Principal component analysis in combination with case-base reasoning for detecting therapeutically correct and incorrect measurements in continuous glucose monitoring

Abstract

This paper introduces a data-driven methodology for detecting therapeutically correct and incorrect measurements in continuous glucose monitoring systems (CGMSs) in an intensive care unit (ICU). The data collected from 22 patients in an ICU with insulin therapy were obtained following the protocol established in the ICU. Measurements were classified using principal component analysis (PCA) in combination with case-based reasoning (CBR), where a PCA model was built to extract features that were used as inputs of the CBR system. CBR was trained to recognize patterns and classify these data. Experimental results showed that this methodology is a potential tool to distinguish between therapeutically correct and incorrect measurements from a CGMS, using the information provided by the monitor itself, and incorporating variables about the patient’s clinical condition.