AUTOPILOT-BT: A system for knowledge and model based mechanical ventilation

Computer clinical decision support that automates personalized clinical care: a challenging but needed healthcare delivery strategy

How to deliver best care in various clinical settings remains a vexing problem. All pertinent healthcare-related questions have not, cannot, and will not be addressable with costly time- and resource-consuming controlled clinical trials. At present, evidence-based guidelines can address only a small fraction of the types of care that clinicians deliver. Furthermore, underserved areas rarely can access state-of-the-art evidence-based guidelines in real-time, and often lack the wherewithal to implement advanced guidelines. Care providers in such settings frequently do not have sufficient training to undertake advanced guideline implementation. Nevertheless, in advanced modern healthcare delivery environments, use of eActions (validated clinical decision support systems) could help overcome the cognitive limitations of overburdened clinicians. Widespread use of eActions will require surmounting current healthcare technical and cultural barriers and installing clinical evidence/data curation systems. The authors expect that increased numbers of evidence-based guidelines will result from future comparative effectiveness clinical research carried out during routine healthcare delivery within learning healthcare systems.

Prediction of high airway pressure using a non-linear autoregressive model of pulmonary mechanics

Background

For mechanically ventilated patients with acute respiratory distress syndrome (ARDS), suboptimal PEEP levels can cause ventilator induced lung injury (VILI). In particular, high PEEP and high peak inspiratory pressures (PIP) can cause over distension of alveoli that is associated with VILI. However, PEEP must also be sufficient to maintain recruitment in ARDS lungs. A lung model that accurately and precisely predicts the outcome of an increase in PEEP may allow dangerous high PIP to be avoided, and reduce the incidence of VILI.

Methods and results

Sixteen pressure-flow data sets were collected from nine mechanically ventilated ARDs patients that underwent one or more recruitment manoeuvres. A nonlinear autoregressive (NARX) model was identified on one or more adjacent PEEP steps, and extrapolated to predict PIP at 2, 4, and 6 cmH₂O PEEP horizons. The analysis considered whether the predicted and measured PIP exceeded a threshold of 40 cmH₂O. A direct comparison of the method was made using the first order model of pulmonary mechanics (FOM(I)). Additionally, a further, more clinically appropriate method for the FOM was tested, in which the FOM was trained on a single PEEP prior to prediction (FOM(II)). The NARX model exhibited very high sensitivity (> 0.96) in all cases, and a high specificity (> 0.88). While both FOM methods had a high specificity (> 0.96), the sensitivity was much lower, with a mean of 0.68 for FOM(I), and 0.82 for FOM(II).

Conclusions

Clinically, false negatives are more harmful than false positives, as a high PIP may result in distension and VILI. Thus, the NARX model may be more effective than the FOM in allowing clinicians to reduce the risk of applying a PEEP that results in dangerously high airway pressures.

The Equilibration of PCO2 in Pigs Is Independent of Lung Injury and Hemodynamics

OBJECTIVES

In mechanical ventilation, normoventilation in terms of PCO2 can be achieved by titration of the respiratory rate and/or tidal volume. Although a linear relationship has been found between changes in respiratory rate and resulting changes in end-tidal cO2 (△PetCO2) as well as between changes in respiratory rate and equilibration time (teq) for mechanically ventilated patients without lung injury, it is unclear whether a similar relationship holds for acute lung injury or altered hemodynamics.

DESIGN

We performed a prospective randomized controlled animal study of the change in PetCO2 with changes in respiratory rate in a lung-healthy, lung-injury, lung-healthy + altered hemodynamics, and lung-injury + altered hemodynamics pig model.

SETTING

University research laboratory.

SUBJECTS

Twenty mechanically ventilated pigs.

INTERVENTIONS

Moderate lung injury was induced by injection of oleic acid in 10 randomly assigned pigs, and after the first round of measurements, cardiac output was increased by approximately 30% by constant administration of noradrenalin in both groups.

MEASUREMENTS AND MAIN RESULTS

We systematically increased and decreased changes in respiratory rate according to a set protocol: +2, -4, +6, -8, +10, -12, +14 breaths/min and awaited equilibration of Petco2. We found a linear relationship between changes in respiratory rate and △PetCO2 as well as between changes in respiratory rate and teq. A two-sample t test resulted in no significant differences between the lung injury and healthy control group before or after hemodynamic intervention. Furthermore, exponential extrapolation allowed prediction of the new PetCO2 equilibrium and teq after 5.7 ± 5.6 min.

CONCLUSIONS

The transition between PetCO2 equilibria after changes in respiratory rate might not be dependent on moderate lung injury or cardiac output but on the metabolic production or capacity of cO2 stores. Linear relationships previously found for lung-healthy patients and early prediction of PetCO2 equilibration could therefore also be used for the titration of respiratory rate on the PetCO2 for a wider range of pathologies by the physician or an automated ventilation system.

Utility of a novel error-stepping method to improve gradient-based parameter identification by increasing the smoothness of the local objective surface: a case-study of pulmonary mechanics

Accurate model parameter identification relies on accurate forward model simulations to guide convergence. However, some forward simulation methodologies lack the precision required to properly define the local objective surface and can cause failed parameter identification. The role of objective surface smoothness in identification of a pulmonary mechanics model was assessed using forward simulation from a novel error-stepping method and a proprietary Runge-Kutta method. The objective surfaces were compared via the identified parameter discrepancy generated in a Monte Carlo simulation and the local smoothness of the objective surfaces they generate. The error-stepping method generated significantly smoother error surfaces in each of the cases tested (p<0.0001) and more accurate model parameter estimates than the Runge-Kutta method in three of the four cases tested (p<0.0001) despite a 75% reduction in computational cost. Of note, parameter discrepancy in most cases was limited to a particular oblique plane, indicating a non-intuitive multi-parameter trade-off was occurring. The error-stepping method consistently improved or equalled the outcomes of the Runge-Kutta time-integration method for forward simulations of the pulmonary mechanics model. This study indicates that accurate parameter identification relies on accurate definition of the local objective function, and that parameter trade-off can occur on oblique planes resulting prematurely halted parameter convergence.

Structural identifiability and practical applicability of an alveolar recruitment model for ARDS patients

Patient-specific mathematical models of respiratory mechanics can offer substantial insight into patient state and pulmonary dynamics that are not directly measurable. Thus, they offer significant potential to evaluate and guide patient-specific lung protective ventilator strategies for acute respiratory distress syndrome (ARDS) patients. To assure bedside applicability, the model must be computationally efficient and identifiable from the limited available data, while also capturing dominant dynamics and trends observed in ARDS patients. In this study, an existing static recruitment model is enhanced by considering alveolar distension and implemented in a novel time-continuous dynamic respiratory mechanics model. The model was tested for structural identifiability and a hierarchical gradient descent approach was used to fit the model to low-flow test responses of 12 ARDS patients. Finally, a comprehensive practical identifiability analysis was performed to evaluate the impact of data quality on the model parameters. Identified parameter values were physiologically plausible and very accurately reproduced the measured pressure responses. Structural identifiability of the model was proven, but practical identifiability analysis of the results showed a lack of convexity on the error surface indicating that successful parameter identification is currently not assured in all test sets. Overall, the model presented is physiologically and clinically relevant, captures ARDS dynamics, and uses clinically descriptive parameters. The patient-specific models show the ability to capture pulmonary dynamics directly relevant to patient condition and clinical guidance. These characteristics currently cannot be directly measured or established without such a validated model.

Dynamically generated models for medical decision support systems

Doctors applying mechanical ventilation need to find the best balance between benefit and risk for the patient. Mathematical models simulating patient's reactions to alterations in the ventilation regime may be employed. A framework is introduced that is able to dynamically combine mathematical models from different model families to form a complex interacting model system. Each of these families consists of submodels differing in complexity of dynamics formulation or anatomical/geometrical resolution. The interaction of model systems reveals qualitatively varying results depending on the complexity of the involved models. Realistic overlaying of respiratory and cardiovascular rhythms can be detected in blood gas concentrations.

Inverse modeling supports quantification of pressure and time depending effects in ARDS patients

The application of respiratory mechanics models combined with standardized ventilation maneuvers enable investigations of patients' lung mechanics at the bedside in order to optimize ventilation therapy. Therefore, the underlying dynamic effects of respiratory mechanics (viscoelasticity, inhomogeneity and recruitment) are uncovered by applying various ventilation maneuvers and subsequently captured by the corresponding model via parameter identification methods. Data sets of patients undergoing quasi-static and dynamic ventilation patterns are available along with a hierarchical model structure for parameter identification and simulation purposes. The applicability of the basic 1(st) order model (FOM) of respiratory mechanics for various flow rates proved to be critical and patient dependent, since distinctive time-depending effects could not be considered. To improve this, a 2(nd) order model (SOM), individualized using data of a SCASS maneuver (Static Compliance Automated Single Step), enables successful simulations of respiratory mechanics in dynamic and quasi-static conditions. Pressure dependent effects such as static recruitment, can be captured by Hickling's nonlinear compliance model. This research illustrates the applicability of various models of respiratory mechanics within the model hierarchy in various circumstances and the ability to distinguish between dynamic and static effects.

Automated mechanical ventilation: adapting decision making to different disease states

The purpose of the present study is to introduce a novel methodology for adapting and upgrading decision-making strategies concerning mechanical ventilation with respect to different disease states into our fuzzy-based expert system, AUTOPILOT-BT. The special features are: (1) Extraction of clinical knowledge in analogy to the daily routine. (2) An automated process to obtain the required information and to create fuzzy sets. (3) The controller employs the derived fuzzy rules to achieve the desired ventilation status. For demonstration this study focuses exclusively on the control of arterial CO(2) partial pressure (p(a)CO(2)). Clinical knowledge from 61 anesthesiologists was acquired using a questionnaire from which different disease-specific fuzzy sets were generated to control p(a)CO(2). For both, patients with healthy lung and with acute respiratory distress syndrome (ARDS) the fuzzy sets show different shapes. The fuzzy set "normal", i.e., "target p(a)CO(2) area", ranges from 35 to 39 mmHg for healthy lungs and from 39 to 43 mmHg for ARDS lungs. With the new fuzzy sets our AUTOPILOT-BT reaches the target p(a)CO(2) within maximal three consecutive changes of ventilator settings. Thus, clinical knowledge can be extended, updated, and the resulting mechanical ventilation therapies can be individually adapted, analyzed, and evaluated.