Objective: To introduce a data transformation that enhances the power of blood glucose data analyses.
Research design and methods: In the standard blood glucose scale, hypoglycemia (blood glucose, < 3.9 mmol/l) and hyperglycemia (blood glucose, > 10 mmol/l) have very different ranges, and euglycemia is not central in the entire blood glucose range (1.1-33.3 mmol/l). Consequently, the scale is not symmetric and its clinical center (blood glucose, 6-7 mmol/l) is distant from its numerical center (blood glucose, 17 mmol/l). As a result, when blood glucose readings are analyzed, the assumptions of many parametric statistics are routinely violated. We propose a logarithmic data transformation that matches the clinical and numerical center of the blood glucose scale, thus making the transformed data symmetric.
Results: The transformation normalized 203 out of 205 data samples containing 13,584 blood glucose readings of 127 type 1 diabetic individuals. An example illustrates that the mean and standard deviation based on transformed, rather than on raw, data better described subject's blood glucose distribution. Based on transformed data: 1) the low blood glucose index predicted the occurrence of severe hypoglycemia, while the raw blood glucose data (and glycosylated hemoglobin levels) did not; 2) the high blood glucose index correlated with the subjects' glycosylated hemoglobin (r = 0.63, P < 0.001); and 3) the low plus high blood glucose index was more sensitive than the raw data to a treatment (blood glucose awareness training) designed to reduce the range of blood glucose fluctuations.
Conclusions: Using symmetrized, instead of raw, blood glucose data strengthens the existing data analysis procedures and allows for the development of new statistical techniques. It is proposed that raw blood glucose data should be routinely transformed to a symmetric distribution before using parametric statistics.