iCVS - Inferring Cardio-Vascular hidden States from physiological signals available at the bedside

Neta Ravid Tannenbaum, Omer Gottesman, Azadeh Assadi, Mjaye Mazwi, Uri Shalit, Danny Eytan

Abstract

Intensive care medicine is complex and resource-demanding. A critical and common challenge lies in inferring the underlying physiological state of a patient from partially observed data. Specifically for the cardiovascular system, clinicians use observables such as heart rate, arterial and venous blood pressures, as well as findings from the physical examination and ancillary tests to formulate a mental model and estimate hidden variables such as cardiac output, vascular resistance, filling pressures and volumes, and autonomic tone. Then, they use this mental model to derive the causes for instability and choose appropriate interventions. Not only this is a very hard problem due to the nature of the signals, but it also requires expertise and a clinician’s ongoing presence at the bedside. Clinical decision support tools based on mechanistic dynamical models offer an appealing solution due to their inherent explainability, corollaries to the clinical mental process, and predictive power. With a translational motivation in mind, we developed iCVS: a simple, with high explanatory power, dynamical mechanistic model to infer hidden cardiovascular states. Full model estimation requires no prior assumptions on physiological parameters except age and weight, and the only inputs are arterial and venous pressure waveforms. iCVS also considers autonomic and non-autonomic modulations. To gain more information without increasing model complexity, both slow and fast timescales of the blood pressure traces are exploited, while the main inference and dynamic evolution are at the longer, clinically relevant, timescale of minutes. iCVS is designed to allow bedside deployment at pediatric and adult intensive care units and for retrospective investigation of cardiovascular mechanisms underlying instability. In this paper, we describe iCVS and inference system in detail, and using a dataset of critically-ill children, we demonstrate its use and power to identify bleeding, distributive states, and cardiac dysfunction, in isolation and in combination.

Introduction

In the Intensive Care Unit (ICU), clinicians act as physiological detectives. When a patient’s blood pressure drops, the immediate challenge is to determine the underlying cause: Is it heart failure, hemorrhage, or excessive vasodilation due to sepsis? To answer this, they must map "observables" like heart rate and blood pressure to "hidden states" such as stroke volume and systemic vascular resistance.

However, this process is fraught with difficulty. Bedside signals are often noisy due to patient movement or medical interventions, and physiological states can shift rapidly. Existing mathematical models are often too complex for practical bedside use, requiring numerous unknown parameters, or they fail to account for the physiological diversity between neonates and adolescents. iCVS (inferring Cardio-Vascular States) was developed to address these gaps by providing a mechanistic framework that translates raw data into actionable clinical insights.

The iCVS workflow: from raw bedside data to hidden physiological parameters.

Methods

The iCVS model is a "lumped parameter" mechanistic model, analogous to an electrical circuit where the heart acts as a pump, blood vessels as capacitors, and organs as resistors.

Model Architecture

The model represents the cardiovascular system using a simplified structure including a single heart chamber, arterial and venous compartments, and organ resistance. It accounts for two types of nervous system responses: the baroreflex (automatic feedback) and an independent component representing external stressors like pain.

The iCVS mechanistic model structure, representing the heart, venous/arterial compartments, and organs.

Key Innovations

  • Two-Timescale Analysis: The model utilizes the "fast" timescale of individual heartbeats to estimate stroke volume and resistance, while performing primary inference on a "slow" timescale of minutes. This approach provides high-resolution insights without excessive computational overhead.
  • Minimal Priors: Unlike many models that require extensive baseline data, iCVS only requires the patient's age and weight. All other parameters are inferred directly from arterial and venous pressure waveforms.

Validation Dataset

The model was validated using data from 10 critically ill children in a Pediatric Intensive Care Unit (PICU), ranging from neonates (<5kg) to adolescents (60kg). The cohort included cases of hypovolemic, distributive, and cardiogenic shock, with ground truth labels provided by expert clinician review.

Results

The iCVS model demonstrated high fidelity in reconstructing physiological signals, with Pearson correlation coefficients ranging from 0.71 to 0.94 for parameters like pulse pressure and heart rate.

Identification of Shock States

  • Hypovolemia: The model successfully identified documented post-operative bleeding by detecting negative volume changes ($\bar{I}_{ex}$). Notably, it identified "silent" bleeding in a neonate where heart rate changes were not clinically obvious.
  • Distributive Shock: iCVS accurately mapped low vascular resistance in patients diagnosed with vasoplegia.
  • Combined States: In complex scenarios involving both heart failure and bleeding, the model successfully decoupled these factors, showing simultaneous drops in contractility and volume.

Discussion and Future Work

By basing its outputs on physics and physiology (Ordinary Differential Equations), iCVS offers inherent explainability, which is crucial for building clinician trust compared to "black box" AI models. Its ability to adapt to patients across a wide weight spectrum highlights its potential for broad clinical application.

Current limitations include the use of a simplified one-chamber heart model and the lack of accounting for mechanical ventilation effects or complex cardiac anatomy. Future work will focus on real-time bedside deployment and the potential integration of mechanistic models with deep learning pattern recognition to enhance predictive power.

Summary updated

Abstract

Abstract Intensive care medicine is complex and resource-demanding. A critical and common challenge lies in inferring the underlying physiological state of a patient from partially observed data. Specifically for the cardiovascular system, clinicians use observables such as heart rate, arterial and venous blood pressures, as well as findings from the physical examination and ancillary tests to formulate a mental model and estimate hidden variables such as cardiac output, vascular resistance, filling pressures and volumes, and autonomic tone. Then, they use this mental model to derive the causes for instability and choose appropriate interventions. Not only this is a very hard problem due to the nature of the signals, but it also requires expertise and a clinician’s ongoing presence at the bedside. Clinical decision support tools based on mechanistic dynamical models offer an appealing solution due to their inherent explainability, corollaries to the clinical mental process, and predictive power. With a translational motivation in mind, we developed iCVS: a simple, with high explanatory power, dynamical mechanistic model to infer hidden cardiovascular states. Full model estimation requires no prior assumptions on physiological parameters except age and weight, and the only inputs are arterial and venous pressure waveforms. iCVS also considers autonomic and non-autonomic modulations. To gain more information without increasing model complexity, both slow and fast timescales of the blood pressure traces are exploited, while the main inference and dynamic evolution are at the longer, clinically relevant, timescale of minutes. iCVS is designed to allow bedside deployment at pediatric and adult intensive care units and for retrospective investigation of cardiovascular mechanisms underlying instability. In this paper, we describe iCVS and inference system in detail, and using a dataset of critically-ill children, we demonstrate its use and power to identify bleeding, distributive states, and cardiac dysfunction, in isolation and in combination. Author summary A common challenge clinicians face across different disciplines is estimating the hidden physiological state of a patient based on partially observed data. Here we describe iCVS (inferring Cardio-Vascular States): a dynamical mechanistic model of the cardiovascular system. We developed iCVS with a translational goal in mind, showing high explanatory power, its inference relies only on routinely available signals, and enables the identification of various clinically important shock states. We demonstrate the use of the model on a dataset that was collected in a pediatric intensive care unit.
BibTeX
@article{tannenbaum2023icvs,
  author = {Neta Ravid Tannenbaum and Omer Gottesman and Azadeh Assadi and Mjaye Mazwi and Uri Shalit and Danny Eytan},
  title = {{iCVS - Inferring Cardio-Vascular hidden States from physiological signals available at the bedside}},
  journal = {PLOS Computational Biology},
  year = {2023},
  doi = {10.1371/journal.pcbi.1010835},
}
Figure