Psychiatric symptomics: a new perspective on mental disorders | Eiko Fried

Psychiatric symptomics: a new perspective on mental disorders | Eiko Fried


Before I start talking about why
analyzing symptom data may be highly relevant for psychiatric disorders, I
would like to introduce a couple of problems in the field. There’s many more
problems, and I will focus on four problems today just briefly.
Starting with genetics and I will try to talk about psychiatric disorders in
general little a little bit but my own research has been on depression mostly
so most of the examples here will be focused on depression. We know that the
heritability of depression is about 40% maybe a little bit higher, depends a bit
which study you read. And so I’ve been looking for genes for depression for
about 20 years and these are just a couple of studies that were published in
the last years. Some of these studies had up to 35,000 participants and the result
is that we’ve had not we’ve not found a single loci. No single Loci. Nothing.
And most studies do read like this. This is an abstract from paper: “We looked at
35,000 individuals. We found nothing so we may need 50,000 individuals to find
something.” And you can already see that this is a somewhat of a problem when it
comes to the effect size, right? So 2015 was a year of discovery and most of the
papers I will show you today are actually just two or three weeks old. So
this was published just very recently and it’s the first study that replicated
two loci in two different samples so that’s a huge breakthrough, at least
that’s how the few considers this. The first sample was about 5,000 Han Chinese
women with recurrent depression, the second sample equal size equal
demographics. They found two loci on chromosome 10, and the new thing they did
is they homogenized the samples. They took only Han Chinese down to the fourth
generation, they took only women, and they took only people with a very special
form of depression. And the authors themselves celebrate this in the
abstract as a success and it’s really a very cool paper. It was a lot of work a
hundred-fifty collaborators, many universities, so kudos to them. And the
field went crazy, first time in 20 years people were really happy about this. There was a huge editorial in Nature “First robust genetic links to depression
emerge: The findings could guide biologists to new drugs based on two loci
and could one day be used to aid diagnosis”. So really major breakthrough. What
neither the paper said in the abstract nor the editorial it at all, is that the
authors actually didn’t replicate the two loci in a third sample of European
ancestry. So you’ve got to read a bit more carefully to actually see that this
was the case. On top of that, of course, if you only stay in the Han Chinese
samples, the two loci don’t explain any variants. There are two alleles they
explain 0.0001 percent of the of the variants for depression. So clinically
this is really useless. It’s a cool insight and it’s a really cool study but
it’s not a breakthrough in any way because it tells us nothing about
depression. And on top of it these two loci are likely not specific to
depression, because we know that the genetic correlation between depression
and schizophrenia or between depression and bipolar disorder is very high: up to
50%. This was a paper in science just published a few weeks ago. So this is bad news. Is there a bright side to this? And I will–because it’s all very upsetting–I
will have a couple of bright side slides in there for schizophrenia. The results
are a little bit more promising, we can explain about 6% of a lot of the
liability for schizophrenia, with hundreds of genes– if we add them up in
some complex way–although it’s a little bit less than non-european samples but
we’ll get there eventually. And to tell you what this means: people at
the high end of risk have an odds ratio of four or five times higher than the–
that’s really cool actually right–but for depression we’re very far from there. Second problem in the field: neuroimaging. Again when we think about depression
most studies so far have been very small samples, totally underpowered, lots of
false positive findings, which have led to a field in which brain regions are
positively and negatively associated with depression. Hippocampus for example. Hippocampal volume has been shown to be larger in depressed patient
and also smaller in depressed patients. Again 2015 major breakthrough, a cool
paper by Schmaal et al. Roughly two thousand patients, structural brain
imaging, lots of controls and they looked at nine very common subcortical regions. They ran two different analyses. The first one is dimensional,
that means they wanted to see in the depressed people whether the severity of
depression was correlated with the volume of these nine cortical regions.
They found nothing. In the next step they looked at categorical analysis. So they
compared the brains of depressed people with the brains of the healthy
people, again for all nine regions, and for one of these nine regions they found
a significant effect. Hooray the hippocampus is slightly smaller in
depressed patients, the volume difference is 1.4 percent. They didn’t provide any
classification accuracy in the paper so they don’t tell us what these one
perfect person 1.4 percent really mean in terms of classifying people based on
their hippocampal volume. Again everybody went crazy so this paper is cited, I
think, 30 times already although it’s published two or three weeks ago. Again it’s a really cool study a lot of people work together on this
it’s the best day that we have currently. There was an interview with one of the
authors–authors in this piece– the authors stated, “I think this resolves for
good the issue that persistent experiences of depression hurts the
brain.” So you see there’s a little bit of over-interpretation going on here in the
field, at least that’s how I think about it. So we wrote a commentary to Molecular
Psychiatry in which we calculated the classification accuracy based on these
hippocampal volumes–so we simulated data. And the blue group is the depressed
group and the red group is the control group and this is the hippocampal volume
and you see it’s 1.4 percent volume difference is really not very much. And
so the prediction accuracy is about 53 percent. So you can toss a coin or you
can scan the brains of 10,000 people. Looking at the current evidence right?
We’ll get there, but at the moment it’s– it’s a bit of a problem. And again
smaller hippocampal volume is not specific to depression, we found it and
basically anything and everything including accelerated aging and lack of
exercise which they didn’t control for in the study–at least this one. Bright side? Well not much, but there’s some functional studies that have reached
sensitivity and specificity values of about 70%, although most studies have
samples of 20 or 30 people–so we need to work on this a little bit harder and
replicate it, hopefully in the future. Antidepressants–I have to tell you right
away there is no bright side to this to this part of the presentation–a study
published just two weeks ago found–and this has been shown many times in the
field that–that antidepressants work for roughly 40 percent of patients and
placebos work equally well for roughly 30% of depressed patients. It was very
disappointing considering we’ve been working on new antidepressants for four
or five decades. It’s quite–quite disappointing. Also of course
antidepressants have side effects that placebos don’t have, right? So last point of the problem introduction, psychologists and psychiatrists use the
DSM it’s our holy book in which we define what disorders are, and what
symptoms they have and so forth. And whenever a new version comes out we run
field trials. And so this came out two years ago, so there was a big field trial
and we want to understand how good diagnoses are. And how we do this is by
doing inter-rater reliability or kappa. How this works is: I will send a
depressed patient “me” to Stephen and Stephen will figure out if I’m depressed
or not. And then I will send the same patient to Randy who will figure out if
I’m depressed or not, but both are blind. Both clinicians are blind regarding the
diagnosis. The kappa value for depression is 2.0 –eh– 0.28 that’s a correlation here, right. Which is above chance but really not
very much, and just for comparison Borderline for example, which is
considered a fairly complex disorder, as a about twice as large of a Kappa and
ADHD is higher and so forth. so depression I think they had 30 diagnoses
and it was in the bottom five or bottom six. Okay so all of these things are big
problems and I have the feeling we’re running
against the same wall with larger samples. Something –I mean–
larger samples sometimes just don’t solve anything.
So in the in the DSM-3 in 1980, in the preamble it stated “by the time the DSM 4
will come out we have biomarkers for all diseases–for all psychotic diseases” and
today we have zero. Nada. At least for the major psychiatric diseases in terms of
reliable biomarkers. and there was a paper in science two weeks ago that
stated exactly the same: you know in a couple of years we’ll have found
everything we just need to do more of the same that we’ve been doing. And so
this is a reviewer and also an editor of the journals I try to publish in and
when I try to say “hey look at this new idea take this ball” they’re like, “no I
will keep my idea” and it’s a bit obnoxious sometimes. Or it’s hard it’s a
challenge It’s really hard to get rid of an idea once you have it. Okay so my
proposal is that we have to look at these problems from a different
perspective by questioning foundations. So let’s go back to the assumptions we
have about how mental disorders work and then how all this comes together. And this is a picture I took yesterday in the new Center for Evolution and Medicine, here. Darwin says “hi” to everybody who comes in
so I guess you’re all welcome to say hi. Maybe a professor Darwin can help us a
little bit with coming up with ideas why we’ve had problems in the field. Alright so using modern evolutionary theory we can separate two types of symptoms of
diseases. The first type maybe–there’s more–I have I usually tend to think of
two types, um the first type is defects. Think about Ebola.
For example, symptoms of Ebola are defects and they kill you and that’s bad,
right? The second type of symptoms are defenses. Think about cough. Think about
fever. Think about pain. These things are triggered by the body in response to
certain threats, to help your body deal with problems, right? With recurrent
fitness threats. Now what’s interesting is that in both cases the disease
precedes the symptoms causally. That’s how we think about symptoms,
right?– So we ask patients about symptoms because we know that the disease causes
the symptoms. We treat the disorders and not the symptoms. And in medicine and in
philosophy of science we call this the “common cause model”. It’s the standard
model in disease model in the field. There’s no other disease model, really.
And, well yes, let’s see how this model works This is measles. You have measles.
It has a circle because I can’t see measles, but I can see red eyes, I can see
fever, I can see other symptoms of measles. And when a patient comes to me I don’t see measles I see only symptoms. But I’m a smart doctor so I figure out–
aha–this patient has measles because these symptoms are somewhat specific to
measles. Not all of them, but some. And my talk today will be structured into three
segments. And these three segments are the three implications that we get out
of this disease model. The first implication is that Peter, Susie and Mark, all of who come to me with these symptoms have the same disease. They have
measles. right? There is no 14 types of measles. There’s measles. The second implication is that the symptoms are roughly interchangeable. It doesn’t
really matter that much if you have the first three symptoms, or if you have the
second three symptoms. A smart doctor will figure out you have measles. And the
third implication, and maybe the most interesting one, is that symptoms are
unrelated beyond their common cause. Think about fever and a runny nose. If
you have a data table of the universe of all people, fever and runny nose–in two
columns–will be correlated somewhat. They will co-occur in nature quite often, but the reason they co-occur is the common cause. It’s not because fever causes the
runny nose or because the runny nose causes fever. That clear? Okay it’s important for today Now this common cost model is the only
one we really have in psychiatry. It’s the only one we will use in depression. And you can see that we use this disease model when you look at how we how we do
research on depression. and how we treat depressed patients. We measure a bunch of
symptoms to indicate the disorder, we add these symptoms to one
total score, to indicate the severity. That’s the a very common questionnaire
for depression, the PDI, it has 21 symptoms, and we ask people about the
symptoms and we add them up to one some score. So Peter has 10 points: he’s
depressed. Susie has five points: she’s not depressed. That’s it. Symptoms are roughly interchangeable otherwise we couldn’t add them up right
mathematically. You can’t add something up if it’s not somewhat equivalent to
each other. And we treat the disease depression, for example the most common
drug, SSRI, targets the serotonin transporter system in the brain. Because
there’s this idea that there’s something wrong with serotonin in the brains of
depressed people. Now I will claim today that is overly simplistic and we need a
new disease model. And I will talk about these three implications now in a little
bit more detail and present a couple of studies we and other labs have conducted
in the last two or three years. So this is all very, very recent really. Ok the
first point is that “all individuals have the same disease” and for depression this
is hugely problematic and you can see that this is problematic just by opening
the DSM and looking at the symptoms that are defined as depression symptoms.
There’s nine symptoms but you see that eight of these symptoms have lots of “Ors”
in them “diminished interest or pleasure”. So this is really more than one symptom,
because interest and pleasure are not the same thing. Three of these symptoms
are actually opposite. “insomnia or hypersomnia” is not the same thing. In
fact it’s as far as apart as it could be and the opposite. So we did a
study last year in which we wanted to find out how many unique symptom
profiles are there in depressed patients. Because nobody had looked at that before.
So it’s a very simple question. We had a database of 3.7 K depressed patients
and all we did was count the number of unique symptom profiles. So this is the
printin profile of Peter, for example. He has the first symptom, he doesn’t have
these symptoms, he has this symptom, and so forth, right? So ones and zeros. Very
simple. And we found 1030 unique profiles in 3700 people. That’s insane.
So that’s yeah that just insane. The most common profile had
a frequency of only 1.8%. So it’s a hugely heterogeneous disorder and the
conclusion from this would be that maybe depression “yes/no” is not a good
phenotype to study genetics, study brain structures. And maybe it’s not a very
good phenotype to treat people with one specific drug. Maybe these people have
very different problems. Okay. Of course it’s technically possible that
all these you these diverse symptom profiles are caused by the same reason.
It’s certainly possible I just find it very unlikely, but we cannot exclude that,
of course. The problem is much more pronounced than our study showed. Because
we only looked at the DSM symptoms for depression, the symptoms defined in the
holy book. But in depression research people don’t use these criteria very
often, what they do is they use questionnaires for depression. These are
seven very common questionnaires for depression: The Beck, the Hamilton and so
forth. These scales were made in the 1960s and 70s mostly. People still use them for
some reason. And the weird thing is that these symptoms–the symptoms these scales
have –have very little to do with the symptoms featured in the DSM actually.
Right, so what I did here–and don’t put this it’s going to be in video, but you
guys don’t put this online because I still need to publish a paper about this–
so we simply counted all the unique proof as symptoms–sorry–all the unique
symptoms that appear in all these seven scales and these are 55 symptoms. And
then we plotted–I plotted–this and you can see here for example this is the CSD
scale, very commonly used today in many social psychology studies and
experiments, it has ten symptoms that are only appearing in this particular scale.
And this the “C” is the use only twenty symptoms, so half of the symptoms are
idiosyncratic to that particular scale. Right so that’s yeah. “What?” It’s really a
problem. Like and it’s not acknowledged in the literature. It’s really weird.
So people believe that you can use this scale or this scale. The result will be
the same. But of course it’s not the same because you ask very different questions
and different scales. Okay why do we do that? Why do we have these beliefs? The reason I think and some philosophers of science, believe that people have
these ideas about mental disorders is that humans have this tendency to
believe things are things in the universe, which we call “natural kinds”.
Right so gold is a natural kind that has the atomic number 79 everything with the
atomic number 79 is gold. It’s very simple. So there’s a an essence to it.
That’s why we call it the essentialism. And this essentialist thinking leads us
to simplistic questions about the world such as: what causes
depression or what are the genes for depression. But maybe depression is not a
very good category to up to ask these questions. And yeah so this is called
essentialist bias, this tendency in the literature. Basic emotions are a good example. I cannot tell you how many dozens of basic
emotion theories I had to learn when I studied psychology. There’s five basic
emotions there’s seven, or eleven, or there’s even one with ninety eight basic
emotions. But that’s not how the world works it’s more complicated than that.
Species are an example you may be more more familiar with. Pre-Darwin people
believe that beavers are beavers, are beavers. There’s like one thing that
defines beavers as beavers, but it’s not how the world works. And even biologists
today sometimes think about species as “natural kinds” because it comes natural
to us to think like that. But it’s wrong. And so again thinking about about Darwin
and his theory may help us to understand that this is a simplified or
a very oversimplified thinking. But then again, psychiatric diseases are not just
“social kinds” as well which is on the other side of the spectrum. “It’s just
made up by people.” I don’t believe that’s the case. There is something distinct
about people with mental disorders right? With severe mental disorders. So an idea
that people have proposed recently are what they call “pragmatic kinds”.
Depression doesn’t have to be real to be useful right? Diagnosis should be useful
things that help us predict treatment response, relapse probability, course of
illness and so forth. And it’s pretty much universally accepted that
depression is not a very good pragmatic kind because the features it
has does do not allow us to predict about the person, or the course, or the
specificity of treatment and so forth. Okay that was the first part heterogeneity of depression, big problem. The second part is the idea that these
symptoms, because we add them up are roughly interchangeable. aAd to make a
question out of this: do symptoms differ in important properties? Because if they
do we can’t add them up to one sum score or we lose a lot of information
if we do. So that’s the–I wrote my PhD on this topic and I’ll present a couple of
very very brief studies. We know that depression causes severe impairment in
in many in 60-70 % of the cases. People can’t go to work anymore, they have a hard
time maintaining relationships, and so forth. So we looked at 3,700 patients
again with depression and wanted to figure out whether symptoms–whether some
symptoms–caused a lot of impairment but other symptoms cause very little
impairment. And here you see that these M symptoms and this is the impairment the
symptoms caused. When you share–when you spread the unique variants to individual
symptoms–so sad mood for example explains 21% of the variants of
impairment, but hypersomnia only explains less than 1%. And this is controlling for
severity so it’s not the most severe symptom that is the most empowering it’s
the nature of the symptom that determines how impaired you are. And so
if you have five symptoms I don’t know much about your impairment but if I know
the nature of your five symptoms for example, these against these, that will
provide a lot of information. Okay. And it’s mind boggling that nobody had done
that before, quite frankly. We also looked at risk factors for depression, and if
you believe into the common cost model, of course risk factors are unrelated to
symptoms. Risk factors increase your probability to get the disorder, an
adverse life event for example, but they have nothing to do with the symptoms. And
so we tested this in a really cool dataset collected by Street & Son and
colleagues at University of Ann Arbor in medical residents. And we know from the
literature that medical residents are a target for depression research because
they have such a bad time in the US. So the median worked. So medical
residents often move away from their own city and work in a hospital somewhere
else. They have no social network, they don’t have family there.
The median work hours per week was hundreds in this data set for a year.
It’s getting better there I think a few laws have been passed since the data
were collected, but anyway it’s a good sample for studying depression. And so we
have a baseline measurement point before enrollment into residency: people are
doing fine, they’re on vacation, they’re having a good time and then six months
later we reassess the the people and they’re having a really bad time. And so
we collected a number of risk factors at baseline and we predicted the increase
of individual depression symptoms. And so we can figure out if risk factors are
associated with specific symptoms. And we start here to explain this graph. So this
is history of depression and we have it again at baseline before terrible things
happen. And these are the nine DSM symptoms and the green lines are
positive associations in terms of prediction. So if you have history of
depression “yes” at baseline, your probability to get these seven symptoms
is quite dramatically increased, six months later. But not a probability to
get these two symptoms which are “psychomotor problems” and “suicidal
ideation”. So it’s specific. Risk factors still are specific things about the type
of symptoms people develop. I can’t go through all of this but we’ll do gender
which is really interesting. If you are a female resident at baseline, I can
predict that sleep problems, fatigue eating problems and concentration
problems will increase over time. But if you’re a male resident you’re more
likely to get suicidal ideation, so we really need to look at these symptoms.
Right? Because they provide crucial information we don’t get if we just sum
this up. In fact, if you sum this up here you will learn that female residents are
more likely to get depression, because you lose that information, here. Okay
same data set. Same dataset:
prospective study of residents. We simply wanted to see how symptoms respond to
stress, and for this specific kind of stressor: medical residency. You see that
some symptoms go up by up to 300%. For example psychomotor problems and
other symptoms don’t increase very much. Again the message here is not we need to
generalize, or we can learn much from this specific study, but looking at
symptoms gives us a lot of insights. Now let’s think about antidepressants for a
second. The way they are studied–and they have been studied always–is you give
people a scale of 20 symptoms, you add them up, and then people that based on
have–let’s say–20 symptoms on average or 20 points on the scale and then you
follow them over time. And you want to find out if the some score decreases
over time and then the treatment is considered to work. Well we have no idea
how individual symptoms respond to treatment and it’s super likely that
certain drugs have certain effects on certain symptoms. So we’re
studying this at the moment but I don’t have the results yet. Another
complication is that most depression symptoms reflect side effects caused by
antidepressants. We know that SSRIs cause sleep problems, weight change,
concentration, and sometimes psychomotor problems. They can cause suicidality, at
least that’s what the research looks like at the moment, and so forth. So this
is hugely confounded. Some symptoms may go up in response to treatment some
symptoms may go down and we don’t know what’s going on. This is all terrible so
I have a bright side slide again. There is a bright side to this and
again maybe professor Darwin can help us out here a little bit. A student of a
friend conducted research during his PhD, Matthew Keller, who’s now a
geneticist, I believe working in Boulder and he came up with a really cool idea.
So he’s looking at symptoms as the evolutionary defense mechanisms not as
defects. Right and he came up with the idea that individual symptoms may serve
as specific responses to specific adaptive fitness threats. Right but it’s symptom specificity. So there were three papers here. This is
in the Journal of American psychiatry, go check it out it’s really awesome and to
summarize his results, that are quite stable across the three papers and
across very different samples of students, and general population, and
veterans and so forth: Different specific recurrent fitness threats are associated
with rather specific symptom profiles. Afterwards so in winter time, for example,
you see that fatigue is pretty high and that pessimism is pretty high and Matt
argues that it may be a very adaptive strategy if there’s no resource
available to tone down optimism and to tone down exploration behavior and to
rather sit around and wait for better– for a better time. Can’t go into details
but check this out it’s I think it’s really promising work. ***** um I don’t know
what I wanted to say with this slide well nobody has looked at defects maybe
we mostly looked at symptoms as defences but it’s quite possible that some
symptoms are also defects and so forth yeah exactly okay last the last point how am i doing
on time plenty of time awesome okay perfect I’m good in time so
oh you’re right you have a brief Intermezzo because you’ve been following
so nicely so you get to see a very cute picture for a second and I get I don’t
even know it’s a small octopus enough okay now we come to a little bit
more statistical part of the talk that’s why you I gave you a treat so we don’t
believe and and we is a group with Danny bourse woman professor in Amsterdam and
and my colleagues at the University of Leuven and in Belgium we don’t believe
that symptoms Co occur due to their common cause and depression we believe
that symptoms Co occur because they cause each other which for depression is
really very very intuitive and this is an example network from a patient
insomnia in this particular Network triggers psychomotor problems fatigue
concentration problems there’s some form of feedback loop here and weight loss
for example for this participant would not be connected to the rest of the
network which we sometimes see right we also don’t buy into the idea that
symptoms are roughly equally important or equivalent things we we know now I
hope I’ve convinced you a little bit with the research I have shown that
symptoms may be distinct properties with different characteristics meaning they
deserve individual attention and that’s why we put all individual symptoms in a
network we don’t ignore information and of course consistent with clinical
theory and intuition I would say I guess in these networks symptoms are allowed
to reinforce each other to fuel each other in vicious circles of problems
that are hard to escape for the patient and there is something clinicians
usually agree with I mean it it happens in in certain disorders so we call it an
attractor state and complex system so the network can can go to a local
minimum and it’s it’s hard to break out of the association chain and from this
network or from this idea important questions arise that we would not get at
if we would not think about these problems as networks and one question is
what symptoms are most central to driving these processes and insomnia in
this case would be a highly central node in the network but weight loss would not
be a very highly central node and if you think in terms of treatment you want to
treat this node and not this not right because it will
Packt on the network as a whole quite fundamentally so this is a study that
was published yesterday yes go check it out so still so three and a half
thousand depressed patients we look at 28 depression symptoms so these are the
DSM symptoms but also other symptoms that we know are quite prevalent among
depressed patients such as anxiety and irritability who knows what gosh and graphical models
are just okay so we estimate these networks using fancy methodology but
it’s actually quite common in other fields like physics and biology this is
not not a fancy thing psychologists don’t know it very well but it’s not
fancy the most important thing is that edges these things here are partial
correlations in the network so if you see this path it’s left over after
controlling for everything else so these are meaningful associations we also use
some form of regularization which is not very important it’s also very standard
what this does is it avoids false positives because in these networks we
estimate thousands of regressions and you want to make sure that you don’t get
false positives and this takes care of that
again it’s really everybody doesn’t it’s not not fancy and I present to you for
the first time the network structure of depression these are 28 depression
symptoms ignore gray and white now they’re just symptoms again these are
partial correlations so the edge between anxiety and panic for example is what is
left over after controlling for everything else and we see some cool
things for example it’s quite a heterogeneous structure if you have a
common call structure then all the edges are roughly equally thick and equally
connected so this is not a common cause network at all some clusters emerge here
for example we learned that if you know about a person’s weight loss or weight
problems you can infer a lot about the person’s appetite problems but you learn
nothing about the rest of the network that’s really cool actually I think it’s
really cool and now we come to the importance of symptoms which we don’t
get it if we don’t look at networks here is the
importance of a symptom we call it note strength centrality and basically this
is just so for sadness the note string is just all the edges added up together
gives you an idea of how relevant the symptom is and what you see is that some
symptoms are very relevant and other symptoms are not connected to the
network and you don’t want to treat these symptoms and in therapy at least
thinking from a network perspective because they didn’t they don’t do
anything with a network is that clear ok thanks for nothing
another really cool thing we did I’m very happy about this is we compared
these M symptoms from the holy book and other symptoms not part of the holy book
right so the gray notes here are these M symptoms and you can already see that
they’re not more central than the white symptoms and it’s all in one big network
it’s all connected and indeed when we run some permutation tests we find that
these M symptoms are not more central than non DSM symptoms and that’s a huge
problem for the DSM I think ok last study do we have three more minutes ok
so this was data collected in Michigan in Detroit in the 1990s by a really
really cool team of appiause I think you were one of them right yeah what they
did is they enrolled married couples aged 60 or older at baseline and they
followed them over time until one of the partners unfortunately died they waited
six months and then they reinvented the surviving partner they also invited a
control participant whose partner had not died so it’s a really cool
prospective study on in which we have a bereaved group and the control group of
where people were still married so we the only difference here is between the
groups is that that the partner died right we can control for this really
well and in baseline indeed people who are going to be bereaved later and
people who are going to be control participants later differ in no aspects
whatsoever so it’s very nicely controlled and so in this study now we
only look at the follow-up part them time point and what we see and this
really wraps up I think all of the presentation because we do symptom based
analysis and we do networks you see that specific symptoms are increased in
various participants but not all four example sleep problems is a common
depression symptom but the two groups are equally high or low in in sleep
problems but slowly nurses increased sadnesses increase they’re less happy
and less enjoying life and so forth so looking at symptoms specifically tells
us something about the problems these people have this is um it’s a reverse
item so the control participants are more happy here meaning they’re they’re
less yes that’s the C is D scale I’m not sure it’s a weird scale but very briefly
hmm we modeled the common cost model in which case bereavement hits depression
which then hits the symptoms and we also modeled how the data are and this is how
the data are there’s actually no significant association here anymore if
you allow the loss note to hit symptoms directly so that’s what we expect right
mm-hmm and this is the the symptom Network here we actually put loss as a
node into the network so this means you are briefed yes or no and what we see
and again this controls for everything else what you see here is that loss
nearly only impacts on being lonely so this is a gateway symptom and then the
activation spreads through the network and of course every clinician in the
room will say obviously that’s the case but we had no idea or okay maybe not but
it’s for me it’s highly intuitive it’s a bit of an sowhat’s but we had no idea
about this before and we had no means to model this statistically before so this
is really the first time we have this insight and we I don’t know if Gateway
symptom is a good idea but people have picked up on it in the literature and so
if you think about intervention or prevention this is where you want to
work not here okay okay so going back to the four problems we discussed at the
very beginning I think you understand very well now why we have these problems
because we add up all the symptoms together and we keep doing it for some
reason conclusions core assumptions of the common cost model are in my opinion
not even remotely tenable for depression summing up symptoms is highly
problematic and and destroys a lot of Dayna we really want to analyze and keep
I think mental disorders are suspicious specifically interesting for this kind
of symptom based network research because it’s very intuitive and because
we’re facing all these problems and open questions are there’s many open more
many more open questions but for now most relevant I think are there other
properties apart from impairment risk factors may
be biomarkers in which symptoms differ from each other we have just opened the
field there’s much more to do are there other groups of disorders apart from
psychiatric diseases and I don’t know I’m not sure maybe there are known for
which these symptom based analysis and the network approach may provide
important insights from measles certainly it’s not a very interesting
question right because the common cause model holds for measles quite well lastly what do these insights if we buy
into the data what do these findings imply for evolutionary models of
diseases and for depression and schizophrenia for example there’s dozens
of hypotheses and maybe a couple of theories on why evolution left us
vulnerable to depression and schizophrenia schizophrenia has a
prevalence rate of 1% and the population that’s a lot for a serious disease right
so why has evolution left us vulnerable and there’s many competing theories and
ideas about it but none of them have been properly reevaluated in the light
of the evidence I just presented so it would be and a cool project to go
through all these ideas again and and see which ideas make sense in light of
this symptom based evidence I’d like to thank my collaborators there’s many more
that fit on the slide and the people who give me funding thank you for through
this place for hosting me and for being awesome and I’ll be here another week
and I’d be happy to talk if you have questions or ideas or projects thank you
so much thank you very much for the talk lots of
insights here I’m curious about this something you tossed off right at the
very end this notion of the Gateway symptom right the symptom through which
the rest of the network is enacted or activated or whatever could that be
thought of as a common cause of the rest of the symptoms I mean if you model it
through the Gateway symptom rather than through loss for instance or higher yeah
I think that’s a very good point I would agree and I would not say that there are
no common causes an attorney you know but but something that’s I mean that’s a
valid criticism of what we say and I should have put this more carefully we
know that giving people certain drugs will activate five or six depression
symptoms so we can do that and we don’t do that anymore but it’s possible so
there are certainly are common causes and the interesting thing about the
network structure is that it tries to control four symptoms being common
causes of other symptoms but certainly loneliness in this case would be a
common cause however a huge problem in the models we have now is that we
average over all participants of course so these are 270 bereaved participants
and 250 control participants and it may very well be that for some there
certainly heterogeneity in how people respond to to be reeseman so in the best
case of course we worked with everybody individually but if we can’t do that for
reasons of time or whatever this model on average tells us that loneliness may
be the most important symptom to work with and in 10 triggers of the symptoms
yeah yeah this is really cool I do research on prejudice and we could
probably take around half of your slides and change the word depression to
prejudice and thought about the different kind of emotional basis of
different prejudices and to call those symptoms and and we would be making we
could have very much a similar kind of message but at the end you’re still
talking about depression I now talk about prejudices yeah right I
don’t believe there’s any such thing as prejudice anymore and you know you look
look at the network model that you that you have and you’re aggregating and
you’re saying okay well these are symptoms that are unimportant these are
centrally important why why are you thinking about it that way given given
the framework and the different profiles in fact you sort of told us about at the
beginning as opposed to really saying there are 38 major different kinds of
depression they’re qualitatively different they have different kind of
causes they manifest in different ways and require different treatments are the
main reason for me and my sorry the main reason that this stage in the career is
this another another big reason is that we don’t know any specific form of
depressions from the data I have now it’s just a mess and that in this
particular field people have been trying I think for 80 years to come up with
subtypes of depression with they use factor scores and screening instruments
and nothing has ever really stuck so people have done what I kind of tried to
do a little bit but nothing has stuck so I it it feels weird to say that I am
able to do what nobody has done in 80 years and I probably can’t and so what I
try to do here is say don’t study depression it’s not a phenotype study
symptoms instead that’s why we call it some tommix which is probably a bad name
but we haven’t found a better name yet we actually we’ve just framed it four
weeks ago so it’s still as in a commentary we’ll see how it goes but um
yeah I agree yeah sure so we try what we do now is we
sub graph so we have these graphs graph structures and we try to sub graph these
structures into symptom groups that hang more together than other other symptoms
but the problem is that most of these depressed patients or about sixty
percent also have anxiety disorders about 40% have eating disorders it’s
really messy and people need many more symptoms than these networks and we
can’t model them at the moment because even populations of 10,000 are not large
enough to have the power to put a lease in there so yeah we’re making small
progress on the symptom scale at the moment but I agree we should say
depressions you have more technical question on your mathematics you know
could you go back to one of your network pictures there towards the end there
that’s good enough I understand how you produce the edges or the links does the
model also generate the 2d the two-dimensional spread there or is that
what you’re mmm what spread do you mean well the overall picture of the oh the
relation between the nodes so the visualization of the visualization okay
so what we use distance between them and where they’re placed right it for you
so what we used as a standard is the for tamanna’ angled algorithm below we use
as standard for this the fourth among Rheingold algorithm no it was a paper by
two mathematicians in 1991 yeah 1991 and it’s a representation based on the
centrality of the notes so what this does is it puts nodes closer together
which are highly higher correlated and it puts nodes in the center it tries to
put note in the center that have more links to other things but I use a
specific seat here and you can get very similar pictures with other seats so
there’s no I was just curious because though the group of four that seemed the
most highly linked there are anglicize yes so I was curious the centrality of
those is actually not very high because they’re only linked with each other oh
so overall I think it’s there somewhere here in the middle mmm but but there’s
many ways to do that so we just use that algorithm because it’s common in
genetics thank you so you talked about the importance of
loss with your special population of bereaved adults when we just think about
the general population of depressed people which I know your point is that
they’re highly diverse but when we think about the list of like the basic nine
symptoms is there one that you can identify as most important for future
interventions in affecting the network one or two or a couple so the problem
here is it’s a very good question the problem here is that this is
cross-sectional and we don’t know if let’s look at this big link here we
don’t know if panic caused anxiety or if anxiety causes panic or more likely they
cause each other over time because this is cross-sectional data so we have
people colleagues working on on phone data so I give all of you you’re all my
depressed patients I give everybody a phone with an app and the app asks you
five times a day 10 symptoms and it does it for every day for three weeks and
then people throw the phone out of the window but it really happens but until
three so for three weeks we get pretty nice data and in these in these samples
we can infer which symptoms are most important to driving processes over time
so be not costal I talked with Katie about this in detail but we can get a
good idea of which symptoms should be treated and but the problem is that
these phone measurements are really expensive to get few people are willing
to do it so we have sample I think if we combine all data in the world we have
150 people so far so we need much more data to answer your question what we do
know is that the core symptoms of depression sad mood and anhedonia loss
of interest have come up in our analysis as quite important both in terms of they
cause a lot of impairment they’re quite central in the networks interest loss and pleasure loss here
these two and sad mood so they’re pretty high up here in terms of centrality and
we see these this also in a couple of other studies where I have data on so
there’s clearly no distinct difference between the core symptoms of depression
and the others of course there’s nothing distinct in the universe about sadness
but if I had to start some I think sadness and loss of interest are
decent candidates but it’s probably very different for each person I just ask a
quick follow-up to that so don’t isn’t the idea of most
antidepressants that they increase serotonin levels to affect sadness and
isn’t the point that your point that you’re making that they don’t actually
work that well so there’s actually I’m not aware of a single recent
meta-analysis showing that certain levels are actually different in brains
of depressed people compared to healthy people so that is there one no I don’t
know I need so the assumption is a problem in itself and I don’t think
serotonin levels are thought to be mostly related to sadness I think
they’re related to depression that’s the assumption and that depression then in
term causes a couple of symptoms of which sadness is one but I I don’t know
what people what that yeah I don’t know hi awesome talk I was just wondering
what your opinion was on the RDoc that was proposed by the National Institute
of Mental Health when the dsm-5 was released yeah so in 2013 the director of
the NIH Tom Insel said they were so upset with the DSM that they were not
going to follow it anymore because of good reasons I think and they
proposed the r-dog actually a little bit early in 2010 already the problem is
that the 10 2010 10 arlok paper by by insulin colleagues says I quote the RDoc
conceptualizes all mental disorders as brain diseases and I’m not entirely
certain I agree with that certainly everything that is in the mind is in the
brain but I’m not I don’t believe that the common cause for a lot of problems
people have is necessarily in the brain there may be types of schizophrenia but
this is certainly the case or the Tourette’s or OCD or maybe even
depression but I don’t think in general that’s a good idea what what is cool
about the r-dog is that they have a lot of things they look at individually most
of which are very biological and I think a couple of social social environment
things are missing but they’re adding them because people have been upset in
the beginning of our talk about it so I think it’s a cool initiative with
many funding opportunities and I just wish they would be a little bit less
everything is caused by dysfunctions in the brain heavy great talk my question is more from a
evolutionary perspective you know you’re talking about
the anxiety and sadness but have you ever looked at or as somebody looked at
the are simple simple five senses that we have and what I mean by that is watch
the Cubs last night go Cubs you know say a hitter you know doing real well and
starts on a on a slump and gets depressed okay coach comes along and
says have you have you gotten your eyesight checked lately he goes in gets
glasses boom he comes out of the slump he’s not depressed anymore so how how
much impact to our and how we take in information that the brain processes is
that I’m searching there for a good question I don’t know if that’s a good
question but I could come up with a hypothesis one could test in it in a lab
for example that is that if you were census for some reason deteriorate
rather dramatically over a short period of time there may be something in the
brain that tells you to rather not do crazy things at the moment in your life
because the information you receive are not most are not very reliable
if you can’t see very well you may run into predators easier and so forth so
coming down for a couple of weeks sitting in the cave being careful of
what exploration may be a smart strategy is that a little bit where you going
after okay I don’t know anybody who’s looked at that I think it’s a cool idea
but I don’t know any dr. Jonas so to take a look at that list of
symptoms one of the things that that’s been crystal clear for a long time I
think in at least in philosophical accounts of what’s going on in
psychiatry is that each individual symptom is not all that particularly
well understood either clinically or otherwise and so is that something that
you that you think your work is gonna help alleviate or is it a problem that
you still have to grapple with but the the sort of fuzziness if you like or the
lack of serious clinical phenomenology to give us a better a better sense of
what the symptoms actually meant so I think it’s a huge problem it’s very
unresolved and in fact all these symptom measures we use here are extremely
unreliable right these people are asked in one question what’s up with your self
blame or often is just yes and no sometimes it’s one two and three or
something these are we have huge problems with them which with
measurement error here um did you sleep well last night yes or no that’s not a
good way to measure sleep quality so it really isn’t and this is already a
comprehensive questionnaire most questionnaires are much shorter because
of time constraints in psychiatry of course I understand so a next step in
terms of phenomenology of symptoms may be to use standard psychometric
approaches in psychology where we measure everything with at least three
or five questions so measuring psychomotor retardation
with five questions and then putting it into a latent variable getting out the
error may get us a step further in terms of just measurement but of course that
doesn’t tackle the the philosophical notion of what is the symptom and what
is the problem and these all these all have different timescales of course if
you think about the course over time sadness for example may be a much more
flexible thing than psychomotor retardation that lasts for weeks and
weeks and weeks sometimes so we’re really struggling that these are very
different properties lots of measurement error and we hope
that this will incentivize people to start looking at it and start measuring
them properly and there’s a couple of so concentration problems for example
there’s a nice test it takes 20 minutes but we have very good idea of
concentration of impairment of concentration after the 20 minute test
right so what suggest what that suggests is that you know
so clinically these tests are not gonna be all that feasible because they’re so
much more extensive but if we can do the research to sort of figure out which
ones might be more appropriate then that could lead to some clinical innovations
at some point that would have a real benefit for patients
this is totally targeted at researchers now I know all of it I understand that
clinics are hard and we have time constraints and so forth and and that’s
why it was so great to collaborate with Randy who has the the all this
experience in the clinic and he keeps telling me I co we can’t do that in the
clinic it’s not feasible and and so this is totally totally targeted researchers
who should look at symptoms in the clinic it just doesn’t work at the
moment and it doesn’t need to work at them I just have a similar idea I just
wondering so all these studies that they don’t think 3000 people and they
aggregate all the data they still have the wrong number so you can parse out
the individual symptoms because I hate to have all that data go to waste with
this possible analysis that could maybe pull something out of it
right so I’ve been trying to get I think since 2010 data from antidepressant
trials and there’s been a couple of those on individual symptoms and in the
the all these trials have to be registered with the FDA and the results
have to be put online there but they only put aggregate scores on there
there’s no symptom based information I’ve written probably 200 emails in the
last four years I’ve got no data set yet work we have one now because people use
the Freedom of Information Act to just sue companies to give us data but I have
a student quoting the data as we speak but it’s in the in the big
epidemiological data sets we have the individual symptoms all I showed you is
reanalysis of existing data i didn’t collect this data myself it’s actually a
good feel to working for me because there’s so much data on depression
there’s a lot so we can look at that yeah I thank you very much great talk
just have a quick methodological question I was wondering how that
network modeling deals with call in arity because I feel like that
many of those symptoms will be highly correlated which made me think that
maybe that might explain some weird looking patterns in there there’s
sadness really high up as a really important symptom and then the the least
important one is smooth quality right which I would think as almost I should
be highly related but this is a terrible question so here there no no no about
here they asked you get is your depression is your sadness similar to
grief yes or no so it’s really not a question that is support that that make
sense to correlate colinearity is not a problem for these models in fact we want
to see so a highly multi multi colinear symptom would be very central the
problem we do have is to figure out what so if you have a network of computer
hubs or a network of people you don’t have to take any notes out of the
network because Susan and Peter have to be in the network the problem here is
whether two symptoms are distinct constructs or whether they’re
differently phrased things that measure the same latent construct we have no
idea how to solve this giulio Constantini has written his PhD on what
he calls topological overlap and that’s a way to look at this in fact these are
three insomnia symptoms and it’s problematic that they’re all three in
there because they boost each other in centrality but there’s some studies
showing that these are distinct so we left them in there because we have no
good thresholds to take anything out what we do now is or what we will do in
the future is to see if these three symptoms have extremely similar
properties in the network not the correlation but we want to see if this
has the same connections as this as this and if they do we call it they overlap
topologically and then we have a good reason to say we should combine these
into one latent node for example okay but this is unresolved at at the moment
thank you so much

Comments

(1 Comment)

  • Philebos XX

    very interesting, thanks for uploading

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