal growth: a study within the Danish National Birth Cohort. Environ Health Perspect. 2007;115:1677–1682.
2. Olsen GW, Butenhoff JL, Zobel LR. Perfluoroalkyl chemicals and human fetal development: an epidemiologic review with clinical and toxicological perspectives. Reprod Toxicol. 2009;27:212–230.
3. Whitworth KW, Haug LS, Baird DD, et al.. Perfluorinated compounds and subfecundity in pregnant women. Epidemiology. 2012;23:257–263.
The 3M Toxicology Laboratory performed all laboratory analyses for free but they have no control over the design, data analysis, and interpretation of the data or writing of this commentary.