Description
Identifying unnecessary lab test is an important target for high value care but lacks systematic methods. By studying electronic medical records on over 167,000 inpatients from three hospitals, we report prevalence of highly repeated lab tests and build machine learning models to predict whether test will yield a “normal” result. The results allow us to identify highly predictable tests of over 100 types. We further show the method is transferable among different hospitals.
Learning Objective: 1) Learn prevalences of unnecessary diagnostic laboratory tests that are unlikely to yield new information for patients.
2) Learn how machine learning can help identify such low-yield lab tests.
Authors:
Song Xu, Stanford university
Jason Horn, Stanford university
Santhosh Balasubramanian, Stanford university
Lee Schroeder, University of Michigan
Nader Najafi, University of California, San Francisco
Shivaal Roy, Stanford university
Jonathan Chen (Presenter)
Stanford university
Presentation Materials: