Diabetes Medication and Birth Defects

Too often, when we talk about issues in fertility or pregnancy, the focus is exclusively on the pregnant woman as the source. This is despite the fact that we know the men can be the source of concern, sometimes in ways that are not well understood. An example is a paper I came across showing that the recurrence of hyperemesis gravidarum in later pregnancies is lower with a different father. It isn’t clear why this would be, but it suggests we have more to understand.

One problem with the focus on women is that it leads to blaming women if things go wrong. But a second problem is that by ignoring the role of men, we ignore the possibility that changes in their behavior might have an impact.

It is this second point that led me to be interested in a paper that just came out, arguing that birth defects are more common among the offspring of men who take metformin, a common diabetes drug, in the months before pregnancy. I want to unpack that paper today, to talk about whether its results are compelling and what it means.

Overview and why this is hard

This paper was published in the Annals of Internal Medicine, and you can read the abstract here. The basic idea is straightforward: the authors would like to compare the incidence of major birth defects for children born to men who took metformin during the period of sperm development (roughly the three months before pregnancy occurred) to the incidence in those who did not.

There are at least two things that make this hard to implement.

First, severe birth defects are, blessedly, relatively rare. To study them convincingly, you need a lot of people in your analysis. Moreover, to study this particular question you need not only a large sample but detailed information about the individuals. The authors here want to connect the medications that men take in the months before pregnancy to diagnosed birth defects among their children. This is asking a great deal of the data, since it is not very common to see this information linked across families.

Second, even if you have enough good data to run this correlation, there is an underlying concern that there are other differences between men who take metformin and those who do not. One obvious difference is they are more likely to have diabetes. But there may be other things, such as demographics or other illnesses, that could be the proximate cause. Correlation and causality are difficult to separate.

Part of what makes this paper compelling is that the authors are able to make significant progress on both of these problems.

Implementation and solutions

In the paper, they fix the first problem — the data problem — by using comprehensive data from Denmark. Denmark is one of several countries (including Sweden and Finland) where it is possible to get detailed data on health for individuals, linked over time and across families. In this paper, the authors can combine one data set with information on adult prescriptions with another that has information on birth defect diagnoses with yet another that tells them who is related to whom. Even better: they can do this for the entire population of Denmark.

The result is 20 years of comprehensive data, with over 1.1 million births, linked with data on parental prescriptions. This is enough to pick up relatively small increases in birth defects. The fact that the data comes from official registries and medical records and not from (say) self-reports allows them to be both more confident in the data quality and more precise in timing.

The second problem is more fundamental and more difficult to fix. What the authors can do, though, is provide a number of pieces of supporting evidence that make it less likely that the main results they see reflect confounding.

The main result of the paper is that, after adjusting for some basic demographic controls, there is an increase in the risk of a serious birth defect — about a 40% increase — if the father was taking metformin or a sulfonylureas in the three months before conception. There is no increase for fathers who are on insulin (who are more likely to be Type 1 diabetics). The increase for fathers taking metformin is larger and more consistently statistically significant than for those taking sulfonylureas.

Without additional evidence, this result is subject to many of the concerns we typically raise with observational data. Relative to the control group, men who are on metformin are much more likely to be on a variety of other medications, and they’re older, less educated, and poorer. These demographic differences are largely not present for men who are on insulin. There is the simple fundamental issue that men in this group have Type 2 diabetes, and from the basic analysis we are hard-pressed to separate out all of these other factors from the impact of metformin.

However: the authors do three things that I think make this result more compelling.

First, they compare men who fill prescriptions for metformin in the exposure period of three months before conception with those who fill them in the two years before or the two years after. The results are shown in the graph below. The increase in birth defects relative to the control group shows up only for those who fill the prescription in the sperm development period. You can think of the groups that fill prescriptions in these surrounding periods as a better control group.

A second test involves comparing across children within families. In a small number of cases, the researchers observe at least two children in the same family where the father took metformin during sperm development for one child but not the other(s). This includes cases both where the father changed prescriptions between pregnancies and cases where the actual father changed (same mother, different fathers). In both of these comparisons, the authors find that the exposed siblings were more likely to have birth defects than the unexposed siblings.

The final result that makes this more convincing is the analysis by birth defect type. When the authors break this down, they find the most significant increase is in genital birth defects for boys. They also find that men in the metformin group are less likely to have male children, and they point to evidence from animal models suggesting a link between metformin and testicular development and similar outcomes. This indicates a more specific mechanism for the link.

My read of this paper is that the authors have pushed the data about as far as possible in terms of making a compelling causal case. There isn’t a randomized experiment here, and doing one would be effectively impossible given the small risks — you’d need too large an effect size to make it feasible. What they have done here is to be extremely thoughtful about what is possible with observational data.

Please note: the effects are small

The results in this paper make a good case for causality. But I think it’s very important to note that the effect sizes are small in terms of absolute magnitudes. Overall, these severe birth defects occur in about 3.3% of births, and the data suggest about a 1.3 percentage point increase in risk as a result of metformin exposure. When we say there is a 40% increase in risk, this can be hard to really contextualize without knowing the baseline. The increase here is comparable in size to increases in birth defects for women over 45.

Many of the conditions considered are treatable with surgery or other approaches. Which isn’t to understate their severity, just to say that there are things you can do.

What’s the action item?

A natural question following this is what the appropriate reaction is. My sense is that if we found similar results for something more optional — say, Botox — there would be a push against using it. Metformin, however, is a key component of diabetes treatment, and untreated diabetes has its own risks to fathers’ fertility and health. The bottom line is that this becomes a situation with some possible trade-offs. The short period of sperm development may provide some help — a break from metformin during conception might have a positive effect — but that is also complicated.

The paper ends on what seems to me an appropriate note, which is to argue for the need for more attention to this connection. One piece of that is more evidence on whether this link is consistent in other data. As I said above, a randomized trial is not likely to be feasible, but one could run a study like this one in another country (Sweden, Finland, maybe even with some medical records in the U.S.). A similar finding in another location would be meaningful. The second piece is a need to think about whether there are reasonable treatment alternatives for men in this situation.

A final thought

Studying the impact of health interventions provides us with many situations in which a randomized trial just isn’t really possible, and too often I feel that papers in this space do not sufficiently exploit observational data. While it’s true that the “gold standard” randomized controlled trial evidence isn’t possible here, the approaches they take allow us to go well beyond a simple correlation. And that’s of tremendous value.