Interobserver reproducibility of cervical cytologic and histologic interpretations: realistic estimates from the ASCUS-LSIL Triage Study

Abstract

Context: Despite a critical presumption of reliability, standards of interpathologist agreement have not been well defined for interpretation of cervical pathology specimens.

Objective: To determine the reproducibility of cytologic, colposcopic histologic, and loop electrosurgical excision procedure (LEEP) histologic cervical specimen interpretations among multiple well-trained observers.

Design and setting: The Atypical Squamous Cells of Undetermined Significance-Low-grade Squamous Intraepithelial Lesion (ASCUS-LSIL) Triage Study (ALTS), an ongoing US multicenter clinical trial.

Subjects: From women enrolled in ALTS during 1996-1998, 4948 monolayer cytologic slides, 2237 colposcopic biopsies, and 535 LEEP specimens were interpreted by 7 clinical center and 4 Pathology Quality Control Group (QC) pathologists.

Main outcome measures: kappa Values calculated for comparison of the original clinical center interpretation and the first QC reviewer's masked interpretation of specimens.

Results: For all 3 specimen types, the clinical center pathologists rendered significantly more severe interpretations than did reviewing QC pathologists. The reproducibility of monolayer cytologic interpretations was moderate (kappa = 0.46; 95% confidence interval [CI], 0.44-0.48) and equivalent to the reproducibility of punch biopsy histopathologic interpretations (kappa = 0.46; 95% CI, 0.43-0.49) and LEEP histopathologic interpretations (kappa = 0.49; 95% CI, 0.44-0.55). The lack of reproducibility of histopathology was most evident for less severe interpretations.

Conclusions: Interpretive variability is substantial for all types of cervical specimens. Histopathology of cervical biopsies is not more reproducible than monolayer cytology, and even the interpretation of LEEP results is variable. Given the degree of irreproducibility that exists among well-trained pathologists, realistic performance expectations should guide use of their interpretations.