THE COMPUTER AS REFEREE IN THE ANALYSIS OF HUMAN SMALL-BOWEL MOTILITY

The aim of this study was to determine whether visual analysis of grap hic records of small bowel motility is a reliable method of discrimina ting pressure events caused by bowel wall contraction from those of ex traenteric origin and to compare this method with computerized analysi s. Each of six independent observers was supplied with the same pair o f records of 1 h of fasting diurnal duodenojejunal motility, acquired with a 3-channel ambulant data-logging system; one record included man y artifacts due to body movement while the other did not. The observer s were asked to identify and classify pressure events and to measure t he duration and amplitude of ''true'' contractions. A computer program for on-line analysis is described; the algorithm was designed to over come the problems of a variable baseline and sudden changes in pressur e due to body movements that are unavoidable in prolonged recording fr om the small bowel of ambulant subjects. For regular contractions (pha se III of migrating motor complex) there was good agreement between ob servers but not for irregular contractions, particularly when movement artifacts were abundant. When the observers were asked to repeat the analysis 6 mo later, there was poor agreement with their original iden tification of irregular contractions and artifacts. There was, however , good agreement between the computer analysis, which was totally repr oducible, and the median decisions of the observer group; this agreeme nt supports the validity of our computer algorithm. We conclude that c omputer analysis is not merely a valuable ergonomic aid for analysis o f large quantity of data acquired in prolonged ambulatory monitoring, but also that, even for brief recordings, it provides a standard of re producibility unmatched by ''expert'' inspection. Visual analysis is u nreliable and thus susceptible to subjective bias; this may, in part, account for conflicting reports of small bowel motility under similar conditions reported by different workers in our own and other laborato ries.