Return to: 2012 MSM CONSORTIUM MEETING

Post-Meeting Feedback

Please enter any post-meeting feedback (e.g.: meeting venue, format, conflicting dates for next year, etc.) to help us make next year's meeting even better!

• The Open Space/Unconference discussions were great addition to the meeting. If they are maintained for future meetings, is it possible to plan for additional meeting space/rooms for side discussions? The auditorium was great for the main part of the meeting (mics and seating worked great!), when it comes to the open discussions the auditorium seemed a bit restricting. Most people probably like to be able to sit around a table or rearrange chairs to get themselves situated for face to face discussions.

TALKS vs. open space: find a balance between topics that are helpful to Grace for getting funding for these RFAs and the science the PIs are doing....more time AFTER EACH TALK for discussion with PI and also can bin the themes for each year based on PI working areas (lung, heart, cancer, signalling, etc)..

• Other meetings I've attended that have been successful in getting the audience/community more involved in contributing to an online forum have incorporated a way to show live feeds of recent questions or comments posted by users (e.g. video or projector screens that automatically update).

• I'm one of the ones who voted for more science and less wide-open discussion. I know that one issue is that we don't all work in the same area, and that means that many of the technical talks aren't so accessible to all. However, this could be addressed if the speakers included a significant introduction and slowed down a bit. I'd really like to hear why a particular area is important, what the significant challenges are, and how the PIs are thinking about them - plusses and minuses to approaches (e.g. how to link different scales) have broader applicability. (Linderman)

• I agree with the above comments: We're a broad audience so the presentations should begin more high-level, but end with technical summaries of the model and the applications of its calculations/predictions. (What other audience is already so familiar with modeling?). Definitely more time for Q&A after talks. We could also begin each panel with a 15 minute presentation, focused on a particular area/topic, followed by a 30 minute panel Q&A. It would structure the Q&A significantly better. (Salis)

From Jacob Barhak:

The meeting was very interesting for me. It is the first time I am attending as a guest and the atmosphere was very welcoming. The wiki is a great idea. However, if some users work concurrently on the same page, they may override each other. If I understand this correctly, all saves are kept and nothing is erased, yet these will be different histories of the page which are not immediately visible. Therefore when posting questions, it may be helpful to do the following:

• If not already done, assign a moderator to view the page history and make sure no questions are lost.
• As someone mentioned in the meeting, ask people to write the text in a word processor and only when the text is complete open the edit for the wiki page and quickly paste the text and save the page.

Here are a few other things that can be improved for next time:

• It will be helpful if Stephanie will give her wiki use talk before the keynote next time. This will help promote the wiki as a working tool.
• Try to test the wiki and network under load, sometimes getting a wiki page was slow - yet it may have been only my machine.

Overall, I was impressed with the positive dynamics. Thank you for hosting.

Moore: I always enjoy the keynote talks, and the Watson presentation was excellent. I would suggest scheduling a second such talk for the second day. In terms of specific science vs open discussion, I thought it leaned too far toward the latter this year. Then again, I don't need to see the same science talks that I see at other conferences (e.g., BMES immediately followed this year's meeting). I suggest that the science talks focus more on methodology than usual. I would like to know what methods people are using to tackle these challenging problems, as they may help with my own research. Thanks to Stephanie and Grace!

Theme Feedback

We would really like your feedback on the overall theme of Multiscale Modeling for Precision Healthcare

National Academies Report on Precision Medicine: Media:precision medicine.pdf

• Let us know if this theme is relevant and timely for the MSM meeting this October
• Please suggest other meeting discussion ideas
• Let us know your thoughts on the questions below
  • Please suggest grouping, removal or adding new questions

Since none of us were experts in this area, having almost a full day and then a follow up day on a topic that was interesting, but we are not experts on, was too long. I think an afternoon of topics that our outside our areas of expertise would be sufficient. I did not like the wrap up summaries of these the next day. They added nothing new.

More time for science presentations by awardees and grouping them by themes each year would be fun..agreed that they should give talks one level up and better introductions to bring us all in when we do not work in a particular area. Also, more focus on what methods people are using to build MSMs...so we can all learn from each other..that is the common ground.

These questions should focus on multiscale modeling

1. How can basic biology models be useful for clinical decision support?

• Comments:

James Schwaber: Please see updated document: Media:Using_models_for_precision_medicine-24aug12-1.doc

GCavelier: If the clinicians understand the models and believe in their power, they can be very useful to them, as a tool that aids towards a better diagnosis and treatment. Clinicians already use this kind of help with much more raw 'models', with their use of imaging for example. But good models would have an extra advantage in their predictive capability, as well as in the possibility of multiple alternative simulations for different parameters or inputs.
Kirschner/Linderman: One way that our models have been used is to conduct what we term “virtual clinical trials”. A group of individuals can be simulated and the output and statistics from the group can be assessed. This allows for focusing actual clinical parameter trials, i.e. assessing the effectiveness of proposed strategies, suggesting new strategies, and eliminating those that are unlikely to be effective. We are not sure how you are differentiating between basic biology and MSM models, so our answer does not distinguish between the two.

2. How can one take a mechanistic model and fit it to an individual patient?

• Comments:

GCavelier: This would be a problem of 'on-line' estimation of parameters, specially some parameters that change or cannot be obtained otherwise. It is an engineering problem.

3. What is still needed to develop computational models for treatment planning?

• Comments:

Moore: For biomechanical modeling, we need better quality imaging. For example, we have all the tools necessary to estimate stress in aneurysms, but the results are extremely sensitive to vessel wall thickness. Determination of wall thickness to a sufficiently accurate degree is beyond current MRI or CT capabilities. For now, we just have to be honest about this limitation, which severely limits the applicability of modeling.

4. What are the remaining critical, fundamental issues that still need to be addressed to apply these models to individual patients?

• Comments:

Kirshner/Linderman: (addressing points 2 - 4) These questions seem to be focusing on making predictions for individual patients in a clinical setting. At this point, our TB models are not yet able to do that. There are many patient parameters that would be necessary to make accurate predictions for a particular patient. Most of these parameters are not available and would not readily be so. However, ongoing experimental and modeling work in our groups is focused on identifying biomarkers in the blood for disease prediction in the host, e.g. is this individual likely to be able to contain the infection, or will s/he develop active TB? Thus at this scale predictions for individual patients may be effective.

5. How can multiscale physics models help make individual predictions on behavior?

• Comments:

6. How do you combine statistical and process models?

• Comments:

7. How can MSM models help move towards clinical decision support?

• Comments:

8. What is stopping us from applying the MSM models towards POC (Point-of-Care) applications?

• Comments:

Kirschner/Linderman: In our area at least, it is too soon to be thinking about POC application.

9. How can predictive models move forward to be useful in clinical practice?

• Comments:

10. How do we determine which types of models, or combination of models for the intended use of the model, for the specific clinical setting?

• Comments:

11. What’s stopping our models from being predictive now?

• Comments:

Kirschner/Linderman: In our area, we continue to gather additional data and build more biological understanding into our models. Computational resources limit the number of simulations, especially the repeated simulations necessary for uncertainty and sensitivity analyses and thus testing of statistical significance.

12. What is the threshold for clinical credibility (V&V, robustness) for a model, depending on the intended use of the model in a clinical setting?

• Comments:

3. What is the MSM response to the recent report, "Assessing the Reliability of Complex Models: Mathematical and Statistical Foundations of Verification, Validation, and Uncertainty Quantification"?

• Comments:

Kirschner/Linderman: Based on the flurry of emails on a related topic in March 2012, this would be a great topic for discussion at the meeting. Copies of the report might be sent ahead of time to the participants.

14. Can we perform a gap analysis of V&V for MSM models – establishing credibility for models in different settings (e.g. for clinical prediction)?

• Comments:

15. What are the factors for uncertainty that are unique to the clinical setting?

• Comments:

Kirschner/Linderman: Unknown time and source of infection. Uncertainty regarding key parameter values for an individual patient.

16. What, if anything, is unique about MSM in the medical domain?

• Comments:

17. Can we develop general MSM tools that work in multiple domains? Can we use such tools already developed in other domains?

• Comments:

Kirschner/Linderman: Certainly. One can imagine a very detailed model used in research, but also versions of the model with less resolution (more coarse-grained) that might be adapted for setting up clinical trials or suggesting particular drug combinations for an individual patient. Doing this for some MSM models might illuminate more general patterns that could be useful to all.

18. In addition to the MSM models, what other infrastructure is needed (hardware, data, networks, software, services, etc.)? Does all the infrastructure that we need exist? Is it useable in both development and clinical settings? If not, what are the technical, policy, or other issues that prevent it from being usable?

• Comments:

Kirschner/Linderman: As noted above, as MSM models get more biologically accurate, computer infrastructure becomes more of an issue.

Sluka/IU: Issues of model and data provenance, long term code persistence (even after the code or model is updated), code "transparency" (closed source or proprietary code may be problematic) will all have to be addressed. In a clinical setting, the liability issues can be quite onerous. A lawsuit occurring years after a clinical decision is made may require that the code and all data that went into a particular clinical decision is still available. Civil liability may extend to the authors of a particular MSM model or code.