Reference Interval Estimation from Mixed Distributions using Truncation Points and the Kolmogorov-Smirnov Distance...

Area of Science:

• Clinical Biochemistry
• Medical Informatics
• Biostatistics

Background:

• Reference intervals are vital for interpreting laboratory test results in clinical practice.
• Traditional methods require large, healthy populations, posing financial, logistical, and ethical challenges, especially for pediatric and elderly cohorts.
• Prevalence of chronic conditions and medication use in older adults complicates traditional reference interval establishment.

Purpose of the Study:

• To implement and validate an indirect method for estimating reference intervals using routine clinical data.
• To overcome limitations associated with conventional reference interval determination methods.
• To provide a robust algorithm for reference interval estimation from existing laboratory information systems.

Main Methods:

• An indirect estimation method was implemented, utilizing mixed physiological and abnormal test results from clinical information systems.
• The algorithm employs the Kolmogorov-Smirnov distance to minimize discrepancies between a hypothetical Gaussian distribution and the observed, Box-Cox transformed test result distribution.
• Simulations and real-world data from a university hospital's laboratory information system were used for validation.

Main Results:

• The indirect method demonstrated reliable reference interval estimations even with less than 20% abnormal test results in simulations.
• Reference intervals generated from clinical data remained stable despite including samples from patient populations with a high prevalence of pathologies.
• The method proved robust across various challenging simulation scenarios.

Conclusions:

• The indirect method provides a feasible and reliable approach for establishing reference intervals using readily available clinical data.
• This method circumvents the challenges of traditional approaches, offering a practical solution for diverse patient populations.
• An open-source C++ implementation is available, facilitating wider adoption and application.