STATISTICAL METHODS FOR MAINTAINING MEASUREMENT STABILITY IN CLINICAL LABORATORIES

Abstract

This dissertation aims to develop statistical methods for maintaining stability of the measuring process in clinical laboratories. We propose novel approaches for problems that are poorly solved by current clinical laboratory quality control (QC) protocols. We first investigate the problem where undetected small lot-to-lot shifts accumulate to a significant drift. We propose a student’s t-test based approach on the regression coefficients of parallel results to detect such drifts. Besides using parallel results to maintain measurement stability, two control schemes on normal population results are further proposed to detect small but systematic mean and imprecision shifts in the measuring process. Lastly, we provide recommendations on how to respond to an error detected by “Average of Normal” control chart. The superiority and applicability of these methods are explored through numerical studies. Practitioners can combine these methods with current QC protocols for better error detection.