Page 72 - Screening for Cervical Cancer: Systematic Evidence Review
P. 72
Chapter IV. Discussion
New Methods for Preparing or Evaluating Cervical Cytology
This portion of our SER evaluated the performance of new technologies for preparing or
interpreting the cytology specimens; these included liquid-based cytology collection systems,
neural-net screening and rescreening tools, and computer algorithms for selecting slides for
screening or rescreening. This literature is fundamentally limited by lack of histologically
confirmed performance measuresno gold standard is used for comparisons. Of the very few
studies using colposcopy and histology to verify the diagnostic test characteristics of new tools,
the most common shortcoming was failure to apply the gold standard to test negative population
or a subset to allow estimation of specificity.
To receive approval from the Food and Drug Administration, each of the new
technologies currently in use (ThinPrep , PapNet , and AutoPap ) had to demonstrate improved
sensitivity over conventional Pap testing. As noted in the introduction of this chapter, improving
sensitivity, especially if it amplifies detection of low-grade lesions, may not be a benefit at a
population level in terms of reducing the burden of morbidity and mortality associated with
cervical cancer. This is especially true if increased sensitivity is accompanied by decreased
specificity, requiring evaluation of a greater number of false positives and increasing costs. It is
precisely these parameters that are most poorly measured in this literature.
Overall, the quality of this literature is poor for the purposes of making decisions about
choice of screening systems in US populations. No randomized trials or prospective cohort
studies relate use of a screening modality over time to outcomes for individual women. The
cost-effectiveness of use of new technologies has only been estimated, not measured directly.
The most sophisticated and thorough cost model to date is significantly hindered by the
limitations of the literature. Nonetheless, it demonstrates that, at present, new technologies are
78
