Impact of missing data on the accuracy of glucose metrics from continuous glucose monitoring assessed over a two-week period

OBJECTIVE: To explore the impact of missing data on the accuracy of continuous glucose monitoring (CGM) metrics collected over a 2-week period in a clinical trial.

RESEARCH DESIGN AND METHODS: Simulations were conducted to examine the effect of various patterns of missingness on the accuracy of CGM metrics as compared to a 'complete' dataset. The proportion of missing data, the 'block size' in which the data were missing, and the missing mechanism were modified for each 'scenario'. The degree of agreement between simulated and 'true' glycemic measures under each scenario was presented as R2.

RESULTS: Under all missing patterns, R2 declined as the proportion of missing data increased, however, as the 'block size' of missing data increased, the percentage of missing data had a more pronounced effect on the agreement between measures. For a 14-day CGM dataset to be considered representative for percent time in range, at least 70% of CGM data should be available over at least 10 days (R2 > 0.9). Skewed outcome measures, such as percent time below range and coefficient of variation, were more affected by missing data than the less skewed measures (percent time in range, percent time above range, mean glucose).

CONCLUSIONS: Both the degree and pattern of missing data impact upon the accuracy of recommended CGM-derived glycemic measures. In planning research, an understanding of patterns of missing data in the study population is required to gauge the likely effects of missing data on outcome accuracy.