Predicting Plateau Pressure in Intensive Medicine for Ventilated Patients - New Contributions in Information Systems and Technologies - page 169

3 Knowledge Discovering Process

The process of knowledge discovery using Data Mining techniques is complex, in that
sense, Cross Industry Standard Process for Data Mining (CRISP-DM) methodology
was followed to carry out the present study.

3.1 Business Understanding

This is the first phase of the CRISP-DM methodology. In this phase it was analysed
the problem and defined the project requirements and its main goals. First, there is a
need for developing this work because the ICU of CHP deals with mechanical
ventilated patients and do not have any mechanism to prevent Barotrauma. Being the
plateau pressure an important indicator and having in consideration all the evidences
above mentioned, this consitutes the main reason to develop this work. In this context,
the Data Mining goal was definied as: to predict and identify the variables (only
monitored by ventilators) which are most correlate with the Plateau pressure. The data
used for performing the present study were derived from the above-mentioned
hospital. The DM models were induced by using a set of variables collected by the
ventilators. The target is the variable corresponding to the Plateau pressure, which is a
numeric variable grouped in two classes of values (<30 cm O and >=30 cm O ).

3.2 Data Understanding

The initial data sample contains several records without patient identification (PID),
i.e., the data collected does not correspond to anyone. This happened bacause
sometimes the patients are admitted (for a few hours) in the ICU but is not assigned to
Electronic Health Record. These records were discarded from this study. The sample
used was collected from the ventilators and comprises the period between 01.09.2014
and 11.07.2014 and a total of 18 377 records. Each record contais sixteen fields:

• Date: corresponds to the day, month and year of the value collected by the

ventilator;

• Bed: is the bed number where the patient was being ventilated;

CDYN - (F_1): Dynamic compliance in mL/ cm O ;

•

CSTAT - (F_2): Static compliance from inspiratory pause measured in mL/
cm O ;

•

• FIO2 - (F_3): Fraction of inspired oxygen (%);

• Flow - (F_4): Peak flow setting in litters per minute;

• RR - (F_5): Respiratory rate setting in berths per minute;

• PEEP - (F_6): Positive End-Expiratory Pressure in cm O ;

• PMVA - (F_7): Mean airway pressure in cm O ;

• Plateau pressure - (F_8): End inspiratory in cm O ;

• Peak pressure - (F_9): Maximum circuit pressure in cm O ;

• RDYN - (F_10): Dynamic resistance in cm O ;