A program for automated summarization of on-line medical records

From Semantic Portal Wiki

{{#vardefine:category|Publication}}{{#vardefine:templatename|i.publication}}{{#vardefine:package|smwbp_instance_templates}}

Edit

[[abstract::The ability to automatically create patient summaries of arbitrary and appropriate complexity would represent an important advance. Such summaries would be a useful adjunct to a patient record for clinical decision making, real-time patient monitoring, or for surveillance of quality of care.The automated summarization program described in this thesis infers major events from time-oriented medical records of systemic lupus erythematosus patients and displays them in a time-oriented graph. The graph is a mouse driven user interface in which the events are represented as active regions selectable by the user. Selecting such a region brings up an explanation of the event in terms of the underlying data in the medical record. Though the knowledge domain for this implementation is medical, the approach can be generalized to other time oriented databases.The program is hypothesis driven. Specific abnormalities of clinical significance are sought in the database. Each abnormality is associated with a set of disease hypotheses. When an abnormality is found, its associated disease hypotheses are evoked. The evoked disease frames are confirmed or refuted by seeking supporting evidence in the patient record using a temporal querying syntax. The evidence is associated with likelihood ratios that are used for Bayesian updating of the evoked disease hypotheses. A knowledge base of definitions of medical objects in terms of their probabilistic links to the data in the medical record underlies the program. New objects may be defined using the temporal querying syntax built into the system.The program is implemented in Interlisp on Xerox 1108 workstations, using the Knowledge Engineering Environment (KEE [TM]) [Fikes 85]. It recognizes five diagnoses in patient records and produces summaries that, on informal examination, resemble those generated by physicians using the same data.|]]

Reference: {{#vardefine:pagename|a program for automated summarization of on-line medical records }}

1. [[]]

bibtex

{{#vardefine:pagename|A program for automated summarization of on-line medical records }}{{#vardefine:key| }}

abstract: The ability to automatically create patient summaries of arbitrary and appropriate complexity would represent an important advance. Such summaries would be a useful adjunct to a patient record for clinical decision making, real-time patient monitoring, or for surveillance of quality of care.The automated summarization program described in this thesis infers major events from time-oriented medical records of systemic lupus erythematosus patients and displays them in a time-oriented graph. The graph is a mouse driven user interface in which the events are represented as active regions selectable by the user. Selecting such a region brings up an explanation of the event in terms of the underlying data in the medical record. Though the knowledge domain for this implementation is medical, the approach can be generalized to other time oriented databases.The program is hypothesis driven. Specific abnormalities of clinical significance are sought in the database. Each abnormality is associated with a set of disease hypotheses. When an abnormality is found, its associated disease hypotheses are evoked. The evoked disease frames are confirmed or refuted by seeking supporting evidence in the patient record using a temporal querying syntax. The evidence is associated with likelihood ratios that are used for Bayesian updating of the evoked disease hypotheses. A knowledge base of definitions of medical objects in terms of their probabilistic links to the data in the medical record underlies the program. New objects may be defined using the temporal querying syntax built into the system.The program is implemented in Interlisp on Xerox 1108 workstations, using the Knowledge Engineering Environment (KEE [TM]) [Fikes 85]. It recognizes five diagnoses in patient records and produces summaries that, on informal examination, resemble those generated by physicians using the same data.

download:

• paper:
• slides: