Biology Reference
In-Depth Information
value of the t-statistics and the p-value corresponding to this value
in the t-distribution. Recall that, in general, a p-value of 0.05 or
smaller is considered to reflect statistically significant differences
between the group averages, although any smaller p-value
diminishes the possibility of incorrectly rejecting the null
hypothesis.
More specifically, Table 5-2(A) presents comparisons of T1DM versus
T2DM on glycemic control averages, such as mean BG derived from
SMBG readings and HbA
1c
. As expected, no significant difference
between the groups is observed for these variables. However, the
standard deviation is markedly and significantly higher for the T1DM
group. This is not surprising, given that patients with T1DM more
frequently experience substantial deviations from the safe target BG
range and the average BG levels.
Table 5-2(B) presents BG characteristics and demonstrates that T1DM
subjects have both significantly lower and significantly higher BG
readings than T2DM subjects.
Finally, Table 5-2(C) presents risk characteristics of the SMBG data in
terms of the LBGI and the HBGI. Table 5-2(C) demonstrates that T1DM
subjects had significantly increased risk for severe hypoglycemia and
increased risk for hyperglycemia.
Note that the significance level for the HBGI is exactly 0.05.
Given the multiple parallel t-tests made in this study, such a significance
level cannot indicate a rejection of the null hypothesis (i.e., cannot
signify the HBGI in T1DM is greater than the HBGI in T2DM).
The reason is that when multiple parallel comparisons are
performed on the same data, simply by chance one of these
comparisons may turn out to be significant, because (if we
try many times) a low-probability event may actually happen.
This fact was mathematically formulated by the Italian mathematician
Carlo Emilio Bonferroni (1892-1960), who introduced an inequality,
stating that the probability of a sum of events is less than the sum of the
probabilities of these events. Based on this inequality, statisticians
introduce Bonferroni corrections for the significance level of
multiple parallel tests, dividing the significance level by the
number of tests. In our case, we have nine parallel tests, so it will be
prudent to only reject null hypotheses that meet a significance level
of 0.005.
Based on Table 5-2, we now have statistical results that provide
initial validation of the low and high BG index as markers reflecting
a medical reality—the differences between the BG patterns in T1DM and
T2DM. However, these statistical analyses do not provide immediate
evidence that these measures would be useful in assessing the risk
for hypo- and hyperglycemia.
