Researchers are investigating the link between paracetamol use during pregnancy and possible abnormalities in the reproductive development of baby girls

This study requires extremely cautious interpretation.  The study design means it’s impossible to make any causal claims.  The small sample size and large numbers of outcomes make false positive findings very difficult to distinguish from any potential true effect.  We cannot conclude from this study that paracetamol is causing these effects.

When you hunt really hard for a pattern you inevitably find one.  I am worried that the authors are over-interpreting the one they found.  They set six primary outcomes – that gives you a 25% chance of false positives right off the bat.  And it gets worse with secondary outcomes without even considering causation, biological plausibility, effect magnitude or the heterogeneity of individual responses.

I am very concerned that this exploratory study will be misinterpreted.  Pregnant women already have a limited choice of pain medication, and paracetamol has an extremely high safety profile.  They should not conclude from this study that paracetamol will harm their child.

The press release is very responsibly written and at pains to point out (correctly) that this does not yet mean that women should stop taking paracetamol when indicated during pregnancy.

It is good quality research, although it is important to point out that while 67-140 people per group seems like a lot, it is actually not such a large sample, considering that the authors measured 13 different outcomes. To have total confidence that we are not looking at chance connections, one would probably ideally have wanted more like 250-650 people per group, not 300 people in total.

The authors argue that it fits well with existing pre-clinical literature, and they see something similar in a replication cohort. One caveat to all this is that when we approach large, complex datasets with expectations in mind, it can be remarkably easy to find individual aspects that support prior work or somehow fit our expectations. These days, more and more scientists therefore try and ‘pre-register’ their analysis plans, i.e. specify ahead of time exactly what they want to do. The authors of the current study did log their trial on a repository, but the log does not contain enough statistical detail to ascertain that they did not accidentally pick up noise.

In the early life exposure group, there were a lot more smokers than in the other two groups. Smoking has a huge effect on pretty much every health outcome.  This and other confounders are adjusted for in the initial cohort, but it is not clear to what extend confounders were adjusted for in the replication cohort.

Right now, I think it is too early to recommend a change in behaviour – as correctly indicated by the authors in the press release.  

Pregnant women should not be worried, as very nicely laid out in the original press release.

This study has several notable weaknesses.

Even at face value, the statistical evidence that paracetamol usage is linked to fetal outcomes is weak. The headline result - that paracetamol usage is linked to reduced ovarian volume - only just achieves the minimal conventional threshold to be declared as a scientific finding. As part of this analysis, the authors considered 11 other outcomes - only one other outcome achieved the same threshold. The scientists appear to have tested a lot of different hypotheses, and focus their presentation on the small number of findings that show associations - this is selective reporting. The associations are all based on small sample sizes, and so there is a considerable risk that these are chance findings. There is also a substantial amount of missing data - only two-thirds of girls in this study had measurements of ovarian volume.

Mothers who took paracetamol during pregnancy are likely to differ substantially from those that did not – this is known as confounding. Confounding makes it difficult to know whether differences between outcomes are attributable to paracetamol itself or to other characteristics of the mothers or their pregnancies.

Even if these differences are statistically robust and even if they are attributable to paracetamol, this study cannot establish paracetamol as a cause. It may be that these findings are driven by infection or fever - women take paracetamol due to mild sickness, and this sickness is the cause of differences in early-life ovarian volume. It may be that impaired fetal development and its causes are what lead to increased paracetamol usage, not that increased paracetamol usage leads to impaired fetal development.

Finally, it is not clear whether the relatively small differences in ovarian volume, follicle number, and hormone levels observed in infancy have any meaningful clinical consequences. It is also not clear whether they are sustained into adulthood, or whether they have any effect on fertility or reproductive lifespan.

Paracetamol has been taken for over 100 years, and so the burden of proof that it is harmful should be high. It is one of the only pain-relieving medications that can be taken during pregnancy. Even if there are negative consequences of taking paracetamol during pregnancy (which this study only provides weak evidence to support), there are also likely future negative consequences attributable to taking away one of the only remaining available and widely-tolerated medications from pregnant women.

This is a small study which performed many different analyses, the majority of which did not show associations with paracetamol usage. If you perform many analyses, you will observe some associations solely due to chance alone. But even if we accept the associations as meaningful, they fall short of showing that paracetamol has a harmful effect: paracetamol usage may be an indicator of impaired fetal development, rather than its cause.

Current Australian advice remains in place. That is, paracetamol can be used as directed for pain or fever during pregnancy, and this study alone should not prompt anyone to stop medically indicated use.

Residual confounding by indication is plausible and, in my opinion, likely. It is one of the major barriers to causal interpretation for this study. Although the authors examined fever specifically, the analyses do not adequately rule out confounding by the broader conditions for which paracetamol was taken, such as headache, pain or infection, which may themselves be associated with fetal development.

Our Epidemiology Research Laboratory applies specialised designs to address issues like this. What we want to know is whether it's the paracetamol or the reason it was taken. The informative comparison is within indication. Among women with the same condition (like headache or migraine), did those who took paracetamol differ from those who didn't?