Clinical evaluation of a ventilator alarm expert system;
This thesis describes the clinical evaluation of the Servo Supervisor (SSV), a computerized ventilator alarm system designed to monitor mechanically ventilated patients. The SSV obtained information from a Siemens 900C ventilator used to ventilate the patient. The performance of the SSV was evaluated by monitoring nine patients in a hospital intensive care unit for a total of 69.37 hours, while the author observed and recorded all pertinent events that occurred during the time intervals. The SSV’s alarm messages were stored in a computer disk file. The SSV’s file of alarm messages was not available to the observer until after the study was completed. The observer monitored the patient”s condition , and recorded any events that might have affected the patient’s ventilation status. In addition, the observer recorded the output from the alarm system of the Siemens 900C ventilator operating on the patient. The SSV recorded 103 alarms, 2 alerts, 97 carbon dioxide warnings, and 58 changes in the ventilator’s settings. Of the 103 alarms, 49 agreed with the observer’s record of events. Forty-two times the SSV correctly identified a disconnection, six times the SSV correctly identified an instance where the patient fought the ventilator, and once the SSV correctly identified a leak in the ventilation system. The SSV’s alarm messages were correct 48% of the time. Only 13 SSV alarms were recorded when the observer noticed no alarm event. Of the 54 false alarms recorded by the SSV system, 31 could be remedied by relatively straightforward modifications to the software. The most frequent cause of false positive alarms was airway suctioning, which accounted for 12 false alarms. The interpretation rule for discerning between an obstruction and an instance where the patient fought the ventilator accounted for 8 incorrect alarm messages. Eight times the leak alarm was transiently activated due to a disconnection. After the disconnection, the leak alarm was activated due to the re-establishment of the patient’s functional residual lung capacity. It is too soon to tell what maximum performance level of the SSV might be. The alarm rules and critical count filters require further refinement, but the SSV exhibits credible performance considering the early stage of its development cycle. The rule-based approach of the SSV, combined with the two alarm filters, and informative alarm messages appear to adequately address many of the problems that currently exist in medical device alarm systems.