The amount of quantitative data available to the clinician at the bedside, in particular in data-rich environments such as the intensive care unit (ICU), has grown tremendously due to advances in medical monitoring and imaging technology. Yet, these advances have largely failed to improve outcomes and significantly impact medical decision making by bedside clinicians, particularly in the acute care setting.
Computerized decision support technologies based on quantitative, mechanistic mathematical models of physiology might help alleviate this situation. Their application, however, has been hindered by the fundamental difficulties encountered when solving the inverse problem of finding model parameters and states best compatible with the observations in individual patients. We have recently shown in a simplified simulation setting how Bayesian inference may use such models to quantitatively interpret clinical measurements and observations, and eventually predict the result of acute interventions and optimize therapy in individual patients.
Rather than attempting to find a single, best parameter vector, we compute the full posterior probability distributions of parameters and states of a mechanistic model conditional on the available data. These distributions integrate uncertainty arising from measurement error and the fundamental non- uniqueness of the solution of the underlying parameter/state estimation problem, and translate into physiologically meaningful probabilistic, yet quantitative interpretations of clinical measurements. We have shown that a direct mapping between the multimodal structure of the inferred distributions representing estimated patient condition and the clinical concept of differential diagnoses may exist.
The overarching goal of this project is to explore the practical usefulness of this novel approach in a cohort of critically ill patients. The proposed research plan will focus on cardiovascular pathophysiology, with the objectives of:
(1) Using mathematical models to provide quantitative estimates of patient condition that incorporate more of the routinely acquired high density data than can be processed by the clinician.
(2) Formally validating the accuracy and predictive power of these estimates.
It will thus provide a first quantitative evaluation of the potential usefulness of this approach as a quantitative, physiology based decision support tool in critical care medicine.
This research is supported by the National Library of Medicine - 5R21LM009936-02