Nonstationary Gaussian Process Regression for Evaluating Clinical Laboratory Test Sampling Strategies

Area of Science:

• Biostatistics
• Clinical Pathology
• Medical Informatics

Background:

• Accurate clinical laboratory test sampling is crucial for monitoring patient health.
• Physiologic functions are often nonstationary, making appropriate test timing difficult.
• Existing research often focuses on oversampling costs, neglecting undersampling issues.

Purpose of the Study:

• To quantify undersampling and oversampling in repeated clinical laboratory tests.
• To assess if current sampling strategies adapt to the volatility of underlying physiologic functions.
• To develop a novel method for analyzing nonstationary data in clinical testing.

Main Methods:

• Analysis of hundreds of observation sequences from four distinct clinical laboratory tests.
• Development of a Gaussian process fitting method for nonstationary latent functions.
• Inclusion of explicit estimation of the latent function's time-varying volatility.

Main Results:

• Undersampling occurs up to ten times more frequently than oversampling on average.
• A small minority of observed test sequences utilize adaptive sampling strategies.
• The developed method provides data-driven estimates of sampling adequacy.

Conclusions:

• Current clinical laboratory test sampling practices are frequently inadequate, particularly regarding undersampling.
• Adaptive sampling strategies are rarely employed, despite their potential benefits.
• The novel Gaussian process method offers improved analysis of nonstationary clinical data.