IOE-STAGE SITE
In the United States over four billion prescriptions are dispensed every year. Of those four billion around 2.4 million are incorrectly dispensed, which can be a fatal error. A team of researchers from the University of Michigan looks to machine intelligence to help humans reduce their dispensing errors.
In the United States over four billion prescriptions are dispensed every year. Of those four billion around 2.4 million are incorrectly dispensed, which can be a fatal error. A team of researchers from the University of Michigan are looking to machine intelligence to help humans reduce their dispensing errors.
The research team includes U-M College of Pharmacy Assistant Professor Corey Lester with his team of researchers from the Lester Laboratory as well as U-M School of Information Ph.D. student, Jiazhao Li and U-M Industrial and Operations Engineering (IOE) Assistant Professor Raed Al Kontar and Associate Professor Xi Jessie Yang.
Their first task in creating a model was to teach the machine intelligence algorithm how to identify anomalies in medical dispensing. Using a database of thousands of medical dispensing images they programmed the machine to detect which ones were bad and good.
“If you give the algorithm a new image, the algorithm can predict whether this image belongs to… the good group or the bad group,” Kontar said. “Fundamentally it’s a matching decision done in a fancy mathematical way.”
After testing the algorithm they found that its predictions were over 99.5 percent accurate. While that may seem exceptional, medical errors are the third leading cause of death in the United States with the largest proportion of medical errors involving medications. The researchers cautioned that the .05 percent of errors in their algorithm should not be ignored.
“Our hope is that in those cases when the algorithm is wrong, the algorithm can tell us that it is not confident in its results,” said Raed.
When the algorithm is not confident a pharmacist will check the prescription manually to ensure that it’s correct. Therefore the algorithm will not replace pharmacists but rather support them in a double check of the work they’re already doing.
“Our next step is to conduct the experiment with pharmacists,” said Yang. “We are going to track how their trust [in the algorithm] changes over time.”
The researchers will be tracking pharmacists’ trust levels through an eye tracking system and survey questions. This data will be put together to see how quickly and confidently humans adjust to the system.
“As an engineer I believe our ultimate mission is to propose solutions that can enhance people’s lives,” said Yang. “Being part of this project is doing just that.”