Processing Algorithm for Large Scale Clinical CGM Data
Deep Blue (University of Michigan) · 01.11.2024
DOI: 10.7302/24886 ↗Kurz gefasst
Fachartikel in der Zeitschrift Deep Blue (University of Michigan), erschienen am 01.11.2024. Verfasst von Williamson und 4 weiteren Autor:innen.
Abstract (Originalsprache)
Background
Widespread adoption of continuous glucose monitor (CGM) technology has led to a proliferation of CGM data and the adoption of CGM metrics in both clinical and research settings. Unlike curated CGM data from clinical studies, CGM data stored in research data warehouses may have periods of duplicated data uploads from the same patient, often accompanied by time shifts, compromising data quality. These have received limited attention in the literature, leading to the computation of CGM metrics on data "as is" or ad-hoc data processing approaches such as averaging measurements from the same time stamp. Proper identification and correction of data quality issues due to duplication and time shifts is required to ensure the accuracy of resulting CGM metrics.
Objectives
To develop and validate an algorithm that resolves duplication and time shift errors in CGM data, enhancing data accuracy, and to evaluate and quantify the difference in CGM metrics before and after processing.
Methods
To develop and validate an algorithm for duplication and time shift correction, we use CGM data from n = 2,308 Michigan Medicine patients collected over a two week period. We identify erroneous data by flagging measurements spaced more frequently than the expected 5-minute interval. To resolve and identify a single glucose profile, we use groupwise comparison of problematic observations, comparing patient ID, device ID, observation ID, glucose level, and timestamp values. We compare consensus CGM metrics before and after processing to evaluate effect on accuracy, including time in range (TIR), mean glucose, CV%, GMI, and glycemic episodes.
Results
Duplication errors were present in 25.9% out of n = 2,038 patients. The proposed algorithm effectively resolved these errors. Comparison of profiles before and after data processing revealed differences in all CGM metric values among individuals with problematic data. The magnitude of impact was greater among individuals with higher rates of duplication errors. Among the n = 528 patients with duplication errors, we observed 213 individuals with a measurable difference in TIR, CV% or GMI, and within this subset the difference in these values had significant clinical implications for 12 individuals.
Conclusions
Our results underscore the utility of processing CGM data and the need for a standardized approach. The varied types of duplication and time shift errors observed, ranging from exact duplication to mismatched timestamps and glucose values, underscore the complexity of the issue. Our algorithm addresses these issues, removing uncertainty introduced by duplication errors and offering a reliable solution with potential for applications in research and clinical settings.
Schlagwörter
Zitieren
APA
Williamson, W., Lee, J., & Gaynanova, I. (2024). Processing Algorithm for Large Scale Clinical CGM Data. Deep Blue (University of Michigan). https://doi.org/10.7302/24886
Vancouver
Williamson, W., Lee, J., Gaynanova I. Processing Algorithm for Large Scale Clinical CGM Data. Deep Blue (University of Michigan). 2024. doi:10.7302/24886
BibTeX
@article{Williamson2024,
author = {Williamson and W. and Lee and J. and Gaynanova, Irina},
title = {Processing Algorithm for Large Scale Clinical CGM Data},
journal = {Deep Blue (University of Michigan)},
year = {2024},
doi = {10.7302/24886}
}
EUVIMED