Continuous Glucose Monitoring Data Compression Using Peak-Nadir Encoding in Diabetes: Method Development and Evaluation.

Background: Continuous glucose monitoring (CGM) generates high-frequency time-series data, creating challenges for efficient storage, transmission, and analysis. Objective: This study aimed to develop and evaluate a CGM-specific compression method that achieves high compression ratios while preserving signal fidelity and clinically relevant glycemic metrics. Methods: We introduce a content-based encoding approach (PN+) that represents CGM profiles using physiologically salient landmarks: glucose peaks and nadirs and a small set of optimally selected support points. Reconstruction is performed using piecewise cubic Hermite interpolation. PN+ was evaluated against peaks and nadirs only, uniform downsampling, piecewise aggregate approximation, and autoencoder-based compression. Performance was assessed across multiple compression ratios using 2 complementary datasets: 40,000 synthetic CGM profiles and real-world CGM data from a randomized crossover trial (558 days from 30 patients). Performance was evaluated using compression ratio, mean absolute error, and R2 between original and reconstructed CGM-derived clinical metrics. Results: At a compression ratio of 13 (22 points per 24-hour profile), PN+ achieved substantially lower reconstruction error than comparator methods (mean absolute error=0.77 vs 2.75-3.45) and consistently higher R2 values across glycemic metrics. Improvements were most pronounced for excursion-sensitive measures such as mean amplitude of glycemic excursions, where PN+ reduced error by more than 4-fold compared with downsampling, piecewise aggregate approximation, and autoencoders. These performance advantages were preserved in heterogeneous real-world data. Encoding and decoding required less than 0.2 seconds per profile, supporting practical scalability. Conclusions: PN+ enables robust CGM data compression by explicitly preserving physiologically meaningful glucose dynamics. The method outperforms generic compression techniques in reconstructing clinically relevant metrics while maintaining low computational overhead, making it well suited for large-scale CGM storage, interoperability, and downstream analytics.