RESUMO
BACKGROUND: To compare lung cancer detection rate by sputum cytology, we need some assurance that the estimates do not vary widely if different observers evaluate the same specimens. The aim of this study was to determine inter-rater agreement of sputum cytology diagnoses. METHODS: Slides of sputum cytology from 150 subjects were selected from a pool of slides held by six of the laboratories that had participated in a population-based lung cancer screening program over the last ten years in Japan. The cytotechnologists in these laboratories had considerable experience with sputum cytology. Each case was re-evaluated six times. Cases that were diagnosed as the same category by all six laboratories were selected as consensus cases to serve as standardized sputum cytology cases. Thirty-seven cytotechnologists with various levels of experience in sputum cytology then re-evaluated these consensus cases. Inter-rater agreement was calculated by kappa statistics including Fleiss' kappa. RESULTS: All pairs of interlaboratory agreement for the 150 cases showed statistically significant kappa values, most pairs showing substantial agreement. Fleiss' kappa value across the six laboratories was 0.5. Fourteen cases were identified as the consensus cases, and the agreement among observers with less experience of sputum cytology showed significantly lower than the agreement among those with considerable experience (Fleiss' kappa value 0.27 vs. 0.45, P < 0.05). Moreover, cytotechnologists with less experience under-diagnosed the slides significantly more often than those with considerable experience. CONCLUSION: When the observers have considerable experience with sputum cytology, inter-observer agreement is good.