MIDAS Steering Committee Meeting Minutes

January 17, 2006


Bryan Grenfell (chair)

Steering Committee Members:
Don Burke,Robin Bush, Stephen Eubank, Paul Glezen, Marc Lipsitch, Ira Longini, Ellis Mckenzie, Farzad Mostashari, Alan Perelson Richard Platt, Gary Smith, Diane Wagener, Dyann Wirth

Jim Anderson, Jeremy Berg, Phil Cooley, Derek Cummings, Jamie Cuticchia, Steven Frank,George Ghneim, Peter Highnam, Ken Kleinman, Rob Pennington, Jennifer Villani, Bill Wheaton


Approved future Steering Committee meeting dates
Approved revised Vision and Mission Statement (attached)
Approved Data Sharing Plan (attached)

Action Items

Send out email asking what each group is doing in model simplification and validation.
Seek approval of the new mission and vision statements from NAGMS Council

Data Committee
Don Burke, Derek Cummings, Diane Wagener, Marc Lipsitch, Ira Longini, Farzad Mostashari, Robin Bush
Set priorities for influenza data collection, both historical and current interpandemic

Non Human Influenza CommitteeDon Burke, George Ghneim,Gary Smith
Identify experts for April 5 consultation

MIDAS Priorities for 2006

  • A major goal should be model simplification and validation.
  • Continue influenza projects and look for connections among projects
  • Seek additional data on influenza outbreaks, possibly focusing on 1957-58.
    • First, collect a subset of data to see what is available
    • Second, define data collection project to answer specific questions
  • Look at the question of antibiotic resistance
    • This is particularly interesting for influenza because we can see viral evolution and drug use at the same time
  • Consider the issue of immunogenic sin - consequences of vaccination with a poor vaccine
  • Explore a collaboration with the NIAID Centers for Modeling Immunity

Introduction of New Research Groups

Robin Bush works with Steve Frank (UCI) and Nancy Cox (CDC) and proposes to study 1) pathogen adaptation to new hosts using structural homology modeling, 2) vaccination against an evolving pathogen, and 3) interference competition between co-circulating pathogen strains. The group brings expertise in phylogenetics of influenza A and B, mathematical modeling, molecular modeling, and the natural history of influenza.

Marc Lipsitch (HSPH) works with James Robins (HSPH), Carl Bergstrom (UW), Jacco Wallinga (), and Gabriel Leung (Hong Kong). His recent work is on estimating R0 for SARS and for the 1918 flu. His goals are to study ways to estimate R in real or near-real time, to develop algorithms to detect epidemics, to study pre-emergence evolution, and to develop hidden Markov models for evaluating interventions.

Richard Platt (Harvard Medical School and Harvard Pilgrim Health Care) collaborates with Martin Kulldorff (HMS). Others in his group include Ken Kleinman and Katherine Yih (HMS), Deborah Yokoe, Susan Huang, Thomas O'Brien, John Stelling (Brigham and Women's Hospital), and Louise Ryan (HSPH). This group has agreements for data collection with the Massachusetts Department of Public Health, Kaiser Permanente of Northern California, and the National Institute for Infectious Diseases in Argentina. Dr. Platt and his colleagues will study and develop models of syndromic surveillance and the emergence of antibiotic resistance.

Gary Smith (Bolger Center of Penn State University) works eith Helen Aceto and Chris Rorres (Bolger Center) and with Matt Keeling (Warwick). He hopes to build a hierarchical system of models to explore the dynamics and control of infectious diseases in agriculture. He is also interested in how to build the most simple model to answer a question.


Informatics Group - RTI (Diane Wagener)
The IG's goal is to support all of the research groups. Based on their needs and recommendations of the steering committee, the IG has doubled the original capacity of compute notes to 63 and provided for 4 terabytes of storage. The IG has contracted with the National Center for Supercomputing Applications (NCSA) for compute support especially in an urgent situation. RTI and NCSA have ported and tested models. Additional compute resources may come from the Texas Advanced Computing Center (TACC) and the TeraGrid. Recently, NSF has provided MIDAS 30,000 compute hours, primarily for the VBI group to develop and test the architecture for its models.

The model repository holds models and model results. The Emory SE Asia model is in, and EpiSims is nearly done. The Imperial / JHU model has been submitted. The IG has collected considerable data, which is available to MIDAS through the portal. For the U.S. models, the IG created a synthetic population in synthetic households, completely indistinguishable from the U.S. Census. This, in itself, is a very useful resource. A new source of data will be the New York City annual survey, which comes with local population characteristics. MIDAS added some specific questions to the survey.

The IG has also explored analytic methods for comparing model behaviors. Working with Josh Epstein, the IG developed analytic techniques, based in AnyLogic, for stepdown capabilities which incorporate multiple levels of granularity.

The group has collected and analyzed pandemic influenza response plans from all states and finds substantial variation, especially with regard to social distancing methods. The plans are available on the portal.

The portal itself is being redesigned, and the IG staff have surveyed MIDAS participants to see how the portal can become more useful. The portal will continue to have a public and private section, with the private section housing access to the cluster, MREP, the metadata server, and documentation.

Johns Hopkins Bloomberg School of Public Health (Don Burke)
Don has developed a formal collaboration with the Thai Bureau of Epidemiology. The most recent studies have focused on three questions: Can we contain a local epidemic? If not, can we slow the spread between countries? If not, can we slow the spread within a country?

Viable policies may include stamping out outbreaks, treating cases, household prophylaxis, school and work closure, school prophylaxis, and travel restrictions.

The results show that travel restrictions may slow, but will not prevent, introduction. Use of geographically targeted prophylaxis may slow an outbreak but will soon exhaust supplies of antivirals. Combinations of strategies may yield a modest reduction in attack rate and slow the peak. Like the SE Asia model, this model underscores the need for very rapid response.

The model results are particularly sensitive to several parameters including R0, the proportion of transmission occurring in different contexts, the proportion of people who are symptomatic and report to healthcare, the behavior of symptomatics, the natural history of infection, and details of population structure.

Fred Hutchinson Cancer Research Center (Ira Longini)
Working with Catherine Macken and Tim Germann at Los Alamos National Laboratory, Ira has modeled large-scale transmission of pandemic influenza, focusing on estimating antiviral efficacy. His model suggests that an outbreak with an R0 of 1.9 or larger would require a minimum of 162 million courses of Tamiflu. For an R0 of less than 1.6, vaccination with a low efficacy vaccine combined with school closure could be effective. Between 1.6 and 1.9, a combination of methods is required.

Results also suggest that if vaccine stocks are limited, , vaccination of school children is better than random vaccination. Prevaccination of the critical workforce and use of antivirals is critical. Rapid use of targeted antiviral prophylaxis may preserve a limited stockpile of drugs.

Virginia Bioinformatics Institute (Stephen Eubank)
Stephen has focused on understanding the aspects of social network structure that determine dynamics and learning what is required to estimate those structures in real networks. He is also interested in improving understanding the stochastic properties of small networks and how to support state and local planning.

The model assumes that the reproductive value of an introduction is distributed rather than being a fixed quantity, and the goal is to reduce the number of infections that a single person causes.

If models are to support local planning efforts, situations should be represnted as naturally as possible, all assumptions should be explicit, the specifics of local situation must be taken into account, and the impact of decisions on society should be modeled. The model shows how various interventions can affect the dynamics of a social network and lower R0.

Resource Sharing

Data collected by MIDAS comes in several forms. Some is restricted and cannot be released, and some is widely available. The steering committee discussed how data resources should be made available outside of the MIDAS Network. The committee concluded that
  • MIDAS should make data as widely available as possible.
  • Within MIDAS, no data sets should be held privately. All MIDAS investigators should have access to the same data.
  • Once a publication has appeared, the data should be placed in the public domain.
  • Data that were restricted when we received them should be identified. MIDAS will not place any additional restrictions on the data.
Models can be shared in several ways. Open access provides for sharing code but does not allow others to profit from the use of that code. Alternatives include providing source code with no restrictions or making only executables available. It is not likely that undocumented code would be useful; however, the models are a resource to a much broader community of scientists and their availability can help push the field forward. A significant stumbling block in systems biology has been that people are continually reinventing solutions and models. The same thing could happen to ID modeling if we keep information and models private.

The recommendation of the committee was to work toward the highest possible level of sharing both within and outside of the Network. The committee also agreed that MIDAS should begin work on a 5-year plan that would address development of a product, among other things.

Documentation of Code
Documenting code is a lengthy and expensive task, and the steering committee agreed that the significant investment was not among the highest priorities. It is better to create a useful library of information, including information on model characteristic, assumptions, and properties.

Mission and Vision

The original MIDAS mission was focused on developing models to be used to respond to emergencies. Our experience is that the best use of models is in planning and preparedness. The new mission and vision clarify this, and will be taken to the National Advisory General Medical Sciences Council for approval.

Take a 5 year view of what we want for the field. It's a development issue.

The steering committee approved the new mission and vision with the change of "collection of data" to "acquisition of data."