Mechanical ventilation of prematurely born infants suffering from hyaline membrane disease (HMD) has been largely responsible for the steady decrease in mortality in this population during the last twenty years. Yet morbidity associated with mechanical ventilation brings a host of long-term problems for survivors of this aggressive therapy, including chronic lung disease, blindness, cerebral palsy, and severe developmental and intellectual deficits. There is evidence that some of this morbidity is the result of episodes of inappropriate respiratory support, particularly during the acute stage of disease when rapid changes in respiratory status may tax the ability of clinicians to respond quickly with appropriate therapeutic changes. The objective of this research is to extract more information from existing clinical data (including arterial blood gas data, ventilator parameters, physical signs, and diagnostic information) by combining it with expert heuristic and physiological knowledge to assist front-line clinicians in making respiratory support decisions.
The aim of this project is a major expansion and refinement of an existing experimental computer-based respiratory therapy decision support system, until it is complete and reliable enough for routine use in our Neonatal Intensive Care Unit. Collaboration with Vanderbilt's Center for Intelligent Systems will bring techniques from the field of knowledge-based systems and the use of object-oriented programming languages to this project. This approach will facilitate development of a modular system which will be easier to expand, test, and maintain than the present prototype. A core of 27 heuristic rules for correcting abnormal blood gases and mean airway pressure will be extended by identifying those clinical contexts where special treatment rules apply and adding new modules containing knowledge for detecting and managing each situation. These modules will contain target ranges for blood gas and ventilator variables, rules for therapy changes expected to correct out-of-range or dangerous parameters, rules regarding transitions to other contexts, and rationales for each such rule. Each new module will be tested retrospectively and prospectively before on-line clinical testing. The system will be useful as a teaching tool and will accumulate patient and treatment data essential for testing and refining system performance.
| Dawant, B M; Manders, E J; Lindstrom, D P (1994) Adaptive signal analysis and interpretation for real-time intelligent patient monitoring. Methods Inf Med 33:60-3 |
| Uckun, S (1994) Intelligent systems in patient monitoring and therapy management. A survey of research projects. Int J Clin Monit Comput 11:241-53 |
| Uckun, S; Dawant, B M; Lindstrom, D P (1993) Model-based diagnosis in intensive care monitoring: the YAQ approach. Artif Intell Med 5:31-48 |
