Data-driven health

Data is critical. Not for its own sake, but to simultaneously inform three feedback loops of health:

1. Self-care
This is often a matter of helping people answer the question ‘what works for me?’ Data can make visible what one might not otherwise be able to perceive. For example, it is not straightforward to confidently detect 10 per cent less pain, and so know to stick with a new behavior or therapy. Or that better sleep hygiene really does give one better quality of life.

2. Clinical care
For example for drug and treatment plan adaptation. The statistical clumping of traditional research studies hides important variability. As my colleague Curtis Cole points out, once you are on more than two medicines, there is no evidence base.

3. Evidence
Today, we can affordably and scalably generate evidence where there has never been before by engaging tools such as ResearchKit, and Research Stack for Android users.

What type of data are we talking about?

We can do a tremendous amount with what is already captured. We still need traditional clinical data, available from large populations of patients, such as insurance claims data, Electronic Health Records, and suchlike. However, the new opportunity is in the individual scale ‘small data’ that is captured and analysable from our mobiles and other online digital interactions.

The richest current example is activity and location, which tells much more than just if you have reached a targeted ten-thousand steps that day. Trends like shifts in location patterns over time (what time you leave for work, hours spent out of the house on weekends) can provide early signs of relapse, diverse side effects, or an indication that the patient really is improving steadily.

We can add to this our interaction in and through other applications, whether it’s detailed typing patterns or detection of laboured breathing or pressured speech, as well as the digital traces of other kinds such as what we buy, the language we use, and the self-medication we do through excessive Netflix binging. These are all sources of near continuous real-time signals that can close both behavioural feedback loops and clinical ones.

Some things we cannot detect and need to ask the patient their perception. Even there we can ask in new and personalised ways rather than generic text- based questions.

Finally, new and sophisticated wearables will, over time, offer up rich physiological signals and a convenient interface.

Making sense of the data

All this raw data is full of noise and confounding influences. It is useless without the statistical techniques and models to interpret them. To make this raw data actionable, we have to move up the information food chain to summarise, filter and fuse these data into measures of patient function. Establishing those analytic techniques means undertaking clinically anchored studies which to develop and iterate across conditions and individuals.

The key to this is to work together, collaboratively sharing data and evolving analytics. This requires a collaborative community, and the open architecture that supports it. A key role of Open mHealth is to be such an infrastructure. Just as with the internet and Web, modular open architectures can float all the boats higher and clearly don’t in any way interfere with commercialisation, as evidenced by the growth of the companies such as Google.

Focus on patient need, not institutional convenience

There is one final piece of the puzzle. At a meeting recently, someone asked me why, with all the years of work already put into clinical informatics, did I still have such great expectations for this area. My response was that we have a new shot at transformation because of the growing focus on creative, adaptive, smart technologies designed for patients to solve patients’ problems rather than starting with a mind-set that we are solving clinicians’ problems.

Pulling this together, we need data, made meaningful, for real people. For example, GPS handheld devices existed for years, but they were mostly used by the military and some nerds. Now most of us use Google maps - an app that is informed by our small, personal, location data. We can see health data make the same transition in coming years.

Author

Professor Deborah Estrin

Professor Deborah Estrin, Cornell Tech and co-Founder, Open mHealth.