Monitoring laboratory performance by statistical analysis of rescreening cervical smears

The rescreening of cervical cytology smears, although inadequate both for checking the reliability of an individual screener and for the detection of false negatives, nonetheless represents an indispensable tool for assessing morphologic diagnostic criteria and monitoring laboratory performance. Interobserver and intraobserver concordance in a given laboratory determine the reliability of diagnostic criteria and disclose the possibility of improving diagnostic consistency. The various aspects of rescreening are discussed, and a simple statistical procedure for a quantitative comparison of screening reproducibility is described. This procedure, with calculation of phi-values and mean prevalence rates, was applied to screening and rescreening data from the Cyt-U-Universitair Cytology Laboratory from 1978 to 1983, in which the morphologic findings in each smear were recorded in detail using the QISO (Quality, Infection, Squamous epithelium and Other abnormalities) coding. The reproducibility of the reported presence of endocervical cells not only steadily improved over the years assessed, but also reached a sufficiently high level that diagnostic variability should not hamper its clinical utility. The same was true of the diagnosis of Trichomonas infection and moderate atypia of squamous cells. On the other hand, the reproducibility of the diagnosis of endometrial cells in smears did not significantly improve. However, since the intraobserver concordance on this criterion was good, it seems likely that the definitions of screeners in this area can still be brought into closer agreement. Conversely, the diagnosis of "abnormal endocervical cells" had such a high level of variability even at the intraobserver level as to question its use as a diagnostic criterion. Similar variability was seen in the distinction between atypical squamous metaplasia and mild dysplasia. Use of this statistical technique for the quality control of cytologic diagnoses frees that assessment from histopathologic control, which does not address the problem of variation within the cytologic or histologic diagnoses themselves.