Previous Abstract | Next Abstract
Printable Version
A2018
October 04, 2020
10/4/2020 12:00:00 PM - 10/4/2020 1:00:00 PM
Room Virtual
Artificial Intelligence Enabled Control Of Hemodynamic Responses By Vasopressors And Narcotics In Cardiac Surgery Patients
Daniel Garcia, M.D.,Ph.D., Jinyoung Brian Jeong, B.S., Michael Zargari, Student, Chih-Ming Ho, Ph.D., Jure Marijic, M.D., Maxime Cannesson, M.D.,Ph.D., Soban Umar, M.D.,Ph.D., Jacques Neelankavil, M.D.
University of California, Los Angeles, Los Angeles, California, United States
Disclosures: D. Garcia: None. J. Jeong: None. M. Zargari: None. C. Ho: None. J. Marijic: None. M. Cannesson: None. S. Umar: None. J. Neelankavil: None.
Background: Precise management of the hemodynamics of patients undergoing cardiac surgery is challenging due to a myriad of patient-specific factors. Artificial intelligence (AI) has been successfully used in the analysis and intervention of diseases. Our group has developed an AI-based phenotypic response surface (AI-PRS) platform to prospectively determine personalized drug and dose combinations for liver transplant immunosuppression, prostate cancer chemotherapy and HIV treatment. The goal of the study was to develop the AI-PRS platform as a personalized therapy and data-driven tool to tailor the anesthetic and hemodynamic management of patients based on their unique physiology and biochemistry response profiles. Methods: After IRB approval, we conducted a retrospective analysis of adult patients undergoing cardiac surgery and collected hemodynamic data in response to dosing of different vasoactive medications and fentanyl. To isolate the hemodynamic effect of the medications, we collected data between the time at which the pulmonary artery catheter was inserted and surgical incision. The generation of the AI-PRS equation to guide medication administration requires a set of inputs and outputs. Here, “inputs” consisted of medications, such as phenylephrine and fentanyl, while mean arterial blood pressure (MAP) was the primary phenotypic “output”. In addition, we started preclinical testing in rats anesthetized with isoflurane and measured left ventricular systolic pressure (LVSP) in response to nitroglycerin and phenylephrine administration. Results: The MAP of a representative patient undergoing a CABG was plotted as a function of time along with the administration of phenylephrine and fentanyl (Fig. 1A). The data was then used to construct a three-dimensional PRS representing the patient’s response to the medications (Fig. 2A). The PRS is governed by the AI-PRS equation (Fig. 2B) relating the MAP to the phenylephrine and fentanyl doses. The planar shape of the PRS suggested that fentanyl and phenylephrine act independently. This important finding enables us to have a simple and effective control of the MAP. Based on the clinical data (Fig 1A) and the AI-PRS platform, a conceptual drawing (Fig. 1B) shows how we properly control the phenylephrine and fentanyl dosing and can keep the MAP in a narrow desired range. In preclinical tests performed in rats (Fig. 1C), the guided medication dosing can gradually decrease MAP fluctuations (red curves) and maintain the MAP within the desired range. Conclusion: The emergence of AI has created a new paradigm for personalized and data-driven patient management. Here, our analysis served as an initial step in validating the AI-PRS platform for clinical use (patent pending). Subsequent studies will incorporate the AI-PRS to prospectively guide medication administration; it will also examine more medication “inputs” such as inotropes and hemodynamic “outputs” such as pulmonary artery pressures, cardiac output, and mixed venous oxygen concentration. Lastly, we will aim to determine if our AI platform can outperform experienced anesthesia providers in maintaining tight hemodynamic control.
Figure 1
Figure 2

Copyright © 2020 American Society of Anesthesiologists