Affiliation: Department of Physiology and Biophysics, Center for Computational Medicine, University of Mississippi Medical Center, Jackson, Mississippi, United States of America, 39215
Auto-generated profile, based on publicly available information
William A. Pruett
Affiliation: Department of Physiology and Biophysics, Center for Computational Medicine, University of Mississippi Medical Center, Jackson, Mississippi, United States of America, 39215
A surrogate model is a black box model that reproduces the output of another more complex model at a single time point. This is to be distinguished from the method of surrogate data, used in time series. The purpose of a surrogate is to reduce the time necessary for a computation at the cost of rigor and generality. We describe a method of constructing surrogates in the form of support vector machine (SVM) regressions for the purpose of exploring the parameter space of physiological models. Our focus is on the methodology of surrogate creation and accuracy assessment in comparison to the original model. This is done in the context of a simulation of hemorrhage in one model, “Small”, and renal denervation in another, HumMod. In both cases, the surrogate predicts the drop in mean arterial pressure following the intervention. We asked three questions concerning surrogate models: (1) how many training examples are necessary to obtain an accurate surrogate, (2) is surrogate accuracy homogeneous, and (3) how much can computation time be reduced when using a surrogate. We found the minimum training set size that would guarantee maximal accuracy was widely variable, but could be algorithmically generated. The average error for the pressure response to the protocols was -0.05±2.47 in Small, and -0.3 +/- 3.94 mmHg in HumMod. In the Small model, error grew with actual pressure drop, and in HumMod, larger pressure drops were overestimated by the surrogates. Surrogate use resulted in a 6 order of magnitude decrease in computation time. These results suggest surrogate modeling is a valuable tool for generating predictions of an integrative model’s behavior on densely sampled subsets of its parameter space.
We present a small integrative model of human cardiovascular physiology. The model is population-based; rather than using best fit parameter values, we used a variant of the Metropolis algorithm to produce distributions for the parameters most associated with model sensitivity. The population is built by sampling from these distributions to create the model coefficients. The resulting models were then subjected to a hemorrhage. The population was separated into those that lost less than 15 mmHg arterial pressure (compensators), and those that lost more (decompensators). The populations were parametrically analyzed to determine baseline conditions correlating with compensation and decompensation. Analysis included single variable correlation, graphical time series analysis, and support vector machine (SVM) classification. Most variables were seen to correlate with propensity for circulatory collapse, but not sufficiently to effect reasonable classification by any single variable. Time series analysis indicated a single significant measure, the stressed blood volume, as predicting collapse in situ, but measurement of this quantity is clinically impossible. SVM uncovered a collection of variables and parameters that, when taken together, provided useful rubrics for classification. Due to the probabilistic origins of the method, multiple classifications were attempted, resulting in an average of 3.5 variables necessary to construct classification. The most common variables used were systemic compliance, baseline baroreceptor signal strength and total peripheral resistance, providing predictive ability exceeding 90%. The methods presented are suitable for use in any deterministic mathematical model.
Water homeostasis is one of the body's most critical tasks. Physical challenges to the body, including exercise and surgery, almost always coordinate with some change in water handling reflecting the changing needs of the body. Vasopressin is the most important hormone that contributes to short‐term water homeostasis. By manipulating vascular tone and regulating water reabsorption in the collecting duct of the kidneys, vasopressin can mediate the retention or loss of fluids quickly. In this study, we validated HumMod, an integrative mathematical model of human physiology, against six different challenges to water homeostasis with special attention to the secretion of vasopressin and maintenance of electrolyte balance. The studies chosen were performed in normal men and women, and represent a broad spectrum of perturbations. HumMod successfully replicated the experimental results, remaining within 1 standard deviation of the experimental means in 138 of 161 measurements. Only three measurements lay outside of the second standard deviation. Observations were made on serum osmolarity, serum vasopressin concentration, serum sodium concentration, urine osmolarity, serum protein concentration, hematocrit, and cumulative water intake following dehydration. This validation suggests that HumMod can be used to understand water homeostasis under a variety of conditions.
4 June 2020 in Annual Review of Biomedical Engineering
In this review, we discuss the science of model validation as it applies to physiological modeling. There is widespread disagreement and ambiguity about what constitutes model validity. In areas in which models affect real-world decision-making, including within the clinic, in regulatory science, or in the design and engineering of novel therapeutics, this question is of critical importance. Without an answer, it impairs the usefulness of models and casts a shadow over model credibility in all domains. To address this question, we examine the use of nonmathematical models in physiological research, in medical practice, and in engineering to see how models in other domains are used and accepted. We reflect on historic physiological models and how they have been presented to the scientific community. Finally, we look at various validation frameworks that have been proposed as potential solutions during the past decade.