OBJECTIVE: Our objective was to compare statistical and clinical methods for the evaluation of five self-monitoring blood glucose (SMBG) meters. RESEARCH DESIGN AND METHODS: Two successive capillary blood glucose measurements were performed, and a simultaneous laboratory venous glucose measurement was used as the reference value. Accuracy was studied by comparing each of the two successive meter values with the reference value by 1) a Spearman's correlation test, 2) a Wilcoxon's paired test, 3) the percentage of values within the 10% interval of the reference value according to the American Diabetes Association consensus statement, and 4) the error grid analysis. RESULTS: The first two methods did not discriminate between the SMBG systems: r was >0.92 for the five meters, and a significant difference between the meter and reference values was found for all but one meter. The two other methods allowed classification of the devices into three groups according to their accuracy: good (two meters), acceptable (two meters), and unacceptable (one meter). These two methods gave consistent results and both had a good reproducibility, because the classification was similar for the two successive measurements. CONCLUSIONS: Both the Spearman's and Wilcoxon's paired tests, although commonly used, are inappropriate to evaluate SMBG systems. The percentage of SMBG values within the +/-10% interval and the error grid analysis are more accurate, because they consistently classified the five glucose meters tested in our study with a high degree of reproducibility.
OBJECTIVE: Our objective was to compare statistical and clinical methods for the evaluation of five self-monitoring blood glucose (SMBG) meters. RESEARCH DESIGN AND METHODS: Two successive capillary blood glucose measurements were performed, and a simultaneous laboratory venous glucose measurement was used as the reference value. Accuracy was studied by comparing each of the two successive meter values with the reference value by 1) a Spearman's correlation test, 2) a Wilcoxon's paired test, 3) the percentage of values within the 10% interval of the reference value according to the American Diabetes Association consensus statement, and 4) the error grid analysis. RESULTS: The first two methods did not discriminate between the SMBG systems: r was >0.92 for the five meters, and a significant difference between the meter and reference values was found for all but one meter. The two other methods allowed classification of the devices into three groups according to their accuracy: good (two meters), acceptable (two meters), and unacceptable (one meter). These two methods gave consistent results and both had a good reproducibility, because the classification was similar for the two successive measurements. CONCLUSIONS: Both the Spearman's and Wilcoxon's paired tests, although commonly used, are inappropriate to evaluate SMBG systems. The percentage of SMBG values within the +/-10% interval and the error grid analysis are more accurate, because they consistently classified the five glucose meters tested in our study with a high degree of reproducibility.