Session PB8.1

Utilizing Histogram to Identify Patients Using Pressors for Acute Hypotension

CT Ho*, X Chen

Institute for Infocomm Research
Singapore, Singapore

In our approach, we propose to use a histogram to hash the mean arterial pressures to identify patients who are using pressors to treat their acute hypotension. After studying the data provided, it was observed that the use of pressors did not fully elevate their mean arterial pressure (MAP) above 60 mmHg. Hence, our research made use of this fact to separate patients using pressors to treat their acute hypotensive episodes (AHE) from patients who did not suffer from AHE. The initial data was filtered to minimize the noise. This noise contained MAP values that were well above the 200 mmHg threshold as seen in the graphs provided by the challenge. These high 200 mmHg values were clipped so that all data could be seen. Unavailable data in the form of a ‘-‘ was replaced with 0. A histogram was created with the filtered data. Each bin in the histogram corresponded to a MAP value that might exist in each data set and also contained the number of occurrences of the MAP value with respect to the data set. It was observed that when a cut-off point was placed at the 60th bin (MAP value 60 mmHg), the number of MAP values below 60 mmHg were still significant with respect to the total number of MAP values for each data set. Comparing the data in the 15 data sets from patients that were using pressors, it was observed that these data contained from 1.88% to 17.18% of MAP values that were below 60 mmHg. Therefore, it can be concluded that the use of pressors did not completely elevate the patients’ mean blood pressure above the safety level. For patients who did not have AHE, their data contained 0% to 0.00004% of MAP values lower than 60 mmHg. Given that the difference from 0.00004% to 1.88% is significant, we can say that these 2 groups of patients are distinct and thus separable. The first 10 hours of the challenge data set was run against the algorithm. The 10 hours worth of data for each patient was segmented into 10 distinct and consecutive 1-hour data segments. Each hourly segment was tested for AHE. Each patient was concluded to have AHE, if and only if, 9 or more segments tested positive. This was in line with the fact that from our research, each AHE patient’s data exhibited at least 1.88% of data that was below 60mmHg. Anything less would be due to changes in blood pressure because of bodily functions and reactions, and not AHE. Hence, over a 10-hour period, a patient who was using pressors would have most (>90%) of their hourly segments indicating the presence of AHE.

(Abstract Control Number: 279)