| Clinical
Trials (blood sugar) |
|
Figure
1
presents simulations and measurements which were performed on
a diabetic subject. The solid line represents the simulation output
for the entire day, while the dots represent the blood glucose
measurements.
Figure
2 and Figure 3 respectively shows the predictive
quality of the AIDA software(the only competitor) and our software.
|
Figure
1 Our accurate blood sugar
predictions (type 1)
|
|
Clinical
Trials (insulin) |
|
Let us investigate the quality of predictions
for insulin requirements using the well-known CHO mass or GI methods.
The results for the CHO and GI experiments for one test subject
are shown in Figure 4 and Figure 5 respectively.
Pearson’s R²- values were calculated for linearized
trend fits through the plotted data.
The R²-values
for the CHO and the GI methods were 0.603 and 0.558 respectively
for the example test subject. For the CHO method the worst spread
was found at 50 g of CHO, namely a factor of 12, while for a GI
at 65 the spread factor was close to 3.
From
the figures the need for a better insulin prediction method according
to ingested carbohydrates is obvious. Therefore, our
method is proposed as an alternative. The method is theoretically
derived from energy principals and is very easy to use. A theoretical
approach is preferred to an empirical one since theory often leads
to better insight.
The
quality of insulin predictions according to the
method is investigated by using the Lee and Wolever measurements
for the same test subject as in Figure 4 and Figure
5. The results are given in Figure 6.
The
linear trend line for the
method yields an R²-value of 0.929 for this test subject.
The value is significantly higher than those calculated for the
other two methods and therefore indicates that the
predictor is better than both CHO counting and the GI methods. |