Session S91.4

Predicting Acute Hypotensive Episodes from Mean Arterial Pressure

P Langley*, S King, D Zheng, EJ Bowers,
K Wang, J Allen, A Murray

Newcastle University
Newcastle upon Tyne, UK

We present an algorithm for identifying patient susceptibility to acute hypotensive episodes (AHE) using mean arterial blood pressure (MAP). MAP, calculated as the average MAP over 1 minute intervals, was obtained for two data sets (training set and test set) provided for the 2009 PhysioNet/Computers in Cardiology Challenge. The challenge was to i) identify 5 patients with and 5 patients without AHEs and ii) predict patients with and without onset of AHE within one hour of the end of the recordings. Using the training set we derived two indices of AHE: 1) average MAP across each recording (aMAP) and 2) the number of episodes of MAP falling below a threshold for a proportion of a 30 minute period. For the training set all the patients (15/15) in the non-hypotensive group and only 20% (3/15) of the hypotensive patients had aMAPs greater than 80 mmHg. However, in the test set only 30% (3/10) had aMAPs greater than 80 mmHg so we were unable to identify the 5 patients with and without AHE using this index. There were no episodes of MAP falling below 65 mmHg for 75% of any 30 minute period across the whole recording of patients without hypotension, but all patients with hypotension had episodes, with an average of 24 episodes per patient. Increasing the threshold to 70 mmHg, there were 2 episodes in the non-hypotensive group while the average number of episodes for each hypotensive patient was 35. Defining the positive AHE index as the occurrence of at least one episode of MAP falling below 65 mmHg for 75% of a 30 minute period we achieved for the test set scores of 0.8 (8/10) in event i) and 0.7 (28/40) in event ii). In conclusion, using MAP alone, we achieved modest accuracy for classifying patients with AHE and predicting the onset of AHE.

(Abstract Control Number: 270)