Hypoglycaemia in children is a major risk factor for adverse neurodevelopment with rates as high as 50% in hyperinsulinaemic hypoglycaemia (HH). A key part of management relies upon timely identification and treatment of hypoglycaemia. The current standard of care for glucose monitoring is by infrequent fingerprick plasma glucose testing but this carries a high risk of missed hypoglycaemia identification. High-frequency Continuous Glucose Monitoring (CGM) offers an attractive alternative for glucose trend monitoring and glycaemic phenotyping but its utility remains largely unestablished in disorders of hypoglycaemia. Attempts to determine accuracy through correlation with plasma glucose measurements using conventional methods such as Mean Absolute Relative Difference (MARD) overestimate accuracy at hypoglycaemia. The inaccuracy of CGM in true hypoglycaemia is amplified by calibration algorithms that prioritize hyperglycaemia over hypoglycaemia with minimal objective evidence of efficacy in HH. Conversely, alternative algorithm design has significant potential for predicting hypoglycaemia to prevent neuroglycopaenia and consequent brain dysfunction in childhood disorders. Delays in the detection of hypoglycaemia, alarm fatigue, device calibration and current high cost are all barriers to the wider adoption of CGM in disorders of hypoglycaemia. However, machine learning, artificial intelligence and other computer-generated algorithms now offer significant potential for further improvement in CGM device technology and widespread application in childhood hypoglycaemia.
Keywords: children; continuous glucose monitoring; hyperinsulinism; hypoglycaemia; machine learning.
© 2020 John Wiley & Sons A/S. Published by John Wiley & Sons Ltd.