The Short Story:
D’Onofrio et al. (2014)
is a study of associations between paternal age at conception and offsprings’
risks for developing mental illnesses and academic problems. It was widely
covered in the world media in late February and March, 2014. Most media reports
about the findings of the study—for example, this
one—included risk statements that were direct quotes from the following
sentence in Dr. D’Onofrio’s original
press release:
When compared to a
child born to a 24-year-old father, a child born to a 45-year-old father is 3.5
times more likely to have autism, 13 times more likely to have ADHD, two times
more likely to have a psychotic disorder, 25 times more likely to have bipolar
disorder and 2.5 times more likely to have suicidal behavior or a substance
abuse problem.
These
alarming numbers are completely WRONG.
What D’Onofrio et al. actually showed was much less alarming:
When compared to a child born to a
24-year-old father, a child born to a 45-year-old father is 1.5 times more
likely to have autism, .7 times less likely to have ADHD, 1.3 times more likely
to have a psychotic disorder, 1.5 times more likely to have bipolar disorder,
.9 times less likely to have suicidal behavior, .8 times less likely to have a
substance abuse problem, .7 times less likely to have failing grades in school,
.9 times less likely to have low educational attainment, .8 times less likely
to have a low IQ; and 1.5 times more likely to have some higher education.
Note the
shocking misrepresentations of the risks in Dr. D’Onofrio’s original press
release. For example, the study actually found that children of older fathers
were at 1.5 times the risk of developing bipolar disorder compared to children
of younger fathers, not 25 times the risk. The study also showed that children
of older fathers were LESS likely to develop ADHD, not 13 times more likely. In
absolute terms, the risks seem even less alarming. For example, approximately 6
of 1000 children born to younger fathers (aged 20 -24) versus 9 of 1000
children born to older fathers (aged 45+) will develop bipolar disorder.
The correct
risk estimates do not appear in numerical form anyplace in the original report or
the appendices. They can be visually estimated from the “Baseline” curves in the
Figure
that appears in the report. They can be
precisely calculated from the data that appears in eAppendix
1 of the report’s Supplemental
Content. For example, eAppendix
1 shows that the rate of bipolar disorder among children of the fathers in
their study who were aged 20 – 24 was .58% and was .85% for fathers aged 45 or
more. This means that the risk of bipolar disorder for children of the older
fathers was 1.47 (=.85/.58) times higher than the risk for children of younger
fathers, not 25 times higher.
I am posting
this corrective analysis because I am concerned that the errors in Dr.
D’Onofrio’s press release have caused unwarranted, and potentially severe, psychological
harm—worry and guilt—to thousands of older parents who saw the media reports
that quoted the erroneous risk estimates in the press release. See, for
example, this
report on the study by an older father. Some older parents may even decide
to take actions that could actually harm their children in a misguided effort
to try to prevent a problem that their children are at very low risk of
developing.
The Long Story:
I am Steve
Herman, PhD, a Stanford-trained psychologist, and an associate professor of
psychology at the University of Hawaii at Hilo. I also happen to be an older
father. I was shocked and worried when I saw one of the hundreds of media
reports about the results of D’Onofrio et al. (2014) claiming that
children like mine are at high risk of developing bipolar disorder, ADHD, and
other serious life problems. The risk estimates seemed unbelievably high. When
I read the original report of the study in JAMA
Psychiatry article, it became clear that the actual findings of D’Onofrio
et al. were much different, and much less alarming, than what was reported in
the media.
When I
discovered that the erroneous statements about risks were direct quotes from a
misleading press release put out by the lead author, Dr. D’Onofrio, I became
angry, and determined to do something to try to correct these errors.
My initial
attempts to try to correct these alarming errors have not been successful.
First, I wrote a letter to the editor of JAMA
Psychiatry. My letter was rejected on the (somewhat justified) grounds that
media misinterpretations of studies published in JAMA Psychiatry should be addressed with the media outlets where
the mistaken reports appeared. I tried contacting some of the journalists who
wrote the erroneous reports, with little success so far. I asked Dr. D’Onofrio
to correct his press release to clearly show the real risks, and to inform
journalists who had relied on his press release that their reports were
mistaken. He declined to do so, although he did make an unhelpful correction to
his original press release (see below).
The
incorrect, alarming risk estimates in Dr. D’Onofrio’s press release are wildly
inconsistent with findings from most other studies. However, the actual
population risks found by D’Onofrio et al. are
consistent with findings from other studies. For example, Frans et al. (2008) found a 1.37 times
greater risk of bipolar disorder for children of men aged 55+ compared to men
aged 20 – 24 using a dataset that overlapped with, or included, the dataset
used by D’Onofrio et al. Another large study, McGrath et al. “A Comprehensive
Assessment of Parental Age and Psychiatric Disorders” (2014) found the
following increased risks for paternal age, comparing men aged 45+ to men aged
25-29: 1.8 for autism, 1.24 for bipolar disorder, and 1.47 for schizophrenia.
In another large study, Parner et al. (2012) found that
children of men aged 40+ were 1.35 times more likely to have autism as children
of men who were less than 35. Hultman et al. (2011) found a 2.2 times
increased risk of autism for offspring of men aged 50+ compared to offspring of
men aged 29 or less.
The numbers
that Dr. D’Onofrio cites in his misleading press release come from what are
described as “within-sibling,” “sibling comparison,” or “fixed effects”
analyses. These are covariate analyses, and the numbers produced by such
analyses do not represent the real risks to real people in the study
population, but are “adjusted” for variables called covariates. In essence, the
numbers from the within-siblings analyses represent hypothetical risks in a
counterfactual world in which many relevant variables could be held constant,
but paternal age would be allowed to vary. For example, in this counterfactual
world, only fathers, but not mothers, grow old; all fathers have the same level
of higher education; all fathers have the same likelihood of having been
arrested for a crime; all families have the same income; and family income does
not change over time. To present these adjusted statistics as if they
represented real risks to real people is misleading.
Although it
was not clearly stated in the original press release, a few media reports
correctly noted that the numbers cited in the press release referred to the
within-sibling analyses, but incorrectly interpreted what that meant. Some
reports interpreted the press release as implying that the risk estimates referred
only to the risks for offspring who had siblings. That this is an incorrect
interpretation can be seen from the “baseline” graphs in Author’s Appendix B
(available here).
After I
complained to Dr. D’Onofrio about the misleading press release, he revised one
version of the press release to read as follows:
Among the numerous findings are the
following, based on sibling comparisons: When compared to a child born to a
24-year-old father, a sibling born to the same father at the age of 45 is 3.5
times more likely to have autism, 13 times more likely to have ADHD, two times
more likely to have a psychotic disorder, 25 times more likely to have bipolar
disorder and 2.5 times more likely to have suicidal behavior or a substance
abuse problem. For most of these problems, the likelihood of the disorder
increased steadily with advancing paternal age, suggesting there is no
particular paternal age at childbearing that suddenly becomes
problematic.
The New
York Times report about the study was also revised to reflect this new
press release wording. Note that this revised risk statement is very different
from the original, which referred to all fathers. This revision refers to a
very small, unusual subset of offspring—offspring who have a sibling from the
same father who is at least 20 years older, probably fewer than 1% of the
population. Although the risks for the offspring in this small subset may be
somewhat higher than those of most children of 45+ year old fathers, I do not
believe that even this revised risk statement is accurate, because the numbers
are adjusted for some of the most important advantages associated with
advancing paternal age. The true population risks for this tiny subset of
offspring cannot be calculated from the data available in the original article
or the appendices.
In summary,
the risk numbers cited in Dr. D’Onofrio’s press release do not describe the
real risks to offspring in the entire study population or to offspring with
older siblings. They do not even seem to correctly describe the risks to offspring
who have siblings from the same father who are 20 or more years older.
D’Onofrio et
al. focused on estimating increases in risk due to disadvantages to offspring
that are associated with advancing paternal age. But offspring of older fathers
also have advantages compared with offspring of younger fathers. D’Onofrio et
al. attempted to estimate the risks due solely to the disadvantages by
statistically removing the effects of some of the advantages. In his press
release, Dr. D’Onofrio cited risk estimates based on these disadvantage-only
analyses. Some of the advantages that may accrue to offspring of older are
probably the result of the higher incomes and greater stability of older
fathers. Older fathers are more likely to have some higher education than
younger fathers, and less likely to have been arrested for crimes. They may
spend more time with their children than younger fathers. Children of older
fathers may be healthier and live longer than children of younger fathers (Eisenberg et al., 2012).
Disadvantages of older fathers with respect to risks for problems in offspring
include consequences resulting from an increased risk of genetic mutations in
the sperm cells of older fathers, as pointed out by D’Onofrio et al., but may
also include other psychosocial factors such as the increased likelihood that a
father will die during an offspring’s childhood.
The
population risk estimates that can be calculated from eAppendix
1 make it clear that the advantages of having an older father compensate
for most of the disadvantages with respect to the risks for autism, bipolar,
and psychosis, so that offspring of 45+ year old men compared to 20 – 24 year
old men are at only moderately higher risks (1.3 to 1.5 times more) for those
three problems. The advantages of having an older father outweigh the disadvantages with respect to the risks of ADHD,
suicide attempts, substance abuse problems, failing a grade, and low
educational attainment, so that the risks for these five problems in the 45+ vs
20 - 24 comparison are actually lower
(.7 to .9 times less) among offspring of the older fathers. None of this is
made clear in either the original or revised versions of Dr. D’Onofrio’s
alarming and misleading press release.
I have
confirmed that my analysis of the real population risks is correct with a
number of colleagues, psychologists and biostatisticians. If you find it
difficult to believe all of this, as I did at first, I suggest that you take a
close look at eAppendix
1. You could also ask Dr. D’Onofrio (bmdonofr@indiana.edu)
to confirm that what I am telling you about the real population risks to real
offspring in their study population is accurate.
I do not
know why Dr. D’Onofrio chose to put out such an alarmingly misleading press
release, you can ask him that yourself. I do know that exaggerations and misrepresentations
of research findings in order to attract media attention are quite common. This
problem is so common that it is explicitly addressed in guidelines
for the ethical dissemination of research findings published on the website
of Sweden’s prestigious Karolinska Institute, the institutional home of Dr.
D’Onofrio’s coauthors:
Communication
with the media
When communicating with the media researchers at the Karolinska Institute should provide balanced information on the significance and applicability of the study.… When risks are described it is similarly prudent to state how many individuals may be at risk and how much the risk increases in absolute terms upon exposure to a certain situation or treatment.… Scientists should not create unfounded expectations or fear among laymen via the media.
When communicating with the media researchers at the Karolinska Institute should provide balanced information on the significance and applicability of the study.… When risks are described it is similarly prudent to state how many individuals may be at risk and how much the risk increases in absolute terms upon exposure to a certain situation or treatment.… Scientists should not create unfounded expectations or fear among laymen via the media.
Please email
me if you have any questions.
Steve
Herman, PhD
Associate
Professor of Psychology
University of Hawaii at Hilo
hermans@hawaii.edu
University of Hawaii at Hilo
hermans@hawaii.edu