Extended Data Fig. 6: The effect of preprocessing steps employed on glucose time-series. | Nature Biomedical Engineering

Extended Data Fig. 6: The effect of preprocessing steps employed on glucose time-series.

From: Prediction of metabolic subphenotypes of type 2 diabetes via continuous glucose monitoring and machine learning

Extended Data Fig. 6

(A) Shows the number of missing glucose timepoints for each participant. These timepoints were imputed using linear interpolation. As illustrated, 87.5% (28 out of 32 participants) of participants had complete time-series, 9.4% (3 out of 32 participants) were missing only one timepoint, and 3.1% (1 out of 32 participants) were missing 2 timepoints. (B) All smoothed glucose curves overlayed on top of the non-smoothed curve. As illustrated in the figure, smoothing has a minor effect on the curve in most cases. However, in cases where there are strong fluctuations (for example, S55), smoothing helps capture the general pattern rather than those immediate spikes and valleys.

Back to article page