Identification of intraoperative hypotension endotypes and revolution with a temporal deep learning algorithm

Area of Science:

• Anesthesiology and Critical Care Medicine
• Biomedical Engineering
• Data Science in Healthcare

Background:

• Intraoperative hypotension (IOH) poses challenges in clinical practice.
• Understanding hemodynamic alterations during IOH is crucial for effective treatment.
• Current methods lack detailed insights into the diverse causes of IOH.

Purpose of the Study:

• To identify distinct hemodynamic endotypes of intraoperative hypotension (IOH).
• To apply deep learning to high-resolution intraoperative hemodynamic data for endotyping IOH.
• To potentially enable more precise, causal treatment strategies for hypotension.

Main Methods:

• Retrospective analysis of surgical patients with continuous hemodynamic monitoring (SVRI, SVI, SVV, CI, HR).
• IOH defined as mean arterial pressure < 65 mmHg for at least 1 minute.
• A long short-term memory (LSTM) autoencoder was used for data compression, followed by k-means clustering to identify endotypes.

Main Results:

• Analysis included 1304 hypotensive episodes from 184 patients.
• K-means clustering identified five distinct IOH endotypes: severe vasodilation with high CI, hypovolemia, myocardial depression, bradycardia, and mild vasodilation with preserved CI.
• Specific endotypes showed characteristic occurrences during different surgical phases, with some intra-event transitions observed.

Conclusions:

• The study successfully identified five distinct hemodynamic endotypes of intraoperative hypotension.
• These endotypes offer a basis for developing more targeted and causal treatment strategies for IOH.
• Further validation is needed to confirm the clinical utility of these identified endotypes.