Meeting Brochure and registration form      SMDM Homepage

Monday, October 22, 2007
P2-47

LIQUID-BASED CYTOLOGY TECHNIQUES FOR CERVICAL CANCER SCREENING: A SYSTEMATIC REVIEW OF TEST PERFORMANCE CHARACTERISTICS

Ba Pham, MSc1, Murray Krahn, MD, MSc2, Meg McLachlin, MD3, Barry Rosen, MD4, Beate Sander, RN, MBA, MEcDev2, Paul Grootendorst, PhD5, George Tomlinson, PhD2, Ava John-Baptiste, MHSc6, Maraki Frikemerid, MSc2, Maggie Hong Chen, MMath2, Gloria Woo, MSc5, Andrea Anonychuk, MSc4, Steven M. Carcone, MSc4, Holly Witterman, MSc2, Wendong Chen, MD2, Karen Liu, BA2, Margaret Sampson, PhD, (Candidate)7, and Andrea C. Tricco, MSc8. (1) THETA - Toronto Health Economics & Technology Assessment Centre, Toronto, ON, Canada, (2) University of Toronto, Toronto, ON, Canada, (3) Pathology, University of Western Ontario, London, ON, Canada, (4) University Health Network, Toronto, ON, Canada, (5) Faculty of Pharmacy, University of Toronto, Toronto, ON, Canada, (6) Department of Health Policy, Management and Evaluation, Toronto, ON ON, Canada Canada, (7) Chalmers Research Group, University of Ottawa, Ottawa, ON, Canada, (8) Institute of Population Health, U. of Ottawa, Ottawa, ON, Canada

Purpose: Liquid based cytology (LBC) may reduce unsatisfactory samples in cervical cancer screening programs. LBC may also be more sensitive than conventional cytology (CC), but this claim has been refuted in a recent systematic review (SR). LBC performance characteristics are estimated here for their potential use in cost-effectiveness analyses of screening programs.

Methods: Primary studies, SRs, and HTA reports evaluating LBC as a replacement for CC (both read manually) were identified via searching multiple sources (DIALOG®, PubMed, Cochrane, and HTA agency websites), back-referencing and author searching (cutoff June 2006).

Data pertaining to study characteristics, test accuracy, cytology classifications, and unsatisfactory samples were extracted independently by two reviewers. Accuracy data were extracted from 2×2 tables that cross-classified between cytological readings and reference standard (e.g., histology or consensus cytology). A Bayesian hierarchical summary receiver-operating characteristics curve model (MCMC simulations in WinBugs) was used to synthesize study results and estimate differences in test accuracy.

Results: The literature search identified 14 relevant SRs (including 7 HTAs). Six SRs (published 2000-2004) concluded that LBC is more effective (e.g., higher sensitivity, less unsatisfactory samples) than CC; five (2002-2006) concluded that there was insufficient evidence to draw conclusions regarding LBC effectiveness, and three (2000-2003) were equivocal in their conclusion.

108 primary studies [reporting LBC-versus-CC accuracy (n=20 studies), some accuracy data (n=49), and unsatisfactory sample data (n=66)] were also identified, including split-sample studies (n=47) and cohort studies (n=31). Data from 20 head-to-head studies (n=68,114 participants) showed a sensitivity difference of 6.4% [95% credible interval: -6.5% to 18.8%], and a specificity difference of -4.0% [-19.8%; 10.6%]. The posterior probability that LBC is more sensitive {less specific} than CC was 83% {72%}.

In 42 split-sample studies, most cytological classifications were concordant, with 2% to 4% of samples giving discordant results (e.g., LSIL+ by LBC but < LSIL by CC). In LBC (n=1,302,378) {CC (n=2,008,724)} samples, the unsatisfactory rate was 0.71% {1.16%}.

Conclusion: Decisions to replace CC by LBC in cervical cancer screening programs entail a 6.4% gain in sensitivity, a 4.0% loss in specificity and a small reduction in the number of unsatisfactory specimens. The likelihood of these differences having arisen by chance is not zero, but neither is it very large. Small absolute differences with moderate uncertainty may explain discordance between previous systematic reviews.