Model Behavior: Creating Virtual Worlds to Solve Real World Problems

January/February 2013

 1  |  2  |  Next > 



The fight against HIV is now being waged in virtual worlds that exist only inside a computer. Making use of statistics and other data, these simulations recreate real world conditions to help us better understand drug resistance, predict the outcome of prevention and treatment efforts, and determine the best use of resources. They can't predict the future, but computer models are helping us to answer the question, What if?

Mixing a Better Cocktail

For a person on HIV medication, strictly following their drug regimen is critical to treatment success. Research has shown that the relationship between adherence and the resistance the virus can develop is different for each medication in a patient's regimen. A new computer model developed by Harvard grad students in collaboration with a similar team at Johns Hopkins School of Medicine is the first to predict drug resistance that takes adherence into account.

"The mathematical model we built looks at the growth rate of the virus in a person who's being treated," said Daniel Scholes Rosenbloom, one of the students who developed the model. "To know whether viral load will grow or decay, you need to know both how strong each drug in the regimen is and how often the patient is taking or skipping their pills."

The Harvard simulation builds on years of accumulated data and mathematical models explaining how viral load changes over time and at different phases of infection. But Rosenbloom and his colleagues also made use of another research team's work. At the Johns Hopkins School of Medicine, Robert F. Siliciano, M.D., Ph.D., worked with graduate student Alireza Rabi to study how HIV reacts to varying doses of medications.


Using data from Johns Hopkins, the Harvard team created a simulation that predicts whether the virus was growing or if different strains were emerging, based upon a patient's adherence level. If drug levels are very low, the virus is able to grow even without being resistant to medication, and so drug resistance does not emerge. If drug levels are very high (as they are in a patient taking 100% of their pills), then even a drug-resistant virus may not be able to grow. However, there is a "danger zone" of intermediate drug levels where drug resistance emerges. The new model computes the size of this danger zone, which is different for each drug. (Rosenbloom noted that boosted protease inhibitors tend to have a very small danger zone.)

The Harvard students fed the data and their equations that simulated more than 1 million patients -- about 50,000 for each of 23 HIV medications -- into the school's computer cluster. These virtual patients varied in adherence and viral load over the course of 48 simulated weeks. "We tried to simulate a diverse cohort [that] might participate in a real-world clinical trial," said Rosenbloom. It took the computers up to two days each time the simulation was run.

Rosenbloom notes that the model focuses on monotherapy and that only some simple drug combinations were simulated. However, it proves that resistance can be predicted. While the current model only examines concentrations of drug in blood plasma, Rosenbloom says future models will look at drug concentrations in other parts of the body, as well as simulating combination therapies. The hope is that their computer model will lead to better, cheaper, and more effective HIV medications.

Led by Martin A. Nowak, Ph.D., professor of mathematics and biology and the director of Harvard University's Program for Evolutionary Dynamics, the Harvard model is discussed in a paper appearing in the September 2, 2012 issue of Nature Medicine.

Sim City, New York

These sims, however, act like real people -- they have sex, sometimes engage in risky behaviors, and sometimes contract or transmit HIV.

Public health officials and policymakers need to identify the best strategies for combating the spread of HIV, especially in a world of limited resources. However, determining the effectiveness of these strategies, either separately or in combination with others, can take years. Brandon Marshall, Ph.D., assistant professor of epidemiology at Brown University, presented a computer model at the AIDS 2012 conference last summer that accurately recreates the spread of HIV in New York and can make predictions into the year 2040 based on a given scenario.

New York City data about drug use, sexual orientation, access to treatment, treatment effectiveness, probabilities regarding risk behaviors, and information about other behavioral, social, and medical factors were used to create the model. The simulation was run and constantly adjusted until it could match actual infection rates that were known to have occurred in New York between 1992 and 2002 among injection drug users.

"With this model you can really look at the micro-connections between people," Marshall said in a Brown press release about his work. "It reflects what's seen in the real world."

Marshall's model of New York is a virtual reality of 150,000 "agents" -- simulated individuals derived from statistical data. These sims, however, act like real people -- they have sex, sometimes engage in risky behaviors, and sometimes contract or transmit HIV.

Six scenarios, each featuring a different HIV prevention policy, was tested by the model: expanding needle exchange programs, expanding substance abuse treatment programs, expanding HIV testing, starting people on HIV treatment earlier, a combination of these strategies, and not changing the current policies.

Simulating one year's time with the six scenarios and the 150,000 agents took Brown University's massive computer array 72 hours to run. To ensure accurate results, each scenario was run several times, providing predictions through 2040.

The single most effective strategy was to start HIV treatment earlier, which lowered the rate of new infections by 45%. Increasing the number of people who get tested for HIV by 50% would reduce new infections among injection drug users only by about 12% through 2040, according to the model. Combining all four strategies would cut infections by 62%.

Marshall was disappointed that the strategies, as shown by the model, would not lead to a greater drop in the infection rate.

Role model: Representation of HIV, made of HIV medications. A version of the image was used to illustrate the cover of <i>Nature Medicine</i> which featured a paper by Daniel Rosenbloom and his colleagues on their model for predicting drug resistance.

Role model: Representation of HIV, made of HIV medications. A version of the image was used to illustrate the cover of Nature Medicine which featured a paper by Daniel Rosenbloom and his colleagues on their model for predicting drug resistance.

"I actually expected something larger," he said. "That speaks to how hard we have to work to make sure that drug users can access and benefit from proven interventions to reduce the spread of HIV."

Marshall plans to expand work on his model. "What we are moving towards now is actually implementing costing data into the model so we can examine the cost-effectiveness of various scenarios," Marshall told Rhode Island Public Radio. "That's the next step that I think will be of most interest to policy makers."

The National Institutes of Health (NIH) and the Lifespan/Tufts/Brown Center for AIDS Research are providing financial support for the model's continued development.

 1  |  2  |  Next > 

This article was provided by Positively Aware. It is a part of the publication Positively Aware. Visit Positively Aware's website to find out more about the publication.

No comments have been made.

Add Your Comment:
(Please note: Your name and comment will be public, and may even show up in
Internet search results. Be careful when providing personal information! Before
adding your comment, please read's Comment Policy.)

Your Name:

Your Location:

(ex: San Francisco, CA)

Your Comment:

Characters remaining:


The content on this page is free of advertiser influence and was produced by our editorial team. See our content and advertising policies.