Explaining Medical Model Exceptions - Computational Intelligence in Healthcare: Advanced Methodologies - page 264

Individual base. The patient suffers from a chronic disease, namely from asthma.
Adaptation. Since the patient started with a fitness program already after five months
of dialysis, the second general solution can be excluded. The first general solution
might be possible, though the individual base does not contain any information about
PTH. Further lab tests showed PTH = 870, which means that PTH is a solution.
Since an additional disease, bronchial asthma, is found in the individual base, this
solution is checked. Asthma is not contained as a solution in the case base, but the
expert concludes that asthma can be considered as a solution. Concerning the remain-
ing general solutions, the patient is not too old and proclaims that she was active at
fitness.
Adapted case. The solution consists of a combination of two factors, namely a high
PTH concentration and an additional disease, asthma.
3 Illustration of ISOR's Program Flow
ISOR's main dialogue menu is shown in figure 3. At first, the user sets up a model
(steps 1 to 4), subsequently he/she gets the result and an analysis of the model (steps 5
to 8), and then he/she attempts to find explanations for the exceptional cases (steps 9
and 10). Finally, the case base is updated (steps eleven and twelve). For illustration
purposes the steps are numbered (in figure 3) and in the following they are explained
in detail.
At first the user has to set up a model. To do this he/she has to select a grouping
variable. In this example CODACT was chosen. It stands for “activity code” and
means that active and non-active patients are to be compared. Provided alternatives
are the sex and the beginning of the fitness program (within the first year of dialysis
or later). In another menu the user can define further alternatives. Furthermore, the
user has to select a model type (alternatives are “strong”, “medium”, and “weak”), the
length of the time period that should be considered (3, 6 or 12 months), and the main
factors have to be selected. The list contains the factors from the observed database.
In the example, three factors are chosen: O2PT (oxygen pulse by training), MUO2T
(maximal oxygen uptake by training), and WorkJ (work in joules during the test train-
ing). In the menu list, the first two factors have alternatives: “R” and “T”, where “R”
stands for state of rest and “T” for state of training.
When the user has selected these items, ISOR calculates the table. “Better” and
“worse” are meant in the sense of the chosen model, in the example of the strong
model. ISOR does not only calculate the table but additionally extracts the excep-
tional cases from the observed database. In the menu, the list of exceptions shows the
code names of the patients. In the example, patient “D5” is selected and all further
data belong to this patient.
The goal is to find an explanation for the exceptional case “D5”. In step 7 of the
menu it is shown that all selected factors worsened (-1), and in step 8 the factor values
according to different time intervals are depicted. All data for twelve months are
missing (-9999).