@phdthesis{598, keywords = {data abstraction, guideline-based care}, author = {Andreas Seyfang}, title = {An integrated system for temporal data abstraction to facilitate guideline execution}, abstract = {
Clinical guidelines and protocols improve the quality of care efficiently, if translated to computer-interpretable models using languages such as Asbru and integrated into the information flow at the place of care. This demands for temporal data abstraction, which bridges the low-level data from monitoring devices and laboratory results to the high-level concepts used in guideline or protocol. To support this, I developed the following.<br />A versatile abstraction algorithm called Spread to handle noise of varying intensity and to generate steady, qualitative values based on such input. An efficient implementation for matching abstract pattern descriptions with input data.
An integrated framework comprising all the involved modules for temporal data abstraction, monitoring, and plan execution.<br />A mapping of Asbru to these modules, and a mapping of the abstraction modules to Asbru.
The above was evaluated on a practical and on a theoretical level. On the practical level, ideas from this thesis were implemented in a system to the control of oxygen supply in a neonatal intensive care unit at the level of a human expert dedicated to the job, and superior to clinical routing, as shown in a clinical study. The integrated framework for temporal data abstraction, monitoring, and plan execution forms the basis of the Asbru interpreter which was also used successfully in international research projects aiming at improving the quality of protocols, and the reduction of patient risks.
On the theoretical level, this thesis discusses the computational effort associated with the proposed algorithms, how they meet the objectives, and their benefits and limitations.
Although the field of application of the work described in this thesis is medicine, with a focus on intensive care, the described methods apply to data abstraction and plan execution in any field in which heterogeneous, time-oriented data must be matched with complex domain knowledge.
}, year = {2011}, journal = {Institute of Visual Computing and Human-Centered Technology}, pages = {212}, publisher = {TU Wien}, address = {Vienna}, }