mining the intensive care unit
DESCRIPTION
A research paper's presentation at the "Data Mining in Bioinformatics" conference, that took place in 7-8 May in Athens, GreeceTRANSCRIPT
Mining the Intensive Care Unit: Knowledge
Extraction out of Medical Scoring Systems
Eirini Lygkoni & Georgios Tziralis, NTUADMINBIO 2009, May 08-09, Athens
a course by blog
mineknowledge
the problem
• patients admitted to intensive care units
• need to reliably monitor their status
• track the expectability of overpassing their incident
given solution
• Scoring Systems - tracking the heaviness of an ilness
• APACHE II (Acute Physiology and Chronic Health Evaluation)
• APACHE III
• SAPS II (Simplified Acute Physiology Score)
• SOFA (Sequential Organ Failure Score)
scoring systems variables
dataset
• 361 patients, *small*
• women 58.9%
• mean age 68.5
• death rate 11.6%
scoring systems distribution
enter data mining
• 23 attributes
• 2887 instances
• 361 patients, > 4 days hospitalization
• repeated measurements (every 3 hours)
• algorithms used: OneR, C4.5, PART
most valuable variables
some rules
and a tree
discussion
• introduced a novel approach to assessing the status of patients in intensive care unit
• quality results with less variables needed
• more easily comprehensible & discrete outcomes
• though maybe need to combine them somehow
future work
• more extended, generalizable dataset needed
• formalization of a simplified and more descriptive new scoring system, out of mining outcomes
• reach an accuracy rate close to 100%
thank [email protected]