For hospitals that use sensors to monitor heart rate, blood pressure, temperature, and other vitals, data mining could prove valuable. Unfortunately, due to privacy and volume concerns, hospitals can only record and store the data directly around medical episodes, making it difficult to perform any sort of predictive analytics that could save lives and aid recovery.

But according to researchers at the University of California-Riverside, the volume concern may be diminishing. A new technique outlined in the paper, “Searching and mining trillions of time series subsequences under dynamic time warping,” by Thanawin Rakthanmanon et al, may make it possible to search a dataset with a trillion or more entries.

Today, the UC-Riverside team is working with Dr. Randall Wetzel of the Children’s Hospital Los Angeles to test the technique and garner insights.

As of right now, the technique can only provide correlations, only some of which are useful. For example, the dataset may reveal that the people giving birth in the hospital are gender female and that people who are six feet tall are all at least five feet tall: facts we hope experienced health care professionals already know. But some of these correlations could become life-saving insights, enabling doctors to predict events such as asthma attacks minutes or hours in advance based on a patient’s heart function.

"This data has the potential to be a gold mine of useful – literally life- saving – information," said Eamonn Keogh, one of the computer science professors working with the team. While the big data-based predictions the system is able to make may pale in comparison to those found in the business or financial circles, it is not particularly fair to compare the two. The healthcare industry, for one, has privacy standards which limit the ability to cull all medical data into one place. Keogh noted that they are confident they will be able to find a way around that limitation, although it is not indicated how quite yet.

The second, and potentially more restrictive limitation is the considerable scale of the data involved in just one hospital. A single patient could have up to 30 vitals measured at a given time, with each being recorded at least every 30 seconds. Naturally, for measurements such as heart rate, the recording is nearly constant.

That creates a lot of data. Keogh estimates that more than a billion measurements have been taken at the Children’s Hospital Los Angeles alone.

Despite the challenges, finding just one useful correlation out of a hundred would be worth it. The National Science Foundation recognized the importance of this work by awarding the team a four-year $1.2 million grant, which the team hopes to use to obtain more useful results and take multiple measurements per second.

Related Articles

Can Hype Spell Hope for Predictive Healthcare Analytics?

Brown University Advances Genomics, Big Data Research

Adding Autonomy to Personalized Medicine