We evaluated the performance of the Papanicolaou smear in screening and diagnostic settings.
We analyzed Papanicolaou smear results of 1,850 women recruited into a clinical trial to evaluate an emerging technology for the detection of cervical cancer. Screening and diagnosis groups were based on the history of previous Papanicolaou smear results. We calculated sensitivities, specificities, positive and negative likelihood ratios (LR+ and LR−), receiver operating characteristic curves, and areas under the receiver operating characteristic curve (AUC).
In the screening group, by defining disease as cervical intraepithelial neoplasia (CIN) 2,3/cancer or worse and using high-grade squamous intraepithelial lesion (HSIL) as the test cutpoint, the AUC was 0.689, and the LR+ and LR− were 39.25 and 0.67, respectively. In the diagnosis group, the AUC was 0.764, and the LR+ and LR− were 3.79 and 0.56, respectively. By defining disease as human papillomavirus/CIN 1 or worse and HSIL as the test cutpoint, the AUC was 0.586, and the LR+ and LR− were 17.01 and 0.92 in the screening group; in the diagnosis group, the AUC was 0.686, and the LR+ and LR− were 2.77 and 0.75, respectively.
In a screening setting, a Papanicolaou smear result of HSIL or worse is 39 times more likely in a patient with CIN 2,3/cancer than in a patient without it. This compares to 4 times more likely in the diagnostic setting. The magnitude of the positive likelihood ratio observed in the screening group indicated that abnormal Papanicolaou smear results obtained in the screening setting should have more impact on clinical decision making than those from results obtained in the diagnostic setting.
An analysis using likelihood ratios shows much greater discriminative ability of the Papanicolaou smear to detect CIN2-3 in the screening setting compared to the diagnostic setting.
1Departments of Biostatistics and 2Gynecologic Oncology, The University of Texas M. D. Anderson Cancer Center, Houston, TX, 3Division of Gynecologic Oncology, Department of Obstetrics and Gynaecology, University of British Columbia, Vancouver, Canada, and 4Division of Population Science, Fox Chase Cancer Center, Philadelphia, PA
Reprint requests to: Scott B. Cantor, PhD, Section of Health Services Research, Department of Biostatistics, The University of Texas M. D. Anderson Cancer Center, Unit 447, 1515 Holcombe Blvd, Houston, TX 77030-4009. E-mail: email@example.com
This study was conducted by the Departments of Biostatistics and Gynecologic Oncology, The University of Texas M. D. Anderson Cancer Center; Department of Obstetrics and Gynaecology, University of British Columbia; and Fox Chase Cancer Center Division of Population Science. This study was conducted at sites in Houston, TX, and Vancouver, British Columbia, Canada.
Financial support for this study was provided by grant number CA82710 from the National Cancer Institute. The funding agreement ensured the authors' independence in designing the study, interpreting the data, and writing and publishing the report.