IMON DE ONTIGNY H PSWC 2020
Transcript of IMON DE ONTIGNY H PSWC 2020
SIMON DE MONTIGNY, PHD
Primer on AI for pharmaceutical scientists
October 2, 2020
PSWC [email protected]
Disclosure
I have no relationships with commercial interests.
I hold a grant from the Réseau de recherche en santé respiratoire du Québec, a thematic network supportedby the FRQS.
Research database
Data gathering process of the high resolution database (HRDB) in the Pediatric Intensive Care Unit (PICU) at CHU Sainte-Justine
Brossier et al. Creating a High-Frequency Electronic Database in the PICU: The Perpetual Patient. Pediatr Crit Care Med. 2018 Apr;19(4):e189-e198.
>3,000 patients
>1.2x109 data points
The ‘’perpetual patient’’
Challenges of intensive care data
Compartmentalization Corruption Complexity
Privacy
Integration
Harmonization
Erroneous Data
Missing Data
Imprecise Data
Multimodal Data
State Estimation
Events Prediction
Validation process of HRDB
Mathieu et al. Validation process of a high‐resolution database in a paediatric intensive care unit—Describing the perpetual patient's validation. J Eval Clin Pract. 2020; 1– 9.
Quantitativevalidation
Reference DatabaseMean difference Agreement R2 ICC
(CI95%)Median [min‐max] Missing data Median [min‐max]
Monitors' data
Heart rate (bpm) 117 [49‐199] 25 (2%) 117 [49‐199] −0.019 99.7% 1 1 (1–1)Respiratory rate (bpm) 28 [11‐89] 8 (1%) 28 [11–89] −0.001 99.9% 1 1 (1–1)Pulse oximetry (%) 100 [74‐100] 10 (1%) 100 [74‐100] 0 100.0% 1 1 (1–1)… … … … … … … …
Ventilators' data
Positive end expiratory pressure (cmH2O)
7 [5‐13] 0 7 [5–13] 0 100% 1 1 (1–1)
Positive inspiratory pressure (cmH2O)
18 [8–35] 0 18 [8–35] −0.022 95.5% 1 1 (1–1)
Respiratory rate (rpm) 34 [11‐56] 0 34 [11–56] 0.008 94% 1 1 (1–1)… … … … … … … …Infusion pumpsRate of infusion (ml/h) 1.3 [0‐100] 23 (9%) 1.3 [0–100] 0 100% 1 1 (1–1)
Validation process of HRDB
Brossier et al. Qualitative subjective assessment of a high‐resolution database in a paediatric intensive care unit—Elaborating the perpetual patient's ID card. J Eval Clin Pract. 2020; 26: 86– 91.
Qualitative validation
… …
Patient monitoring
Özcan et al. Alarm in the ICU! Envisioning Patient Monitoring and Alarm Management in Future Intensive Care Units. In: Pfannstiel M., Rasche C. (eds) Service Design and Service Thinking in Healthcare and Hospital Management. Springer, Cham, 2019.
Patient profilingBrossier et al. Perpetual and Virtual Patients for CardiorespiratoryPhysiological Studies. J Pediatr Intensive Care. 2016 Sep;5(3):122-128.
AI in the ICU• Applications
– Severity scoring
– Prediction of sepsis
– Decision support in mechanical ventilation
• Benefits
– Rapid analysis of many physiological variables
– Reduce inter-clinician variability
• Challenges
– Fully electronic data (no paper!)
– Data security and privacy
Lovejoy et al. Artificial intelligence in the intensive care unit. Crit Care 23, 7 (2019).
AI techniques• Recurrent Neural Networks (and variants) for modeling
real time data streams
• Automatized modeling and simulation: System EntityStructure approach to build virtual patients
Rim B et al. Deep Learning in PhysiologicalSignal Data: A Survey. Sensors (Basel). 2020 Feb 11;20(4):969.
Zeigler B et al. Artificial Intelligence in Modeling and Simulation. In: Meyers R. (eds) Encyclopedia of Complexity and Systems Science. Springer, New York, NY, 2009.
Real time analysis
• In a HRDB, even data preprocessing steps can be time consuming
• Algorithm performance is critical
• Example– Synchronization (learning) of
virtual patient with real patient
– Might be needed at differenttimes during an intervention
Thank you!