Sunday, March 08, 2015

Will these truths be handed down from Mt. Sinai?

As I read this person's quote at the start of the article, my thought was that the only thing lacking to achieve success was modesty:

“We’re pursuing problems that are computationally and intellectually exciting, and where there is the potential to change how doctors treat patients in two or three years,” Mr. Hammerbacher said.

The line shows up in a Steve Lohr New York Times story about  Mt. Sinai hospital where:

[T]he goal is to transform medicine into an information science, where data and computing are marshaled to deliver breakthroughs in the treatment of cancer, Alzheimer’s, diabetes and other chronic diseases. Mount Sinai is only one of several major medical schools turning to data science as a big part of the future of medicine and health care. They are reaching out to people like Mr. Hammerbacher, whose career arc traces the evolution of data science as it has spread across the economy.

The impetus comes from other fields:

Chronic diseases, Dr. Schadt explained, are not caused by single genes, but are “complex networked disorders” involving genetics, but also patient characteristics such as weight, age, gender, vital signs, tobacco use, toxic exposure and exercise routines — all of which can be captured as data and modeled.

“We are trying to move medicine in the direction of climatology and physics; disciplines that are far more advanced and mature quantitatively,” he said.

Oh, climatology, where models remain in formative stages after years of research.  Or, physics, where the hoped-for general laws that describe the universe or quanta or both are in constant flux.

Don't get me wrong.  I love that they are trying.  But, please let's be realistic about both the development of the science and the speed with which diffusion of new diagnostic and therapeutic regimes infuse the health care system.

As I came to the end of the article, I found myself hoping that there might be a touch of that modesty after all:

[Mr. Hammerbacher] is optimistic about his initiative’s prospects, but has come to appreciate that the mysteries of the human body may be more resistant to math than finance or social networks are. Today he speaks less about quants taking over than about their lending a hand. “We’re not the most important people,” he said, “but we can help.”


fairhavenhorn said...

It's too bad he isn't trying to advance medicine as an experimental science. It has a lot potential there, and experimental sciences like Physics and Chemistry are much easier to expand. My experience with healthcare is that they are bothered by and have difficulty with the mathematical skills needed to design and understand experimental science in the highly noisy context of medicine.

Observational sciences like Astronomy and Meteorology require much more mathematical sophistication. Very few medical researchers are confortable with the much more complex mathematical concepts needed to design and understand observational studies. The IT folks are usually just as unfamiliar and don't understand how much is wrong with so many of their observational results.

It's not hopeless. There are people who write good books and teach these subjects. But the medical and IT staff typically hate those courses and don't like that kind of work. It's a personality mismatch that is hard to overcome.

(Angrist and Pischke's recent "Mastering Metrics" is an example of trying to present at highly simplified view of the concepts needed. I've seen even this high level overview scare off readers.)

Don't Believe The Hype said...

This NYT article was also troubling to me.

As far as I can tell, the only thing special about this research lab is that its investigator is a wealthy person from the tech industry. I wasn't able to find a single scientific paper that Hammerbacher and his team have produced. What does it say about our media and the scientific community when an investigator who hasn't published anything, much less anything important, merits space in the NYT just because he is affiliated with a "hot" sector in the economy?

The article didn't even articulate what the scientific goals of the lab are. Something about statistics and big data in cancer care? As if that's not what biostatisticians and cancer researchers have been doing all these years?

Medical Quack said...

Everyone missed this part of the story, as he's a diagnosed bipolar and he's working on some very complex models so with all the data selling and analytics that goes on out there, is he medication compliant? It's a good question to ask as his work could vary greatly if he were in a "manic" or "depressed" mode and not under control.

It's bad enough we have folks that are just normal that go off the deep end with modeling. Does he have a secret FICO medication score like the rest of us probably have with proprietary code nobody can verify that only requires your name and address? I think this is important due to what his work load is coupled with the fact that he worked for a banks and Facebook prior, financial and clicking models.

Health insurers have packed their payrolls with Quants too in the last few years, no secret just look at classifieds anywhere, so you have former high frequency and bank quants everywhere doing healthcare. Models are not that different but the difference is "people" attached to those numbers and that's what's scary when you get a former HFT person modeling in healthcare. United Healthcare is the king of that stuff and for making somewhat spurious correlations the Pvaled Morcellator data with c-section justifications and expect to see more of that as it's money drive.

So yeah is this guy medication compliant..I think inquiring minds at some point want to know:)

Susan D. said...

At the end of the day, every patient is cohort of one. We can use data to create general hypotheses, but treatment will always have to focus on that one person's unique biochemistry that is as much a product of their genetics as their environment. If 70% of cancers are ideopathic, data will only be helpful for the other 30%. First rule of epidemiology and biostatistics, correlation is not causation.