Previous studies suggest a minimum duration of ~14 days of continuous glucose monitoring (CGM) data are required for robust assessment of glucose control metrics in people with type 1 diabetes (T1D). However, this was derived from older CGM devices and the employed techniques required big datasets (>150 subjects) to achieve reliable results. In this work, we present a robust technique using data from newer generation sensors to determine if the minimum duration remains ~14 days. We also evaluate a larger set of glycaemic metrics.
CGM data from a 6-month randomised clinical trial in 25 adults with T1D was analysed. Eight glucose metrics were evaluated on different sliding time windows of durations ranging from one to 90 days. Then, the absolute percentage error was computed for each window incidence against the entire duration. Finally, the median (interquartile range) for each window length was calculated.
The figure below shows the results corresponding to the metrics: percentage time in [70,180] mg/dL, and percentage time below 70 mg/dL. Note that the latter presents a much higher error and variance.
The duration of ~14 days of CGM data remains the minimum required for most of the evaluated metrics, but not for the ones related to hypoglycaemia, which require a much longer duration. The proposed technique can be employed in a smaller dataset to attain a high level of reliability.