Can computer simulators accurately represent the pathophysiology of individual COPD patients?

Developing an Artificial Intelligence-Based Representation of a Virtual Patient Model for Real-Time Diagnosis of Acute Respiratory Distress Syndrome

Acute Respiratory Distress Syndrome (ARDS) is a condition that endangers the lives of many Intensive Care Unit patients through gradual reduction of lung function. Due to its heterogeneity, this condition has been difficult to diagnose and treat, although it has been the subject of continuous research, leading to the development of several tools for modeling disease progression on the one hand, and guidelines for diagnosis on the other, mainly the "Berlin Definition". This paper describes the development of a deep learning-based surrogate model of one such tool for modeling ARDS onset in a virtual patient: the Nottingham Physiology Simulator. The model-development process takes advantage of current machine learning and data-analysis techniques, as well as efficient hyperparameter-tuning methods, within a high-performance computing-enabled data science platform. The lightweight models developed through this process present comparable accuracy to the original simulator (per-parameter R² > 0.90). The experimental process described herein serves as a proof of concept for the rapid development and dissemination of specialised diagnosis support systems based on pre-existing generalised mechanistic models, making use of supercomputing infrastructure for the development and testing processes and supported by open-source software for streamlined implementation in clinical routines.

A computational cardiopulmonary physiology simulator accurately predicts individual patient responses to changes in mechanical ventilator settings

We present new results validating the capability of a high-fidelity computational simulator to accurately predict the responses of individual patients with acute respiratory distress syndrome to changes in mechanical ventilator settings. 26 pairs of data-points comprising arterial blood gasses collected before and after changes in inspiratory pressure, PEEP, FiO₂, and I:E ratio from six mechanically ventilated patients were used for this study. Parallelized global optimization algorithms running on a high-performance computing cluster were used to match the simulator to each initial data point. Mean absolute percentage errors between the simulator predicted values of PaO₂ and PaCO₂ and the patient data after changing ventilator parameters were 10.3% and 12.6%, respectively. Decreasing the complexity of the simulator by reducing the number of independent alveolar compartments reduced the accuracy of its predictions. Clinical Relevance- These results provide further evidence that our computational simulator can accurately reproduce patient responses to mechanical ventilation, highlighting its usefulness as a clinical research tool.

Modeling Mechanical Ventilation In Silico-Potential and Pitfalls

Computer simulation offers a fresh approach to traditional medical research that is particularly well suited to investigating issues related to mechanical ventilation. Patients receiving mechanical ventilation are routinely monitored in great detail, providing extensive high-quality data-streams for model design and configuration. Models based on such data can incorporate very complex system dynamics that can be validated against patient responses for use as investigational surrogates. Crucially, simulation offers the potential to "look inside" the patient, allowing unimpeded access to all variables of interest. In contrast to trials on both animal models and human patients, in silico models are completely configurable and reproducible; for example, different ventilator settings can be applied to an identical virtual patient, or the same settings applied to different patients, to understand their mode of action and quantitatively compare their effectiveness. Here, we review progress on the mathematical modeling and computer simulation of human anatomy, physiology, and pathophysiology in the context of mechanical ventilation, with an emphasis on the clinical applications of this approach in various disease states. We present new results highlighting the link between model complexity and predictive capability, using data on the responses of individual patients with acute respiratory distress syndrome to changes in multiple ventilator settings. The current limitations and potential of in silico modeling are discussed from a clinical perspective, and future challenges and research directions highlighted.

What links ventilator driving pressure with survival in the acute respiratory distress syndrome? A computational study

Background

Recent analyses of patient data in acute respiratory distress syndrome (ARDS) showed that a lower ventilator driving pressure was associated with reduced relative risk of mortality. These findings await full validation in prospective clinical trials.

Methods

To investigate the association between driving pressures and ventilator induced lung injury (VILI), we calibrated a high fidelity computational simulator of cardiopulmonary pathophysiology against a clinical dataset, capturing the responses to changes in mechanical ventilation of 25 adult ARDS patients. Each of these in silico patients was subjected to the same range of values of driving pressure and positive end expiratory pressure (PEEP) used in the previous analyses of clinical trial data. The resulting effects on several physiological variables and proposed indices of VILI were computed and compared with data relating ventilator settings with relative risk of death.

Results

Three VILI indices: dynamic strain, mechanical power and tidal recruitment, showed a strong correlation with the reported relative risk of death across all ranges of driving pressures and PEEP. Other variables, such as alveolar pressure, oxygen delivery and lung compliance, correlated poorly with the data on relative risk of death.

Conclusions

Our results suggest a credible mechanistic explanation for the proposed association between driving pressure and relative risk of death. While dynamic strain and tidal recruitment are difficult to measure routinely in patients, the easily computed VILI indicator known as mechanical power also showed a strong correlation with mortality risk, highlighting its potential usefulness in designing more protective ventilation strategies for this patient group.

Electronic supplementary material

The online version of this article (10.1186/s12931-019-0990-5) contains supplementary material, which is available to authorized users.

Inhaled sGC Modulator Can Lower PH in Patients With COPD Without Deteriorating Oxygenation

This study uses a highly fidelity computational simulator of pulmonary physiology to evaluate the impact of a soluble guanylate cyclase (sGC) modulator on gas exchange in patients with chronic obstructive pulmonary disease (COPD) and pulmonary hypertension (PH) as a complication. Three virtual patients with COPD were configured in the simulator based on clinical data. In agreement with previous clinical studies, modeling systemic application of an sGC modulator results in reduced partial pressure of oxygen (PaO2 ) and increased partial pressure of carbon dioxide (PaCO2 ) in arterial blood, if a drug-induced reduction of pulmonary vascular resistance (PVR) equal to that observed experimentally is assumed. In contrast, for administration via dry powder inhalation (DPI), our simulations suggest that the treatment results in no deterioration in oxygenation. For patients under exercise, DPI administration lowers PH, whereas oxygenation is improved with respect to baseline values.

Capturing complexity in pulmonary system modelling

Respiratory disease is a significant problem worldwide, and it is a problem with increasing prevalence. Pathology in the upper airways and lung is very difficult to diagnose and treat, as response to disease is often heterogeneous across patients. Computational models have long been used to help understand respiratory function, and these models have evolved alongside increases in the resolution of medical imaging and increased capability of functional imaging, advances in biological knowledge, mathematical techniques and computational power. The benefits of increasingly complex and realistic geometric and biophysical models of the respiratory system are that they are able to capture heterogeneity in patient response to disease and predict emergent function across spatial scales from the delicate alveolar structures to the whole organ level. However, with increasing complexity, models become harder to solve and in some cases harder to validate, which can reduce their impact clinically. Here, we review the evolution of complexity in computational models of the respiratory system, including successes in translation of models into the clinical arena. We also highlight major challenges in modelling the respiratory system, while making use of the evolving functional data that are available for model parameterisation and testing.

Investigating the effect of cardiac oscillations and deadspace gas mixing during apnea using computer simulation

Gaseous mixing in the anatomical deadspace with stimulation of respiratory ventilation through cardiogenic oscillations is an important physiological mechanism at the onset of apnea, which has been credited with various beneficial effects, e.g. reduction of hypercapnia during the use of low flow ventilation techniques. In this paper, a novel method is proposed to investigate the effect of these mechanisms in silico. An existing computational model of cardio-pulmonary physiology is extended to include the apneic state, gas mixing within the anatomical deadspace, insufflation into the trachea and cardiogenic oscillations. The new model is validated against data published in an experimental animal (dog) study that reported an increase in arterial partial pressure of carbon dioxide (PaCO₂) during apnea. Computational simulations confirm that the model outputs accurately reproduce the available experimental data. This new model can be used to investigate the physiological mechanisms underlying clearance of carbon dioxide during apnea, and hence to develop more effective ventilation strategies for apneic patients.

Hemodynamic effects of lung recruitment maneuvers in acute respiratory distress syndrome

BACKGROUND

Clinical trials have, so far, failed to establish clear beneficial outcomes of recruitment maneuvers (RMs) on patient mortality in acute respiratory distress syndrome (ARDS), and the effects of RMs on the cardiovascular system remain poorly understood.

METHODS

A computational model with highly integrated pulmonary and cardiovascular systems was configured to replicate static and dynamic cardio-pulmonary data from clinical trials. Recruitment maneuvers (RMs) were executed in 23 individual in-silico patients with varying levels of ARDS severity and initial cardiac output. Multiple clinical variables were recorded and analyzed, including arterial oxygenation, cardiac output, peripheral oxygen delivery and alveolar strain.

RESULTS

The maximal recruitment strategy (MRS) maneuver, which implements gradual increments of positive end expiratory pressure (PEEP) followed by PEEP titration, produced improvements in PF ratio, carbon dioxide elimination and dynamic strain in all 23 in-silico patients considered. Reduced cardiac output in the moderate and mild in silico ARDS patients produced significant drops in oxygen delivery during the RM (average decrease of 423 ml min^-1 and 526 ml min^-1, respectively). In the in-silico patients with severe ARDS, however, significantly improved gas-exchange led to an average increase of 89 ml min^-1 in oxygen delivery during the RM, despite a simultaneous fall in cardiac output of more than 3 l min^-1 on average. Post RM increases in oxygen delivery were observed only for the in silico patients with severe ARDS. In patients with high baseline cardiac outputs (>6.5 l min^-1), oxygen delivery never fell below 700 ml min^-1.

CONCLUSIONS

Our results support the hypothesis that patients with severe ARDS and significant numbers of alveolar units available for recruitment may benefit more from RMs. Our results also indicate that a higher than normal initial cardiac output may provide protection against the potentially negative effects of high intrathoracic pressures associated with RMs on cardiac function. Results from in silico patients with mild or moderate ARDS suggest that the detrimental effects of RMs on cardiac output can potentially outweigh the positive effects of alveolar recruitment on oxygenation, resulting in overall reductions in tissue oxygen delivery.

High PEEP in acute respiratory distress syndrome: quantitative evaluation between improved arterial oxygenation and decreased oxygen delivery

BACKGROUND

Positive end-expiratory pressure (PEEP) is widely used to improve oxygenation and prevent alveolar collapse in mechanically ventilated patients with the acute respiratory distress syndrome (ARDS). Although PEEP improves arterial oxygenation predictably, high-PEEP strategies have demonstrated equivocal improvements in ARDS-related mortality. The effect of PEEP on tissue oxygen delivery is poorly understood and is difficult to quantify or investigate in the clinical environment.

METHODS

We investigated the effects of PEEP on tissue oxygen delivery in ARDS using a new, high-fidelity, computational model with highly integrated respiratory and cardiovascular systems. The model was configured to replicate published clinical trial data on the responses of 12 individual ARDS patients to changes in PEEP. These virtual patients were subjected to increasing PEEP levels during a lung-protective ventilation strategy (0-20 cm H₂O). Measured variables included arterial oxygenation, cardiac output, peripheral oxygen delivery, and alveolar strain.

RESULTS

As PEEP increased, tissue oxygen delivery decreased in all subjects (mean reduction of 25% at 20 cm H₂O PEEP), despite an increase in arterial oxygen tension (mean increase 6.7 kPa at 20 cm H₂O PEEP). Changes in arterial oxygenation and tissue oxygen delivery differed between subjects but showed a consistent pattern. Static and dynamic alveolar strain decreased in all patients as PEEP increased.

CONCLUSIONS

Incremental PEEP in ARDS appears to protect alveoli and improve arterial oxygenation, but also appears to impair tissue oxygen delivery significantly because of reduced cardiac output. We propose that this trade-off may explain the poor improvements in mortality associated with high-PEEP ventilation strategies.

Development of an integrated model of cardiovascular and pulmonary physiology for the evaluation of mechanical ventilation strategies

We describe the development of an integrated cardiovascular and pulmonary model for use in the investigation of novel mechanical ventilation strategies in the intensive care unit. The cardiac model includes the cardiac chambers, the pulmonary circulation and the systemic circulation. The modeling of complex mechanisms for vascular segments, time varying elastance functions of cardiovascular components and the effect of vascular resistances, in health and disease under the influence of mechanical ventilation is investigated. The resulting biomedical simulator can aid in understanding the underlying pathophysiology of critically-ill patients and facilitate the development of more effective therapeutic strategies for evaluation in clinical trials.

Computational simulation indicates that moderately high-frequency ventilation can allow safe reduction of tidal volumes and airway pressures in ARDS patients

BACKGROUND

A recent prospective trial using porcine models of severe acute respiratory distress syndrome (ARDS) indicated that positive-pressure ventilation delivered by a conventional intensive care ventilator at a moderately high frequency allows safe reduction of tidal volume below 6 ml/kg, leading to more protective ventilation. We aimed to explore whether these results would be replicated when implementing similar ventilation strategies in a high-fidelity computational simulator, tuned to match data on the responses of a number of human ARDS patients to different ventilator inputs.

METHODS

We evaluated three different strategies for managing the trade-off between increasing respiratory rate and reducing tidal volume while attempting to maintain the partial pressure of carbon dioxide in arterial blood (PaCO2) constant on a computational simulator configured with ARDS patient datasets.

RESULTS

For a fixed sequence of stepwise increases in the respiratory rate, corresponding decreases in tidal volume to keep the alveolar minute ventilation and inspiratory flow constant were calculated according to standard formulae. When applied on the simulator, however, these sequences of ventilator settings failed to maintain PaCO2 adequately in the virtual patients considered. In contrast, an approach based on combining numerical optimisation methods with computational simulation allowed a sequence of tidal volume reductions to be computed for each virtual patient that maintained PaCO2 levels while significantly reducing peak airway pressures and dynamic alveolar strain in all patients.

CONCLUSIONS

Our study supports the proposition that moderately high-frequency respiratory rates can allow more protective ventilation of ARDS patients and highlights the potential role of high-fidelity simulators in computing optimised and personalised ventilator settings for individual patients using this approach.

Evaluation of lung recruitment maneuvers in acute respiratory distress syndrome using computer simulation

Introduction

Direct comparison of the relative efficacy of different recruitment maneuvers (RMs) for patients with acute respiratory distress syndrome (ARDS) via clinical trials is difficult, due to the heterogeneity of patient populations and disease states, as well as a variety of practical issues. There is also significant uncertainty regarding the minimum values of positive end-expiratory pressure (PEEP) required to ensure maintenance of effective lung recruitment using RMs. We used patient-specific computational simulation to analyze how three different RMs act to improve physiological responses, and investigate how different levels of PEEP contribute to maintaining effective lung recruitment.

Methods

We conducted experiments on five ‘virtual’ ARDS patients using a computational simulator that reproduces static and dynamic features of a multivariable clinical dataset on the responses of individual ARDS patients to a range of ventilator inputs. Three recruitment maneuvers (sustained inflation (SI), maximal recruitment strategy (MRS) followed by a titrated PEEP, and prolonged recruitment maneuver (PRM)) were implemented and evaluated for a range of different pressure settings.

Results

All maneuvers demonstrated improvements in gas exchange, but the extent and duration of improvement varied significantly, as did the observed mechanism of operation. Maintaining adequate post-RM levels of PEEP was seen to be crucial in avoiding cliff-edge type re-collapse of alveolar units for all maneuvers. For all five patients, the MRS exhibited the most prolonged improvement in oxygenation, and we found that a PEEP setting of 35 cm H₂O with a fixed driving pressure of 15 cm H₂O (above PEEP) was sufficient to achieve 95% recruitment. Subsequently, we found that PEEP titrated to a value of 16 cm H₂O was able to maintain 95% recruitment in all five patients.

Conclusions

There appears to be significant scope for reducing the peak levels of PEEP originally specified in the MRS and hence to avoid exposing the lung to unnecessarily high pressures. More generally, our study highlights the huge potential of computer simulation to assist in evaluating the efficacy of different recruitment maneuvers, in understanding their modes of operation, in optimizing RMs for individual patients, and in supporting clinicians in the rational design of improved treatment strategies.

Electronic supplementary material

The online version of this article (doi:10.1186/s13054-014-0723-6) contains supplementary material, which is available to authorized users.