Automatic Outlier Detection in Laboratory Result Distributions Within a Real World Data Network

Area of Science:

• Health Informatics
• Data Science
• Biostatistics

Background:

• Accurate comparison of laboratory test results across healthcare organizations requires data interoperability.
• Standardized terminologies, such as LOINC (Logical Observation Identifiers, Names and Codes), are essential for unique identification of laboratory tests.
• Real World Data (RWD) often contains outliers and abnormal values that necessitate exclusion from analysis.

Purpose of the Study:

• To analyze two automated methods for selecting histogram limits to sanitize lab test result distributions.
• To compare Tukey's box-plot method and a 'Distance to Density' approach within the TriNetX Real World Data Network.
• To evaluate the impact of these methods on data sanitization for RWD.

Main Methods:

• Implementation of Tukey's box-plot method for outlier detection and exclusion.
• Application of a 'Distance to Density' algorithm for identifying and excluding abnormal values.
• Analysis of laboratory test results within the TriNetX Real World Data Network.

Main Results:

• The study compared the generated histogram limits from both methods using clinical RWD.
• Tukey's method generally produced wider limits, while the 'Distance to Density' approach yielded narrower limits.
• Both methods' outcomes were significantly influenced by the chosen algorithm parameters.

Conclusions:

• Automated methods can effectively sanitize lab test result distributions by excluding outliers.
• The choice between Tukey's method and 'Distance to Density' depends on desired limit width and parameter sensitivity.
• Further research is needed to optimize parameter selection for robust RWD analysis.