It’s a beautiful morning here in Vancouver!
First order of business
Find the press room, get my badge and get online! Somehow couldn’t get my ARVO account activated correctly on current email address but that’s all fixed and their handy-dandy app will tell me me exactly how many minutes to get from where I am now to wherever the heck that course is…
Big Data
Delectable, simply delectable. I have dreams of what big data could do in the dry eye world. It’s far too immature as a science for the data to be all that useful for research per se (for example we are nowhere near having core outcomes established) but it doesn’t have to be completely ready for the big-time for it to be useful. Trends identified in big data could point the way to suspicious gaps, could validate known problems, could help patients contextualize their experiences, could help patients formulate questions to ask in order to more effectively advocate for themselves, long before the data and the big data analysis methodologies are ready for it to become a useful adjunct to clinical trials.
I can’t geek out too much about the presentations this morning because I’ll get hopelessly behind - past experiences have taught me that if I want to sleep between now and Wednesday, I have to both blog during the day and keep it snappy. Not that I know enough to truly geek out anyway, but I do like pondering stuff.
Incidentally, I know perfectly well I’m writing a blog post no one will read, but writing helps me sort out and summarize my own thoughts about what I heard, and that’s good enough for me today.
Laura Balzer (UMass) gave the opening presentation. A lot of it could best be described as approaching a room full of highly sophisticated balloons with two fistfuls of even more sophisticated pins. Kablooey. You think you’re a scientist? Watch her find and expose all the wrong assumptions you may be making about the soundness of your approach to big data. Among the many take-home messages: Big data does not solve any of the usual fundamental research problems (eg correlation vs causation), and actually causes new problems. For example, you might be really surprised at what’s actually lurking in what seem to be fairly trivial missing data, so don’t get ahead of yourselves.
Michael Chiang (Casey Eye OHSU) brought the discussion into ophthalmology. He highlighted the problem of the very, very long lag time between when science discovers things and when the new knowledge actually changes patient care. (It’s 17 years.) Then showed us how they tried to use ‘big data’ to improve patient wait times in their clinic. It was really kind of fun to hear about - not just the technical parts but the extent to which data analysis contradicted opinions. For example, some patients take a long time right? (And did you know that they know who you are, by the way?) So do you schedule them at the beginning of the day (which is what they used to do but then they discovered through big data how badly that threw everything off for the whole day) or at the end of the (in which case you piss off the entire stuff because nobody goes home on time)? Big data helped them optimize between not messing up the schedule and not having a ridiculously long day.
After data analysis, he went into the question of data registries like IRIS (covered later) and more twists on the notion that observation changes things… in this case, that measuring things improves quality, as they’re finding in electronic medical records (paper is in press now about this).
Joshua Stein (UMich) then got into the real meat of the matter as regards big databases and some of the issues going on. He started with the medical claims databases. I loved this because we’ve been seeing more studies published lately in dry eye that were based on claims and I wonder sometimes about the implications and limitations. He outlined the key issues and challenges with this type of data, such as a how “messy” it is (80% of medical data is unstructured, for example, text descriptions that you can’t always accurately extract data from automatically. He talked about the areas of research that this type of data is useful for, and what it’s not useful for. One of the problems is that everything’s based on insurance codes and there are a lot of issues with the quality of the data.
Next source he mentioned is the American Academy of Ophthalmology’s IRIS database (covered by Dr Coleman below), and last, his SOURCE database project, which currently has about 300,000 patients. It’s a collaborative thing that taps into EPIC, software removes patient data, pools it with diagnostics, test data, genomic data, etc, then makes it available to participating institution researchers. His bottom lines were that the data, to be more useful, needs to move beyond insurance codes to being disease-based, and needs to be more granular. He sees the role of big data as preparing the way for clinical trials - learning as much as possible before the huge investments in those trials. Audience member pointed out that it could also be hugely helpful for clinical trial recruiting.
Christopher Hammond from St Thomas’, London went into the big data genomics in biological research. Super interesting, just not to me right now. I was mildly amused when an audience member attempted to raise the specter of patient privacy. Her hypothetical: “If an employer knew that based on your genomics you are likely to have X eye disease within 10 years, they won’t hire you. How do we prevent this?” did not elicit a more nuanced response than “Well, they agreed to let their data base used.” Ah well.
Anne Coleman (Jules Stein/UCLA, also president-elect of AAO) then presented on the IRIS registry, which is a database system created in March 2014. They have data on 53 million patients now and it’s integrated with 55 different electronic health records systems. The goal of IRIS was described to be closing the loop in the process of establishing care standards, so as to substantially shrink that 17-year gap we talked about as taking place between discovery and new care standards actually being adopted. As she was presenting data on glaucoma, I couldn’t help but think (warning: soapbox) about how long we have had data on the harmful effects of benzalkonium chloride in glaucoma medications and yet clinical practice is moving at a glacial pace towards broader use of either preservative free formulations or less toxic preservatives in glaucoma medications. Sigh. Anyway. One of the many points she made was that big data can help in research by finding small, statistically significant differences that can only be found in very very large data samples.
Dr Coleman talked about all the different research opportunities using IRIS, including (1) IRIS registry analytics teams (5 teams in process or completed?) (2) Joint project between Research to Prevent Blindness and IRIS (3) Hoskins Center IRIS registry research fund (4) American Glaucoma Society IRIS Registry research initiative and (5) Knights Templar Eye Foundation Pediatric Ophth IRIS reg research fund
In terms of impact, there have been several studies published already with IRIS data, and it also has the potential to influence policy - she mentioned an example of the FDA wanting to restrict use of drug bevacizumab (because of compounding errors) but they were able to demonstrate with IRIS numbers how small the actual impact was so the FDA didn’t. (Incidentally I noticed later on in the poster session that there was a poster about using this drug successfully in PROSE devices to treat neovascularization.)
An audience member posed the question of whether she could envision a world where patients have some kind of access to this data, for example, to see data on their provider in a greater profession-wide context, and the answer was a simple no. Another asked whether there could be synergies between what AOA (i.e. optometry) collects versus AAO’s IRIS. Answer: It’s something we’re interested in.
Dr Chiang wrapped up the session by saying that bringing together different people from different fields is what advances the field, and that that is his vision for the role of big data.
Moving on!
Blog post #1 at ARVO… somehow this does not appear to qualify as “keeping it snappy”. Next!