Decision models for surveillance of Barrett's oesophagus (BO) are governed by the grade of dysplasia on endoscopic biopsy, but subjective grading is prone to observer variation. Computerized morphometry and immunoquantitation can objectively discriminate between different grades of dysplasia in oesophagectomy specimens with BO. The present study evaluated the feasibility of such quantitative analysis on surveillance biopsies of BO. Biopsy criteria for quantitative analysis were defined, excluding 101 (21%) of 472 archival BO surveillance biopsies. In the remaining haematoxylin and eosin (H&E) sections, 105 areas that distinctively displayed no dysplasia (ND), low-grade dysplasia (LGD) or high-grade dysplasia (HGD) were demarcated. Agreement on double-blind examination by two experienced pathologists was reached in 66 areas (63%; kappa: 0.44). For 21 ND/LGD and 11 LGD/HGD disagreement areas, corresponding sections for p53 and Ki67 immunohistochemistry were available. The best combination of two discriminating features was stratification index (SI) with p53 area % for ND versus LGD (89% correct classification), and SI with Ki67 area % for LGD versus HGD (91% correct classification). Fifteen of the 21 ND/LGD disagreement areas could be classified uniquely as either ND or LGD by SI and p53, and eight of the 11 LGD/HGD disagreement areas as either LGD or HGD by SI and Ki67. Correlation coefficients for repeated measurements of SI, Ki67, and p53 by the same observer were 0.94, 0.92, and 0.86, and by two independent observers 0.86, 0.93, and 0.92, respectively. Computerized quantitative pathology on BO surveillance biopsies is feasible provided that well-defined biopsy criteria are used. Using a combination of features associated with cellular differentiation and proliferation, such as SI, p53, and Ki67, quantitative pathological analysis assists in reducing diagnostic variability in the grading of dysplasia during surveillance of BO.
Copyright 2000 John Wiley & Sons, Ltd.