Comparison of the cervical cytology test using the PAPNET method and conventional microscopy

From August 1994 to June 1995, laboratories in 28 Mexican states and the Federal District submitted a total of 10098 diagnosed Pap test slides to Mexico's National Institute of Epidemiologic Diagnosis and Reference (INDRE) for reexamination by conventional methods and also by the automated PAPNET system in Suffern, New York, U.S.A. The aim was to determine the degree of agreement obtained by these various methods. Most of the slides examined (at least 78%) yielded negative results or merely indicated an inflammatory process; 8% to 14% indicated mild or moderate cervical dysplasia; and 2% to 3% indicated conditions ranging from severe dysplasia to invasive cervical cancer. Comparison of the state laboratory and INDRE diagnoses yielded a Kappa correlation coefficient of 0.62, near the lower limit of agreement, the agreement being poorest in cases where it was necessary to distinguish between degrees of abnormality. Although state laboratory underestimation appeared lowest with respect to cases of atypia and of mild, moderate, and severe dysplasia (between 12% and 20%), these percentages are alarming because it is at these stages that the patient may be treated to prevent evolution to carcinoma. While the Kappa correlation coefficient was better (0.80) when the INDRE and PAPNET diagnoses were compared, PAPNET showed only limited ability to distinguish between various pathologic alterations, and the percentages of underestimates (false negatives) obtained with PAPNET were also high. Overall, the results indicate a need to improve the quality of cervical cytology diagnoses at state public health laboratories in Mexico through stepped-up training and supervision. They also indicate that the use of PAPNET involves greater difficulty than does manual microscopic examination of cervical smears, and that a way still needs to be found to detect and review the false negative results generated by PAPNET before approving use of this technology.