Á. Silva, J. Pereira, M. Santos, L. Gomes, and J. Neves (Portugal)
Data Mining, Knowledge Discovery from Databases,Artificial Neural Networks, Organ Failure, MortalityPredicting Models based on Intermediate Outcomes,Intensive Care Units.
Medical prognosis has played an increasing role in health, namely in the critical care medicine. These factors have induced the medical community to take a more active interest in developing models for mortality prediction based on Artificial Intelligence (AI) techniques [1], that make possible the doctors pro-active action. In this context, the existence of large Databases (DB) containing Intensive Care Units (ICU) clinical information, motivate and enable the application of Data Mining (DM) techniques, in a Knowledge Discovery Database process (KDD), to induce prediction models of organ failure in a much more efficient way than other approaches (e.g., Logistic Regression)[2]. In this study we used a clustering approach as a strategy to predict the organs failures, based on the occurrence of Clinical Adverse Events (CAE) during the patient's stay in the ICU. It follows other studies made in this area (Santos e Silva [3,4]), where DM techniques were used to predict hospital mortality. The clinical database used was created from part of the database originated from the EURICUS II [5] study, after being subjected to processes elaborated with SPSS and SPSS Clementine tools[6]. Various models were created by applying the C5 Algorithm [7] in order to generate decision rules for easy interpretation, of each organ, for each day of stay, and for each cluster defined. This approach enables a better accuracy on some models, a homogeneous segmentation of patient's (for further medical analysis) and a tracking model of the patients' evolution during his stay. In such a way it is possible to identify the cluster where he fits and, consequently, to make use of the adjusted model to predict the status of a certain organ on the following day.
Important Links:
Go Back
