A Better Way to Find Patient Zero

Computer simulations and a little math could help narrow the search for an outbreak’s origins.

When George Soper, then a little-known sanitation engineer, was finally able to track down Mary Mallon, he was pretty sure he’d found the woman responsible for a series of typhoid outbreaks in early-20th century New York. Discovering patient zero—Typhoid Mary, in Soper’s case—takes tireless effort even today, and while the concept has been criticized, identifying an outbreaks’ origins can help scientists understand and prevent the next one. Now, health officials might get a helping hand: Researchers have found a new way to whittle down the list of likely candidates for that dubious title of patient zero.

Ordinarily, health officials try to gather the information necessary to finding patient zero through extensive interviews with patients. Often, that information isn’t complete or entirely accurate, so Nino Antulov-Fantulin, a computational biologist at the Ruđer Bǒsković Institute in Zagreb, Croatia, and his colleagues tried to see if they could at least narrow the search.

Their method begins with real-world information about a disease outbreak—who’s come into contact with whom, who’s already gotten sick, and who hasn’t. Step two is to pick someone at random, whom they assume is patient zero, and use computer simulations of a standard model called susceptible-immune-recovered (SIR) to see what happens next. In the SIR model, each sick person infects a fraction of the healthy people they come into contact with, and sick people recover at a rate determined by the disease in question.

It’s a good start for an epidemiologist searching a population of 15,000 or more for patient zero.

For each potential patient zero, the team simulated 20,000 hypothetical epidemics and computed the fraction of those that matched the real-world list of people who’d contracted the disease—in other words, the likelihood a given person could have started an actual, observed outbreak. Compute those likelihoods for every possible patient zero, and the researchers can infer each person’s probability of having been the source of the sickness.

The team tested its approach using data on sexual contacts between 6,642 escorts and 10,106 clients in Brazil between 2002 and 2008. Unfortunately, that data came from clients’ online reviews of their escorts, so there wasn’t much mention of sexually transmitted infections. The fix: Cook up 500 simulated epidemics and treat those as if they were real-world data.

With those caveats in mind, how well did the technique perform? The method couldn’t reliably identify the true patient zero, but it did narrow things down quite a bit: In nearly all of the test runs, the technique identified the originator as either the true patient zero or someone within four degrees of separation on the contact network—friends of friends of friends of friends, so to speak. That’s not perfect, but it’s a good start for an epidemiologist searching a population of 15,000 or more for patient zero.

Even if the method isn’t infallible, it has broad applicability, the team writes in Physical Review Letters; for example, it could help trace certain computer viruses back to their seedy hacker homes.

Quick Studies is an award-winning series that sheds light on new research and discoveries that change the way we look at the world.

Related Posts