Shortly after a backpacking trip in Michigan in 2009, 20-year-old Sarah Sheridan came down with what seemed to be a nasty case of the flu. Unlike the flu, however, her symptoms only got worse with time. Blood tests, MRI scans, spinal taps and other investigations came back normal or inconclusive.
Sheridan spent the next three years in and out of hospital, all to no avail. Her insurance claims swelled to over $100,000. It wasn't until a chance encounter with someone who'd had Lyme disease that she finally found relief.
New web-based tools seek to spare others from a similar ordeal. CrowdMed, launched on 16 April at the TedMed conference in Washington DC, uses crowds to solve tough medical cases.
Anyone can join CrowdMed and analyse cases, regardless of their background or training. Participants are given points that they can then use to bet on the correct diagnosis from lists of suggestions. This creates a prediction market, with diagnoses falling and rising in value based on their popularity, like stocks in a stock market. Algorithms then calculate the probability that each diagnosis will be correct.
In 20 initial test cases, around 700 participants identified each of the mystery diseases as one of their top three suggestions.
Among them was Sheridan's case, which was solved in a week – around 100 users proposed Lyme disease as the top ranking diagnosis. "To get an answer in just a week is exciting, astounding and incredibly frustrating," Sheridan says. "I keep thinking, 'Where were you three years ago?' It really would have changed the course of my life."
The goal is to help people who come down with any of around 7000 "rare diseases" as defined by health agencies in Europe and the US. In Europe alone, 30 million people have a rare disease, 40 per cent of whom either go undiagnosed or are misdiagnosed at some point.
Once the crowd arrives at a consensus, CrowdMed gives patients a list of the top three, which they can then explore with their physician. Those whose diagnosis proves to be correct earn more points, which can be used to bet on future cases.
Frustrated patients and doctors can also turn to FindZebra, a recently launched search engine for rare diseases. It lets users search an index of rare disease databases looked after by a team of researchers. In initial trials, FindZebra returned more helpful results than Google on searches within this same dataset.
"Medical students and doctors can't learn about all of these thousands of diseases with very low prevalence," says Radu Dragusin, one of FindZebra's developers and a computer scientist at the University of Copenhagen in Denmark. "It's very important to give clinicians an aid, be it FindZebra or CrowdMed, to help make these diagnoses." He adds, however, that he thinks CrowdMed would work best if most of its users had some medical expertise or research knowledge.
IBM's Watson artificial intelligence system is already being used to help doctors wade through mountains of fast-changing medical research on cancer. Claudia Perlich, who helped develop Watson, imagines it teaming up with CrowdMed to work on tough diagnoses.
"If Watson could get hold of what people submit to CrowdMed, I would love it," she says. "I absolutely agree with the premise that the big problem of the medical system is that we don't have sufficient information sharing."
Diagnoses without borders
Early in Sarah Sheridan's quest to diagnose her mystery illness, her doctors often focused on a trip she took to East Africa, rather than a backpacking excursion, where she likely encountered a tick that gave her Lyme disease (see main story).
Diagnostic tools like CrowdMed, can also attract worldwide input that confuses the issue. When initial CrowdMed tests included users from outside the US, Sheridan's case was flagged as malaria, which is nearly nonexistent in North America, though it has similar symptoms.
To avoid this problem, CrowdMed's founder, Jared Heyman, has limited input on cases to users from North America for now, though he plans to expand to different regions soon.
Post a Comment