The Physiological Deep Learner: First application of multitask deep learning to predict hypotension in critically ill patients

Ményssa Cherifa; Yannet Interian; Alice Blet; Matthieu Resche-Rigon; Romain Pirracchio

doi:10.1016/j.artmed.2021.102118

The Physiological Deep Learner: First application of multitask deep learning to predict hypotension in critically ill patients

Artif Intell Med. 2021 Aug:118:102118. doi: 10.1016/j.artmed.2021.102118. Epub 2021 May 29.

Authors

Ményssa Cherifa¹, Yannet Interian², Alice Blet³, Matthieu Resche-Rigon⁴, Romain Pirracchio⁵

Affiliations

¹ Université de Paris, ECSTRRA team, Center of research in epidemiology and statistics (CRESS) - INSERM UMR 1153, 1 Parvis Notre-Dame - Pl. Jean-Paul II, Paris 75004, France.
² Data Analytic Program, University of California San Francisco, 101 Howard St, San Francisco, 94105, CA, United States.
³ Department of Anesthesiology, Critical Care and Burn Center, Lariboisière-Saint-Louis Hospital, AP-HP Nord, 1 Avenue Claude Vellefaux, Paris 75010, France; Université de Paris, Cardiovascular MArkers in Stress Conditions (MASCOT) - INSERM UMR-S 942, 41, boulevard de la Chapelle, Paris 75010, France; University of Ottawa Heart Institute, University of Ottawa, 40 Ruskin St, Ottawa, Ontario, Canada.
⁴ Université de Paris, ECSTRRA team, Center of research in epidemiology and statistics (CRESS) - INSERM UMR 1153, 1 Parvis Notre-Dame - Pl. Jean-Paul II, Paris 75004, France; Department of biostastistics and medical information, Lariboisière-Saint-Louis Hospital, AP-HP Nord, 1 Avenue Claude Vellefaux, Paris 75010, France.
⁵ Université de Paris, ECSTRRA team, Center of research in epidemiology and statistics (CRESS) - INSERM UMR 1153, 1 Parvis Notre-Dame - Pl. Jean-Paul II, Paris 75004, France; Department of Anesthesia and Perioperative Medicine, Zuckerberg San Francisco General Hospital and Trauma Center, University of California San Francisco, 1001 Potrero Ave, San Francisco, 94110, CA, United States. Electronic address: [email protected].

PMID: 34412841
DOI: 10.1016/j.artmed.2021.102118

Abstract

Critical care clinicians are trained to analyze simultaneously multiple physiological parameters to predict critical conditions such as hemodynamic instability. We developed the Multi-task Learning Physiological Deep Learner (MTL-PDL), a deep learning algorithm that predicts simultaneously the mean arterial pressure (MAP) and the heart rate (HR). In an external validation dataset, our model exhibited very good calibration: R² of 0.747 (95% confidence interval, 0.692 to 0.794) and 0.850 (0.815 to 0.879) for respectively, MAP and HR prediction 60-minutes ahead of time. For acute hypotensive episodes defined as a MAP below 65 mmHg for 5 min, our MTL-PDL reached a predictive value of 90% for patients at very high risk (predicted MAP ≤ 60 mmHg) and 2‰ for patients at low risk (predicted MAP >70 mmHg). Based on its excellent prediction performance, the Physiological Deep Learner has the potential to help the clinician proactively adjust the treatment in order to avoid hypotensive episodes and end-organ hypoperfusion.

Keywords: Critical care; Multitask learning; RNN; Shock hypotension.

MeSH terms

Arterial Pressure
Critical Care
Critical Illness
Deep Learning*
Humans
Hypotension* / diagnosis