Abstract: Knowledge-based formulation of dynamic decision models in medicine

• Main Authors: Wang, C. G., Tze-Yun LEONG, Leong, A. P. K., Seow, F. C.
• Format: text
• Language: English
• Published: Institutional Knowledge at Singapore Management University 1998
• Subjects: Computer Sciences; Health Information Technology
• Online Access: https://ink.library.smu.edu.sg/sis_research/3059
• Institution: Singapore Management University

We present a knowledge-based model construction approach to support automated and interactive formulation of dynamic decision models, e , models that explicitly consider the effects of tune We mcorporate a hybnd knowledge representation scheme that mtegrates categoncal knowledge, probabilistic knowledge, and determlrustlc knowledge The categoncal knowledge captures the definitional and structural relations of the domam concepts, e g , diseases, tests, treatments, and other entities related to the decision problems Tlus type of knowledge provides the power of abstraction and mhentance, it supports modeling at multiple levels of details The probabilistic knowledge captures the interactions among the concepts These relations are needed to support automated derivation of nussing information m model construction The determlnlstic knowledge expresses the deterinaustic rules and declaratory constraints about the domam The target decision models are expressed m terms of DynaMoL, a high-level dynamic decision modeling language that supports multiple graplucal perspectives and allows simple and explicit specification of decision factors and constraints The modeling process begins with the decision context information elicitation, followed by the action space and the state space defimtion A transition view, which corresponds to the Markov state transihon diagram, is then constructed for each action defined m the action space If the action-specific trans1tlon functions among the states cannot be directly determined from the knowledge base, the model constructor will try to identify a set of event or chance vanables that constitute the effects of the action and a set of probabilistic influences among the event vanables These entities constitute an action-specific mfluence view, wluch corresponds to the dynamic mfluence diagram without decision nodes or value nodes, they can be translated mto the correspondmg action-specific transition functions A set of knowledge-based model modification operations are provided for automatic and interactive generation, abstrachon, and refinement of the model components The final model is evaluated or solved by a prototype iznplementation of DynaMoL to determme an optimal course of action We have conducted a comprehensive case study m colorectal cancer management The results demonstrated the practical promise of the proposed design through the correctness of and the ease with which some reasonable dynamic decision models m medicine can be constructed.