Page 50 - Screening for Cervical Cancer: Systematic Evidence Review
P. 50
Chapter III. Results
biopsies were identified. They could not capture those women who had normal colposcopy, and
therefore no biopsy, among those with abnormal cytology. Additionally, ThinPrep and
conventional specimens were from separate groups of women, not one of each specimen type
from each individual. As a result, the sensitivity and specificity of ThinPrep (reported as 95%
and 58%, respectively) are not valid and cannot be appropriately compared with conventional
cytology (reported as 85% and 36%).
The earlier study obtained split samples for ThinPrep and conventional cytology from a
52
cohort of 35,560 Australian women. Within the Australian health care system, colposcopy is
the recommended immediate evaluation for inconclusive slides in which high-grade
abnormalities cannot be excluded and for cytology diagnoses of CIN1 and above. High-grade or
inconclusive cytology results were reported for 433 ThinPrep specimens and 430 conventional
cytology specimens, of these, 325 (75%) and 319 (74%), respectively, had histology results.
This again excludes women who had colposcopy without biopsy. However, the focus on high-
grade lesions makes the probability of biopsy high, and so the proportion with follow-up is
adequate. In this context, the relative true-positive and false-positive rates of the test can be used
54
to compare performance. The relative true-positive rate for ThinPrep compared to
conventional cytology for detecting high-grade histologic abnormalities is 1.13, suggesting a
modestly higher sensitivity of Thin Prep ; the relative false-positive rate is 1.12, suggesting a
modestly lower specificity.
Neural-network Rescreening
The literature about neural-network rescreening (and screening) technology is also
fundamentally limited by the lack of histologically confirmed performance measures. Applying
50
