Return to 2012 MSM CONSORTIUM MEETING
Panel Discussion Agenda:
- Panel questions (discussions will include solicitation of input from the audience):
- What areas of clinical practice are most/least likely to be amenable to precision medicine via modeling? Note that this question is not meant to rule in/out any disease entities, but rather to discuss opinions of the precision medicine landscape.
- What hurdles need to be overcome before modeling can broadly impact clinical practice via precision medicine?
- Once modeling is used for a precision-medicine application, how can its impact be tested?
- Are there inherent advantages/limitations of data-driven versus mechanistic/physiological modeling in the context of precision medicine?
Note: the agenda may change if the Panel 1 and 2 discussions earlier in the day sufficiently cover any of the above and/or stimulate ideas for more fruitful discussion points.
Panel Member Discussion
A new taxonomy could be of interest if the direct connection between clinical findings and model characteristics/space parameter is improved. Some thoughts on the implication of clinical data on modeling for precision healthcare are below (points to be elaborated during the discussion).
- a. Decide which of the relevant clinical findings are descriptive of the condition we are trying to model. Example: HEMI effects. Superficial burns resulting from currents are not likely to be the reason behind the behavior of the human and a model that predicts that in association to HEMI models fails to capture the true reason behind the effect itself, albeit being correct.
- b. Establish connections between clinical findings and physiological/structural models. Often we have one or the other, with limited availability of both. Which easily attainable clinical data has a direct impact on what part of the model is often not easy to determine.
- c. Establish arrays of clinical data that are objectively useful for the development of modeling strategies and models. What parameters appear to have the most significant impact on clinical findings? Correlation between parameter variations and clinical measurable is a challenge. Often, there is not a one to one variation of findings per individual parameter.
- d. Investigate clinical findings that do not have a clear explanation through modeling using available structural models.
- e. “Explain the unexplainable”. Different clinical results are not predicted by modeling. How not immediately related responses affect the primary response we are looking for? (e.g., panic – it may affect readings but how will our models account for that? e.g., “spinal reflex” role in electromuscular incapacitation: investigate the impact on “peripheral” clinical findings on localized/specific models of actuation).
- f. Multiple existing conditions: differentiating the clinical findings and provide a unique association to the modeling of the process of interest is a significant challenge.
- g. “Subjective” clinical finding in terms of patient feedback should be included in modeling. For example, in an artificial retina to restore vision to the blind, patient feedback is what their percept looks like (color, size, position of the stimulation).
- h. Known conditions that seek a very specific answer on few parameters to control and require very specific clinical information (e.g, fields and gradients magnitude for neurostimulation) have advantages in precision healthcare modeling compared to modeling for diagnostics or treatement.
A) I am strongly against the new taxonomy and nomenclature that comes with it that is proposed in the report. I think there is enough jargon and partitioning in science between disciplines, and I don't think adding to it will help our cause.
B) everyone should read the book: The creative destruction of medicine by Eric Topol CEO of Scripps...
POINT: the digital revolution can spur unprecedented advances in the medical sciences
1) a technology-enhanced future where new tools are integrated into diagnosing and treating patients, transforming the handling of common medical problems
2) healthcare reform as discussed in government is not about delivery of care, but payment of care (BTW--real revolution would reduce cost automatically)
3) "no single innovation will have a more profound effect than the conversion of biological data. With the aid of technology medical progress may well begin to resemble modern computers' own astonishing surge in processing power and data storage.
4)in the book he continuously checks his blood sugar with an implantable meter, he goes to bed wearing a "Zeo clock" that monitors brain function to help analyze sleep patterns. When he tries to fake sleep so that he can disregard his wife's bedtime chatter, he learns that "it's hard to play possum with a sensor displaying your real-time brain waves.
5)focuses much of his attention on the development of "theranostics," or the integrated use of treatments and diagnostics (especially genomic and protein information) to better guide therapy. These tools, he says, will enable treatment systems that combine the constant monitoring of a patient's biological information and the infusing of targeted medicines.
6)The FDA, he says, should allow the testing of drugs on patients who are selected for their prospect of deriving a benefit. Right now, the FDA usually requires drugs to be tested in a scattershot fashion on large populations. With drugs being tested on cancer patients, he notes, the "FDA insists on a body count to be able to quantify how much and how long the new drug improves survival"—even though diagnostic markers can sometimes reveal in advance which patients are unlikely to gain a benefit.
7)Dr. Topol worries that doctors will resist technologies that empower patients because the tools will also diminish the doctors' gatekeeper role. The American Medical Association, for example, battled firms that provide genetic information directly to patients. "This arrangement ultimately appears untenable," the author writes, "and eventually there will need to be full democratization of DNA for medicine to be transformed."
8) innovation that enables real-time diagnosis and personalized treatments is a certainty, though not because reluctant doctors accept it or because Washington wills it into being. A seductive technology that works like a dream and improves lives will set off a consumer clamor, whether the new tool is an iPhone 4S or an implantable blood-sugar meter.
C) Final topic, Machine learning. Basically best we will do is blood data. If we create large large large interconnected databases with blood data from patients yearly (or more often) for their life with 1000s of analytes and track the changes in time and then the patients outcomes (disease)..take 5-10 years to create biomarkers for every known disease and can then track and predict future events based on changes in these analytes..(stats and modeling obviously important)
Audience Questions and Comments
Can we learn from an example from a high-technology that has had a big influence and big success -- imaging with CT and MRI. These modalities give a clear literal picture of the patient that is represents a transform (a simple model) of direct data measurements.
Raj Vadigepalli: Eric Green, Director of NHGRI wrote a perspective piece on Genomic Medicine last year. Green ED, Guyer MS, National Human Genome Research Institute. 2011. Charting a course for genomic medicine from base pairs to bedside. Nature 470: 204–213. http://www.nature.com/nature/journal/v470/n7333/full/nature09764.html
Figure 2 of this paper talks about “game changing blue dot accomplishments” http://www.nature.com/nature/journal/v470/n7333/fig_tab/nature09764_F2.html
What are the “blue dots” in the Multi-scale Modeling in the context of Precision Medicine? Is that something we as a group should develop and work towards?
For example, Pathologists have initiated such projects to examine the issues involved directly. Tonellato PJ, Crawford JM, Boguski MS, Saffitz JE. 2011. A National Agenda for the Future of Pathology in Personalized Medicine: Report of the Proceedings of a Meeting at the Banbury Conference Center on Genome-Era Pathology, Precision Diagnostics, and Preemptive Care: A Stakeholder Summit. Am J Clin Pathol 135: 668–672.
Question from Louis Gross: What evidence is there that we can find biomarkers that will be effective in assisting medicine given the great between and within-individual variability in genetics and exposure and the multi-factor nature of many complex diseases? What is your opinion of a focus of the NRC report on linking information about individuals that can allow us to potentially look at the distribution of individual responses?
Raj Vadigepalli: The definition of “clinical data” could change…. Consider the following study from Stanford.
Chen R, Mias GI, Li-Pook-Than J, Jiang L, Lam HYK, Chen R, Miriami E, Karczewski KJ, Hariharan M, Dewey FE, et al. 2012. Personal omics profiling reveals dynamic molecular and medical phenotypes. Cell 148: 1293–1307. http://www.sciencedirect.com/science/article/pii/S0092867412001663
There is data on genomic sequence, transcriptomics, miRNA, proteomics, cytokines, autoantibodies, metabolomics, physiological/clinical parameters, etc etc data from one individual at several time points over a period of 14 months.
Is this the kind of data that would be useful for a Blue Dot MSM project?
Bill Lytton: A traditional clinical way to formulate some of these issues is the question of lumpers vs splitters. We are moving towards greater and greater splitting: EBM->Precision->Personalized. This typical medical description may help us to reach out to clinicians. Another typical teaching point is 'horses & zebras' -- precision med is helping us find the zebras -- here a correlation with suggestions of focus on the rare disease mentioned earlier.
ErdemirA 14:54, 22 October 2012 (EDT) Clinical data are available, possibly depending on the disease and discipline. One good example is Osteoarthritis Initiative which can be access through http://oai.epi-ucsf.org/datarelease/
Roger Mark: PhysioNet contains large quantities of physiologic and clinical data. A particularly valuable resource for certain modelers is MIMIC II, a highly granular critical care database containing comprehensive clinical data for approximately 40,000 ICU patients. See http://physionet.org/mimic2
Scott Berceli, Univ of Florida: Treatment of abdominal aortic aneurysm has all the elements needed for successful application of multiscale modeling techniques. It is a disease process in which only 5% of the patients are at risk of rupturing, yet we surgically treat all patients because of the severe consequences of rupture. A biology/mechanics based model that provides an improved prediction of rupture risk would be readily accepted by clinicians, who currently use computer-based reconstructions for surgical planning. The major hurdle in making this a reality is the validation of the model that would be required to make it widely acceptable for clinical use. A decade long, multi-site trial would be required.
- Panel started with brief introductory comments from panel members
- David Christini introduced the panel and discussed how we would have a largely free-flowing discussion because many of the main themes we might talk about were discussed by panels 1 and 2.
- Gianluca Lazzi: discussed his points detailed above in Panel Member Discussion
- Denise Kirschner: discussed her points detailed above in Panel Member Discussion
- Terry Sanger: discussed difficulties inherent to getting modeling to impact clinical medicine. He then asked for a show of hands of how many clinicians were in the room (about 5), noting that the result is somewhat indicative of the disconnect between clinical practitioners and modelers.
- The discussion focused for a considerable time on how to get physicians to buy in to models.
- Martin Kohn noted that physicians have heard over and over again how the latest new technology would solve all of their problems, only to see it come up short. Thus, we have to recover from a poisoned well. That being said, he believes that the younger generation of physicians is eager for new technologies and would be happy to work with modelers.
- A common concern is the lack of access to data. While some of it is a result of HIPAA, much of it is due to a perceived lack of interest on the part of physicians working with modelers (which, of course, when it exists is probably due to the problem described by Martin)
- The need for data brought attention to the need to collaborate with physicians. In addition to leading to data access, Denise noted the importance of modelers not working in a vacuum; i.e., collaboration helps to keep modeling grounded in clinical reality.
- We as a community can also push for investment in data repository resources containing large amounts of de-identified patient data. If pooled from multiple sites, it should be possible to maintain patient privacy.
- The audience was asked to discuss “What areas of clinical practice are most/least likely to be amenable to precision medicine via modeling?”
- I (David) was struck by how the three applications discussed (orthopedic implants, cardiac valves, vascular surgery) are all physical problems for which the personalization is largely based on imaging. There was little to no discussion of cancer, metabolic disorders, etc., for which I believe the term “precision medicine” is really envisioned.
- Several other questions were discussed, as indicated above in the “Audience Questions and Comments” section.
- Summary / Path Forward:
- If you can't make tools easily usable by clinicians, they won't get used.
- Although data may not be available for some of the predictions/tools currently, concerted efforts (collaborations and/or pushing for de-identified data repository resources) can help that.
- MSM community still needs to identify distinct diseases that models can impact as a starting point.
- There was near unanimity in the audience that forging better/closer working relationships with clincians would be mutually beneficial in getting MSM to impact clinical practice.
- The discussion is particularly relevant to the Data Sharing Working Group and the Clinical and Translational Issues Working Group.