Elham Doratabadi
Image source: York University
A new study from York University published in the Jama Network Open Journal found learning strategies that are important for AI models to mitigate data shifts and subsequent harm, with continuous, continuous and forwarded learning strategies.
To determine the effectiveness of the data shift, the team built and evaluated an early warning system to predict the risk of in-hospital patient mortality and strengthened patient triaging at seven large Toronto-region hospitals. This study used Gemini, Canada's largest hospital data sharing network, to assess the impact of data change and bias on clinical diagnosis, demographics, gender, age, and hospital type. 143,049 patient encounters were included, including lab results, transfusions, imaging reports, and management functions.
“As AI use in hospitals increases to predict something from the occurrence of mortality, sepsis, and sepsis diagnosis, they need to work as predicted and not do harm.” “However, building reliable, robust machine learning models has proven difficult, as they generate system reliability as data changes over time.”
