![](bg4.png)
Table 3: CE Data Ranges (T2).
EvId Descr MinEC MaxEC MinVal
axVal
inAny
axAny
1011 BP 90 180 0 300 60
3000 SPO2 90 100 0 100 80
2009 HR 60 120 0 300 30 180
DIU UR 30 1000 0 1000 10
According to the Table 3, value can be normal (0),
critic (1) or too critic (2). This process is executed
through a cascade of trigger which is performed at
the moment when the value is collected.
To an event be critical is necessary achieving one
of the two characteristics (defined continuously out
of range or anytime). This procedure also calculates
the ACE and the ratios associated to each variable.
Finally, it calculates the total results of the hour. For
the real values (ACE and Ratios) a discretization
technique is used. The values are grouped and
categorized in accordance to a minimum and
maximum. Using this technique, the sets are defined
according to some rules using the respective average
(R1) or maximum (R2) of the values collected.
These ranges are flexible and are updated according
to the values collected in the ICU.
The ranges were created using a 7-point-scale
adapted from Clinical Global Impression - Severity
scale (CGI-S) (Guy, 1976). Table 4 presents the
rules to create the ranges. In the case of R1 (ratios
using elapsed time) is used the average of the values
collected. In the case of R2 (ratios using max
number of ACE) is used a percentage of the
maximum value obtained in the range.
Table 4: Discretization rules.
SET
R1 R2
Definition Average Maximum
> <= > <=
0 - 0% - 0% Inexistence
1 0% 25% 0% 10% Normal condition
2 25% 50% 10% 25% Borderline condition
3 50% 100% 25% 50% Mild condition
4 100% 150% 50% 75% Moderate condition
5 150% 200% 75% 90% Marked condition
6 200% 300% 90% 100% Severe condition
7 300% 1000% 100% 200% Extreme condition
Using Table 4 the ranges were obtained
according to the importance /significance of the
value to ICU. Table 5 presents the discretization
rules defined for each continuous value. At the top
of the table is the identification of the set. The left
column identifies the variable. In the middle of the
table are defined the ranges for each set. The R2min
and R2max are used by R2 (max number of ACE).
According to the percentage of the value, it is
categorized. These values were defined by ICU
doctors, but can be modified in the future.
Table 5: Discretization set of Data Mining Input.
SET 0 1 2 3 4 5 6 7
R1
BP
Min -0,1 0,000 0,011 0,021 0,042 0,063 0,084 0,126
Max 0 0,011 0,021 0,042 0,063 0,084 0,126 2,000
R1
O2
Min -0,1 0,000 0,017 0,034 0,068 0,102 0,136 0,204
Max 0 0,017 0,034 0,068 0,102 0,136 0,204 2,000
R1
HR
Min -0,1 0,000 0,005 0,010 0,019 0,029 0,038 0,057
Max 0 0,005 0,010 0,019 0,029 0,038 0,057 2,000
R1
TOT
Min -0,1 0,000 0,000 0,000 0,000 0,000 0,000 0,000
Max 0 0,000 0,000 0,000 0,000 0,000 0,000 2,000
R2
Min -0,1 0,000 0,100 0,250 0,500 0,750 0,900 1,000
Max 0 0,100 0,250 0,500 0,750 0,900 1,000 2,000
ACE
Min -0,1 0 3 5 8 10 12 15
Max 0 3 5 8 10 12 15 50
During the processes described above, a
procedure is responsible to get all the data generated
and store them in into the knowledge base for the
Data Mining. Finally, and after having all the values
correctly inserted in DM Input table, another
procedure runs to clean the bad values. All tasks are
executed by Data Mining agent.
6 MODELLING
In this phase 126 models were developed (6 targets
(renal, hepatic, coagulation, cardiovascular,
respiratory and outcome) x 7 models x 3 techniques
(DT, NB, SVM). During the modelling process the
neurologic system weren’t considered due to the
existence of high number of GSC data in fault. Data
mining models are a junction of the groups detailed:
M1 = CM ⊲⊳ACE
M2 = CM ⊲⊳ACE ⊲⊳ R
M3 = CM ⊲⊳ ACE ⊲⊳ R1
M4 = CM ⊲⊳ ACE ⊲⊳ SOFA
M5 = CM ⊲⊳ ACE ⊲⊳ SOFA ⊲⊳ R
M6 = CM ⊲⊳ ACE ⊲⊳ SOFA ⊲⊳ R2
M7 = CM ⊲⊳ ACE ⊲⊳ SOFA ⊲⊳ R1
With the purpose of automating this process, some
researches were done to know how to induce DM
models automatically. As a result it has been
possible to develop a procedure which executes the
DM engine in real time.
DataMiningforReal-TimeIntelligentDecisionSupportSysteminIntensiveCareMedicine
273