For young Emily Hutt, it was a holiday trip that turned tragic.
While visiting Virginia in December 2015, the Summerville, SC teenager decided to try snowboarding. Her mom, Christine, videotaped the inaugural run. To her horror, she watched as Emily missed a turn and crashed, hitting her head on an exposed metal pole.
Thankfully, Emily survived. But recovery – which began with learning how to walk and talk again – was long and arduous. It required months of rehabilitation at Shepherd Pathways in Atlanta.
Shepherd Pathways specializes in helping people overcome traumatic brain injuries, severe spinal cord injuries, and other serious medical conditions. It is complex work. For each patient, Pathways must provide the right mix of physical, speech, occupational, and other kinds of therapies.
Here’s something else that is incredibly complex: scheduling these therapies.
At any given time, Shepherd Pathways has dozens of patients with varying injuries, requiring an assortment of skilled staff, in sessions spread out over different periods of time. Their way of orchestrating it all is to hold weekly assemblies of up to 25 therapists, gathering them in a room with clip-boarded sheets and pencils — a scene that, when described, evokes the old trading floor of the Chicago Mercantile Exchange.
“Every week, they come into this big room and create a whole schedule on paper,” said Meena Iyer, senior process improvement advisor for Shepherd. “It is a major time commitment.”
They knew there had to be a better way. And to find it, Shepherd turned to Pinar Keskinocak, William W. George Chair and Professor in the Stewart School of Industrial and Systems Engineering (ISyE) and the director of the Center for Health and Humanitarian Systems.
Once a year, Keskinocak teaches a graduate-level course in which student teams apply their industrial engineering, operations research, and data analytics skills to solve real-world problems in health care, humanitarian systems, and other public sector or societal applications. During spring 2017, one of the project teams focused on streamlining and automating the scheduling process at Pathways.
“Pinar was actually one of my professors,” said Kristin Goin (MSHS 08), Shepherd’s former senior improvement consultant who teamed up with Pathways colleague Iyer on the project. “I knew that optimization and scheduling was a key skill set that industrial engineers had.”
“One of the first steps was to answer the question, ‘What is ideal care?’” Keskinocak recalled. “The Pathways schedule was somewhat provider-centric. As we started working, we asked them how they would provide an ideal care to patients if they had a magic wand. This question was helpful in thinking about each patient’s unique needs.”
Defining the problem also meant examining the current state of scheduling in detail. “We created a map of every aspect of the manual scheduling process,” Goin said. “We put that across an entire wall of a room, like a war room, and went over every step.” They also surveyed Pathways staff to identify pain points and inefficiencies.
Under the guidance of Goin, Iyer, Keskinocak, and the Pathways staff, the six-member student team first streamlined how Pathways collected information on both the needs of the patients and the availability of the staff. They created two tools to do this, each to ensure that complete and consistent sets of data were available.
“After that, we developed heuristic algorithms,” explained Kirthana Hampapur, who was a master’s student in ISyE at the time, and a member of the student team. “The heuristic would go through the two tools – the patient info, which we called an Ideal Plan of Care tool, and the staff availability data – then assign each patient to therapists across multiple time slots.”
The students could have declared victory, as patient scheduling was now much closer to automation. But Hampapur remembered, “We felt like we hadn’t left the project in a good place.” The process was still not optimized to provide patients with as many therapy appointments as possible.
So after consulting with Keskinocak, she started a summer internship with Shepherd Pathways to continue the work. Using Python, the high-level programming language employed in application development, Hampapur continued to develop and implement the heuristic. Then in the fall semester, as she began pursuing her Ph.D., she teamed up with another student, Idil Arsik, also a Ph.D. student in ISyE, to develop a model of mixed integer-programming to improve the solution.
The result was something Shepherd nicknamed “The Wizard.”
“This was the actual software that took into account all of the input on patient needs, all of the staff availability, and then several other rules, such as when a certain patient has to go with a certain staff member,” Goin said. “The Wizard then generated the schedule.”
“Given the specified objectives and constraints, it creates the best schedule possible,” Hampapur said, “because every patient gets what he or she needs, and the number of therapy appointments is maximized.”
The outcome was happy for Shepherd Pathways. The time required for scheduling was reduced by 60 percent, and more than 90 percent of scheduling requirements were met.
“With 150 patients who have different needs, there are millions of possible schedule combinations,” Goin said. “The number of constraints is phenomenal, so it’s like peeling an onion. Georgia Tech has never failed to deliver the onion as peeled.”