- Breast cancer prediction tool personalizes treatment
- Facilitating trust between virtual and local ISUs
- Encoding events improves product service
Breast cancer prediction tool personalizes treatment
No woman wants to hear a radiologist say an abnormality has been spotted on her mammogram, but if an abnormality is present, many radiologists recommend a biopsy. A final determination on whether the abnormality is cancerous can take weeks—an agonizing amount of time for a woman and her family.
Assistant Professor Oguzhan Alagoz is working to give radiologists more information that could alleviate some of this distress and prevent unnecessary biopsies. In collaboration with Radiology Associate Professor Elizabeth Burnside, Alagoz has created diagnostic models that produce a probability of cancer based on a woman’s individual attributes and risk factors.
Using this model, a radiologist could tell a woman that the abnormality has, for example, a 10 percent chance of being cancerous, so if the woman is advised to undergo a biopsy, she will understand the realistic likelihood of cancer.
Supported by the National Science Foundation and the National Institutes of Health, Alagoz’s model goes beyond calculating cancer probability and actually helps radiologists determine whether to recommend a biopsy. The model relies on sequential decision techniques, which account for decisions that are made multiple times and have a cascading effect, such as an abnormal mammogram prompting a woman to have another mammogram in six months. The model has shown that in general, radiologists recommend far more biopsies and short term follow-ups than necessary, especially for older women, who are more at risk for biopsy complications than are younger women.
Alagoz’s overall objective is to expose medical researchers to engineering tools and techniques to help clinicians tackle complex healthcare issues. Along with the breast cancer model, Alagoz is also researching optimal, personalized screening schedules for women and why certain demographics of women suffer from higher breast cancer mortality rates.
Additionally, Alagoz has begun to develop models for colorectal cancer screening, and, like his breast cancer research, he will investigate how often and with which technologies screenings should be conducted.
“My models will help clinicians provide both evidence-based and personalized medicine,” says Alagoz. “We have to remove inefficiencies from the healthcare system, and I think industrial engineering can play a big role in that. We can truly make a difference.”
Facilitating trust between virtual and local ISUs
for a nurse monitoring patients in an intensive care unit, or ICU, an extra pair of eyes is likely welcome—even if those eyes belong to a doctor or nurse in another room.
A virtual ICU is a room either on-site at a hospital or in another hospital or company, where physicians and nurses monitor computer screens displaying information about patients. Virtual ICU physicians and nurses, who have critical care certification and experience, are connected to patients via video and audio feeds that allow them to carefully watch patients 24 hours a day and alert local ICU staff of any changes. Virtual ICUs, which have emerged in the last decade, act as a supplement to local hospital ICU staff members. Approximately 300 virtual ICUs exist and are connected to more than 2,000 ICUs nationwide.
As the technology continues to gain popularity, virtual ICU producers and users are asking questions about how best to integrate virtual ICUs with local ICUs, especially when one virtual ICU serves multiple hospitals. Procter & Gamble Bascom Professor in Total Quality Pascale Carayon is researching this issue in collaboration with Associate Professor Doug Wiegmann, Professor of Medicine Ken Wood and Research Scientist Peter Hoonakker.
Supported by the National Science Foundation, Carayon and her team from the Center for Quality and Productivity Improvement are gathering data from five virtual ICUs about how virtual and local ICU staff develop trust and share information about ICU patients. Carayon and her team have developed a software tool that monitors ICU tasks, and the team will interview nurses and virtual ICU managers to learn how patient care decisions are made, how the two staffs manage conflict and whether nurses believe patient care is what it should be.
After the data are gathered, the team will evaluate how the use of virtual ICUs can be improved. Examples could include having the virtual and local ICU staffs meet in person or making video and audio feeds two-way to further facilitate communication.
Encoding events improves product service
Selling a product isn’t the only way manufacturers can generate income. Many also rely heavily on revenue from after-sale service plans, which are maintenance plans to prevent and fix product malfunctions. After-sale service care can make up as much as 50 to 70 percent of a company’s total revenue, and accurate tools to help manufacturers develop appropriate maintenance schedules are crucial. Associate Professor Shiyu Zhou is researching fundamental, cost-efficient methodologies for manufacturers to deliver optimal after-sale service care to their customers.
Zhou has developed techniques for GE Healthcare to help the company conduct aftersale service for complex medical equipment such as MRI or CT machines. These types of machines generate data logs that record everything that happens, from turning on the machine and taking an image to a critical failure in a specific mechanical component. Zhou, in collaboration with University of Iowa Mechanical and Industrial Engineering Assistant Professor Yong Chen, encodes the data from those logs into a mathematical model that can predict a failure. The prediction allows maintenance technicians to either prevent the failure or have a spare part on hand to quickly repair the machine when a failure occurs.
The specific model Zhou has developed is a kind of survival model, which is a statistical technique widely used in reliability engineering. The model can quantitatively describe the relationship between non-failure events, called benign events, and critical failure events recorded in the logs. Manufacturers often assume that benign physical events affect machine failures, and Zhou’s research has proven this to be true. Supported by the National Science Foundation and GE Healthcare, Zhou is currently working to improve the model to identify the exact benign events that affect prediction accuracy. He is also working to identify optimal maintenance strategies based on the model’s predictions.