Using decision thresholds to improve medical diagnosis
U-M IOE PhD student, Gian-Gabriel Garcia receives student prize from the Society of Medical Decision Making.
IOE-STAGE SITE
U-M Industrial and Operations Engineering (IOE) PhD student, Gian-Gabriel Garcia has received the Lee B. Lusted Student Prize from the Society of Medical Decision Making (SMDM) for his recent research presentation on data-driven frameworks for improving risk-based medical diagnosis.
“The growing availability of medical data has led to the increased development of risk estimation models for guiding medical diagnosis decisions. However, one major challenge in implementing these models in practice is determining appropriate decision thresholds,” explained Garcia.
“One major challenge in implementing these models in practice is determining appropriate decision thresholds.”
Gian-Gabriel Garcia
PhD Student, U-M Industrial & Operations Engineering
As part of a small team of researchers from U-M, Indiana University and the Medical College of Wisconsin, Garcia helped to develop a data-driven framework that combines optimization and predictive modeling to identify practical decision thresholds. The decision thresholds developed identify those likely to receive a positive or negative diagnosis – a positive diagnosis is when a disease or medical condition is assessed to be present, a negative diagnosis is when it is assessed not to be. Additionally, they also identify when there is not enough data available to use risk estimation models to make an accurate diagnosis.
A lack of sufficient data means there is a need for additional information. This need could lead to the creation of new models that are better suited to cases that are categorized as hard to diagnose.
“If the information leads to the construction of new models, just being able to identify hard cases could be useful,” said Garcia. “In other cases, it may be better for a clinician to use their experience or additional resources to make the diagnosis decision, especially if the information required is not easily incorporated in existing modeling approaches.”
The team applied their framework to acute concussion assessment and showed improvement upon methods which are commonly used today.
“It has been great to work with my collaborators, who span a wide range of technical and clinical expertise and share our work with others at conferences like the Society of Medical Decision Making’s annual meeting. I am optimistic that this research has the potential to make a positive impact on clinical practice for acute concussion assessment and many other disease areas,” said Garcia.