The practical management of artifact in computerised physiological data.

Computerised physiological data contains artifact that needs to be identified and possibly removed. Whilst computers may eventually satisfactorily perform this function, at present only manual removal is possible for the majority of intensive care computer groups. We assessed the effects of artifact and its removal on the physiological data of 3 patients. Artifact was manually removed from 7 days of data in 4 parameters (heart rate, respiratory rate, systolic blood pressure [sbp] and transcutaneous oxygen [tcpO2]) by 3 independent observers. Six hour time periods were analysed. Median and mean values before and after the manual removal of artifact were compared. Overall 6.5% of data was removed as artifact. This was greatest for tcpO2 (9.9%) and sbp (10.6%), with smaller amounts for respiratory rate (2.8%) and heart rate (2.4%). Sbp showed a marked difference in the amount of data removed between patients, whereas tcpO2 data contained quite large volumes of artifact, but this was fairly consistent between patients. Removal of artifact affected mean values more than median values. One observer considered that both physiological and non-physiological artifact should be removed, whereas the other two observers removed only non-physiological artifact. Agreement in results between the latter was good. Our results suggest that inter-observer variability should have a minimal effect on values, once rules identifying the type of artifact to be removed are agreed. Removal of artifact did not have a clinically significant effect on results, but may be an important consideration in the statistical analysis of computerised physiological data.