Prediction of Hemodynamic Response to Epinephrine via Model-Based System Identification

A Lumped-Parameter Model of the Cardiovascular System Response for Evaluating Automated Fluid Resuscitation Systems

Physiological closed-loop controlled (PCLC) medical devices, such as those designed for blood pressure regulation, can be tested for safety and efficacy in real-world clinical settings. However, relying solely on limited animal and clinical studies may not capture the diverse range of physiological conditions. Credible mathematical models can complement these studies by allowing the testing of the device against simulated patient scenarios. This research involves the development and validation of a low-order lumped-parameter mathematical model of the cardiovascular system's response to fluid perturbation. The model takes rates of hemorrhage and fluid infusion as inputs and provides hematocrit and blood volume, heart rate, stroke volume, cardiac output and mean arterial blood pressure as outputs. The model was calibrated using data from 27 sheep subjects, and its predictive capability was evaluated through a leave-one-out cross-validation procedure, followed by independent validation using 12 swine subjects. Our findings showed small model calibration error against the training dataset, with the normalized root-mean-square error (NRMSE) less than 10% across all variables. The mathematical model and virtual patient cohort generation tool demonstrated a high level of predictive capability and successfully generated a sufficient number of subjects that closely resembled the test dataset. The average NRMSE for the best virtual subject, across two distinct samples of virtual subjects, was below 12.7% and 11.9% for the leave-one-out cross-validation and independent validation dataset. These findings suggest that the model and virtual cohort generator are suitable for simulating patient populations under fluid perturbation, indicating their potential value in PCLC medical device evaluation.

Unifying cardiovascular modelling with deep reinforcement learning for uncertainty aware control of sepsis treatment

Sepsis is a potentially life-threatening inflammatory response to infection or severe tissue damage. It has a highly variable clinical course, requiring constant monitoring of the patient's state to guide the management of intravenous fluids and vasopressors, among other interventions. Despite decades of research, there's still debate among experts on optimal treatment. Here, we combine for the first time, distributional deep reinforcement learning with mechanistic physiological models to find personalized sepsis treatment strategies. Our method handles partial observability by leveraging known cardiovascular physiology, introducing a novel physiology-driven recurrent autoencoder, and quantifies the uncertainty of its own results. Moreover, we introduce a framework for uncertainty-aware decision support with humans in the loop. We show that our method learns physiologically explainable, robust policies, that are consistent with clinical knowledge. Further our method consistently identifies high-risk states that lead to death, which could potentially benefit from more frequent vasopressor administration, providing valuable guidance for future research.

Machine Learning Prediction Models for Mechanically Ventilated Patients: Analyses of the MIMIC-III Database

Background: Mechanically ventilated patients in the intensive care unit (ICU) have high mortality rates. There are multiple prediction scores, such as the Simplified Acute Physiology Score II (SAPS II), Oxford Acute Severity of Illness Score (OASIS), and Sequential Organ Failure Assessment (SOFA), widely used in the general ICU population. We aimed to establish prediction scores on mechanically ventilated patients with the combination of these disease severity scores and other features available on the first day of admission.

Methods: A retrospective administrative database study from the Medical Information Mart for Intensive Care (MIMIC-III) database was conducted. The exposures of interest consisted of the demographics, pre-ICU comorbidity, ICU diagnosis, disease severity scores, vital signs, and laboratory test results on the first day of ICU admission. Hospital mortality was used as the outcome. We used the machine learning methods of k-nearest neighbors (KNN), logistic regression, bagging, decision tree, random forest, Extreme Gradient Boosting (XGBoost), and neural network for model establishment. A sample of 70% of the cohort was used for the training set; the remaining 30% was applied for testing. Areas under the receiver operating characteristic curves (AUCs) and calibration plots would be constructed for the evaluation and comparison of the models' performance. The significance of the risk factors was identified through models and the top factors were reported.

Results: A total of 28,530 subjects were enrolled through the screening of the MIMIC-III database. After data preprocessing, 25,659 adult patients with 66 predictors were included in the model analyses. With the training set, the models of KNN, logistic regression, decision tree, random forest, neural network, bagging, and XGBoost were established and the testing set obtained AUCs of 0.806, 0.818, 0.743, 0.819, 0.780, 0.803, and 0.821, respectively. The calibration curves of all the models, except for the neural network, performed well. The XGBoost model performed best among the seven models. The top five predictors were age, respiratory dysfunction, SAPS II score, maximum hemoglobin, and minimum lactate.

Conclusion: The current study indicates that models with the risk of factors on the first day could be successfully established for predicting mortality in ventilated patients. The XGBoost model performs best among the seven machine learning models.

Accuracy assessment methods for physiological model selection toward evaluation of closed-loop controlled medical devices

Physiological closed-loop controlled (PCLC) medical devices are complex systems integrating one or more medical devices with a patient’s physiology through closed-loop control algorithms; introducing many failure modes and parameters that impact performance. These control algorithms should be tested through safety and efficacy trials to compare their performance to the standard of care and determine whether there is sufficient evidence of safety for their use in real care setting. With this aim, credible mathematical models have been constructed and used throughout the development and evaluation phases of a PCLC medical device to support the engineering design and improve safety aspects. Uncertainties about the fidelity of these models and ambiguities about the choice of measures for modeling performance need to be addressed before a reliable PCLC evaluation can be achieved. This research develops tools for evaluating the accuracy of physiological models and establishes fundamental measures for predictive capability assessment across different physiological models. As a case study, we built a refined physiological model of blood volume (BV) response by expanding an original model we developed in our prior work. Using experimental data collected from 16 sheep undergoing hemorrhage and fluid resuscitation, first, we compared the calibration performance of the two candidate physiological models, i.e., original and refined, using root-mean-squared error (RMSE), Akiake information criterion (AIC), and a new multi-dimensional approach utilizing normalized features extracted from the fitting error. Compared to the original model, the refined model demonstrated a significant improvement in calibration performance in terms of RMSE (9%, P = 0.03) and multi-dimensional measure (48%, P = 0.02), while a comparable AIC between the two models verified that the enhanced calibration performance in the refined model is not due to data over-fitting. Second, we compared the physiological predictive capability of the two models under three different scenarios: prediction of subject-specific steady-state BV response, subject-specific transient BV response to hemorrhage perturbation, and leave-one-out inter-subject BV response. Results indicated enhanced accuracy and predictive capability for the refined physiological model with significantly larger proportion of measurements that were within the prediction envelope in the transient and leave-one-out prediction scenarios (P < 0.02). All together, this study helps to identify and merge new methods for credibility assessment and physiological model selection, leading to a more efficient process for PCLC medical device evaluation.

Controlled infusion of intravenous cardiac drugs using global optimization

OBJECTIVES:

The objective of the study is to develop an automatic drug infusion control system during cardiovascular surgery.

MATERIALS AND METHODS:

Based on the clinical drug dosage analysis, the modeling of cardiovascular system with baroreceptor model is mathematically modeled using compartmental approach, considering the relationship between the volume and flow rate of blood during each heartbeat. This model is then combined with drug modeling of noradrenaline and nitroglycerine by deriving the volume and drug mass concentration equations, based on pharmacokinetics and pharmacodynamics of the drugs. The closed-loop patient models are derived from the open-loop data obtained from the physiology-drug model with covariate as age. The proportional-integral controller is designed based on optimal values obtained from bacterial foraging-oriented particle swarm optimization algorithm. The controllers are implemented individually for each control variable such as aortic pressure and cardiac output (CO), irrespective of varying weights based on the relative gain array analysis which depicts the maximum influence of cardiac drugs on control variables.

RESULTS:

The physiology-drug model output responses are simulated using MATLAB. The controlled responses of aortic pressure and CO with infusion rate of cardiac drugs are obtained. The robustness of the controller is checked by introducing variations in cardiovascular model parameters. The efficiency of the controller during normal and abnormal conditions is compared using time domain analysis.

CONCLUSIONS:

The controller design was efficient and can be further improved by designing switching-based controllers.

Regulatory Considerations for Physiological Closed-Loop Controlled Medical Devices Used for Automated Critical Care: Food and Drug Administration Workshop Discussion Topics

Part of the mission of the Center for Devices and Radiological Health (CDRH) at the US Food and Drug Administration is to facilitate medical device innovation. Therefore, CDRH plays an important role in helping its stakeholders such as manufacturers, health care professionals, patients, patient advocates, academia, and other government agencies navigate the regulatory landscape for medical devices. This is particularly important for innovative physiological closed-loop controlled (PCLC) devices used in critical care environments, such as intensive care units, emergency settings, and battlefield environments. CDRH's current working definition of a PCLC medical device is a medical device that incorporates physiological sensor(s) for automatic manipulation of a physiological variable through actuation of therapy that is conventionally made by a clinician. These emerging devices enable automatic therapy delivery and may have the potential to revolutionize the standard of care by ensuring adequate and timely therapy delivery with improved performance in high workload and high-stress environments. For emergency response and military applications, automatic PCLC devices may play an important role in reducing cognitive overload, minimizing human error, and enhancing medical care during surge scenarios (ie, events that exceed the capability of the normal medical infrastructure). CDRH held an open public workshop on October 13 and 14, 2015 with the aim of fostering an open discussion on design, implementation, and evaluation considerations associated with PCLC devices used in critical care environments. CDRH is currently developing regulatory recommendations and guidelines that will facilitate innovation for PCLC devices. This article highlights the contents of the white paper that was central to the workshop and focuses on the ensuing discussions regarding the engineering, clinical, and human factors considerations.

Reduced Rank Least Squares for Real-Time Short Term Estimation of Mean Arterial Blood Pressure in Septic Patients Receiving Norepinephrine

Norepinephrine (NE), an endogenous catecholamine, is a mainstay treatment for septic shock, which is a life-threatening manifestation of severe infection. NE counteracts the loss in blood pressure associated with septic shock. However, an NE infusion that is too low fails to counteract the blood pressure drop, and an NE infusion that is too high can cause a hypertensive crisis and heart attack. Ideally, the NE infusion rate should maintain a patient’s mean arterial blood pressure (MAP) above 65 mmHg. There are a few data-driven, quantitative models to predict the MAP, and incorporate NE effects. This paper presents a model, driven by intensive care unit (ICU) measurable data and known NE inputs, to predict the future MAP of an ICU patient. We derive a least square estimation model for MAP based on available ICU data, including heart period, NE infusion rate, and respiration wave. We learn the parameters of our model from initial patient data and then use this information to predict future MAP data. We assess our model with data from 12 septic patients. Our model successfully predicts and tracks MAP when the NE infusion rate changes. Specifically, we predict MAP 3 to 20 min in the future with the mean error of less than 4 to 7 mmHg over 12 patients. Conclusion: this new approach creates the potential to advance methods for predicting NE infusion rate in septic patients. Significance: successfully predicted patients’ MAP could reduce catastrophic human error and lessen clinicians’ workload.