UCHealth taps machine learning to boost capacity

LeanTaaS, UCHealth looking at managing OR time and more
May 25th, 2016

UCHealth is using the technology at the heart of Google’s self-driving cars and Netflix’s recommendation engine to help maximize the use of chemotherapy infusion beds, operating room time and, soon, much more.

The technology is called machine learning. It lets computers learn from vast data sets and uncover hidden patterns and insights as they do it. While UCHealth employs no shortage of technical talent, machine learning is a highly specialized field that combines data science and advanced mathematics.

Steve Hess, UCHealth’s chief information officer, says predictive analytics and machine learning can help boost capacity by optimizing operations – a key to increasing patient and provider satisfaction, improving efficiency and delaying the need for physical expansion.

To tap the potential of machine learning to manage complex hospital operations, Steve Hess, UCHealth’s chief information officer, turned to Santa Clara, Calif.-based LeanTaaS for its expertise. The company’s chief technology officer led the cloud computing team at Yahoo as well as distributed computing at Oracle. LeanTaaS’s team of data scientists and mathematicians quickly customized the company’s iQueue data analytics software for University of Colorado Cancer Center’s Infusion Center. The system went live in October 2015, analyzing the Infusion Center’s operations.

The results were striking. Patient wait times plunged 60 percent during peak hours and were an average of 33 percent lower throughout the day – despite increases of 16 percent and 7 percent, respectively, in patient peak-time and overall volumes, Hess said. There was also a 28 percent drop in Infusion Center staff overtime hours.

“So it was win-win-win,” Hess said.

Complex puzzle

The team, including Tom Gronow, chief operating officer of University of Colorado Hospital, considered other applications for the iQueue software and settled on operating rooms, where freeing up capacity, minimizing patient wait time and optimizing surgery-team time is a constant challenge. ORs are a prized hospital resource, too: Each percentage-point increase in a single OR’s utilization adds $100,000 to a hospital’s bottom line, Hess said.

There are 38 ORs at UCH alone, Hess said, including 25 inpatient, eight outpatient, three Eye Center, and two Lone Tree Surgery Center spaces. Squeezing extra capacity through efficiency could delay or avoid having to build out new ORs, he added.

UCH’s OR block utilization – the share of actual OR time used of the total time allocated – is about 65 percent on a given day. Getting that number to a higher percentage could make a big difference, Hess said.

“It means patients are getting elective surgery done in a more timely fashion, and it would be a huge win for docs,” Hess said.

Getting to that goal has been a challenge. Ashley Walsh, University of Colorado Hospital’s perioperative business manager, serves as UCHealth’s point person on the iQueue OR project. A big part of her job has been reporting OR utilization at UCH. That requires tapping into Epic electronic health record data analytics and reports, SharePoint cubes, web-based reports from the hospital’s anesthesia partners and other inputs.

“We go to many different places to get the numbers we want to see,” Walsh said.

The goal is to manage OR blocks – chunks of time allocated to particular surgeons for particular procedures – as efficiently as possible. It requires fitting OR blocks to procedure times (so that, for example, a two-hour surgery doesn’t occupy a four-hour block), allocating blocks among the many ORs, and minimizing start-time delays as well as the time it takes to turn a room over between surgeries. It’s complex puzzle even before taking into account surgeon preferences, surgeries that turn out to be more difficult than expected and a host of other surprises.

Hess and Walsh saw the ability of LeanTaaS to help meet these challenges. But getting there has required plowing new ground. LeanTaaS hadn’t yet tailored its software to ORs. With UCHealth as its pilot customer and partner in development, the company launched a project to develop “iQueue for Operating Rooms” in February. As of mid-May, iQueue for ORs was analyzing live UCHealth OR utilization data as part of the project’s first phase, which centers on gathering and analyzing operational data.

In the project’s second phase, iQueue for ORs will start applying what it has “learned” about the hospital’s OR operations and will recommend changes to OR block scheduling and other possible efficiency improvments. The third phase will add “mobile block exchange,” which will allow surgeons and administrators to release or add OR blocks on the fly using their mobile phones so that valuable surgical time isn’t wasted.

Rapid deployment

It’s all happening fast. The plan is to have mobile block exchange in place at UCH by the end of July. By about October, Hess and colleagues will know whether iQueue for ORs is making a difference; if it is, they’ll assess the potential for rollout across other UCHealth hospitals, he said.

From what Walsh has seen so far, it’s a job made for LeanTaaS’s machine learning.

UCHealth’s partnership with Silicon Valley firm LeanTaaS is symbiotic: LeanTaaS brings PhD-level expertise in machine learning and data analytics; UCHealth serves as both customer and partner in developing solutions that meet the needs of diverse health care settings.

“LeanTaaS turns around advanced analytic outputs in a week where it could take an in-house team a month or more to produce,” she said.

The benefits flow both ways, too. Walsh shares UCHealth subject-matter expertise, which informs iQueue for ORs.

“I’m telling them what I do, what I need, and why I need it, and they’re turning it around,” she said. “They’re taking it to the next degree – to where we’ve wanted to take it.”

Peter Witt, MD, a University of Colorado neurosurgeon, said he’s looking forward to using iQueue for ORs.

“It will allow for very immediate – almost ‘live’ – access to usage data related to my block time in the OR and will let me see if I am effective in using the allotted time,” Witt said in an email.

Assuming LeanTaaS can do for ORs what it has done for infusion centers, Hess would like to take it further yet. The next step is to apply machine learning to managing inpatient bed capacity. For example, based on analysis of scads of historical capacity data, Hess hopes LeanTaaS’s systems can pinpoint ahead of time expected constraints in particular units that frequently blossom into broader capacity problems.

Ambulatory clinics, radiology, pharmacy, the clinical lab – any high-demand hospital resource could benefit from the technology, Hess said.

“The idea is to figure out what the constraints are going to be tomorrow or even two days from now so you can start moving levers today,” he said.