The Emerging Field of Health Data Science in Boston | Hacker News Reader

The Emerging Field of Health Data Science in Boston

insighthealthdata.com

96 pointsre_jena15 comments

Strapping a fitbit to everyone so docs can better monitor their life choices reminds me of the logic behind a classic Simpsons scene: A more precise measurement of an already accurately-understood condition does not equate to greater understanding of how to actually fix the problem.

"Marge: Can't you do something for him?

Dr. Hibbert: Well, we can't fix his heart, but we can tell you exactly how damaged it is.

Homer: What an age we live in."

I think it's a little more optimistic than that. Recently there was a Kaggle competition that analysed enzyme make-up of saliva of animals in order to predict epileptic seizures (I might be mincing specifics here but the point remains). So while it's true that we won't be able to treat some of the things that a fitbit etc. highlights (bad decision making, cancer, etc.) it can definitely potentially be used in preventative diagnoses and early detection warning systems. What an age we live in indeed!

I worked with a couple of the Insight Data Fellows mentioned in this blog post, on the UCSF Health eHeart study. Mike Klein (Fitbit data to predict heart failure) and Yancheng Liu (Apple Watch data to predict atrial fibrillation) both did a fantastic job!

If anybody here has questions about the program or the projects, happy to give the "customer's" point of view on what it's like to work with Insight.

A bit late but can you tell me how exactly walking patterns change when you are about to have heart attack? I think it is all meaningless without a described mechanism that can be tested in a scientific way. Big data analyses are nice to aid in hypothesis generation but until a testable mechanism behind the observation is described, to me the "mechanism" is just as likely to be over fitting as real. The FDA also wants mechanistic explanations before approving applications.

When looking at enough parameters, there are always parameter subsets that nicely match the phenomenon you are looking for. Whether that is a heart attack or a running nose.

I don't think data science for health is in any way limited to wearables.

I think this is the first time we've had the opportunity to do real data science at a massive scale with patient data of all sorts, including electronic medical record (EMR) data. This is because we now have:

- Widespread use of EMRs

- Decent standardized schemas for representing that data (i.e. FHIR https://www.hl7.org/fhir/)

- Not-terrible medical ontologies (i.e. RxNorm for drugs, LOINC for labs, Snomed for concepts etc.)

- Reasonably accessible tools to process the massive datasets (<insert big data toolkit here>)

Many of these things emerged or matured fairly recently, so I think we're going to see a real explosion in medically relevant data science. Fitbit/wearable sensor data is only the beginning, I think things like meaningfully informing diagnoses of patients based on medical history beyond what a doc or nurse could do on their own is within reach, and the sky's the limit!

(Full disclosure I work for a health DS startup, and was an Insight Fellow, although prior to the health program existing)

We're getting there. We still have huge problems on the data input side. Standards like fhir might be okay (though limited). Problem is most of the data we would love to analyze is never input as discrete data in the first place. We are far too dependent on the free text note. Also, such data that is available tends not to be available through, say, an API call, as a FHIR message...yet

I think we're just scratching the surface when it comes to the types of health data we can retrieve. There is a lot of power in bringing together data sets from the sensors/apps (more of a continuous feed of wellness data) and correlating that with discrete medical data models.

The startup I work at is stitching together all of this data to gain a better profile or health state of a person, and it's pretty amazing some of the early insights we've been able to uncover.

"What if you could pick out early warning signs of heart conditions out of somebody's Fitbit data? It turns out that you can. "

Yes, in the future everyone will wear a smartwatch:

https://h4labs.wordpress.com/2015/07/28/in-the-future-everyo...

Why wear a smartwatch when we might be able to embed more accurate, useful sensors instead given advances in communication and battery technology?

Probably not in the next few decades. But if you want extrapolate way out into the future, perhaps you don't even need a internal sensors. Give it 2 or 3 hundred years.

What if you could pick out early warning signs of heart conditions just by looking at somebody and observing that they are overweight? It turns out that you can.

So, you're arguing that measuring other data points doesn't add value? Just look at someone and if they are overweight that's enough of a data point?

It's an empirical question. I would predict that you could do pretty darn well with height, weight, resting heart rate, blood pressure, bloodwork, and asking the patient about their physical activity. Fitness trackers are cool, but I'm just arguing against breathless health claims without justification.

This is going to have a huge impact especially when combined with human longevity data science directions.

http://calicolabs.com

http://genopharmix.com

http://humanlongevity.com

Reply on news.ycombinator.com