Paul J. Chang, MD, loves live jazz, hates the ballet—and extends his taste to two different means of wringing useful business intelligence from medical imaging informatics.
“If I go to the ballet three days in a row, I will see the same performance every day. It’s an example of choreography,” explains the vice chair of radiology informatics at the University of Chicago. “The dancers respond to a certain broadcast signal, just like an HL-7 or DICOM message—the beat of a drum, the change of a key—and they jump in the air knowing that someone will catch them. Every single time.”
He’ll pick a jazz concert over the ballet every time, he said during a presentation at RSNA last November, because the jazz will give him a different experience every time even when the same songs are played. That’s because a jazz concert is not choreographed but, rather, orchestrated.
“When you look at other business verticals”—that is, those outside healthcare—you see that, “without exception, they’ve gone beyond the primordial ooze of hardwired choreography to orchestration,” he said. “And by committing intellectual arbitrage, applying orchestrations that have been proven in other industries and applying them here, we can take advantage of a great risk-mitigation opportunity.”
Chang, who cited Amazon as his primary example of a master orchestrator—“I guarantee you my Amazon landing page looks very different from yours, because I buy different things from Amazon than you do”—presented “Interoperability and Integration—From HL7, DICOM and IHE to SOA,” on December 1, at the 2014 meeting of the Radiological Society of North America in Chicago.
“This is an opportunity for us to understand that we should be a little less arrogant about what we do with health IT,” he said. “We have to understand that we’re actually years behind other industries.”
Chang described how service-oriented architecture (SOA), and homegrown alternatives to it, can help radiology practices orchestrate integration and interoperability among and between disparate information systems—including those that don’t natively “dance with” each other.
Loose coupling, tight connections
The end game of all the computing, said Chang, is supporting imaging-informatics workflow to provide better, safer patient care at lower cost.
There’s actually nothing new about the concept behind SOA, said Chang, adding that early iterations were referred to as, for example, end-tier architecture or middleware. He likened the process to the human anatomy. “The concept is, instead of every cell broadcasting to every other cell, you have pipes that consolidate or aggregate information to a central bus or spinal cord,” he said. “Eventually, over time, that spinal cord develops a ‘brain.’ We call that business intelligence analytics.”
Enter orchestration. Instead of using only one way to consume information in the EMR—through the EMR client—information is extracted from sources throughout the enterprise. “We can make all of that usable and repurposable so that we can mash up an appropriately localized, idiosyncratically relevant experience,” Chang said. “That’s how our brain works. That’s how our spinal cord works. That’s how SOA works.”
Here Chang stated that healthcare is rife with major misconceptions about SOA. “When I go around the world and talk to CIOs, a lot of them say, ‘Oh yes, we have SOA.’ And they don’t. What they have are a whole bunch of web services.”
He said that, unlike web services, SOA is component-based architecture that supports composite applications—mashups—that are created by orchestration of loosely coupled services that are universally exposable, self-improvising and consumable. SOA, he added, requires disciplined governance, security, semantics, quality and service.
By contrast, “web services is the flavor of the month,” said Chang. “It’s an implementation technology.” You can do SOA using various transfer protocols, he said. “It doesn’t matter how you do it. The key point is loose coupling.”
‘Let the kids play’
One advantage of loose coupling in radiology is that end users can enter and access data without damaging data integrity, making the process attractive from a security point of view, said Chang.
“I can let the kids play, build whatever they want, without fear that they’ll screw something up,” he said. “That’s very different from today. Today when I want my EMR to just move something just over to the right, there’s a potential of breaking the system or the database. Why? Because it’s tightly coupled with the data schema. With SOA, you have a loose coupling.” Outside of healthcare, he stressed, “that’s basically how most of the world works.”
Having driven home that latter point, Chang offered something of a confessional. He told how, years ago—after observing SOA in other industries and seeing it succeed at the University of Chicago—he preached it far and wide as a must-do.
“I am much more chastened now,” he said. “I know that SOA is very hard [for many healthcare enterprises] to do because it’s not something you buy, install and support. SOA is an architectural commitment. It is a governance commitment. It’s not challenging from a technical perspective, but it is challenging from a governance perspective. It requires significant cultural change. And so within the few years I’ve been giving talks and saying ‘You must do SOA,’ I have basically come to where I’m now saying, ‘You know what? You probably won’t be able to do SOA.’”
The good news: There are alternatives.
Cookie-cutter equals commoditized
Chang talked about some ways a sort of virtual SOA can be set up using, for example, an off-the-shelf state aggregator appliance. “You can put in an appliance that basically can extract information into a lot of what SOA can do, but without the formal commitment of governance and a defined architecture,” he said, although he did not name any such appliances by make, model or type. “It will get you to almost 90 percent of what you would do with SOA.”
Chang wrapped up by describing do-it-yourself success stories at his institution, including a mashup for breast imaging that uses a webpage portal and an SOA-based protocoling tool to save time accessing clinical context.
“How many of you deal with your EMR?” asked Chang. Most hands in the room went up. “I could go anywhere in the world and the EMR, once it’s up, looks exactly the same. I can go to any PACS implementation and it’s identical to everybody else’s. This is cookie-cutter. By definition, if you use cookie-cutter tools, you’re cookie-cutter yourself. There’s another word for that: commodity.”
“We need to move from this model where we are being commoditized and peripheral-ized,” Chang insisted. “You should have tools that allow you to go beyond commodity-level service. And that’s the key: One size does not fit all.”
“The key point is, BIA—business impact analysis—requires [some version of] SOA. It requires the ability to arbitrarily extract information from various sources and orchestrate something useful.”