BayesLENS Software

Description

BayesLENS is a software developed by Jason Doctor and Greg Strylewicz to detect clinical laboratory errors using probabilistic (Bayesian) methods.

Technology Benefits

Over 180,000 certified laboratories perform 7 billion laboratory analyses each year, providing 70% of the data used to make medical diagnoses and treatment decisions. In the process, clinical laboratories fail to detect an estimated 70 million errors. Adverse events, due in part to preventable laboratory errors, cost the nation between $17 billion and $29 billion annually.

Methods employed by a typical clinical laboratory to detect errors in laboratory results have not changed substantially over the years even though technology, in general, has grown rapidly. The primary methods for validating patient results include the manual review of results by seasoned laboratory experts or by rule-based systems. An automated system used to review and report laboratory data is called an autoverification system. When reviewing data, both the autoverification system and the laboratory experts estimate the believability of results based on the internal consistency of the data and against delta checks. BayesLENS uses a Bayesian network learner to infer relevant variables. The key benefit of using Bayesian networks is that they offer a formal calculus for quantifying belief an error has occurred.

Features of BayesLENS

• More accurate identification of errors.
• Faster turn-around time of results than laboratory experts.
• Uses Bayesian networks for reasoning under uncertainty.
• Uses a patent-pending method to overcome the class-imbalance and to hypothesize as to the source of error.

Development Background

BayesLENS team consists of Dr. Jason Doctor and his graduate student Greg Strylewicz. Dr. Doctor's experience in health-related decision-making began when he was a graduate student working with noted health policy and health value measurement expert Robert M. Kaplan, Ph.D. at UC San Diego. Dr. Doctor's group has been actively involved in advancing work on decision theory in health and the application of graphical models in health including Bayesian Networks and influence diagrams. Currently, Dr. Doctor is funded as principal investigator by the CDC to study the use of Bayesian networks for syndromic surveillance on Northwest regional health information data. Greg holds a Master's Degree in Computer Science and has ten years of experience in the software engineering field. For the last six years Greg has been working at the University of Washington at the Northwest Lipid Metabolism and Research Laboratories (NWRL) where he developed their laboratory information management system.

Future Goals

• Perform an extensive comparison of BayesLENS against laboratory experts.
• Implement BayesLENS in a clinical laboratory.
• Expand the knowledge base used by BayesLENS to include additional analyses.

Scientific Contact

Greg Strylewicz
Biomedical and Health Informatics
University of Washington
Email: gstry@u.washington.edu