The therapy prediction model uses quantitative constraint equations to specify the relations among the physiologic parameters and consists of a subset of the parameters in the diagnostic model. Since the effects of therapy are determined by the parameters that govern the short-term hemodynamic state of the patient, these are the only ones that are needed in the model. These relations capture the changes that take place in minutes, hours, and to some extent days. Including mechanisms that function over longer time periods would be difficult because of the external factors that influence longer terms changes and not as important for the acute management decisions that are the primary domain of the program.
Like the diagnostic model, the prediction model has a fixed part
specified by the file that defines the model, an enhanced part that is
computed when the model is loaded, and additions to the PSM that are
computed for each patient case. In the fixed part of the model,
equations are specified for each parameter that relate the parameter to
other parameters in the model. For example, the primary slots of the
definition of blood pressure are as follows:
This definition provides the equation for the relationship between blood pressure, cardiac output, systemic vascular resistance, and the right atrial pressure. This equation includes right atrial pressure (which is often ignored) because it has significant effects in many heart failure situations when the blood volume is high. Parameters are included in the equations if they are important for predicting the clinical hemodynamic effects of the important therapies. The equations in the model conform to the usual notion of causality in the cardiovascular system (basically related to the direction of blood flow), even though the equations are actually directionless. This makes the relations easier to understand for the physician. The measure clause indicates that the current value of blood pressure is determined by the computed mean arterial pressure from the input data on arterial pressure (which provides the systolic and diastolic pressures). This is the source of the quantitative value used in the equations.
Notice that these equations must be consistent with the causal probability relations of the diagnostic model, but neither is a substitute for the other. The causal probabilities represent a summary of experience with what happens in actual cases that is not apparent from the equations and the equations provide constraints that make it possible to combine the effects of many mechanisms in the feedback system.
The equations are used by the therapy prediction operators to determine the gains on the links between the parameters. The gains are computed as the partial derivative of the equation with respect to the parameter on the link and the link structures become part of the enhanced model. The enhanced model also includes all of the paths through the parameters and the feedback loops that are in the model. These are analogous to the causal paths in the diagnostic model and allow for the rapid computation of the prediction of changes given therapies. The PSM also has parts that correspond to the prediction model, but these are implemented as slot values for the parameter values and the computed changes and are not separate structures.
The model includes equations from several sources. Some equations were readily available from the physiology literature, such as the relationship between cardiac output, vascular resistance, and pressures. Others were determined from data in the literature, such as the relation between heart rate and systolic time. Basic hemodynamic relations were also borrowed from existing models such as Coleman's Human Program[2], including the relation between blood pressure and vagal stimulation. Included are equations for the hemodynamic effects of congestive cardiomyopathy (primary muscle disease) and four of the valvular diseases (aortic stenosis, aortic regurgitation, mitral stenosis, and more recently mitral regurgitation). The valvular diseases are modeled by equations relating the pressure drops or regurgitant volumes to the effective valve area from the original hydraulics equations developed by Gorlin[7][6]. The therapies in the model include representative drugs from most of the major classes of cardiovascular therapies. Each therapy is represented by its direct effects on parameters, with proportions determined by comparing the model predictions to published results. The parameters included in the model are those likely to be measured in the patient or reported in the literature, plus the parameters needed to account for the actions of the usual cardiovascular agents.