Aaron Sorensen is director of informatics at Temple University School of Medicine of Philadelphia, PA.
Tell me about your job.
I’m at Temple University at the School of Medicine with an affiliated health system. Our new leadership is keen on creating a robust infrastructure to support clinical research. I’m heading up the informatics aspects of that.
What is the informatics influence in the School of Medicine?
Within the health system, you have the IT shop that runs a myriad of clinical systems. There’s a feeling from the researchers that all this data exists, but it’s hard to get at. What do you do with it once you have it? What are the appropriate safeguards regarding compliance and privacy?
The School of Medicine is trying to make it so that every time a clinical researcher wants to ask a question of the clinical data, it doesn’t become a maze that you get lost in, with different people are telling you different things. There’s this straightforward way to do it and you can go to a central team of people that will guide you through the path and help you along your way.
Describe how PCORnet came about and what it does.
My understanding is that over 10 years ago, when the NIH was originally thinking about redoing the way they fund clinical research extramurally at academic medical centers, the PCORnet idea was floated. The feeling was that it would be costly and it would be hard to achieve. They had other priorities, so instead of doing that, they funded the CTSA awards.
PCORI, the Patient-Centered Outcomes Research Institute, is not a federal organization, but it’s funded through the Affordable Care Act. It’s federal dollars, but it itself is a independent non-profit. The feeling was that it was worth pursuing the idea of creating a network of hospitals that have the ability to share de-identified patient data for the purposes of clinical research.
Although they have grants that fund all different kind of things, just like the NIH does, I believe the crown jewel within the PCORI portfolio is PCORnet. It has 29 funded groups, some of which focus more on general health system patient populations, whereas others are more focused on particular patient groups with specific diseases.
What Temple systems are contributing data to PCORnet?
In terms of our electronic medical record, we’ve been on Epic outpatient for about three years. We’re just now kicking off the project to go with Epic inpatient. Epic, as most EMRs, receives a number of feeds from different systems. When you get to the back-end Epic reporting database, you not only have the data that originated in Epic, but from a number of different systems.
For our contribution to PCORnet, we are only using our Epic back-end database that gets feeds from cardiology systems, pulmonary medicine systems, and billing type of data. It’s a wide range of things. For the purposes of this project, we are only using what comes into our central EMR.
Can researchers query data from any or all of those 29 contributing organizations?
Yes and no. The 29 break out into two groups.
The patient-focused ones that are disease specific are called PPRNs, the Patient-Powered Research Networks. The health system ones, of which Temple plays a role, are called CDRNs, or Clinical Data Research Networks.
I don’t know 100 percent what the PPRN plan is. I think it’s slightly less ambitious than the CDRN plan of which I’m a part. I can speak to the 11 funded groups that are part of the CDRN and that cross the country.
There are two aspects to the PCORI contract. Our network is the University of Pittsburgh, Johns Hopkins University, Temple, and Penn State Hersey. Within our network, we have been funded to create the ability to share data for two different diseases. One is rare disease – idiopathic pulmonary fibrosis. Then a more common disease, for which we chose atrial fibrillation.
At the CDRN level, at the national PCORnet level, we have to support two cohorts. One is what they originally called an obesity cohort, but then they decided they wanted to expand beyond people who are already obese to include people who are at risk of becoming obese. They’re now calling it the weight cohort. We’re going to support a weight cohort.
Then we have to have a randomly chosen one million plus patient pool from which PCORnet can do centralized queries. Each of the 11 groups has to make available at least one million randomly selected patients, or else their whole patient population, for these centralized queries. As well as a subset of that which will be used specifically for to measure issues around obesity. For that group, you have to have collected good data on weight, height, calculating BMIs, and things around diabetes, coronary artery disease, and certain co-morbidities associated with obesity.
Do researchers have to file paperwork for what they’re looking for? Can you tell how they are using the system?
Yes. Within our network, we have IRB protocols that have been set up to allow for the researchers to ask certain questions. That’s specified ahead of time and is pretty locked down.
For PCORnet, they have the ability to ask anything. The data is always de-identified. You’re not typically ever sharing patient-level information. You’re aggregating it so that they can get an understanding within a given population how it breaks out — what the demographics are, what the prevalence or incidence of a given disease is, etc.
For those questions, they are not pre-established. It’s not like at the beginning of the project that we know, “We will ask these 100 questions over the next year and a half.” Each funded site will have the ability to not respond to a given query, assuming that they have good justification not to do so.
The advantage to the researcher is that they might need to reach outside of Temple to identify a patient cohort large enough for their project, right?
Exactly. For our rare disease, idiopathic pulmonary fibrosis, at the time we submitted the grant, we estimated that we only had about 70 living patients with that disease. If you went to Pitt, which was the highest, they maybe had about 350 or so.
With only 70 patients, maybe you don’t even have the number to show any statistical significance in certain differences between drugs or other interventions you’re trying to assess. Whereas if you were to combine all the centers together and you get above 500 patients, then all of a sudden potentially you have the ability to make a finding that will stick with the general population.
Is there a plan to add organizations or conditions or to use the data more widely?
Yes. We were initially funded for 18 months. That 18 months is supposed to be used largely to build an infrastructure to support future research. There will be some research done during the 18 months, but the idea is to make sure you can set up this robust network for the future.
PCORI has said that they will be having a Phase 2 in which no longer will they be paying to help you set up this infrastructure, but instead they will want specific questions answered. You have the ability to then apply for Round 2 funding, in which you will potentially participate in clinical trials where, using the network, you identify certain patient profiles and you go out and enroll them in certain studies, or for large-scale retrospective studies, where you harness the power of the longitudinal data you have for your one million plus cohort of randomly selected patients times 11.
So at least 11 million patients that you can then query to say, over the last 10 years, patients with this profile who were given this type of therapy, how did they fare over the last 10 years compared to this other therapy? There will be a Phase 2 where we can extend the funding to actually try to answer certain questions.
In terms of being awarded the contract, everyone was being asked, to what level is your institutional leadership committing to making this sustainable over the long run? Should the money dry up tomorrow, do you have strategies and do you have commitments from your top leaders to make sure that this stays in place and that you extend it to anyone outside of the network so that any non-funded investigators have the ability to ask any center and consortium … my consortium is called PATH , the initials of all the participating institutes. Geographically, we’re the mid-Atlantic CDRN. So anyone in our geographic area who is not at a funded institution has the ability to request access to our data and to collaborate with any of our investigators on any particular study.
Is there anything else you’d like to talk about?
The one really neat thing that’s come out of this that’s linked to PCORnet is the use of i2b2. It stands for Informatics for Integrating Biology in the Bedside. It’s an open source software package created at Boston Children’s. It is used extensively throughout the Harvard-Partners HealthCare network. It allows you in an open, non-proprietary way to take data out of any clinical system, merge it with other data you might have – such as genetic data from other systems — and to make it queryable, both at your institution or potentially teaming up with other institutions. The adoption rate has been growing by leaps and bounds.
Temple was not an i2b2 user before this initiative. While we are implementing it for the purposes of PCORnet, as are many of the other CDRNs, we also are using it as a springboard to create an internal tool that our investigators can use for any patients of any disease asking potentially any questions using the EMR data.
A lot of times when an institution implements a new clinical data warehouse, they take their time and go step by step. It evolves over a period of years, potentially. Whereas because of this PCORI initiative, we had to go from zero to 60 quickly. Phase 1 lasts 18 months, and at the end of 18 months, you have to show that you’ve successfully created this infrastructure which can be used for robust clinical research.
The i2b2 prevalence within academic medical centers over the US has been growing. As I dug into it, I realized that people use it in different ways. If you are trying to share data with another institution via i2b2, one approach is to try to convert all your data to the same standard. If you have internal lab codes and the other institution has their own internal lab codes, you could try to convert all your codes to a standard like LOINC. Or, you could allow them to stay as they are and then you have some lookup table that converts on the fly from your local ones to a standard.
As I was experiencing this and going through the baptism by fire of getting our institution using i2b2, not only for PCORI but for ourselves, it became clear that there should be a boot camp that helps you think about all these things. It needs to give you what I call the mental scaffolding, so that from the beginning of a project, you can consider all of the types of decisions you’ll have to make and the potential downstream ripple effects.
I contacted Harvard, the folks that created i2b2 and the accompanying SHRINE software that allows you to connect other institutions. I gave them some ideas about how it would have been great if I had been able to take this intensive boot camp before our project started. We went back and forth and we’re going to offer a pilot i2b2-SHRINE boot camp at Harvard in early 2015.
Harvard is trying to assess what type of a demand would there be for such a boot camp after the pilot. We’ll try to fill maybe 25 spots with the pilot, but then whether there is enough hunger and demand to offer it regularly. If any of your readers have any thoughts about that, I’d love feedback in order to gauge whether it’s a minor niche thing or if it has wide applicability.