Latest / EHA Unplugged / Accelerating Clinical Trials with AI (with Dr Jan Moritz Middeke)

Transcript

Accelerating Clinical Trials with AI (with Dr Jan Moritz Middeke) · EHA Unplugged

• 0:03
  Greetings and welcome to EHA Unplug, the official podcast
• 0:06
  channel of the European Hematology Association EHA.
• 0:10
  Hello, I am Isabella Rivera, a medical writer for EHA.
• 0:14
  Today we're thrilled to have Doctor Jan Moritz Medeka, a
• 0:18
  distinguished hematologist and AI specialist.
• 0:22
  We're going to be talking today about AI to accelerate clinical
• 0:26
  trials. Welcome, Doctor.
• 0:28
  Medeka, can you please introduce yourself?
• 0:31
  Hi, yeah, thanks for having me. I'm, I'm Moritz Malika from the
• 0:35
  University Hospital in Dresden. I'm working there as a
• 0:38
  hematologist and have a. Research Group.
• 0:41
  On AI in hematology. So we want to understand how you
• 0:47
  can accelerate clinical trials by generating data with AI.
• 0:52
  Can you please explain first what is synthetic data?
• 0:56
  Synthetic data is as the name says.
• 0:59
  Data which is. Made-up and.
• 1:02
  The idea is that we face a. Big problem with clinical trials
• 1:06
  in general, but especially in hematology.
• 1:11
  Where a lot of the diseases are rare diseases for.
• 1:14
  Example acute myeloid leukemia is a rare.
• 1:16
  Disease per seat. And we already know that.
• 1:20
  There are several subgroups with targeted therapies, so the.
• 1:24
  Number of patients we can include in clinical trials are
• 1:28
  even lower and therefore we need to think about new ways how we
• 1:34
  can. Accelerate how we.
• 1:37
  Perform clinical trials in order to get sufficient data to get
• 1:43
  the knowledge about novel. Therapies.
• 1:47
  So what types of synthetic data are you generating that could be
• 1:52
  used for clinical trials? So the idea.
• 1:55
  Is that you have synthetic? Data which is.
• 1:57
  Reassembling the original cohort in a A.
• 2:00
  Most perfect way. But without being able.
• 2:05
  To get back. To the original data.
• 2:09
  So you're generating images. This can be used.
• 2:12
  Used both for. Images.
• 2:14
  But also for tabular. Data and what we did in recent
• 2:17
  study and also other groups have shown is that you can reassemble
• 2:22
  real clinical trial data in terms of baseline data, but also
• 2:27
  in terms of outcome results based on cohorts of of.
• 2:34
  Real. Patients and the synthetic data
• 2:36
  is reassembling all the information about the original
• 2:40
  patient which is like. Age.
• 2:42
  Sex. But also molecular changes and
• 2:46
  then the outcome after a given intervention.
• 2:51
  And how do you generate the synthetic data?
• 2:53
  So for this. We used guns, Generative
• 2:57
  Adversarial Networks, which is a known form of AI.
• 3:04
  Which has been. Used for image generation.
• 3:06
  Quite successful. Already.
• 3:08
  But can also. Be used for tabular data and in
• 3:12
  our specific work we use two types of these, so.
• 3:15
  Nflow and C Top Gun. So these are types of deep
• 3:21
  learning and it's generative modelling.
• 3:23
  You have described 4 scenarios in which synthetic data
• 3:28
  generation can be useful and I would like to walk through this
• 3:33
  for the audience. So first, AI generated data
• 3:40
  could help to prevent privacy concerns about clinical trial
• 3:48
  date data of the patients. Exactly so.
• 3:51
  The generated synthetic data can be shared without any.
• 3:55
  Privacy issues, so we checked that.
• 3:58
  There was no single patient in the synthetic cohort.
• 4:02
  Which was exactly. Matching the original patient.
• 4:06
  So there's no way that you can go back to the original patient
• 4:10
  which. Of course is the case if you.
• 4:13
  Just have an anonymized data, so this makes.
• 4:17
  It really easy. You can just upload these data
• 4:20
  and make them available. Also for other research.
• 4:22
  Group and see public. So how does this compare to the
• 4:27
  de identification or anonymisation of patient data?
• 4:31
  So anonymisation of these data isn't 100% sure, and with the
• 4:37
  synthetic data it's 100% sure. So there's no way that you can
• 4:41
  go back to the original patient. Because see, there's.
• 4:44
  No original data coming from one single patient.
• 4:50
  So it's basically 100% sure that you can't go back to the.
• 4:56
  Characteristics of. One single patient.
• 4:59
  So this is a big step forward for this kind of data.
• 5:02
  Also, you would create new patients and this would lead to
• 5:07
  either cohort augmentation, so some synthetic patients among
• 5:12
  patients that are real or completely a substitution and
• 5:17
  you have a complete new cohort. How robust is this process?
• 5:25
  So for the. Generation it's.
• 5:28
  Very robust. We showed and also other groups.
• 5:30
  Matteo de la Porta did this for MD's.
• 5:33
  Showed that it's. Reassembling the whole.
• 5:37
  Crestrix of the. Patients for a given disease.
• 5:40
  So it's not only age and sex in terms of the distribution, but
• 5:45
  it also comes down to the molecular changes, which is very
• 5:49
  important because we know that there are certain relationships
• 5:52
  between certain mutations. So for example, if you have an
• 5:55
  NPM 1 mutation, you're more likely to also have inflat 3
• 5:59
  mutation. And these correlations are
• 6:02
  preserved in the synthetic. Data set.
• 6:05
  And this. Holds also true.
• 6:06
  For correlation between age and certain cytogenetic
• 6:10
  abnormalities like complex. Karyotypes with with.
• 6:13
  Higher age and so on so. Biologically.
• 6:16
  Important correlations are. Preserved.
• 6:19
  Within this data and even more important, this also.
• 6:22
  Holds true for the. Correlation to the outcome data,
• 6:26
  so known factors for. Poor or good?
• 6:29
  Prognosis are also preserved in the synthetic.
• 6:33
  Data so if. You have a patient with some
• 6:37
  complex karyotype. He.
• 6:39
  Also has a worse prognosis in the.
• 6:41
  Synthetic group of patients, and this is super important because
• 6:46
  if we want to use him as a real control group, we have to make
• 6:51
  sure that. This is also.
• 6:53
  Like in the. Real world.
• 6:56
  So there's also A use for exploratory analysis.
• 7:00
  So synthetically enlarged data sets, what is the interest?
• 7:04
  Of this, we know that there are several important subgroups when
• 7:09
  we stay in the case of acute myeloid leukemia, certain
• 7:12
  molecular subgroups which are recurrent.
• 7:14
  But are rare. And.
• 7:16
  So for these group. Of patients, it's of course very
• 7:20
  important to know. How they fare with a certain.
• 7:23
  Therapy and. With a synthetic.
• 7:28
  Generation of data you can. Substitute or.
• 7:31
  Enlarge. These rare subgroups.
• 7:36
  How reliable is this process? We have all heard about
• 7:40
  hallucinations in AI. Can this generative AI generate
• 7:48
  misleading data? Well.
• 7:51
  Of course, you always have to show it.
• 7:53
  For a given group, but for the examples I mentioned.
• 7:57
  This is very robust, so. There are some small
• 8:02
  differences, but in general. This is.
• 8:05
  Very robust in terms of how the baseline distribution of.
• 8:11
  Characteristics is but also. For the outcome data, however,
• 8:16
  especially when the outcome events get lower in the training
• 8:21
  group. This leads to a lower.
• 8:24
  Robustness of the model. So there is still open question
• 8:28
  where we have to work on. And the last case you mentioned
• 8:33
  where this kind of synthetic data can be used is in model
• 8:36
  training or benchmarking. Can you explain this further?
• 8:40
  So the. Idea is that.
• 8:42
  We also have to generate. Data for rare subgroups.
• 8:47
  And for that we can use synthetic data to enlarge the
• 8:50
  group of. Patients, basically.
• 8:53
  For in clinical trials this generated patients are called
• 8:59
  now synthetic patients. Can you summarize what the
• 9:02
  advantages to use this kind of synthetic patients?
• 9:06
  We have to point out that. So far synthetic.
• 9:10
  Patients have not been used within clinical trials and.
• 9:13
  I believe that there's still a lot of work to do, also from a
• 9:16
  regulatory. Aspect to.
• 9:19
  Show that this is really feasible.
• 9:21
  On the other hand, we know that the generation works and the.
• 9:25
  Problem is so big that. We have to think of new ways how
• 9:29
  we can accelerate clinical testing.
• 9:32
  So for this example we used. Only intensively treated acute
• 9:37
  myeloid leukemia patients almost with.
• 9:41
  Our targeted therapy, so this has to.
• 9:43
  Be shown with. Targeted therapies this has.
• 9:45
  To be shown with. Real life data.
• 9:48
  And with certain intervention and there's a lot of.
• 9:51
  Work to do. Before we can really.
• 9:53
  Say OK. So we one group of patients
• 9:57
  within a clinical trial for in the end leading to an approval
• 10:02
  or identification of certain efficacy of a of a given drug.
• 10:08
  So really is work in progress, how closely does these cohorts
• 10:13
  match real world cohorts? OK.
• 10:17
  That's a difficult question because.
• 10:19
  It hasn't been shown yet. So for the moment you don't know
• 10:24
  really how well and it's being tested.
• 10:26
  We know how well it matches. For.
• 10:29
  A given group of patients we studied.
• 10:32
  Or other groups. Studied and there it is working
• 10:34
  well. However, we have to.
• 10:36
  Yeah. Prove this for real world
• 10:39
  cohorts. And and certain interventions.
• 10:43
  So the idea would be to use these cohorts mostly in the
• 10:47
  control group in a clinical trial at the beginning at least.
• 10:52
  Yes, I, I. I think this is the most
• 10:54
  realistic and and closest scenario that we use this as a
• 10:57
  control group, an enlarged control group.
• 11:01
  Yes. With a certain intervention.
• 11:04
  Which would be interesting because then you don't have to
• 11:07
  assign patients to a control group.
• 11:10
  So you mentioned the regulatory considerations.
• 11:13
  We have some tools at the moment, like GDPR or HIPAA.
• 11:18
  Are they enough to address the potential issues that this kind
• 11:22
  of synthetic patients would bring?
• 11:30
  Well, I think this. Requires.
• 11:33
  Complete new thinking about the approval process and the.
• 11:37
  The way we look at. Clinical trials.
• 11:40
  We need a. Way to prove that this is.
• 11:43
  Really working and and I envision this in the 1st.
• 11:46
  Place to go. Alongside clinical.
• 11:48
  Trials and this is also. What we are currently doing, so
• 11:51
  we're going back to clinical. Trials we already.
• 11:54
  Did and we are going to see how many patients we can substitute
• 12:01
  by synthetic patients. To get the results, we.
• 12:06
  Already know, so for me this is an important step.
• 12:10
  To show how. We can incorporate A synthetic
• 12:15
  patients in real clinical trials and I think this step has to be
• 12:20
  done in close contact. With the authorities.
• 12:24
  That we together define criteria.
• 12:28
  Where we say OK this is. Working and we can move on
• 12:33
  further. So if you had a crystal ball and
• 12:40
  could see what the future holds in five years time, do you think
• 12:44
  it's going to be used a lot? Yeah, what 5?
• 12:46
  Years is close in in clinical testing, but I I know.
• 12:51
  That the way we. Did clinical trials in rare
• 12:56
  disease like acute myeloid leukemia in the.
• 12:58
  Past is over. So there won't be clinical.
• 13:01
  Trials with like. 1000 patient randomized with a certain drug
• 13:07
  because in the meanwhile other drugs are coming up we we gain
• 13:12
  further knowledge. So this is getting harder and
• 13:16
  harder to perform these large trials.
• 13:18
  So we have to. Adopt new methods and I think
• 13:21
  synthetic data is 1 very promising way.
• 13:25
  But there are several steps. Before we can really say, OK,
• 13:28
  that this is working and enhancing our clinical testing.
• 13:32
  Are there other uses of this data that you are now developing
• 13:39
  for prognostic capabilities in hematological malignancies?
• 13:47
  So coming back to images, there's another interesting
• 13:53
  field where we can enhance the amount of images.
• 13:57
  For rare subgroups. Of patients, so basically doing
• 14:02
  the same thing for small subgroups within clinical.
• 14:05
  Trials. We can also do this for.
• 14:07
  Images of of bone marrow smears, for example.
• 14:11
  This is already, it's been used. Well, I wouldn't say it's been
• 14:15
  used, but it's like we and others have shown that this is
• 14:19
  working and that you can generate synthetic images based
• 14:25
  on real bone marrow smears and that you can't really tell which
• 14:29
  is a synthetic or. A real bone marrow smear
• 14:33
  picture. And these have been used to
• 14:36
  train. Well, it hasn't been used, but
• 14:40
  that's. The idea.
• 14:40
  So the idea. Is that you can generate more
• 14:43
  data and then increase model development.
• 14:49
  So in the future you think that you will go beyond the control
• 14:53
  group, so you would use the synthetic data also in the test
• 14:58
  group. The improvement of the
• 15:01
  techniques is so fast that we can really, like simulate what
• 15:06
  is going on within a single cancer.
• 15:09
  Cell with a. Certain intervention and with
• 15:13
  that I believe that in I don't know how many years, but in the
• 15:18
  future. We will be able to.
• 15:19
  Also simulate what will happen to a certain group of patients
• 15:24
  when we. Give the drug X.
• 15:28
  Yes, I'm I'm a biologist by training and I find difficult to
• 15:34
  imagine that you will be able to reproduce this complex system in
• 15:40
  its totality, but I look forward to it.
• 15:43
  It would be fantastic. Yeah.
• 15:46
  Agree on that but like. Five years back, you probably
• 15:51
  would. Have said that in predicting
• 15:52
  protein structure which is now working almost.
• 15:56
  Perfectly. So yeah.
• 16:01
  I think it's really promising and I really hope this works
• 16:04
  because it would help enormously in accelerating clinical trials
• 16:10
  and treating the other patients. So thank you very much, Doctor
• 16:15
  Mideke, for sharing your insights and your experience
• 16:18
  with us. Thank you to everybody for
• 16:22
  listening. And if you like this episode,
• 16:25
  don't forget to like it and share it with your colleagues.
• 16:28
  And stay tuned for our next episode.
• 16:31
  Thank you. Thank you.