News
Projekt von Prospero am 19th Annual Congress
der European Society of Intensive Care (ESICM),
24. - 27. September 2006
Mortality predictions calculated using scoring scales are often not accurate in populations other than those in which the scales were developed. Read more
INTRODUCTION
. Mortality predictions calculated using scoring scales are often not accurate in populations other than those in which the scales were developed. Bio Intelligence helps to generate outcome models, based on information of SAPS2. Weighting of the attributes can help identifying the case-mix, effects of new treatment regimen like insulin therapy.
METHODS
. 14 bed interdisciplinary ICU, acute care teaching hospital. TISS-28: 33,43. 1277 patients, 2003/4, subset neuro 332 patients. p.MEDIS tool: Self Organizing Feature Map, Evolutionary Learning Algorithms. The perfomance was measured by ROC-analysis.
RESULTS
. The overall fit of the SAPS2 was dramatically improved in our population (ROC: AUC SAPS2: 0.90140; p.MEDIS: 0.97624). The predictive power for survival was strongly associated to the GCS; the organ dysfunction, the admittion type and the age were less predictive.. A model based on neuro patients could gain a very high predictive power for the outcome (AUC=0.989). The variables of the SAPS2 and there predictive power for the outcome were slightly different for the neuro patients. The age and the bilirubin become more important, systolic blood pressure is less in the subset of the neuro patients.
CONCLUSION
. The predictive power of the SAPS2 can be improved dramatically by using the tool, the quality of care can be monitored more precisely without using more data input than used for the SAPS2 score. The differences of case-mix can be identified, therapeutic regimens like the insulin therapy change the predictive power of the attributes for the outcome, e.g. critical ill polyneuropathy. Decisions can possibly be guided by using the data input like the attributes of SAPS2. Patients at risk can be identified. REFERENCE(S). 1.Reis Miranda D, de Rijk A, Schaufeli W 1996 TISS-28 Crit Care Med 24 64-732.Kharishman Dmitry 2006: Evaluation of a data mining tool, Master of Science Thesis3.Data-Mining-Tool-Suite Prospero AG, Gewerbeweg 15,FL-9490 Vaduz.
