CPMS Webinar

Thank you everyone who participated in the Committee for Credible Practice on Modeling & Simulation in Healthcare webinar. If you had a question that did not get answered, please feel free to ask here, or, if you prefer, email the team directly. We will go through the recorded webinar as soon as possible and address any questions posted there that did not get answered.

Thank you all,
Rodney Metoyer
rmetoyer@ara.com

Comments

  • Question: What kind of uncertainty quantification do you do perform in your physiology model?

    In order to provide as much information as possible, we will decompose the question and answer it in two parts. First, how accurately does BioGears describe the true biological, physical, and/or physiological processes? Second, what is the model uncertainty and what is the impact of that uncertainty on model outputs?

    To quantify how accurately BioGears describes true physiology, we first define parameters for comparison, and then we collect experimental data. Both of these steps present challenges. We validate as many clinical endpoints as possible to meet the needs of the user community, and we also validate some intermediate measures to demonstrate validity of the more mechanistic models. However, there are plenty of measures which do not get compared to experimental values. The engine produces very large data sets, so we have to be selective about which measures are validated.

    Once an output parameter is identified for validation, we attempt to collect experimental data, primarily through literature search. The data from the literature usually has some measure of uncertainty associated with it. Sometime the literature contains a mean and standard deviation. Other times the validation target is a range of values. Each chosen BioGears output is compared to either the mean or to a range and then it is classified as either less than 10% from the target, 10 to 20% in deviation, or more than 30% deviated from the target validation value. We do not report the uncertainty in the reference data, we only report the deviation of the BioGears output from the mean of the reference data. We cite all sources of validation data in our documentation, which is available at biogearsengine.com.

    The BioGears model is deterministic. It is an integrated model of linear and non-linear sub-models. The two foundation sub-models within BioGears (cardiovascular and respiratory) are linear physics-based lumped parameter models that are solved by inverting matrices. There is numerical uncertainty introduced through the matrix solver. BioGears currently uses a bi-conjugate gradient method specific for sparse square systems (using the Eigen third party packages). This is an iterative method and we use the default tolerance for their solver, which is as close to zero as reasonable (around 1e-16). If this solver fails during a simulation the user is alerted and the resultant error is logged. Other numerical uncertainly comes from modeling the distribution of blood gasses (dissolved and bound oxygen and carbon dioxide as well as the bicarbonate species). We try to model these reactions as mechanistically and with as much detail detailed as possible. As a result, we must solve a system of non-linear equations using the (unsupported, we include a number of unit tests to maintain solution confidence) Eigen hybrid solver. This solver uses Powell’s method for rapid convergence to a local minimum, and the numerical uncertainty in this method is set to 1e-5 digits of accuracy.

    We do not currently have the resources to conduct a formal sensitivity analysis for this non-linear integrated model. We would welcome any effort by the community to identify sensitive parameters. We know that the engine is sensitive to initial conditions, but we have not quantified that sensitivity or identified the key parameters. We would certainly be willing to support, in as much as we are able, any uncertainty quantification effort, both in process definition and execution.
  • Question: Can you tell me about a relationship with the Virtual Physiological Human (VPH) project?

    The goal of the US Government-funded BioGears project is to lower the cost of developing medical training tools by providing an open source physiology simulation engine (please see biogearsengine.com/documentation/index.html for details). The Virtual Physiological Human is a European initiative with the eventual goal of producing a complete mechanistic model of the entire human body. With BioGears, we are trying to simulate whole-body physiology with reasonable accuracy for a target population. In other words, we are attempting to model a generic individual within a reference population. In contrast, the eventual goal of the VPH project is individualized medical simulation. VPH will eventually be able to model the physiology of a specific individual, and thus could be used as a predictive tool for clinical trials. Individualized simulation is not within the current scope of the BioGears project, but we believe that some of our development processes could be used toward that end. We will be presenting some of our ideas at the VPH conference later this month.

    We have tremendous respect and reverence for the work being done by VPH researchers, and we try to use the knowledge that they have generated as much as possible.
  • Question: Do you support arbitrary chemical reactions? I suppose no diffusion simulated or subcellular level?

    BioGears does not simulate any process below the tissue level. Although physicochemical properties are inputs to some of the models (specifically the PBPK and membrane diffusion models), we do not simulate any chemical reactions. BioGears is designed with modularity and extensibility in mind in order to facilitate the incorporation of more detailed models. It is possible to add models with higher resolution, but the current focus is at the organ level, moving down to the tissue level when possible and necessary for mechanistic models, and integrating out to the whole body.
Sign In or Register to comment.