Histological response to treatment is an important outcome in patients with ulcerative colitis (UC). The accuracy of biopsy-based measurements of inflammation may be limited by error imposed by natural microscopic heterogeneity on the scale of individual biopsies. We determined the magnitude of this error, its histological correlates, and the density of biopsy sampling within mucosal regions of interest required to meet specified benchmarks for accuracy.
A total of 994 sequential 1-mm digital microscopic images (virtual biopsies) from consecutive colectomies from patients with clinically severe UC were scored by 2 pathologists. Agreement statistics for Geboes subscores and Nancy (NHI) and Robarts Histological Indices (RHI) between random samples from 1 to 10 biopsies and a reference mean score across a 2-cm region of mucosa were calculated using bootstrapping with 2500 iterations.
The agreement statistics improved across all indices as the biopsy density increased, with the largest proportional gains occurring with addition of the second and third biopsies. One biopsy achieved moderate to good agreement with 95% confidence for NHI and RHI corresponding to scale-specific errors of 0.40 (0.25-0.66) and 3.02 (2.08-5.36), respectively; and 3 biopsies achieved good agreement with 95% confidence corresponding to scale-specific errors of 0.22 (0.14-0.39) and 1.87 (1.19-3.25), respectively. Of the individual histological features, erosions and ulcers had the greatest impact on the agreement statistics.
In the setting of active colitis, up to 3 biopsy samples per region of interest may be required to overcome microscopic heterogeneity and ensure accurate histological grading.