Page 52 - Screening for Cervical Cancer: Systematic Evidence Review

Chapter III. Results


of PapNet to conventional screening ranges from 1.08 to 1.16, supporting the claims that


PapNet reduces false negatives. Unfortunately, the research team missed the opportunity to
evaluate all performance measures using the 2-year normal follow-up group as a fairly well-

documented confirmation of test negatives. The authors do not comment on the classification of

the normal slides, focusing only on the 516 slides from women with known abnormalities.

Using data from the centralized Dutch national pathology reporting program, Kok and

colleagues evaluated cervical cancer outcomes and risk of screening method failure in 2 large

groups: 109,104 smears evaluated by conventional screening and 245,527 slides submitted for

 57
neural-network based screening (not rescreening as in the United States) with PapNet . They

used the reporting system to identify 71 women with a diagnosis of biopsy-confirmed squamous

cell carcinoma and to locate the screening smear obtained prior to the biopsy. Of the 71 smears,


19 had been evaluated by conventional methods and 52 by PapNet screening; there were four
false negatives among the conventional screening group and five false negatives among the


PapNet slides. The balance of the 71 smears were all reported with a level of abnormality
ranging from LSIL to carcinoma that prompted clinical follow-up. The false negatives were


subjected to masked detailed review by both standard microscopy and PapNet . All five


PapNet failures were confirmed to be sampling errors; i.e., the slides did not contain

identifiable abnormal cells as assessed by hand review or PapNet . The conventional screening

failures did contain scant abnormal cells and constitute screening failures. Two independent,


masked PapNet re-evaluations of the slides classified three of the four as “suspicious” and the

fourth as CIN 3, suggesting that the screening failures could have been averted because none of

these individuals would otherwise have been returned to the 5-year routine screening interval.

52