Computers and … Feelings? Exploring New Sources of Health Data

Health care is, in its essence, about people – people caring for, confiding in, supporting and listening to one another. Clearly, technical knowledge and skills are crucial, but so too is the ability to “read between the lines.” Good providers must interpret not only what is said but also what is not said, interpreting the feelings that lie beneath the words. That entails focusing on the context and the details, but also being able to fit those disparate pieces into a cohesive, multi-dimensional picture.

In an era when the collection and analysis of data are becoming increasingly important, is it possible to capture all – or even some – of those amorphous pieces of information that are so important to the treatment and care of human beings? Sentiment analysis – the mining of written words for feelings as well as information – is one such attempt. It relies on sophisticated computer algorithms to interpret what a person is saying and feeling through his or her use of specific words, as well as how that person constructs sentences or phrases. A phrase such as, “I would have liked it if the doctor had spent more time with me,” is a good example of how a short sentence can be loaded with both information and feeling. Natural language processing software can extract from those 14 words the simple fact that the doctor did not spend very much time with the patient (at least in the patient’s perception). It can then “bucket” that information with other patient comments about doctors, about the adequacy of the amount of time spent, about providing information to patients. In addition, based on the way the patient strung those words together (i.e., “I would have liked it if…”), the software also interprets the feeling being expressed. It is clear that the patient was disappointed, perhaps irritated, saddened or confused. The patient is also making a suggestion for improvement: patients need more time with their doctors. When these types of facts and feelings are extracted from thousands of comments, the result is a window into patients’ experiences, their expectations and their needs that helps to round out the picture given by purely quantitative measurements.

Similar strides are being made in computer-based interpretation of verbal language. Work being done at Massachusetts Institute of Technology’s Media Lab and by some start-up companies is beginning to mine spoken language not just for information but for feelings. In a Boston Globe article, Joshua Feast, CEO of Cogito Health in Charlestown, Mass., stated that his company’s software can “pull out what we call social signals underlying interactions, hidden below the actual words spoken.”

The software being developed can use the dynamics of speech – its speed, rhythm and patterns – as indicators of states of mind. Is the person depressed, is she interested, is he actually listening to what you are trying to tell him? One of the early uses of the software is by companies providing wellness services; by analyzing their client’s responses, counselors can measure how well they are emotionally connecting with their clients to help them stop smoking or manage their diabetes. They can see if they have engaged clients’ interest or even whether clients exhibit signs of depression.

In both applications – sentiment analysis and the analysis of verbal interchanges – technology is being used to put the personal back into analyses that have up to now been largely limited to quantitative measures. And while those quantitative measures are absolutely critical to the practice of medicine, it’s nice to know that computer technology can now provide additional insight into the “people” side of medicine as well.