Let’s call him Naskar. Naskar grew up in a small village outside Calcutta, and moved to the city when he was 22. He is homosexual, and quickly developed a likeminded group of friends. He doesn’t do drugs, though he sometimes drinks. He sees his family infrequently, and spends most of his time inside the city at his minimum-wage job. Naskar has a boyfriend and a few lovers on the side; he assumes his boyfriend has lovers too, and doesn’t mind. In a few years he will marry, but he will continue to have sex with multiple men, as a giver and as a receiver. He contracted HIV a while ago, but does not yet realize it. By the time he does, he will have given it to his wife, and will have no idea who gave it to him.
These are not the only facts of Naskar’s life – far from it. But to those who study networks as a means of combating HIV, they are the most important. When plugged into a network of similarly profiled individuals – who will have different answers to the same questions – he becomes a data point. A probability. A pattern. A likelihood. An average.
As an individual, he is just Naskar. As a data point, he could save lives.
* * *
Circumcision, methadone treatment and marriage counseling might seem like oblique ways to combat HIV, but they are all part of the rapidly growing arsenal of responses to the worldwide epidemic, which infected 2.2 million new adults in 2009, according to international HIV & AIDS charity AVERT. Even here in the United States, where treatment is more available than almost anywhere in the world, 1 in 5 HIV-positive people don’t know they are infected. A full range of treatment options can cost up to $12,000 a year, making prevention more crucial than ever. And despite steps like President Obama has taken – asking people to get tested no matter what, even doing so himself – the sneaky and mutative disease still defies science’s attempts to see the big picture. At last, though, they may have found one.
Sexual network modeling is when medical engineers and healthcare workers use data gathered from individuals about their romantic interactions with others to build maps of connections within a certain society or subset of that society. Doing so can provide guidelines for where to treat HIV and how.
“A network model is useful because it captures homogeneity and behavior and susceptibility differences among person in the population,” says Elisa Long, assistant professor of operations management at Yale University. “Diseases spread in the same way that rumors spread or a new product is diffused in the market.” The way a particular disease, like HIV, moves through a group of people will dictate how treatment and prevention efforts should progress. And whether it’s vaccines, vaginal gels and more traditional remedies (like drug treatment regimes) or circumcision and counseling, healthcare workers must offer preventatives and remedies that mesh well with the societies in which they’re offered and the problems they are meant to address.
For instance, circumcision can prevent men from getting HIV during intercourse by removing tissue both easily torn (providing an entry to virus) and hospitable to virus growth. It is not, however, a viable preventative in countries religiously opposed to the practice. Similarly, methadone treatment can reduce the number of intravenous drug users in the population, but is useless without a means to target them, says Benjamin Armbruster, assistant professor of industrial engineering and management sciences at Northwestern University’s McCormick School of Engineering.
One of the biggest challenges to HIV’s eradication is that, so far, completely reliable methods of treatment and prevention continue to elude scientists and medical practitioners. When a person is first infected, the amount of virus in their bloodstream is extremely high. This period, which is known as acute infection and follows a 2-4 week incubation stage, is the most infectious and lasts roughly two months. By the time a person finds out they are infected, chances are excellent they have already transmitted the disease to someone else, often several.
And the problem of identification always remains. “Not enough people know their status and that’s part of the story of how it’s getting transmitted: people just don’t know they have it,” says Armbruster. Even in high-risk populations, people often lack the resources or ability to get tested, or simply choose not to. In fact, even groups classified “at increased risk” often don’t perceive themselves as such, according to Shannon Galvin, associate director of research at Northwestern’s Center for Global Health. “That’s just basic psychology,” she says. “Despite everyting we know, there are still approximately 50-60,000 new cases in the US each year,” says Robert Murphy, director for Northwestern University’s Center for Global Health. “That’s more people that get killed in automobile accidents per year.”
* * *
Providing testing, prevention and treatment options tailored to specific groups of people, combining interventions and convincing people to get treated are all ways network modeling can help, Armbruster says.
Modeling can offer glimpses into a society at several different stages of the HIV epidemic. How HIV enters a population, for example, depends on how it is constructed. When a large population of sex workers exists, for example, the disease is likely to appear there first, spreading quickly among them and then to their clients, after which it will trickle into the population from clients to their spouses and other partners. Where polygyny (men keeping multiple wives) is practiced, these partners are likely to be infected quickly. The women, however, are not nearly as likely to infect others as they would be in a society that condoned women having more than one lover. Once these women get it, they are unlikely to spread it further. By constructing these maps, medical workers can begin to tell who to target for treatment and how.
It’s easy to confuse the issue, says Armbruster, who cautions against thinking of models as accurate representations of any given group. It’s not like Facebook, he says, where anyone you might wish to treat is already plugged in to the network with his/her social relationships neatly laid out and on view. Rather, constructing a model of a given area can tell medical practitioners and engineers about the ground rules of that society: How do its members tend to relate sexually? (Are they polygamous? Promiscuous? Do they wait until marriage?) What are the most common sexual interactions? (Homosexual? Heterosexual? Bisexual? Does sex with different partners often occur within a short time period?) Which sexual interactions do not happen? What taboos need to be respected? (In societies with an underground homosexual scene, how can treatment occur without publicizing certain activities? What are the alternatives to circumcision for the religiously opposed?)
Through his work with sub-Saharan African communities, Armbruster has learned that partners of infected people have a drastically higher chance of being infected themselves. Whereas the population-wide prevalence in Malawi hovers around 10 percent, network modeling has shown that the chances an infected individual has an HIV-positive partner are closer to 50 percent. This has major implications for treatment.
“With the screening program, at least to some extent, people get tested because they’re sick,” Armbruster says. “Once they’re sick, it’s pretty late. In the early stages is where treatment helps the most.” HIV has fairly mild, flu-like symptoms, which can be easy to miss – more severe, opportunistic infections don’t generally develop for a number of years, and are almost always characteristic of the development of AIDS. If a person is coming to a testing center for symptoms like these, then they have likely been living for years with undiagnosed HIV. So once they’re sick, in other words, they’ve almost certainly transmitted it, and there is a good chance that at least some of their partners have done so in turn.
Finding the people most likely to have been infected or to get infected in the near future is a major preventative strategy. “Early treatment can reduce HIV transmission by about 96%,” Long says. This makes getting the infected on regimes one of the most reliable methods of intervention available.
The question then becomes: what is the best way to find and treat people early? Armbruster proposes contact tracing, a fancy term for a simple process: asking people who their partners are or were. This isn’t as simple as it sounds. “First of all asking people who their sexual partners are is not easy,” Armbruster says. “Second of all you have to find them. So it’s not obvious that this is a cheap way of going about things.” However, the chance that the partner of an HIV-positive individual is also positive is extremely high, Armbruster adds, which makes finding them a better idea. “It mitigates the increased costs,” he says. “Instead of a one in 10 chance, we have a one in two or so.”
* * *
HIV can be a difficult infection to treat simply by virtue of its sexual nature: the more private a society considers it, the more difficult it becomes to discuss. In India, where homosexual roles have historically been clearly defined but are now changing, the prevention and treatment picture is pretty difficult to define. John Schneider, assistant professor of medicine & epidemiology at University of Chicago, uses network data to figure out how these changing roles affect intervention options. “In India there’s traditionally been these set sexual roles that men take on, where men identify as a certain sort of person that participates in a certain type of behavior,” he says. This is not unique to India – it happens in South America and other places as well, Schneider says. There are “insertive” and “receptive males” – similar to tops and bottoms in the U.S. – whose roles have been prescribed for centuries. But that is now changing to include a versatile role – a “double-decker.”
“What’s happened so far is identity is strongly tied to behavior, but increasingly, there are breaks between identity and behavior,” says Schneider. This makes intervention difficult. If you tell a receptive individual to put on a condom during sex, but his insertive partner does not, you’ve done nothing but waste a condom. Similarly, if you circumcise a receptive individual, that might not help much.
Modeling can help, by indicating which individuals engage in which behavior, how and how often they are likely to be infected, and which treatments are effective. Having really flexible models is important here, Schneider adds. What happens to the population if the numbers of insertive, receptive and versatile change? Transmission methods and rates would alter accordingly, inevitably changing the necessary preventatives and treatments.
* * *
Georges Reniers, assistant professor of sociology and public affairs at Princeton University, studies marriage patterns in societies with a high prevalence of HIV, specifically how people stay in unions or choose new ones based on anticipated risk on infection. His work is closely related to Armbruster’s.
Several factors play in to how a sexual network will end up looking. Age of marriage is an issue, since populations with early marriage ages for females look different than later ones. “In populations that practice polygyny, you don’t have that very long interval between first sex and first marriage, so sex in these populations is much more quickly channeled in the context of a marriage.” Frequency of sex is another issue. “Models predicting the effects of partnership concurrency on the epidemic have always assumed that someone who has two partners will have twice as much intercourse as someone who has only one, so the number of coital acts per partnership remains constant.” As it turns out, this is usually untrue, Reniers says. What actually happens is that women in polygynous unions end up having significantly less sex than women in monogamous unions, which changes the network quite a bit. Yet another issue is concurrency, or having multiple sexual partners at one time.
A debate regarding the role of partnership concurrency currently rages within the scientific community. Some organizations, such as USAID, have already deployed anti-concurrency campaigns in countries such as Zambia, Swaziland and Botswana based on the assumption that multiple partners contribute to higher rates of infection. But the empirical evidence for this is actually fairly limited, says Georges Reniers. “We found this negative correlation between HIV and concurrency and we started looking for reasons. One possible explanation is the network structure. Most people who have advanced this theory of concurrency have assumed that both men and women have multiple concurrent partners, and that creates a very different sexual network than is the case where only men have multiple concurrent partners. This network is not as conducive to the spread of HIV.”
Essentially, a network where both men and women have multiple concurrent partners will create a vast web, similar to, say, the flight chart of a popular airline. A network in which only men have multiple partners, however, will tend to be much more limited: many distinct, star-shaped clusters, with a central male and satellite females. Doubtless there will be some connections outside the marriages as well, but far fewer. Studying sexual networks based on sample data may seem academic, but it actually has a huge impact on how funds are spent and treatment and prevention are approached: should treatment and prevention take place within the context of the marriage or not? Who should be targeted for prevention? What would be more effective, marriage counseling or population-wide vaccination attempts?
“Prevention responses also need to take into account the progress of the epidemic,” says the USAID website. Network modeling can inform that progress. “One hypothesis is that in early epidemics, most discordant couples occur when HIV is introduced into a pre-existing relationship, whereas in more mature epidemics, a greater proportion of discordant couples initiate relationships with a new partner who is already infected.” Thus, even within a single society, treatment might vary depending on the disease’s progress. A map could be really useful at a time like that.
* * *
A network cannot be used to plot individuals, stresses Eva Enns, a PhD student in electrical engineering at Stanford University. “Ultimately it’s not likely, in the case of humans to humans on the individual level, that you would map out a network and use it. You’re not going to say ‘Aha! These are the people I should target!’” Instead,
“What you can do is leverage technology to map interactions more easily and more continuously and make structural changes to how people are interacting with each other.”
Enns cites a recent study in which high schoolers were assigned radio frequency identification tags (RFID tags) before going about their daily business. The tags recorded the face-to-face interactions students had with each other, allowing researchers to create a detailed network of the high school. Once they’d mapped it out, the map became a tool in times of crisis. “From there what you could do is to look at that network and say, ‘Are there things in the structure of the school that we could change if a pandemic were detected? Maybe we should change how we have our lunch hour, change how classes enter and exit.’ You could change characteristics of the network, but you wouldn’t be targeting specific individuals.
Sabina Alistar, a PhD student working on resource allocation for control of infectious diseases at Stanford, approaches modeling similarly. For her, though, flexibility is key. “Instead of having a network that models how people interact, you group people by characteristics and group them in buckets where you assume an average of how people interact,” Alistar says. “This is a pretty realistic way of modeling the epidemic.”
There are also issues of combining treatment, Armbruster points out. “If you have two interventions, they are in some sense competing,” he says. “To combine them lessens the sum of the two. What people are just now starting to do is really get the interactions correctly to get a good idea of how you want to distribute your resources.”
The power of the network is best exemplified here, when thinking about where and how people do things, and the combinations between approaches. Health officials can use those high-frequency wheres and hows to direct their intervention efforts. If we think of a network as a vast map of probabilities and likelihoods, then Naskar’s contribution to the cause becomes clear. Sure, treating him will be one small step toward eradicating HIV entirely. More than that, though, Naskar can provide a map to the future: his story is the story of many.