Prediction with local patterns using cross-entropy
KDD '99 Proceedings of the fifth ACM SIGKDD international conference on Knowledge discovery and data mining
Discovery of Frequent Episodes in Event Sequences
Data Mining and Knowledge Discovery
Fast Algorithms for Mining Association Rules in Large Databases
VLDB '94 Proceedings of the 20th International Conference on Very Large Data Bases
Temporal Abstractions for Diabetic Patients Management
AIME '97 Proceedings of the 6th Conference on Artificial Intelligence in Medicine in Europe
Constructing (Almost) phylogenetic trees from developmental sequences data
PKDD '04 Proceedings of the 8th European Conference on Principles and Practice of Knowledge Discovery in Databases
Integrating classification trees with local logistic regression in Intensive Care prognosis
Artificial Intelligence in Medicine
| Hi-index | 0.00 |
Patient outcome prediction lies at the heart of various medically relevant tasks such as quality assessment and decision support. In the intensive care (IC) there are various prognostic models in use today that predict patient mortality. All of these models are logistic regression models that predict the probability of death of an IC patient based on severity of illness scores. These scores are calculated from information that is collected within the first 24 hours of patient admission. Recently, IC units started collecting sequential organ failure assessment (SOFA) scores that quantify the degree of derangement of organs for each patient on each day of the IC stay. Although SOFA scores are primarily meant for recording incidence of organ derangement and failures, the hypothesis is that they contribute to better prediction of mortality. There is virtually no systematic way in the literature to exploit the temporal character of SOFA scores for prediction. This paper adapts ideas from temporal datamining for discovery of sequential episodes and suggests a way to put them into use in the problem of mortality prediction. In particular, we discover frequent temporal patterns, assess their suitability for prediction, and suggest a method for the integration of temporal patterns within the current logistic regression models in use today. Our results show the added value of the new predictive models.