HomeOur ProjectsCompleted ProjectsHIV Network Simulation and the Firewall Effect

HIV Network Simulation and the Firewall Effect

Injection Drug User Network Topologies and HIV Stabilization Dynamics

Bilal Khan (PI), Kirk Dombrowski (PI), Ric Curtis, Mohamed Saad, Kate McLain, and Samuel Friedman

October 2009-November 2011 ($736,858) NIH/NIDA RC1 DA028476-01

This research represents an exploration of well-known but until now unaccounted for HIV-1

Infection patterns among injecting populations. This research models the interaction of actors within a risk network, an evolving approach in HIV research. Because IDU communities remain reservoirs of HIV, and present the potential for infecting the larger communities in which they live and with whom they continue to interact, understanding the short and long term dynamics of infection transmission represents an important public health priority for both the IDU community itself, and the general population that surrounds it.


The key findings and conclusions of the experiments undertaken here include:

  1. Stabilization. Simulation trials modeled on the SFHR network resulted in HIV rate sub-saturation stabilization at a level similar to that observed for injecting drug user networks in New York City in the early stages of the HIV epidemic.
  2. Hypothesis 1. In a majority of cases, and for a range of parameter settings, nodes with mature (i.e. low infectiousness) HIV+ status demonstrated a “firewall effect”, dividing the network into clusters of uninfected nodes which remained relatively stable over time. These nodes play an important role in the non-spreading of HIV in the net- work despite the presence of high numbers of uninfected nodes and the ongoing reappearance of new, high infectiousness nodes.
  3. Hypothesis 2. Local level social network structures within the overall risk network contribute to the non-spreading of HIV in injecting drug user networks despite continuing turnover in network participants, high turnover in network connections, and continuing very high numbers of risk events.
  4. Hypothesis 3. Injecting drug user networks with stable, sub-saturation levels of HIV-1 infection continue to produce new infections over the course of time which, while they fail to propagate full throughout the network, never the less represent reservoirs of high infectiousness that can spread to surrounding populations.
  5. Scale. While networks of size 5,000 through 25,000 behaved within a narrow (and therefore predictable) range of overall characteristics, networks of 1000 nodes showed high variability in their network wide behavior. As a result, known outcomes of interventions in small scale networks may not serve as good indicators of likely outcomes of the same intervention in other small networks, nor in the same networks at a different time. In each case, the effects of random events may render the most successful interventions moot, or the most ill-adapted interventions successful—this without a change in the underlying set of network attributes or dynamics.
  6. Risk. In simulations where the overall rate of actor risk was set intentionally high, the firewall effect was pronounced, while in cases where risk rates were low, stochastic variations from one trial to the next seem to dominate an overall infection rates the firewall effect were difficult to predict.
  7. Virus Transmissibility. Varying rates of virus transmissibility showed high variability in the spread of the virus, including ranges where over- all infection rates failed to stabilize, even in very long simulation runs of 40-50 years. From this, it would appear that for set levels of risk, the sub-saturation stabilization phenomenon is specific to a level of infectiousness. It remains possible that changes in the likelihood of infection (for a given risk event) can have serious effects on overall network infection levels unpredicted by historical data. On the other hand, in simulations where stabilization was achieved, increased virus transmissibility did not produce a greater proportion of new infections through time, showing that topological effects preventing virus trans- mission were, in mature networks, robust against large increases in virus transmissibility.
  8. Transient Members of the Network. Transient participants in an IDU network can have a serious impact on the overall HIV rate. While overall network stabilization remains, even low numbers of transient participants engaging in risk behaviors with long-term network members increases HIV to levels approaching saturation, enlarging the “reservoir” of HIV infection within injecting drug user networks.
  9. Network Participation Duration Stable networks, created by the long-term participation of network members, yield highly stable rates of infection, despite the fact that individuals within the network continue to “churn” their connections to one another. Conversely, high turnover of network participants, even where new participants enter the network in an uninfected state, necessarily produces both higher and continuously rising HIV rates. These dynamics significantly affect the emergence of network structures that mitigate HIV infection. In particular, a high turnover rate of network participants significantly diminishes the firewall effect, while stable networks show consistent effective structural barriers to disease spreading.
  10. Churn Variations in the rate at which network participants changed risked partners seemed to have little effect on HIV rates within the net- work, while increasing the firewall effect. This is caused by the breakup of clusters of mature, low infectiousness nodes whose new connections to uninfected parts of the network increases the isolation of uninfected actors from periodic sweeps of new infection.