Virtual patient with temporal evolution for mechanical ventilation trial studies: A stochastic model approach.

Area of Science:

• Computational biology
• Medical informatics
• Biomedical engineering

Background:

• Healthcare datasets often suffer from data scarcity and class imbalance, hindering research and development.
• Clinically validated virtual patient (VP) models offer a solution for synthetic data generation in critical care settings.
• This research focuses on creating a time-varying VP profile for mechanically ventilated respiratory failure.

Purpose of the Study:

• To synthesize a realistic, time-varying virtual patient (VP) profile for mechanically ventilated respiratory failure using a stochastic model.
• To demonstrate the feasibility of using these VPs for in-silico simulation and validation.
• To reduce reliance on traditional clinical trials for developing mechanical ventilation protocols.

Main Methods:

• Developed a stochastic model using respiratory elastance (Ers) data from clinical cohorts, averaging over 30-minute intervals.
• Generated future Ers data with normally distributed random noise for VP synthesis.
• Validated VPs via Monte Carlo simulation and retrospective Ers profile fitting, comparing synthesized cohorts to independent patient data.

Main Results:

• Generated 120,000 three-hour VPs for pressure control (PC) and volume control (VC) ventilation modes.
• Optimized stochastic simulation with 5-10% noise and 200,000 iterations for realistic Ers profiles.
• Achieved high accuracy in self-validation (MSE < 0.099% for PC, < 0.051% for VC) and demonstrated cohort-level validation through virtual trials.

Conclusions:

• Temporally evolving VPs are feasible for designing, developing, and optimizing bedside mechanical ventilation (MV) guidance protocols.
• Stochastic simulation-based VPs reduce the need for lengthy, expensive clinical trials.
• These VPs facilitate statistically robust virtual trials, ultimately improving patient care in mechanical ventilation.