Section 3. 2 Theoretical Background
3.2.3 Realization
A key requirement for the envisioned model/health IT system is the ability to be real-time and dynamic i.e. administer the intervention when needed. The design of such model, or system
requires a feedback mechanism and ability to exert control as close to the source as possible. We examine the design from the theoretical perspective of control theories that provide a basis for real time control systems (Albus and Barbera 2005; Albus and Rippey 1994; Carver and Scheier 1982). Control theory is an interdisciplinary branch of engineering and mathematics that deals with the behavior of dynamic systems with inputs, and how feedback modified their behavior. The objective of control theory is to control a system, so its output follows the desired control signal, called the reference, which may be a fixed or changing value. A controller monitors the output and compares it with the reference. The difference between actual and desired output, called the error signal, is applied as feedback to the input of the system, to bring the actual output closer to the reference. Mapping this to the problem of medication adherence, we model the control system to have the ability to capture the dosing event. The controller compares the dose event to the reference prescribed dose event. If the dose is not effective or the dose is
undesirable, a feedback is generated. The feedback is the probable value of effective medication adherence at next prescribed dosing event. The probable value depends on the past pattern of medication behavior of the individual. Based on the value of feedback generated, it can be decided to administer intervention.
To develop a system that can adapt to the medication behavior exhibited in real-time also leads us to examine the cognitive foundations for system design. A cognitive architecture is the organizational structure of functional processes and knowledge representations that enable the modeling of cognitive phenomena. The fundamental cognitive system is composed of a behavior generation engine driven by a model updated by a perceptual system and governed by a value system as shown in Appendix A3.3a.
Controllability and observability are main issues in the analysis of a system before deciding the best control strategy to be applied. Controllability is related to the possibility of forcing the system into a particular state by using an appropriate control signal. Observability is measuring the state of a system. There are several control techniques as represented in Appendix A3.3b which are applicable to model and strategy chosen.
Hierarchical Control System is close to our problem domain based on the multidimensional nature of the problem that involves one or more than one feedback influencing the goal or behavior. In a hierarchical control system, intelligent control can be built as units of the control system without having to change the design explicitly.
We leverage the existing hierarchical control system Reference Model Architecture (RMA) theory for Real time Control Systems for its extensible nature (Albus 1993; Albus and Rippey 1994). At its core, the RMA can be mapped directly to the controlled process or real world, avoiding the need for a mathematical abstraction, and in which time-constrained reactive planning is implemented. RMA is cognitive theory based architecture designed to enable any level of intelligent behavior.
RMA consists of a multi-layered multi-resolution hierarchy of computational agents each containing elements of sensory processing (SP), world modeling (WM), value judgment (VJ), behavior generation (BG), and a knowledge database (KD). At the lower levels, these agents generate goal-seeking reactive behavior. At higher levels, they enable decision making, planning, and deliberative behavior. RCS functional modules may add, subtract, multiply, differentiate, integrate, compute correlation functions, recognize patterns, generate names or addresses of symbolic representations, or perform planning functions at a hierarchy of levels. In its most complete theoretical form, the RCS reference model architecture provides a framework for
integrating concepts from artificial intelligence, machine vision, robotics, computer science, control theory, operations research, game theory, signal processing, filtering, and
communications theory.
For the purpose of developing a system for smart intervention, we are considering a single node that is assigned the task to decide if there is a need for intervention, and when and what type of reminder to administer. A single RMA node can be used to model this requirement. Figure 15 shows a node of RMA (Albus 1993; Albus and Rippey 1994). Each node has ability for:
Behavior generation (BG) - BG modules contain job assignment, planning, and control algorithms.
Sensory perception (SP) - SP modules process data from sensors. It contains filtering, masking, differencing, correlation, matching, and recursive estimation algorithms, as well as feature detection and pattern recognition algorithms.
Value judgment (VJ) - VJ modules contain algorithms for computing cost, risk, and benefit, for evaluating states and situations, for estimating the reliability of state estimations, and for assigning cost-benefit values to objects and events. VJ modules may compute statistics on information about the world based on the correlation and variance between observations and predictions.
World Model (WM) - The WM modules models the state space of the problem domain. They contain information storage and retrieval mechanisms, as well as algorithms for transforming information from one coordinate system to another. WM modules use dynamic models to generate expectations and predict the results of current and future actions. WM modules may contain recursive estimation algorithms and processes that compute lists of attributes from
images, graphics engines that generate images from symbolic lists, and recognition and detection algorithms that perform pattern matching operations necessary to verify the identification of features, surfaces, objects, and groups. The WM module maintains a
knowledge database (KD), acts as a question answering system and uses information on from the KD to predict or simulate the future.
Figure 15. Node of reference model architecture (Albus 1993; Albus and Rippey 1994)
The operation of the Node of RMA is a set of steps defined by a planning period or an execution clock cycle:
While the planning period is open {
BG planner hypothesizes a tentative plan;
BG planner checks to see if the probable result value is greater than the previous probable result value of the plan already in the current best plan buffer,
{
if it is,
then the tentative plan replaces the current plan in the current best plan buffer; else continue;
} };
At the top, the highest level task is defined by the highest level (i.e., mission) goal. At each successive level in the hierarchy, commanded tasks from above are decomposed into subtasks and sent to subordinates below. Finally, at the bottom, subcommand outputs are sent to actuators to generate forces and movements. Also at the bottom, sensors transform energy into signals that provide sensory input.
Figure 16 shows the dosing event (DE) being evaluated by a node in RMA model to plan and decide upon intervention based on prescribed dosing regimen (PR), dose not effective (DNE) and undesirable dose event (UDE). The sensory perception (SP) module registers an actual dose event (DE) from the patient. The value judgment (VJ) module evaluates the time of the dosing event by comparing it with the prescribed dosing regimen (PR) and identifying the dose not effective (DNE) or undesirable dose event (UDE). It then evaluates the value of effective medication adherence (EMA) for the next prescribed dose event based on a plan from the world model (WM). The value judgement (VJ) module compares the predicted the value of EMA to the prescribed value of EMA and takes decision to administer intervention (INTVN) based on
DE* - Dose Event, DNE* - Dose Not Effective, UDE* - Undesirable Dose Event, EMA* - Effective Medication Adherence, INTVN* - Intervention, MB* - Medication Behavior, PR* - Prescribed Dosing Regimen, BCT* - Behavior Control Techniques
Figure 16. MEMA implementation of RMA node for smart intervention
The current set of envisioned functionality is to provide a reminder (BCT) to the patients. The world model is enriched by knowledge module and inputs from other sensors, and the capability exists for enhancements in future. The usefulness of the node based on RMA model would be more suited is an enriched use case scenario.