MIDAS Steering Committee Meeting Minutes

May 1, 2006
Hyatt Regency, Reston

Present

Steering Committee Members
Don Burke, Robin Bush, Irene Eckstrand, Stephen Eubank, Skip Garner, Paul Glezen, Brian Grenfell, James Leduc, Tom Marr, Ellis McKenzie, Gary Smith, Diane Wagener

Guests
Jeremy Berg, Giorgi Bobashev, Phil Cooley, Leland Ellis, Josh Epstein, Moshe Feder, Martin Kulldorff, Theresa Lustig, Christina Mills, Steve Naron, Doug Roberts, Jen Villani, Bill Wheaton

Action Items

Executive Committee
Report to Steering Committee in January on implementation of Big Iron report, including specific projects, collaborations, and distributed vs centralized computing.

Informatics Group Site Visit Report

The IG has made excellent progress since the beginning of MIDAS, especially in providing compute resources to the research groups. It is important for MIDAS groups to recognize that the IG responds to many different needs and requirements. This suggests that it is important to set clear priorities in collaboration with the research groups.

The Big Iron Subcommittee recommended that the IG work on a number of clearly defined projects focused on deliverables. Such projects could be collaborative with one or more research groups. An example of a productive collaboration is the global model, championed by Josh Epstein, which may be able to produce results very quickly.

Balance of Science and Support

The steering committee pointed out that there is a need for RTI staff to master running the models and analyzing output. They endorsed (1) mini-sabbaticals of RTI staff in research groups and (2) hiring and supervising postdoctoral fellows jointly.

Service

Each research group has a specific liaison in the informatics group, and these have had variable success in addressing the needs of the researchers. It is important for the PIs to report back to the informatics group if there are liaison problems, and the informatics group need to address these problems quickly.

The steering committee was supportive of having a help desk on the MIDAS portal. The IG should look at the problems and inquiries and decide what general help should be available through the portal.

Both Gary Smith and Josh Epstein have had good collaborations with the IG staff. Gary is working with the IG to create a synthetic poultry map along the lines of a synthetic human population. Josh has collaborated extensively on the global model.

Distributed vs centralized resources

Changes in technology suggest that a combination of distributed and centralized resources may be most efficient. A trusted central resource provides for management of information and backup. Distributed systems are probably better for development purposes.

The steering committee asked the executive committee to report back on these points in January.

Global Model

Josh Epstein
The global model has three versions - stochastic equation based (posted for comment), individual based global (not parallelized), and pure JAVA agent-based (parallelized). The global model is an SEIR coupled patch model that includes travel restrictions, quarantine, vaccination, and combinations. It is updated for the 2004 ppn and global transportation and can be rapidly parameterized. The aim is to submit the model for peer review in September, showing the dynamics of outbreaks across cities, passage time to U.S., and implications of travel restrictions.

It is important to consider at what point a simulation can be turned over to an equation based model. There is considerable interest in this within the infectious disease modeling community and also beyond.

Dr. Leduc pointed out that this model may be useful for economic modeling as well.

There was considerable discussion about how MIDAS could use the global model. All agreed that it needs to be peer reviewed and published, and it may be a good idea to put canned runs with limited knob-turning capability on the home page of the MIDAS portal. This would also provide an opportunity to educate people about what models can and cannot do.

Should models like this one be available to policymakers? Who is a policymaker for this purpose? These questions were raised but not answered and will require further discussion.

Combined Scenarios

Ira Longini, Don Burke, Stephen Eubank
Thanks to Richard Hatchett, this has been an iterative process between science and policy. The runs were designed to test how well combinations of strategies would work to slow transmission until well-matched vaccine is available. The hope is to define strategies that will spread out the cases over a longer period of time, and hopefully reduce the total number of cases.

There was discussion about the meaning of generic social distancing. Generic social distancing is a critical element of success in the scenarios. What are specific activities or contacts that would be affected?

Dr. Burke noted that estimates of reproductive number of past epidemics included some social distancing. It is unclear if additional social distancing would be effective or possible.

It may be more important to focus on relative magnitudes of various interventions instead of or in addition to calculation about amounts of drugs or vaccines needed. This is because there is some variation among replicates.

Results show that

Households contribute greatly to transmission, ahead of schools and workplaces.
Stockpiling vaccine, even if not perfectly matched, may be a valuable strategy. It would need to be delivered within two months of an outbreak.
School closure is potentially effective.
Household prophylaxis can reduce attack rates by 1/3.
Pre pandemic vaccine with 30% efficacy increases impact with a 67-75 % reduction.
Social distancing can be highly effective, depending on how it is implemented.
Utility of intervention strategies depends on exactly how they are implemented.

History is not encouraging that it is possible to achieve these goals; the scenario assumptions lack support from data. Nevertheless, this is a very useful exercise and has raised questions for future research.

Dr. Glezen pointed out that the proportion of asymptomatics is very important and needs to be revisited as data become available.

The steering committee, as a whole, agreed that it is important to educate users of the output about the assumptions and caveats. They also pointed out the importance of calculating the cost of these interventions.

Spatial scaling
Heterogenities in compliance rates, possibly as a function of income
Differential attack rates as a function of compliance

Stephen will do enough runs to calibrate to Ira's and Neil's models and then look more deeply at heterogenities.Neil will look more deeply at the implications of workplace absenteeism.

Addendum

At a meeting on May 2 with Richard Hatchett, the modelers decided on the following additional runs:

Implications of generic social distancing in 10% increments from 0 to 50
Length of time interventions need to stay in place (3 mo vs 6 mo)
Assume 60% of clinical cases are isolated in hospitals
Contact processes (e.g., Is it the same if 1 person stays home full time as when 2 people stay home ½ time?)
Heterogenities in communities and between communities

Nondisclosure forms

This was the first steering committee meeting for which MIDAS required participants to sign nondisclosure forms. The forms require that information presented at the meeting be treated as confidential. The purpose is to facilitate discussion within MIDAS, especially on pre-publication projects.