*Purpose: Sepsis post liver transplantation is a frequent challenge impacting patient outcomes. We aimed to develop an artificial intelligence method to predict earlier the onset of post-operative sepsis.

*Methods:

This pilot study aimed to identify ‘physiomarkers’ in continuous minute-by-minute physiologic data streams to predict the onset of sepsis. 5,748 non-transplant patients across UTHSC-ICUs were monitored over 8-months with 604 patients that developed sepsis. Data from 48 liver transplant patients, of which 13 developed sepsis was also available. We aimed to test those models in liver transplant patients.

*Results: Using an alert timestamp generated by the Sepsis-3 definition as a reference point, we studied up to 24 prior hours of continuous physiologic data previous to the event, totaling 8.35 million data points. 150 features were generated using signal processing and statistical methods. Feature selection identified 52 highly ranked features, many of which included blood pressures. A Random Forests classifier was then trained on the ranked features using 5-fold cross validation on all non-LT (n=5,748) and the optimal model was subsequently tested on the LT patients. We identified that the average sensitivity, specificity, PPV and AUC after 10-iterations of the model was 0.85 ±0.04, 0.66 ±0.03, 0.48 ±0.01, 0.75 ±0.01 on predicting sepsis 12 hours before the event.

*Conclusions: Our data suggests that machine learning/deep learning can be applied to continuous streaming data in the transplant ICU to monitor patients and may predict sepsis.

« Back to 2020 American Transplant Congress