Matthew Gillman – The Arc of Fetal, Developmental and Adult Health: Stanford Childx Conference

Matthew Gillman – The Arc of Fetal, Developmental and Adult Health: Stanford Childx Conference


[MUSIC PLAYING] Hello. Well I’m delighted to introduce
the next session to you. Anybody who has worked
or cared for children, anybody who has
children themselves, or anybody who themselves
were once a child understands the enormous
amount of development that occurs between
infancy and adulthood. Our experiences modulate how
we form and who we become. Nowhere is this more evident
than the brain, which maintains some
mechanisms of development into adulthood to allow ongoing
adaptation and plasticity. Today we will hear from a
spectacularly talented group of scientists who
have elucidated the effects of experience
in infancy and childhood on the body and the brain,
discovered ways to use exercise to heal the brain for
children whose development was interrupted by
cancer, and who have uncovered fundamental mechanisms
of learning and adapting. We’re going to begin this
panel with a keynote address by Matthew Gillman. Dr. Gillman is a
professor and director of the obesity program
in the Department of Population Medicine at
Harvard Medical School. His research interests
include early life prevention of chronic disease,
including obesity, diabetes, cardiovascular
disease, and asthma; individual and policy
level interventions to prevent obesity
and its consequences; and childhood
cardiovascular risk factors. He directs Project Viva,
an NIH funded cohort study of pregnant women and
their offspring focusing on effects of gestational diet
and other factors on outcomes of pregnancy and childhood. He has served in leadership
roles in the US National Children’s study, the
International Society for Developmental Origins
of Health and Disease, the American Heart Association,
and the American Academy of Pediatrics. I’d like to welcome Dr. Gillman. [APPLAUSE] Thanks very much. It’s a real pleasure
to be here, especially coming from a place where it’s
a lot colder and a lot snowier. I understand you’re
having drought, but to me it looks really
beautiful, so thanks. So my story today begins
with David Barker. With pluck, and
serendipity, and a lot of perseverance against
prevailing wisdom, he– almost solely– is
responsible for the emergence of a new research
paradigm, which is now called Developmental
Origins of Health and Disease. And this paradigm emphasizes
the prenatal period and early childhood as important
for the development of health and disease over a lifetime. And teleologically
it makes sense that our early developmental
plastic periods have a lifetime of
consequence, here depicted as the majority
of cell divisions that happen within
the first 1,000 days. And the theory is that
if we can intervene at a time that’s early and
plastic that we can get on trajectories of
health that are better than if we don’t intervene. And that if we try
to intervene later, it’s more difficult because
there’s an inadequate response to new challenges. So that’s the theory. Now let’s go back 20 years. I was invited to–
for The BMJ– review David Barker’s first
book for the lay public, and at that point
I was skeptical. I said, OK here
comes a new paradigm. But in the grip of an
enthusiast, one must be wary. I talked about biases and
confounding and the fact that I didn’t think that
David Barker’s arguments at that point were persuasive. But even in the
following five years there was an
explosion of research that was interdisciplinary,
that related to population level research with
experimental animal models, clinical studies,
in vitro studies. And so by five years later, my
colleague Janet Rich-Edwards and I said, we
approach the field of fetal origins of adult
disease as skeptics. As one of us said,
some of us were rooting for the null
hypothesis for the first time in our lives. But with the recent publication
of epidemiologic studies that have started to overcome
the flaws of the initial work, we’ve become reluctant converts. Now if we forward
ahead 15 years, we have journal articles,
we have journals, we have magazines, we
have books for the public, we have books that are
for academic audiences, we have a society with
increasing membership, called the International
Society for DOHaD. We’re on our ninth
international congress. The next one is occurring this
fall in Capetown, South Africa. And we’ve been able to take
more traditional views of life course health, which talk about
the diminution of function in adulthood leading
to disabilities. And we’ve said that
early life is also important in characterizing
these trajectories and determining which
kind of disability we have late in life. And in this life
course perspective we’ve also been able to not only
take your life and your life prenatally but also talk about
your mother’s and father’s life, going back
generations, and your child’s and your grandchildren’s
life going forward. So this is our life
course perspective. So then the question
is in 2015 if all the scientific
activities happened, why has there been so
little impact on practice in health policy? And maybe it’s because we’re
Barker-ing up the wrong tree. When we think about how
policy decisions are made, there are lots of
things that go into it. And you can see a number of
these elements in this circle. If we have the wrong
evidence, then we’re not going to be having
evidence-based policies, and maybe the
wrong policies will ensue, or maybe no policies. And the idea is how
do we get the evidence to be central in our decisions
about practice and policy? So today I want to
talk about that, and I want to give
some examples about how we can improve etiologic
research, how we can use the science of prediction, talk
about risk-benefit analyses, a bit about implementation
of interventions, long-term effects,
which of course very important here, and finally talk
about evaluation of policies. And I aim to be an equal
opportunity stone thrower, so I’ll throw stones at our own
epidemiologic kind of research as well as others. And I’d like to begin
by throwing stones at my colleagues who do
whole animal experiments that are mostly in the
physiological basis. So how could animal
experiments be more helpful? Well first of all, why
do animal experiments in developmental origins? Well if you’re like me and
you start a cohort study, the cohort study has to outlive
me to be maximally useful. But if you work with
rats or mice, which have a lifespan of at most five
years, probably mostly less, you can go through
several lifetimes, and that’s really an advantage. And in fact animal
models have shown that this perinatal programming
of adult health outcomes exists. And by programming we mean
that some perturbation or [? a cue ?] at a critical
or a sensitive period of development
causes alterations with lifelong, sometimes
irreversible, consequences. So we have classic
animal experiments like this– the
couch potato rat. So this rat is produced by under
nourishing the mother, reducing calories, and by giving
the rat after birth a cafeteria diet ad lib. So we have this combination
of undernourishment before birth,
overnourishment afterwards, and you get a rat
that’s sedentary, that doesn’t do physical
activity, that’s hyperphagic, hyperglycemic,
has high blood pressure. This is also what
Hales and Barker called the thrifty phenotype–
a restriction before birth followed by a surfeit afterbirth
gives us the worst physiology. There’s some other
classic animal experiments in this field. Here we have two
identical twins. So these mice have
the same genotype. They’re clearly really
different in their phenotype. One’s yellow, and
fat, and cancer prone, and doesn’t live very long. And the other one’s
brown, and thin, and lives for a long time. So what’s the difference
between these two? It’s there epigenotype. So just by altering the
amount of methyl donors that the mother gets around
the time of conception, DNA methylation is altered, and
you get these wildly different phenotypes. So animal experiments are
good for exposures for timing, for mechanisms, for
effects on outcomes. They’ve proved the
programming principle. They’ve shown this thrifty
phenotype hypothesis of untoward consequences
of the combination of prenatal restriction,
of postnatal surfeit. So what’s not to like? Well if you look at the animal
experimental literature, we find that often missing
are the basic things that we do in human experiments– source
population, sampling frame, eligibility criteria,
recruitment and retention rates, blinding, intention
to treat analyses, attention to missing data,
and cluster methods. Here’s an example– maternal
high-fat diet has now become a rather common
paradigm in the Developmental Origins of Health and Disease. And a few years
ago these authors tried to do a systematic
review of animal models of maternal high-fat
feeding and offspring glycaemic control. They reviewed 1,483 studies. Only 11 met their criteria,
and among those 11, because of the
criteria I showed you on the previous slide
and other reasons, the quality scores were low. Not only that, there was a large
variability in maternal diet. Some were hypocaloric,
others hypercaloric, others not stated, but
none were isocaloric. And there was a wide range of
fat and carbohydrate content, so you really can’t tell
what a high-fat diet means. They had different
postnatal feeding regimens, different ages at outcome,
different outcome assessments. It’s impossible to summarize
or meta-analyze those data. Another way that
animal experiments can be useful in
this field is go back to what they were
a few years ago. I showed you this sort
of interdisciplinary, multidirectional communication. Right now a lot of
animal experimenters have their own paradigm, and
go deeper and deeper into that, and forget to translate
back up to the things that are important for humans. And so there are very
few animal studies that look at the issues on
this slide, which are really important to health today. So my view is that animal
studies should harmonize interventions and measures. I didn’t mention it, but
there’s an importance of publishing null results
and translating up. Well we’re not so great in
human population studies either. We’ve got our advantages
and disadvantages. And in developmental
origins, a lot of the studies that we have
come from cohort studies. Observational studies
have the big problem, and that’s confounding. Randomized controlled trials are
meant to minimize confounding, and they have some
other advantages, but randomized controlled
trials have their disadvantages as well. Since confounding
is such a big issue in observational cohort
studies, why aren’t we doing more to overcome it? To do the things we know get
us closer to causal reasoning? Many of them are listed on
this slide– more judicious multivariable approaches,
sib-pair designs, cohorts with different
confounding structures, long-term follow ups
of randomized trials, Mendelian randomization,
biomarkers, and quasi-experimantal studies. Each of these has
strengths and weaknesses. We can’t just look
at one approach. We need to consider
them together as a basis for judging evidence. Here’s an example–
breastfeeding and childhood obesity. So for a long time
there’s been a hypothesis that having been
breastfed may reduce the risk of childhood
obesity and obesity throughout the life course. And there are two
major hypotheses. One has to do with
components of breast milk– that there might be hormones
that are programming appetite– and the other is a
behavioral approach– that mother child
diodes who breastfeed the child learns more
about self-regulation of energy intake growing up. But there might
also be confounding because the same sociocultural
factors that lead to a decision to breastfeed may be the
same ones that are associated with obesity, and besides,
fast-growing babies may wean, so-called
reverse causality. Now when we published this
study from the Growing Up Today Study in 2001, there
was a lot of hope that longer duration
of breastfeeding would actually reduce
the risk of obesity, and here you can
see this association between longer duration
of breastfeeding and a lower risk of
overweight in adolescents. Ten years later I was
asked to review this field for the International
Journal of Epidemiology, and I took the opportunity
to do a scorecard in which I summarize the evidence for
and against the hypothesis that having been breastfed
reduces the risk of obesity. And at that point,
you can see that there were a number of different
kinds of studies available. There was early follow-up of
a cluster randomized control trial. We had a number of cohort
studies in developed world; fewer in the developing world
and racial ethnic minorities; the sib-pair analyses; different
confounding structures; reverse causality; biological
effects; behavioral effects; and an ecological analysis. And what you can
see on this slide is they’re somewhat
equal entries in each of these columns
for yes, maybe, and no. I don’t know if that’s the best
way to summarize these data, but in any case I wound up being
on the fence at that point. But this was before we
published the most recent data from the PROBIT study. PROBIT is a cluster
randomized controlled trial in the Republic of Belarus
which randomized hospitals to the Baby Friendly Hospital
Initiative or Usual Care. There were 31 hospitals. And follow up at 11
and 1/2 years involved about 80% of the participants. And what you can see is that
breastfeeding promotion did not reduce adiposity at this
age of 11 and 1/2 years. If you look on the left
hand part of the slide, you can see the
differences in body mass index, fat mass index,
fat free mass index. On the right hand
slide is the odds ratio for either overweight
or obesity. And there’s certainly no
evidence of reduction. In fact, if anything,
there’s evidence of harm. So in breastfeeding-obesity,
earlier studies suggested considerable
protection. More recent studies
cast a lot of doubt through this range
of study designs. And PROBIT is a
very important study because it not only is a
good test of causality, but it can also test the effect
of this Baby Friendly Hospital Initiative policy. Here’s another example. Dr. [? Goodmacher ?]
raised the sort of hot issue of epigenetics. So we’re looking at
epigenetics, and Project Viva, which was mentioned
before as the cohort study that I run in the Boston
area where we recruited women in pregnancy, and we’re
following their kids over time. And we’re interested in looking
at prenatal exposures, DNA methylation in cord
blood, DNA methylation later in childhood, as relates
to adiposity and metabolic consequences. And on the bottom you can see we
have a discovery, a validation, and a replication. So to put it more simply,
we have an exposure, an intermediate, and a health
outcome– prenatal factors, epigenetics, health outcomes. So what is the role
of studies like this in policy-relevant evidence? So we have this intermediate–
DNA methylation– between pre and
perinatal exposures and obesity related outcomes. So for one thing epigenetics
becomes a surrogate outcome. That makes studies
more feasible. We don’t have to have
as long-term studies. But I have to say that in
the new field of omics, and even in precision
medicine, there’s this move towards providing
signatures for prediction. In fact, the term biomarker
now means something about prediction, rather
than a more generic term. But we have to remember
that prediction has a high bar of proof. For prediction to
be useful there needs to be a high sensitivity
and specificity not just a modest elevation
of relative risk. We’re used to thinking
of relative risks in the order of two
or three as strong, but to make good
screening tests, relative risks often have to
be in the order of magnitude of 200 or 300. And besides this
is a medical model. It’s all about individual risk. And as Dr. [? Goodmacher ?]
pointed out, a lot of this is in the service of
developing pharmaceuticals with drug targets. But are we really in the
business of developing drugs for pregnant women and infants? Probably not so much. And the other thing
is if we really go to the extreme in this
individual risk we’ll wind up with messages like this–
“Epigenetics Warning: What You Eat Today Could Harm
The Health of Your Children and Grandchildren.” So how is a pregnant
woman who already is very concerned about
her fetus supposed to take this kind of message and
turn it into positive behavior change? And this is why Sarah
Richardson among other– with a bunch of co-authors,
including myself– published this paper in Nature
last year saying, let’s not blame the mothers. So if we’re not going to
blame the mothers, which– well actually I have
another blame the mothers that you might like. “So I blame you for everything–
whose fault is that?” Those of you who have
teenagers may recognize this. So if we’re not going to do
that, why do we do epigenetics? Because it’s
elucidating a mechanism. And by the way the
best definition I’ve ever heard of
mechanism is from Tom Insel, the director of the National
Institute of Mental Health. He said a mechanism is
one level of reductionism more than you work in. That’s the definition
of a mechanism. So why do we look for mechanisms
in developmental origins? It’s because we want biological
plausibility for causation. This goes back to the classic
definitions of causality by Bradford Hill
some 50 years ago. And these become a rationale
for primordial prevention, and by primordial
prevention we’re talking about optimizing
socio-behavioral milieus starting at
conception or before. We’re trying to avoid maternal
obesity, excess weight gain, gestational diabetes,
smoking in the first place. And it’s not necessarily,
primarily, a medical model. In fact Clyde
Hertzman and Tom Boyce think about epigenetics
as a way to characterize how societal variables
embed themselves in biology irrespective
of behavior. Epigenetic– Experiences
get under the skin early in life that affect the
course of human development. Epigenetic regulation
is the best example of operating principles
relevant to biological embedding of societal influences. I also think that there’s a way
to communicate between science and policymakers in which
mechanisms that aren’t too deep are good– are acceptable
and easy ways to communicate. And I think epigenetics is
that right archaeological level to motivate how
pre and perinatal factors may affect chronic
disease over the life course. So, for example,
even I can understand that after conception there’s
a massive demethylation of our genome and over the
ensuing weeks and months, methylation comes back
at different rates in different tissues. That’s why our liver cells are
different from our brain cells. And I can also understand
why the fat, yellow mouse is different from the
brown, skinny mouse. So I mentioned
prediction a minute ago. When can prediction be useful? So we published this paper just
over a year ago using Project Viva data, and
what we did was we took two prenatal factors and
two postnatal factors that have often been related to obesity. We dichotomized
them for simplicity, and we said, OK what if
you have the optimal levels of these risk factors? Your mother didn’t smoke,
gained less weight, breastfed you for
less than 12 months, and you were a good sleeper. The predicted probability of
obesity at age seven years is 4%. On the other hand, if
you had the adverse level of these factors, your mother
smoked, gained excessively, didn’t breastfeed you
as long, poorer sleeper, and the predicted
probability is 28%. And this slide shows all 16
combinations of those factors. The 4%, the 28%, and if you do
a population attributable risk percent calculation,
you get 20% to 50% depending on where
you put the cut point. So the easy thing to say is, OK
if you ameliorated these four risk factors, you’d save 20%
to 50% of childhood obesity, but of course that depends on
the assumptions of causality and modifiability. That’s why we do interventions. And the implication is
that multiple risk factor interventions hold promise
for preventing obesity, and that’s what a
number of interventions are testing right now. So prediction can quantify
the overall benefit of intervening early and
may be able to distinguish the most important
determinants that may vary by population and subgroup. Just a few minutes about
risk-benefit analyses. So we often look at one
exposure and one outcome, but in reality there
are multiple exposures and multiple outcomes. So rapid weight gain in
infancy, in many studies, is associated with
later obesity and cardiovascular consequences. But remember that weight is
composed of both linear growth and adiposity
growth, that we might be interested in multiple
outcomes, neurocognition in addition to obesity and
cardiovascular disease, and we have both full
term and preterm infants. So Mandy Brown Belfort
and I, a couple years ago, published this review in which
we looked at these features. We had healthy AGA full terms,
preterms, and SGA infants. We looked at these
two outcomes, and we separated into linear
growth and gain and weight-for-length
a proxy for adiposity. You can see on this
slide that there are a lot of question
marks for linear growth. In our rush to
obesity, we sort of forgot about that
element of growth. And you can also see that in
the preterm– in all of them gain and weight-for-length
is associated with obesity, but in the preterms a
gain and weight-for-length is also associated with
better neurocognition. So we need to know the balance. Here’s another example–
fish intake during pregnancy. My colleague Emily
Oken has worked in this space for
about ten years, and the question is what kind of
fish should pregnant women eat? It’s complex because we
have nutritional benefit, we have toxicant
risks, like mercury, we have ecologic
concerns over fishing, and economic influences. Fish is more expensive. And there are a
lot of complexities within and across
these elements. Here’s from Project Viva. One indication of
benefit– the more omega 3 fatty acids
in the maternal diet, or in the blood of the mom, or
in the cord blood, the lower the obesity rate
several years later. But you also have the harms. “For shame! Pregnant and eating fish? Think of the baby!” And Emily was able to
look a couple years ago at 19 published seafood
consumption guides and calculators and seeing
whether they covered this general population
perspective– contaminants,
benefits, ecological, and economic influence. As you can see, none
of them covers all five of these influences. And so you wind up
with recommendations like, “Eat up to 12 ounces a
week of a variety of fish.” And from another
organization, “Consume a minimum of 12 ounces
of seafood per week.” So risk benefit analyses
have to take account of multiple exposures,
outcomes, and scenarios, and they really help with
establishing guidelines. Just a word on interventions. Most interventions
in the DOHaD have been efficacy interventions,
so very highly controlled circumstances. I just want to make
a comment about how we need to focus
on implementation– the upstream effects for
effectiveness, sustainability, and dissemination. And now there are a number of
whole-of-community approaches to obesity prevention. They really work
at higher levels of the socio-ecological
ladder, and they focus on environment and policy. And in a rather new study that
I’m working on collaboratively with a number of national,
international investigators, we’re looking at this
upstream intervention element. We’re trying to ask what
works for whom and under what circumstances for 0-5-year-olds
in community based obesity prevention. As I said, we’re
focusing upstream. We’re thinking about how
do stakeholders relate to each other, and how
do these relationships lead to effective
implementation. And the way we’re doing this,
and I can come back in a couple months when I have
some more information to tell you exactly
how it works, is we’re applying computational
system science simulation modeling, specifically
agent-based modeling, to two completed interventions,
one in the US, called Shape Up Somerville,
one in Australia, with the better name
of Romp and Chomp. We’re doing an iterative
process with community members to refine the model with an
ongoing cluster randomized control trial in
Australia, and then come back and design a new
intervention for under fives called Shape Up Under Five. Long-term effects. Well I just mentioned
simulation modeling, and in developmental
origins we need to know about long-term effects. We want to know
effectiveness, safety, cost, cost-effectiveness. The only way I know to
integrate from multiple sources is simulation modeling. Now this example is
not from earlier, that much earlier
in the life course, this is about cost-effectiveness
of blood pressure screening in adolescence, which
we published a few years ago. And in this we took a
two-stage model structure. There already exists a coronary
heart disease policy model, which takes risk factor
distributions at age 35, and over time talks about the
risks and costs of developing disease and either
going on to death, or having the disease for
a long time, or recovering. That’s a Markov model. And what we’re able to
do is take information that we and others have
produced about blood pressure tracking, predictions,
screening, and treatment from age 15 to 35 and
hook them on together. And we asked several
screen-and-treat and population strategy questions– we
compared different strategies– and we applied this
to a baseline cohort of 15-year-old adolescents. And the bottom line
is that we found that population-wide policy
approaches are both more effective and less
costly than any of the screen-and-treat
strategies. And as you know,
everyone is supposed to measure blood pressure in
pediatric care in routine child visits. But this study suggests
that that might not be a useful strategy. Instead, what about a
salt reduction campaign, which is a really an
environmental and policy intervention. Finally just a couple words
about policy evaluation, natural experiments. This is actually the
first study we ever published from Project Viva. It goes back to the fish story. Emily Oken was the first
author, and we were lucky that the recruitment
into Project Viva straddled the first
federal mercury warning that came out in 2001. And so we were able
to look at fish intake both before and
after the warnings. And you can see in
the top blue line that the intake of all
fish was actually probably going up before the warnings. Right after the warnings, which
were not a really strongly implemented intervention–
they were promulgated through the media, and there
might have been posters in obstetricians’ offices. So it was a modest intervention. But right after
this intervention, there was a drop in fish intake,
and the slope was downward over the next year. So the good news here is
that with not a very strong intervention,
pregnant women may be able to change their behavior. They really care about the
health of their fetuses. The bad news is this was
the wrong thing to change. And then more recently
we’ve looked at associations of tobacco control policies. Summer Hawkins was a
postdoc working with me. And we looked at all births
from 28 states and DC between 2002-2010 and found
that increases in cigarette tax are associated with improved
health outcomes related to smoking, and it was just
among the highest risk mothers. It was among those
with low education. So in conclusion to achieve
evidence-based policies and their
implementation, I think animal studies should have
more consistent methods, harmonization of
designs and measures. And our own studies
of epidemiology, we need to combine observational
intervention studies. We need to use innovative
designs and analyses, we need to compare and
combine across studies, we need to use epigenetics and
other mechanisms like that, not only as our influences
on the kind of interventions we do, but also communication
with policymakers. Prediction models, I
think, are useful for potential intervention targets. And to identify
those, we really have to be careful about
risk stratification. Risk-benefit estimates are
good to inform guidelines. We need to move beyond
efficacy and intervention to talk about implementation. Long-term simulation models
are excellent, especially for cost-effectiveness
and evaluation of current policies for impact. So while good evidence
is not sufficient to make sound policy, it
sure is helpful. And later I’d be glad
to answer any questions. Thank you very much. [APPLAUSE]

Leave a Reply

Your email address will not be published. Required fields are marked *