Machine learning tools effectively identified patients at greater risk for readmission, according to a study published in Anesthesiology and highlighted by the UCLA Anderson Review.
Researchers at UC Los Angeles Medical Center used three machine learning models to predict which patients faced elevated readmission risks. Researchers tested the models using data from 34,532 adult surgical readmissions from April 2013 to December 2016, then narrowed that number down to the 1,942 cases that had emergency readmissions.
Researchers created a database with 1,013 patient variables that the models assessed each patient against. Two of the three models used classification trees that had the ability to discover nonlinear relationships and interactions among variables.
The models had accuracy rates between 85 percent and 87 percent, surpassing past models that had rates between 60 percent and 70 percent. Researchers believe their models can be used to identify at-risk patients while they're still in the hospital recovering.