Context-Sensitive Temporal Abstraction of Clinical Data
Intelligent Data Analysis in Medicine and Pharmacology
Temporal-abstraction in medical domains is the task of abstracting higher-level, interval-based concepts (e.g., 3 weeks of moderate anemia) from time-stamped clinical data (e.g., daily measurements of hemoglobin) in a contextsensitive manner. We have developed and implemented a formal knowledge-based framework for decomposing and solving that task that supports acquisition, maintenance, reuse of domain-independent temporal-abstraction knowledge in different clinical domains, and sharing of
... n-specific temporal-abstraction properties among different applications in the same domain. In this paper, we focus on the representation necessary for creation during runtime of appropriate contexts for interpretation of clinical data. Clinical interpretation contexts are temporally extended states of affairs (e.g., effect of insulin as part of the management of diabetes) that affect the interpretation of clinical data. Interpretation contexts are induced by measured patient data, concluded abstractions, external interventions such as therapy administration, and the goals of the interpretation process. Thus, a runtime induction relation is defined over interpretation contexts and other proposition types. Interpretation contexts also can be composed from certain combinations of induced contexts. Knowledge about clinical interpretation contexts is represented in a domain-specific context ontology that includes four types of interpretation-contexts: basic, composite, generalized, and nonconvex. We present examples of the use of interpretation contexts in several clinical domains, in particular in the diabetes-therapy domain. The explicit separation of interpretation-context propositions from the propositions inducing them and from the abstractions created within them has several distinct conceptual and computational advantages, which we discuss in detail. These advantages are especially pertinent in clinical domains, which are typically knowledge intensive .