Dr. Smartphone, What’s My Diagnosis?
There has been an explosion in the number of apps and smartphone systems that help diagnose and track the progression of a disease. Companies such as Cellscope and uChek are using smartphones to make diagnosing disease something that doesn’t require a doctor’s office. Smartphone diagnostics have the potential to treat patients more effectively and lower costs.
One of the newest and more promising solutions in smartphone diagnostics comes from the team at Lionsolver who recently won the Michael J. Fox Foundation Parkinson’s Data Challenge for their machine learning approach to analyzing smartphone data. A machine learning approach means training “computers to make decisions like humans do. They adapt their computational models based on the output of the data,” explains Drake Pruitt, CEO of Lionsolver. He and his team applied their enterprise software to a smartphone dataset of patients with and without Parkinson’s Disease (PD) provided by the Michael J. Fox Foundation. They were able to correctly distinguish which smartphone users had PD and predict the progression of the disease over a 90-day timeframe.
To determine who had PD, the Lionsolver team classified the phone’s GPS, compass and accelerometer data, sourced from a background app. After training the models with the data, they ran several iterations to filter out irrelevant movement data and then placed the users into clusters based on which ones experienced shaking within a certain hertz range. The team then used these clusters to distinguish which people had PD. Pruitt’s team also found that they could use the accelerometer data to predict whether the disease would improve or worsen in the short term. They also prototyped their own version of a mobile app that would provide clinicians with additional data through daily gameplay by PD patients. This type of app has the potential to feed a dashboard connecting patients and their caregivers to their physicians for a more complete, individualized diagnosis and tracking of PD.
Like other neurodegenerative diseases, the current method for the diagnosis of PD is subjective and often inaccurate. Doctors diagnose PD based on a patient’s medical history, signs and symptoms and a physical examination. These assessments use clinical information provided by a patient on a single day, making them prone to inaccuracy because a patient’s symptoms frequently fluctuate. The long gaps of time between doctors appointments contribute to the incomplete picture of a patient, making accurate diagnosis or treatment more difficult. Compared to these current methods, a machine learning approach that draws on smartphone diagnostics “would bring a level of objectivity to help diagnostic work and would provide a whole new level of data driven communication between the physician and patient,” explains Pruitt. Looking forward, smartphone data collection and diagnostics have the potential to serve as the basis for the patient-physician connection.
Strengthening the patient-physician connection with continuous, informative data will help lower costs by enabling doctors to treat patients more effectively. For example, around $25 billion is spent annually in the US alone to cover the direct and indirect costs of PD, including treatment, lost income from inability to work and social security disability payments. The estimated seven to ten million people worldwide who are living with Parkinson’s each spend an average of $2,500 per year for medication. Since there are currently a plethora of medications to treat the symptoms of PD, a lot of that money is spent trying a variety of medications. Additionally, since many other neurological conditions mimic the appearance of PD, a lot of money is spent on running tests to rule out other diseases. By providing data to accurately diagnose and monitor the progression of Parkinson’s, smartphone tools will lower costs by preventing unnecessary tests and helping to identify the best medications for a specific patient.
While several companies are working on smartphone diagnostic solutions, there is still a lot of room for innovation within the space. Lionsolver’s machine learning solution provides a great launching pad for more companies to expand upon.
Image by Mike Twohy.