Everyone knows that a woman’s eggs don’t improve with age. The arbiter of turn-of-the-millennium pop culture, “Sex and the City,” gave us the image of the single woman in her mid-30s (Miranda) and her maturing eggs. And while fertility may not quite fall off a cliff at 35, it’s hard for women to ignore the idea that things are getting worse as they get older. Obstetricians have a special category for pregnant women over 35: “advanced maternal age.”
So, I challenge any woman in this group not to feel just a teensy bit of schadenfreude at the increasing discussion of aging sperm and its effects on childbearing. I’m sure we can all look forward to the day when television features 25-year-old women on dates with desperate 37-year-old men discussing their sperm-freezing plans. Of course, for this to happen there has to be something to all of this concern. So, are older sperm really worse?
Worries about aging sperm — or, more accurately, sperm from aging men — are the same as the concerns about aging eggs: decrease in fertility, and increase in genetic problems and psychiatric and behavior disorders among offspring. The primary issues in the latter category are autism and attention deficit hyperactivity disorder (ADHD).
Let’s consider fertility first. You might think that it would be easy to figure out how sperm quality declines by comparing pregnancy rates for the partners of older and younger men. And, in fact, some scientific papers do this. But they have a central problem: Older men tend to be matched with older women. And since we know female fertility declines with age, it’s hard to figure out whether male age has any additional effect. This study in the American Journal of Obstetrics & Gynecology is a good example of a case where older men appear to have lower fertility, but the effect is impossible to detect when you also adjust for partner age.
It turns out this problem is surmountable with a simple trick: Observe cases where couples are using donor eggs. There’s no reason that the age of the egg donor would be related to the age of the father (indeed, in the data it typically is not) which means you can examine the impact of the father’s age independent of the mother’s.
Studies that do this — here is one nice review in the journal Fertility and Sterility — tend to find that while semen volume (and, hence, sperm count) does decline with age, there is little overall impact on reproductive success (the ability of the sperm to penetrate the egg, fertilize it, etc.). This is encouraging — but it’s worth noting that while the number of sperm doesn’t matter much when fertilization occurs outside the body, as is the case when using donor eggs or doing in vitro fertilization, it will matter if you’re procreating the old-fashioned way. So more sex may be necessary.
The evidence on autism and other behavioral disorders appears less reassuring. The authors of a 2011 review article in Molecular Psychiatry combined a number of studies comparing children of older and younger fathers (this is called a “meta-analysis”). They concluded that relative to children born to 20- to 29-years-old fathers, those with fathers aged 30 to 39 were 1.2 times more likely to have autism; those with fathers aged 40 to 49 were 1.8 times more likely; and those with fathers over 50 were 2.5 times more likely. The studies used in the meta-analysis were all pretty consistent. They did not all find the same size effect, but virtually all pointed to increased risk of autism with increased paternal age.
For the most part, these studies try to adjust for maternal age — though they can’t do as much as the donor-egg studies — but this analysis also runs into a second problem: The type of men who have children later in life may be different from those who do not, and those differences may be at least partially responsible for the increased autism rates.
We would clearly prefer to have a study that compared children born to the same man at different times in his life. This fixes a few problems. First, since the father’s genetics stay the same among children, there is no longer a concern that, say, men who are autistic themselves have children later. Second, we avoid any concern that some kinds of dads are more likely to have their kids evaluated for autism or other disorders.
An analysis like this is referred to as a “sibling fixed effects” regression, a comparison between or among siblings. It’s been used to good effect in an analysis of breastfeeding, for example.
Last year, the journal JAMA Psychiatry published a study about paternal age that included this design, and, to put it mildly, the results were alarming. The study used a full census of births in Sweden from 1973 to 2001 and merged together information on parent age, siblings, other family members, psychiatric diagnoses, grades in school, and on and on. It’s a Big Data dream — as if you were inside the house of every person in Sweden for 30 years, taking notes on their every move.
The researchers ran regressions where they effectively compared children born to the same father at different times in his life, and they reported enormous changes in psychiatric problems. They found big effects on autism: Children born to men over 45 were 3.5 times as likely to be diagnosed with autism as those born to men aged 20 to 24.
But even more striking, and notable, were the effects on ADHD diagnosis: Relative to children born to men 20 to 24 years old, those who were born to men over 45 were 13 times more likely to be diagnosed with ADHD. Even children born to men 30 to 34 were more than three times as likely to be diagnosed with ADHD as those born to the youngest father group. This is enough to send a lot of men I know straight to the sperm freezer.
But let’s look a little more closely at the data and statistics here. The authors actually ran three analyses. First, they looked at the raw data and asked, on average, are children born to older fathers more likely to have ADHD? Second, they controlled for some demographics, such as education and income, and maternal age. And finally, they ran their “sibling fixed effects” regression. They reported their results in helpful graphs, and below we’ve replicated the one for ADHD.
In the first analysis using raw data, older men do not appear to be more likely to have children with ADHD. In the second analysis, when the researchers’ adjusted for some characteristics of the father, mother, and so on, we see that the children of older fathers are less likely to be diagnosed with ADHD. It’s only when we move to the siblings analysis that we see a huge positive effect of paternal age on a child’s likelihood of an ADHD diagnosis.
One theory for this gap in results is, of course, that the final analysis is the most accurate because it examines children under near ideal conditions: They were born to the same father at different times in his life. But when there is such a big difference in findings across analyses, it’s worth asking if something odd is going on.
To answer this, we need to be clear on what this “sibling fixed effects” analysis really does. Consider a dad with two children, born four years apart. A basic fixed-effects model would ask whether the older child is less likely to have ADHD than the younger, and then attribute any observed difference to paternal age.
Running this basic analysis would have the same problem with maternal age that we discussed earlier. Because mothers often age along with fathers within a family, if we analyzed sibling pairs with the same mother and the same father, it would be impossible to separate the effects of maternal and paternal age.
This paper does claim to separate these effects. How? The data includes half-siblings — children who share a father but not a mother.1 The authors can infer the effects of the father’s age separately from the mother’s age by looking within groups of siblings who share a father but not a mother.
That means the huge positive effect seen in the chart above essentially tells us that a later-born child of a father who has multiple kids with multiple partners is more likely to be diagnosed with ADHD. In the paper, this difference is attributed to paternal age.
Why push so deeply into the statistics here? Seeing what the data is really saying lets us think a little more about what else might be happening. Now that we know the effects are driven by differences across half-siblings, we can start asking what else — beyond paternal age — might be driving the difference. Most obviously, we may wonder whether being a child in a fluid family situation could itself have an impact on ADHD risk (as other studies have found).
Another, if more mundane, explanation for the big effect that paternal age appears to have on a child’s ADHD risk is that by striving so hard to isolate the effect of paternal age, the paper’s authors ended up studying a small number of individuals. And this makes it more likely that the results they saw were just due to chance.
But the data is proprietary, and without getting my hands on it, it’s difficult to know. Our best hope of finding out how a father’s age affects his offspring may be to follow the methodology of the earlier study on paternal age and fertility: Use cases with donor eggs and compare within a family. If you have two children born to the same father via donor eggs, assuming the donor eggs do not age with the father, you could get a clear estimate of the impacts of the father’s age.
But until then, I’d say the jury is out on the relationship between paternal age and child psychiatric problems. Yes, there is some possibility that it matters. But it’s probably not necessary to run off to the sperm freezer just yet.