Is natural language processing useful in predicting organ acceptance?

“The potential to improve existing kidney utilization prediction models is limited if further attempts are based solely on the same traditional OPTN data sources,” said United Network for Organ Sharing data science manager Andrew Placona. “Often the only way to improve model performance is by incorporating new data sources or finding new insights within the data.”

Placona is corresponding author in a new proof-of-concept article published in the American Journal of Transplantation that explores the possibility of using natural language processing to mine the free-text fields of deceased organ donor registration forms for information that might predict kidney utilization rates. Existing predictive algorithms have only leveraged structured data, such as yes or no questions, which may not provide sufficient detail to enable fully informed organ offer acceptance.

“We’re using data that traditionally UNOS has never touched,” Placona said. “No one has really looked at the clinical text yet and it’s gone unnoticed for decades because it is not easy data to work with.”

Read more about how Placona and the UNOS data science team are taking an innovative approach to improving the nationwide kidney utilization rate.

Read the abstract and see the models here

Placona, A., Martinez, C., McGehee, H., Carrico, B., Klassen, D.K. and Stewart, D. (2019), Can Donor Narratives Yield Insights? A Natural Language Processing Proof of Concept to Facilitate Kidney Allocation. Am J Transplant. Accepted Author Manuscript. doi:10.1111/ajt.15705