The Impact of Clinical Information on the Assessment of Endoscopic Activity: Characteristics of the Ulcerative Colitis Endoscopic Index Of Severity [UCEIS].
Travis SPL., Schnell D., Feagan BG., Abreu MT., Altman DG., Hanauer SB., Krzeski P., Lichtenstein GR., Marteau PR., Mary J-Y., Reinisch W., Sands BE., Schnell P., Yacyshyn BR., Colombel J-F., Bernhardt CA., Sandborn WJ.
To determine whether clinical information influences endoscopic scoring by central readers using the Ulcerative Colitis Endoscopic Index of Severity [UCEIS; comprising 'vascular pattern', 'bleeding', 'erosions and ulcers'].Forty central readers performed 28 evaluations, including 2 repeats, from a library of 44 video sigmoidoscopies stratified by Mayo Clinic Score. Following training, readers were randomised to scoring with ['unblinded', n = 20, including 4 control videos with misleading information] or without ['blinded', n 20] clinical information. A total of 21 virtual Central Reader Groups [CRGs], of three blinded readers, were created. Agreement criteria were pre-specified. Kappa [κ] statistics quantified intra- and inter-reader variability.Mean UCEIS scores did not differ between blinded and unblinded readers for any of the 40 main videos. UCEIS standard deviations [SD] were similar [median blinded 0.94, unblinded 0.93; p = 0.97]. Correlation between UCEIS and visual analogue scale [VAS] assessment of overall severity was high [r blinded = 0.90, unblinded = 0.93; p = 0.02]. Scores for control videos were similar [UCEIS: p ≥ 0.55; VAS: p ≥ 0.07]. Intra- [κ 0.47-0.74] and inter-reader [κ 0.40-0.53] variability for items and full UCEIS was 'moderate'-to-'substantial', with no significant differences except for intra-reader variability for erosions and ulcers [κ blinded: 0.47 vs unblinded: 0.74; p 0.047]. The SD of CRGs was lower than for individual central readers [0.54 vs 0.95; p < 0.001]. Correlation between blinded UCEIS and patient-reported symptoms was high [stool frequency: 0.76; rectal bleeding: 0.82; both: 0.81].The UCEIS is minimally affected by knowledge of clinical details, strongly correlates with patient-reported symptoms, and is a suitable instrument for trials. CRGs performed better than individuals.