Blackford Middleton, MD, first came across the term “clinical decision support” in 1983.

That’s when he was in medical school, at SUNY Buffalo, discussing health data management. “I thought that for the medical students and doctors it was ridiculous that we were carrying around clipboards to transfer data between hospital information systems,” said Middleton in a recent interview with Government nHealth IT. Today, Middleton is chief informatics officer at Vanderbilt University Health System and chair of the American Medical Informatics Association board.

Even back then, the year before Apple’s first personal computer went to market, he and his classmates thought it would be worthwhile for information systems to help clinicians with the complexity of diagnosis — and that endeavor continues.

As Vanderbilt and other institutions try to advance one of the most ambitious applications of clinical decision support, the field of pharmacogenomics, at both the molecular and computational levels, in the short-term Middleton is hoping the federal EHR incentive program can help raise the national baseline of evidence-based healthcare by making clinical decision support systems more accessible, intelligent and interoperable.

[See also: AMIA: Regenstrief pumps up its clinical decision support.]

Middleton, who spent more than a decade at Partners HealthCare before landing at Vanderbilt, has had a front-row seat in the evolution of clinical decision support (CDS) systems — and a hand in developing them.

After med school, as an informatics fellow at Stanford in the early 1990s, he worked on the Quick Medical Reference DT system, through which clinicians entered symptoms and received advice on 700 disease profiles.

“We found it could make basic diagnoses pretty well, near to the level of a medical student perhaps, but not the expert clinician,” he said.

It had a number of limitations that continue in CDS to this day, including for what some consider the most advanced intelligence system, IBM’s Watson.

“Reasoning in computers is hard. They don’t know what they don’t know,” Middleton said. “Diagnosis is filled with nuance. That remains a fundamental challenge. It’s much more important to get the knowledge base right, because reasoning methods will evolve.”

Watson and other artificial intelligence systems are capable of learning from mistakes, though, and Middleton thinks the healthcare community should do the same when it comes to CDS systems and health IT in general.

[See also: Study: Clinical decision support reduces unnecessary imaging.]

“The idea of implementing EMRs across the country is all well and good, however, there seems to be a fundamental policy gap,” Middleton said. “We’re asking clinicians to implement CDS, but we’re not helping them with the actual knowledge base.”

For one thing, there’s the current lack of interoperability, which is a problem for medication decision support systems, among other areas. “If my records are actually interoperable, then when I see the patient at hospital a, or clinic b, I’ll have the patient’s full med list. We are trying to get fill history from Surescripts,” he said, referring to a pilot project at the CDS Consortium he heads. “But that’s still just an idea.”

Then there’s EHR design and incorporating CDS systems. “It’s impossible to imagine every which way you might see a patient, so it’s hard for EMRs to design templates because every patient is unique,” Middleton said.

In the second phase of the meaningful use program now getting underway, providers need to have a system to check for drug-drug interactions and allergies, as well as five CDS interventions related to at least four clinical quality measures.

That’s more ambitious than the first stage, but Middleton thinks the program could go further. “My concern is we’re scratching the surface. Most end users are not equipped to do the translation of guidelines into CDS. The average doc, there’s no way he’s going to do it him- or herself.”

For CDS in phase 3 of meaningful use, Middleton argues that regulators and stakeholders “need to make sure that the knowledge base is available in a public and free way,” with rules systems “that can be used in EMRs across the country. “

On that front, a number of providers, vendors and researchers participating in the CDS Consortium — such as Oregon Health & Science University and the Regenstrief Institute — have used the enterprise clinical rules service that Middleton helped developed at Partners.

The challenge, even beyond meaningful use, will be scaling those types of open CDS rules. One pilot study last year suggested a feasible path to bringing CDS rules to clinicians: the cloud.

CDS systems may be able to evolve into what Middleton calls “analytics for what-if questions,” through a sort of “real-time epidemiology.”

Clinicians and their patients, especially those facing complex health problems like cancer, with multiple treatment options, will be able to get an accurate sense of how similar patients responded by asking a system, “What happened to patients like this over the past five years?”

This story was first published on Government Health IT.