DTAI

  • Increase font size
  • Default font size
  • Decrease font size
DTAI News DTAI news DTAI-seminar: Data Mining in Intensive Care (Jelle Van Eyck)

DTAI-seminar: Data Mining in Intensive Care (Jelle Van Eyck)

Monday April 15, 2013 at 16h30 in Celestijnenlaan 200A (auditorium 00.225)

Data Mining in Intensive Care
by Jelle Van Eyck  (PhD student DTAI)

An intensive care unit (ICU) is a very data-rich environment. Demographic data, monitoring data, administered treatments and medication, laboratory measurements, ... are all registered and collected in a central database. This data can be used for a number of purposes (clinical trials, quality assessment, ...), including data mining.

This talk will consist of three distinct parts, each discussing a particular application relevant to the ICU. In the first part, I will focus on acute kidney injury (AKI). This is a condition that affects up to 25% of all ICU patients and is associated with an increased risk of mortality and morbidity. Being able to predict which patients are at risk would allow for the application of preventive treatments. To this end, extensive research has been done on various biomarkers for the early detection of AKI. Here, I will compare one of the most promising biomarkers (NGAL) to a machine learning approach and discuss the results.

The second part of this talk will focus on an algorithm for scheduling elective cardiac surgery patients. Here, the main difficulty is that certain factors required to construct a planning are actually not known at the time of planning. Fortunately, predictive models allow us to obtain estimates of these factors. Here, I will present a planning approach based on Monte Carlo techniques that incorporates these estimates in order to overcome these difficulties.

In the third and final part I will introduce a new data source: progress reports. On a daily basis, intensivists generate these reports describing the current state of a patient, ongoing treatments and possible diagnoses. These reports are a rich source of information which have been rarely analyzed. Here, I will show that, using text mining techniques in combination with predictive models, it is possible to identify features and to relate these features to specific diseases.