With so many variables to consider and a number of limitations on data access, healthcare issues are among the more difficult to tackle for the big data community.
To address this problem and more, MIT launched a big data initiative last year at their Computer Science and Artificial Intelligence Laboratory (CSAIL) called bigdata@CSAIL. Since healthcare does provide challenging and potentially lucrative use cases, the big data research at CSAIL encompasses topics like note text analytics and infection tracking.
The amount of genomics research and related data volumes have grown exponentially over the last decade, leading researchers like Peter Szolovits, professor in the Harvard-MIT division of Health Sciences and Technology (HST) and director of the Clinical Decision Making group at CSAIL, to develop a database that combines clinical and genomic data. Szolovits hopes that by linking the two, doctors would be able to draw on the wealth of genetic data and conduct research on which genetic variations are connected to certain diseases.
According to MIT, Szolovits’s group is also working on applying text analytics to physician notes and inputting them into a database. One of the main challenges for the group was determining proper definitions based on context. Their example was the word ‘discharge,’ which in the medical field can mean a release from either a hospital or from a human body.
In November, the group presented a system that was able to pick out word definitions at 75 percent accuracy. While by no means perfect, it is a step toward incorporating useful doctor thoughts into a workable medical database. In the future, algorithms such as this one that pick out words based on context could strip documents of personal information, allowing those documents to be input and used by other physicians in research.
Another application deals with infections within hospitals that often setback treatments. According to MIT, grad student Jenna Wiens has produced several big data-influenced papers on infection risk. Specifically, she was able, through machine-learning to process several independent variables, such as age and vital signs, to find those with a higher risk of the intestinal bug Clostridium difficile.
Along with tracking risk of infection in admitted patients, it can be useful for a network of nearby hospitals to develop models of disease spread. Sandy Pentland of the MIT Media Lab and bigdata@CSAIL addressed this by adjusting data mining techniques from mobile sensors to fit the epidemiological world. A study produced by Pentland with the help of his students diagrammed the spread of the flu across a group of MIT students by culling both social media and mobile data. The paper, which won the best-paper award in last year’s International Conference on Social Computing, Behavioral-Cultural Modeling, and Prediction, marks a step toward a larger modeling effort that would help hospitals help prepare for upcoming flu epidemics in certain regions well in advance.