$3 Million Dare Asks Data Crunchers to Fix Healthcare
Can an algorithm prevent unneeded hospital stays?
By Morgon Mae Schultz
A network of California doctors is issuing a $3 million dare asking data miners to fix healthcare. The Heritage Health Prize, which stands to be the largest-yet data-modeling competition, will challenge participants to write an algorithm identifying patients most at risk for unnecessary hospitalization—an economically draining component of U.S. healthcare woes. Ultimately, the algorithm will alert doctors to intervene before hospitalization with healthier, far cheaper preventive action.
Hospitals are costly. Jonathan Gluck, senior executive with Heritage Provider Network, the competition’s sponsor, estimates that Americans spend between $30 billion and $40 billion annually on unnecessary hospitalizations. Unneeded admissions also put patients at risk for hospital-borne infections and divert resources from patients who really need them. More to the point, Heritage asserts, they’re symptomatic of a system that treats sickness rather than keeping people healthy.
“With all that’s going on with predictive modeling and data mining, the thought was, well, let’s see if we can kind of think outside the box and get new people involved in trying to solve these problems,” Gluck says. Some feared the algorithm will be used to avoid caring for costly patients, but Gluck stresses that Heritage is not an insurance company and, as a network of doctors, has no say in which patients it treats or doesn’t treat.
To understand why the network decided a huge data-mining prize was the best way to prevent unneeded lay-ups, Gluck says you need to know a little about its founder and CEO, Richard Merkin. Merkin is a medical doctor, philanthropist for youth and medical charities, and a core contributor to the X Prize Foundation, which runs innovation competitions in space exploration, genomics, ecology and other fields. Merkin is genuinely excited to bring new minds to the healthcare table and believes data miners hold great potential, according to Gluck. The contest was Merkin’s idea to not only yield a winning algorithm, but also to grab the attention of data miners globally and raise awareness about competitive innovation. “You could be the best company in the world and if you hired 20 really good minds to work on your problem, you’ve got 20 good minds. If you run a prize, and you’ve got 2,000 entries, I would assume you’ve got a better shot at success,” Gluck says.
The company is putting up substantial money, and serious data, to back up its hopes. The $3 million award is bigger than the Nobel Prize for medicine and the Netflix prize combined. (The former varies but has paid about $1.5 million each of the past 10 years. The latter, a famous data-mining competition, awarded $1 million.) It’s by far the biggest prize Kaggle, the data-contest administrator that will run the competition, has ever handled. The 11-month-old company accelerated a total infrastructure rebuild to accommodate it.
Competitions attract great minds based on three factors, according to Kaggle founder and CEO Anthony Goldbloom: by offering interesting, real-world data to researchers usually stuck with lackluster fictional or public data; by posing meaningful social problems; and with large prizes. And this contest, he says, ticks all boxes. “Firstly, there’s a ginormous prize up for grabs. Secondly, the data set is fantastic. And thirdly, it’s a really, really, really significant, meaningful problem,” Goldbloom says. “I mean the Netflix prize attracted something like 50,000 registrations and it was to make sure people didn’t see movies that they disliked. This is obviously far more significant.”
Goldbloom says the Heritage data set alone could attract scores of participants even if it weren’t linked to a fortune. He cites a chess-rating competition that posed what data miners considered an interesting problem. Two-hundred and fifty-eight teams competed for the top prize, a DVD signed by a chess grand master. A $25,000 prize attached to a less interesting problem attracted far fewer entries.
The Heritage set includes real patient data on perhaps hundreds of thousands of members—doctor’s visits, test results, prescriptions and whether they’ve been filled—scrubbed of identifying details by a group of health-privacy experts. Kaggle will conceal a late portion of the 2005-2010 data and challenge participants to predict hospitalizations based on the earlier portion. When entrants upload their predictions, Goldbloom says, they’ll get real-time feedback on the site’s leader board. “It’s literally like making data science a sport. These people will know in real time exactly how they’re doing.”
Kaggle will launch the contest April 4 and expects it to run for about two years.
Who Will Win?
Goldbloom and Gluck say the $3 million purse could go anywhere. In the contest to design a new chess-rating system, an IBM researcher took first place—no big surprise. But in a competition to predict viral load in HIV patients, the winner had learned to data mine by watching lectures Stanford University had posted online. “This guy just became interested in it. Started watching these Stanford lectures, and learned enough,” Goldbloom says. And the computer scientist who won Kaggle’s first contest was a 25-year-old in Slovenia. The underdog factor is so important to competitions that Kaggle helps hosts whittle down data sets as much as possible so participants with few resources can run the problems on laptops or small systems.
“There’s a lot of low-hanging fruit that I suspect things like this could help to unleash,” Goldbloom says. “For example, you identify somebody as being a really high risk of hospitalization and you get them into an exercise program or you have a nurse call them up every day to make sure they take their prescription. Well, that’s going to have a significant effect on quality of life and it doesn’t involve a $3 trillion drug trial. It’s low-tech but it can have as dramatic an impact as a new blockbuster drug.”
To participate in the contest, visit the Heritage Health Prize website.
Morgon Mae Schultz is a copy editor for MSP TechMedia