A new challenge looks for a smarter algorithm to improve healthcare

The Heritage Healthcare Prize puts up $3 million dollars for a predictive algorithm to identify at-risk patients.

Starting on April 4, the Heritage Health Prize (@HPNHealthPrize) competition, funded by the Heritage Provider Network (HPN), will ask the world’s scientists to submit an algorithm that will help them to identify patients at risk of hospitalization before they need to go to the emergency room.

“This competition is to literally predict the probability that someone will go to the hospital in the next year,” said Anthony Goldbloom at the Strata Conference. Goldbloom is the founder and CEO of Kaggle, the Australian data mining company that is running the competition for HPN. “The idea is to rank how at risk people are, go through the list and figure out which of the people on the list can be helped,” he said.

If successful, HPN estimates that the algorithm produced by this competition could save the country billions in healthcare costs. In the process, the development and deployment of the algorithm could provide the rest of the healthcare industry with a successful model for reducing costs.

“Finally, we’ve got a data competition that has real world benefits,” said Pete Warden, author of the “Data Source Handbook” and founder of OpenHeatMap. “This is like the Netflix Prize, but for something far more important.”

The importance of reducing healthcare costs can’t be underestimated. Nationally, some $2.8 trillion dollars are spent annually on healthcare in the United States, with that number expected to grow in the years ahead. “There are two problems with the healthcare reform law,” said Jonathan Gluck, a senior executive at HPN. “We pay for quantity, not quality. The more services provided, the more the provider gets paid.”

If patients who would benefit from receiving lower cost preventative care can receive relevant treatments and therapies earlier, the cost issue might be addressed.

Why a prize?

HPN is just the latest organization to turn to a prize to generate a solution to a big problem. The White House has been actively pursuing prizes and competitions as a means of catalyzing collaborative innovation in open government around solving grand national prizes. From the X-Prize to the Netflix Prize to a growing number of challenges at Challenge.gov, 2011 might just be the year where this method for generating better answers hits the adoption tipping point.

Goldbloom noted that in the eight months that Kaggle has hosted competitions, they’ve never had one where the benchmark hasn’t been outperformed. From tourism forecasting to chess ratings, each time the best method was quickly improved within a few weeks, said Goldbloom.

As David Zax highlighted in his Fast Company article on the competition, adding an algorithm to find patients at risk might suggest that doctors’ diagnoses or clinical skills are being subtracted from the equation. The idea here is not necessarily to take away a doctor’s skills. Rather, it’s to provide them with predictive analytics that augment those capabilities. As Zax writes, that has to be taken in context with the current state of healthcare:

A shortage of primary care physicians in the U.S. means that doctors don’t always have time to pick up on the subtle connections that might lead to a Gregory House-style epiphany of what’s ailing a patient. More importantly, though, the algorithms may point to connections that a human mind simply would never make in the first place.

Balancing privacy with potential

One significant challenge with this competition, so to speak, is that the data set isn’t just about what movies people are watching. It’s about healthcare, and that introduces a host of complexities around privacy and compliance with regulations. The data has to be de-identified, which naturally impairs what can be done. Gluck emphasized that the competition is HIPAA-compliant. Avoiding a data breach has been prioritized ahead of a successful outcome in the competition. Not doing so, given the sanctions that exist for such a breach, might well have made the competition a non-starter.

Gluck said that Khaled El Eman, a professor at the University of Ontario and a noted
healthcare privacy expert, has been making attempts to de-anonymize the test data sets. Gluck said El Eman has been using public databases and other techniques to try and triangulate identity with records. To date he has not been successful.

Hotspotting the big picture

The potential of the Heritage Health Challenge will be familiar to readers of the New Yorker, where Dr. Atul Gawande published a feature on “healthcare hotspotting.” In the article, Gawande examines the efforts of physicians like Dr. Jeffrey Brenner, of Camden, New Jersey, to use data to discover the neediest patients and deliver them better care.

The Camden Coalition has been able to measure its long-term effect on its first thirty-six super-utilizers. They averaged sixty-two hospital and E.R. visits per month before joining the program and thirty-seven visits after—a forty-per-cent reduction. Their hospital bills averaged $1.2 million per month before and just over half a million after—a fifty-six-per-cent reduction.

These results don’t take into account Brenner’s personnel costs, or the costs of the medications the patients are now taking as prescribed, or the fact that some of the patients might have improved on their own (or died, reducing their costs permanently). The net savings are undoubtedly lower, but they remain, almost certainly, revolutionary. Brenner and his team are out there on the boulevards of Camden demonstrating the possibilities of a strange new approach to health care: to look for the most expensive patients in the system and then direct resources and brainpower toward helping them.

The results of the approach taken in Camden is controversial, as Gawande’s response to criticism of his article acknowledges. The promise of applying data science to identifying patients at higher risk, however, comes at a time when the ability of that discipline to deliver meaningful results has never been greater. If a smarter predictive algorithm emerges from this contest, $3 million dollars of prize money may turn out to have been a bargain.

Related:

tags: , , , , , , ,