Predict or draw blood: An integrated method to reduce lab tests.

Area of Science:

• Artificial Intelligence
• Clinical Informatics
• Healthcare Management

Background:

• Serial laboratory testing is a standard but costly practice in healthcare, particularly in Intensive Care Units (ICUs).
• Repeated lab tests incur significant expenses and pose potential risks to patient well-being.
• Identifying which specific lab tests can be safely omitted presents a complex optimization challenge due to the vast number of tests and decision trajectories.

Purpose of the Study:

• To develop a novel deep-learning framework for predicting and omitting unnecessary serial laboratory tests.
• To jointly forecast future lab test events for omission and estimate their values based on observed data.
• To address the challenge of optimizing test reduction in a time-dependent decision-making context.

Main Methods:

• Proposed a novel, concise deep-learning architecture for joint prediction of lab test omission and omitted event values.
• Utilized observed testing values to train the model on time-series data.
• Focused on modeling the temporal dynamics of clinical laboratory testing decisions.

Main Results:

• Achieved a 15% reduction in the number of serial laboratory tests performed.
• Maintained a prediction accuracy loss of less than 5% despite test omission.
• Demonstrated the efficacy of the deep-learning approach in optimizing lab test utilization.

Conclusions:

• The developed deep-learning method effectively reduces the frequency of serial laboratory testing without significant loss of predictive accuracy.
• The framework offers a generalizable solution for optimizing decision-making in similar business and clinical contexts involving sequential data.
• This approach has the potential to decrease healthcare costs and mitigate patient risks associated with excessive testing.