President Obama’s 2016 budget includes a $215 million investment in research on personalized medicine to provide clinicians with new tools, knowledge, and therapies to select which treatments will work best for which patients. That figure includes $5 million for the Office of the National Coordinator to support the development of interoperability standards and requirements that address privacy and enable secure exchange of data across systems.
Last week I had the opportunity to interview A John Iafrate, M.D., Ph.D., founder and director of the Center for Integrated Diagnostics (CID) at Massachusetts General Hospital, about some of the informatics challenges his organization faces as personalized medicine takes off.
The CID was one of the first centers to look at large panels of genes in cancer to support clinical decision-making, Iafrate said. It looks for mutations and other genetic alterations in patient tumors with the idea of getting those patients on new targeted agents. Its SNaPshot assay screens for well over 100 cancer-associated mutations that have important clinical implications. Iafrate’s organization has begun using the HealthShare health informatics platform from InterSystems to target issues involving large data set management and cross-organizational collaboration in support of genomic research and clinical innovations.
At Mass General, once a tumor is genotyped, the patient’s oncologist receives that information in a plain-text report in the EHR. The oncologist can act on that information if they have a clinical trial open or a drug available, Iafrate said.
I asked him how the oncologists keep track of all the available trials.
“In fact, one of our first projects with InterSystems is a clinical trials locator,” he said. “That is app No. 1.” An oncologist who sees 100 patients, all with different genetics, cannot keep track of it. “If I am in an academic practice group, maybe there are 50 trials. Someone could make an Excel spreadsheet of genome types and trials available,” Iafrate explained. “But how would I know next Wednesday, when I see Mrs. Smith, whether or not she has other clinical parameters that make her ineligible? But a piece of software can have all the entry criteria, know the lab values for all the patients, and in real time know the genotype and entry criteria for trials and whether there is a spot available in those trials.”
Iafrate says that there seems to be some consensus that this “apps” model is the approach of the future. “To get novel analytics, you need a stable database structure and then let people build reliable apps you can put on top,” he said. “That is what we are excited about. I think most people would view that as the most efficient way forward.”
He said InterSystems has helped solve a lot of the problems around data security and data formats. “One of the reasons we liked InterSystems is their focus on building HIEs,” he said. “This is not a research project. We are dealing with identified data that needs the highest level of security. The capabilty to share between sites is critical.”
There are still many informatics issues to address, he said. “How do we get data out of the current data repository and how do we share data between institutions in a safe way that limits the risk?”
There are big macro-issues with genetics, he added. “In this day and age, when we can sequence a genome, is any data de-identifiable? You can de-identify some clinical data, but if you have DNA sequence linked, that is no longer de-identified,” he said. “There is no consensus on how to deal with this issue,” he said, and no national consensus within the healthcare informatics world on how risky someone’s DNA sequence is.
Iafrate said another challenge is all the unstructured data in healthcare settings. “That is the major issue we are dealing with,” he said. “As good as any natural language processing software is, there will always be data quality problems.”
He said the CID hopes to create a physician portal — not just a viewer, but a way for clinicians to generate their notes in a way that is as fully structured as possible. “To do cutting edge research and cutting edge clinical analytics, you really need the highest quality data possible,” he explained, because every data point will have noise associated with it. You can have a physician’s note that says ‘Mrs. Johnson has been receiving chemotherapy and is doing fantastic. She feels great.’ If you want to do research on quality of life, natural language processing will hone in on it, but there is noise associated with it. “What you really want is a scale of 1-10,” he said. “We want to build into a physician portal a way they could enter data that is as high quality quantitative data as possible.”
I asked Iafrate if it was likely that EHR vendors would soon start to build in tools that support genetic data sharing. “Definitely, everyone is moving in that direction,” he said. “Epic has a working group around that. Everyone understands that personalized medicine is important.”
Iafrate is working with the Global Alliance for Genomics and Health, which was formed in 2013 to create a common framework of harmonized approaches to enable the responsible, voluntary, and secure sharing of genomic and clinical data. He said most of the work in genomics has been done by a few large research facilities. “They have an interest in sharing data among large genome centers but not in sharing it widely with community hospitals and primary care physician practices,” he said. “They can agree on one or two large databases they share with each other, but that does not solve the problem of how we democratize it,” he said. “Without standards, you are limited in transporting data and comparing studies. We won’t get companies like Epic to invest a whole lot unless there is a standard format.”
Iafrate said that once data is structured sufficiently and a single database can store large amounts of genetic data and can bring it together with clinical data, then “the sky is the limit.”
“We could create real-time clinical analytics apps that you could put into Epic or another EHR, he said. One future app could be called “Patients Like Mine.”
Here is how Iafrate explained it to me: Twenty years ago oncologists would rely on their medical knowledge and experience to make decisions because they didn’t have so much data. Now when the genetics results come back, they are complicated. “Can we help those clinicians by showing them real-time survival rates?” he asked. “How can you generate a Kaplan-Meier curve, a survival curve, for the patient sitting right in front of you? This is not a research tool, but a clinical real-time tool.” What if you had structured data on every time the patient came in, what drugs and dosage they had, and a CT scan measurement of the size of the tumor at each point in time. You could do a quantitative measure of drug response in that patient — and that is the equivalent of a clinical trial, he said. “Today that is not done in routine clinical care, where you quantitate the response rate or tumor shrinkage, because there is not a need for doing that in the clinic.” But now there would be a reason. If you measure the tumor size of every patient that comes through, the oncologist sitting with that patient could pull up a Kaplan-Meier curve of all the patients in their practice and say ‘query the data by defining 50-year-old females with this mutation and this tumor type. Tell me how my patients have done.’
And providers could toggle between looking at only their own patients or patients in the HealthShare HIE network. “Once you structure that data, if you can de-identify it to some degree,” Iafrate said, “then it could be shared and turned into something really special.”