Early Detection and Diagnosis
Improved sensitivity of vaginal self-collection and high-risk human papillomavirus testing
Article first published online: 16 AUG 2011
Copyright © 2011 UICC
International Journal of Cancer
Volume 130, Issue 8, pages 1855–1860, 15 April 2012
How to Cite
Belinson, J. L., Du, H., Yang, B., Wu, R., Belinson, S. E., Qu, X., Pretorius, R. G., Yi, X. and Castle, P. E. (2012), Improved sensitivity of vaginal self-collection and high-risk human papillomavirus testing. Int. J. Cancer, 130: 1855–1860. doi: 10.1002/ijc.26202
- Issue published online: 22 FEB 2012
- Article first published online: 16 AUG 2011
- Accepted manuscript online: 31 MAY 2011 08:20AM EST
- Manuscript Accepted: 10 MAY 2011
- Manuscript Received: 24 MAR 2011
- cervical cancer
Self-collected vaginal specimens tested for high-risk human papillomavirus (HR-HPV) have been shown to be less sensitive for the detection of cervical intraepithelial neoplasia or cancer (≥CIN 3) than physician-collected endocervical specimens. To increase the sensitivity of self-collected specimens, we studied a self-sampling device designed to obtain a larger specimen from the upper vagina (POI/NIH self-sampler) and a more sensitive polymerase chain reaction (PCR)-based HR-HPV assay. Women (10,000) were screened with cervical cytology and HR-HPV testing of vaginal self-collected and endocervical physician-collected specimens. Women were randomly assigned to use either a novel self-collection device (POI/NIH self-sampler) or conical-shaped brush (Qiagen). The self-collected and clinician-collected specimens were assayed by Cervista (Hologic) and the research only PCR-based matrix-assisted laser desorption/ionization time-of-flight (MALDI-TOF). Women with any abnormal screening test underwent colposcopy and biopsy. Women (8,556), mean age of 38.9, had complete data; 1.6% had ≥ CIN 3. For either HR-HPV assay, the sensitivity was similar for the two self-collection devices. Tested with Cervista, the sensitivity for ≥CIN 3 of self-collected specimens was 70.9% and for endocervical specimens was 95.0% (p = 0.0001). Tested with MALDI-TOF, the sensitivity for ≥CIN 3 of self-collected specimens was 94.3% and for endocervical specimens was also 94.3% (p = 1.0). A self-collected sample using a PCR-based assay with the capability of very high throughput has similar sensitivity as a direct endocervical specimen obtained by a physician. Large population-based screening “events” in low-resource settings could be achieved by promoting self-collection and centralized high-throughput, low-cost testing by PCR-based MALDI-TOF.
The ideal cervical cancer-screening test would be both sensitive and specific for cervical intraepithelial neoplasia (CIN) Grade 3 or early cancer (≥CIN 3). As 85% of the global burden of cervical cancer resides in developing countries,1 this ideal screening test would not require a pelvic examination, which is time consuming and increases the costs by using speculums, or extensive clinical infrastructure and would be inexpensive. A vaginal self-collected specimen tested for high-risk human papillomavirus (HR-HPV) meets some of these criteria as it does not require a pelvic examination, requires fewer resources, and is less dependent on a complex healthcare infrastructure. Therefore, it is probably less expensive and easier to implement than testing using clinician-collected specimens. Still, of course, programmatic models will need to be developed to manage the call-back and tracking necessary for positive cases. Also when compared to endocervical clinician-collected specimens tested for HR-HPV, the vaginal self-collected specimens have lower sensitivity and specificity for high-grade precancers and cancer.2–9 There are several possible explanations for this lower sensitivity and specificity that include sampling fewer infected cells collected from the cervix and upper vagina, more high-risk types identified in the vagina unassociated with high-grade cervical neoplasia, and test cross-reactivity with low-risk viral types.4
We conducted the current trial with the goal of overcoming the performance limitations of self-collection for the detection of ≥CIN 3. Specifically, we examined whether different combinations of sampling devices and HPV tests could increase the sensitivity and specificity of ≥CIN 3 and approach the performance of clinician-based endocervical screening. Such improvements could increase acceptability of cervical cancer screening using self-collection with HPV DNA testing in low- and middle-income countries by providing greater reassurance against cancer after a negative HPV test and/or reduction in overtreatment in those who tested HPV positive. The study components were based on our prior clinical trials that suggested a more cellular sample from the upper vagina and a more sensitive HPV assay would increase the sensitivity of self-collected specimens.4
Material and Methods
Study population. Between April 2009 and April 2010, a multisite, population-based cross-sectional cervical cancer-screening study [The Shenzhen Cervical Cancer Screening Trial II (SHENCCAST II)] was conducted in seven sites in Guangdong Province, China. The human subject review boards of the Cleveland Clinic (CCF) and the Peking University Shenzhen Hospital (PUSH) approved the protocol and all procedures followed were in accord with the ethical standards established by these institutions and are in accord with the Helsinki Declaration of 1975. The methods are detailed in a previous publication.5 Women were eligible if they were 25–59 years of age, not pregnant, had no cervical cancer screening for at least 3 years, no prior hysterectomy and no prior pelvic radiation. Eligible women, after being educated on the goals and specific conduct of the study, signed an informed consent document agreeing to participate.
Participating women were block randomized by the day of screening to self-sample with either the POI/NIH self-sampler or the conical-shaped brush. The block randomization varied by day and the recruitment by day was controlled within tight limits to allow smooth operation of our screening clinic. As shown in Figure 1, the Preventive Oncology International/National Institutes of Health self-sampler (POI/NIH self-sampler) has a tamponlike introducer and a nylon-flocked head. It was designed to obtain a larger specimen selectively from the upper vagina versus those obtained with a “conical-shaped brush” (Qiagen, Gathersburg, MD). Women were instructed to insert the tamponlike introducer containing the brush, high in the vagina, retract the introducer, then to rotate the nylon-flocked head. The nylon-flocked head was then pulled through the tamponlike introducer and removed from the vagina. The women who used the conical-shaped brush (Fig. 2) simply inserted the brush high into the vagina, rotated it three times and then removed it from the vaginal canal. Education in the method of vaginal self-collection of specimens was by video and personal instruction. After self-collection of the vaginal specimen, a physician placed a vaginal speculum and the endocervical sample was obtained using a “broom” sampler (Rovers Cervex-Brush; broom, Rovers Medical Devices, Oss, The Netherlands).
Vaginal self-collected specimens were placed in 20-cm3 vials of PreservCyt medium (Hologic, Bedford, MA) for testing for HR-HPV by the Cervista assay (Hologic)10 and the PCR-based mass array matrix-assisted laser desorption/ionization time-of-flight (MALDI-TOF) mass spectrometry system.11 Endocervical clinician-collected specimens were placed in a 20-cm3 vial of PreservCyt medium. The 20 cm3 was then processed for cytology and three HR-HPV assays. (i) HC2, a nucleic acid hybridization assay with signal-amplification for the qualitative detection of 13 high-risk types of human papillomavirus.12 (ii) Cervista is also a signal-amplification method that uses Hologic's invader technology for the qualitative detection of 14 high-risk HPV types.13 (iii) MALDI-TOF is a mass spectrometry method that uses a multiplex primary PCR also for the same 14 HR-HPV types detected by Cervista. In contrast to the HC2 and Cervista assays, the MALDI-TOF assay individually reports each of the 14 specific HR-HPV types. The MALDI-TOF-based HPV multiplex assay was designed by BGI-Shenzhen (Shenzhen, China). Genotyping was performed using the MassEXTEND methodology on a MassARRAY genetic analysis system (Sequenom, San Diego, CA). ThinPrep® slides were evaluated using the I2 Imager computer assisted diagnostic system from Hologic. The cytology slides were reviewed by a cytologist and then a cytopathologist in China. One of the authors (B.Y.) then reviewed all of the slides with interpretations of atypical squamous cells of undetermined significance or worse (≥ASC-US) and a random selection of 4% of the normal slides. The final cytologic interpretation was that of B.Y. A full analysis of the function and performance of the I2 imager and cytology in this trial will be in a future article.
Women who tested positive by either of the HR-HPV tests used for self-collections (Cervista or MALDI-TOF) or any of the three done on endocervical specimens (HC2, Cervista or MALDI-TOF) or had a cervical cytology of ≥ASC-US were asked to return for colposcopy. Women returning for colposcopy were evaluated using the POI microbiopsy protocol of directed and random biopsies which results in subjects having a minimum of four cervical biopsies plus an endocervical curettage.2 Histology slides were interpreted by a gynecologic pathologist from PUSH (Author C.W). Slides with initial histologic diagnoses of CIN 2, CIN 3 or cancer and a random selection (31%) of those with a diagnosis of CIN 1 were reviewed by a gynecologic pathologist from the Cleveland Clinic (author B.Y.). Immunochemical staining with p16 was selectively obtained to adjudicate difficult cases. The final histologic diagnosis was that defined by B.Y.
This trial was powered around the paired comparisons of the HPV assays. There were no sample size calculations for the analyses of the test characteristics for the novel self-sampling device as there are no estimates of its performance. However, to justify further development of the self-sampler we will assume that it must have a sensitivity of at least 90% when used with the HPV assays in this trial. This required only 56 cases of CIN Grade 2 or greater (≥CIN 2) and less than 2,000 participants, well within the study design.
Differences in sensitivity and specificity for ≥CIN 3 were compared to McNemar's test when there were paired results and Chi-square when there were independent groups. Multiple comparisons (four) of the HPV tests (HC-II, Cervista and MALDI) were performed, therefore to be conservative we applied the Bonferroni correction and set our alpha level for the comparisons of sensitivity and specificity of the assays and brushes at 0.0125. Difference in the partial area under the ROC curve (PAUC) between a specificity of 100 and 85% were compared using the comproc procedure in STATA 10.0, Z-scores were used to test the equality of the AUC. Confidence intervals (CIs) are exact binomial CIs. Differences PAUC with p < 0.05 were considered significant. Subject block (daily) randomization was the framework for the comparison of the self-collection devices. However, the paired samples were the determinants for biopsy verification. Data analyses were performed using STATA 10.0 (StataCorp LP, College Station, TX).
Ten thousand women were screened and entered into the SHENCCAST II study. There were 8,556 women with a mean age of 38.9 years who had all prescribed screening and diagnostic procedures with no missing data, who are included in this analysis. One thousand four hundred forty-four (14.44%) women were dropped from the analysis (some missing more than one test, most related to the available sample based on the order of the assays): 196 (1.96%) were missing cytology data, 3 (0.03%) were missing HC2 endocervical test, 264 (2.64%) were missing Cervista endocervical test, 537 (5.37) were missing MALDI-TOF endocervical test, 106 (1.06%) were missing Cervista self-test and 159 (1.59%) were missing MALDI-TOF self-test. Six hundred twenty-five of 3,328 (18.78%) asked to return for colposcopy did not return. The group that did not return was similar in test positivity and demographics to the group that did return.5
The prevalence of ≥CIN 3 was 1.6% (141/8,556). As shown in Table 1, when vaginal self-collected specimens were tested for HR-HPV by Cervista, the sensitivity for ≥CIN 3 of specimens obtained with the POI/NIH self-sampler (71.6%) was similar to that of specimens obtained with the conical-shaped brush (70.0%, p = 0.92). When vaginal self-collected specimens were tested for HR-HPV by MALDI-TOF, the sensitivity for ≥CIN 3 of specimens obtained with the POI/NIH self-sampler (91.4%) was also similar to that of specimens obtained with the conical-shaped brush (98.3%, p = 0.14). Although vaginal self-collected specimens tested for HR-HPV by Cervista and obtained with the POI/NIH self-sampler had a higher specificity for ≥CIN 3 (86.9%) than those obtained with the conical-shaped brush (85.3%, p = 0.04), there were no differences in specificity of the two sampling devices when tested by MALDI-TOF (87.9 vs. 87.3%, p = 0.42). The overall accuracy of the two brushes is shown graphically in Figure 2 in which the receiver operating characteristic (ROC) curves for diagnosis of ≥CIN 3 for Cervista HR-HPV testing of the specimens obtained with the POI/NIH self-sampler are compared to those of specimens obtained with the conical-shaped brush. The partial area under the Cervista HR-HPV testing ROC curve from specificity of 100 to 85% of the POI/NIH self-sampler (0.8115) is similar to that of the conical-shaped brush (0.8256, p = 0.76).
By contrast, the sensitivity for ≥CIN 3 of vaginal self-collected specimens obtained with the POI/NIH self-sampler or the conical-shaped brush and tested for HR-HPV by Cervista (71.6 and 70.0%) were significantly lower than the same specimens tested by MALDI-TOF (91.4%, p = 0.0002 and 98.3%, p < 0.0001). Figure 3 graphically demonstrates that the HR-HPV testing with MALDI-TOF points for sensitivity and 1-specificity for ≥CIN 3 for vaginal self-collected specimens obtained by the POI/NIH self-sampler (open circle) and those obtained by the conical-shaped brush (solid circle) are separate from the Cervista vaginal self-collected specimen ROC curves.
Table 2 shows that the sensitivity for ≥CIN 3 of vaginal self-collected specimens obtained with either sampling device tested for HR-HPV by Cervista (70.9%) is lower than that of endocervical clinician-collected specimens tested by Cervista (95.0%, p = 0.0001), whereas the sensitivity for ≥CIN 3 of vaginal self-collected specimens obtained with either sampling device tested for HR-HPV by MALDI-TOF (94.3%) is identical to that of endocervical clinician-collected specimens tested for HR-HPV by MALDI-TOF (94.3%, p = 1.0). The sensitivity for ≥CIN 3 of vaginal self-collected specimens obtained with either sampling device tested for HR-HPV by MALDI-TOF (94.3%) is also similar to that of endocervical clinician-collected specimens tested for HR-HPV by Cervista (95.0%, p = 1.0). Regardless of the HR-HPV assay employed, the specificity of the vaginal self-collected specimens is lower than that of the endocervical specimens (86.1 vs. 90.3%, p < 0.001 for Cervista and 87.5 vs. 89.4%, p < 0.001 for MALDI-TOF). The relationship between HR-HPV testing of vaginal self-collected specimens obtained with either sampling device and endocervical clinician-collected specimens is shown graphically in Figure 4 in which the HR-HPV testing with MALDI-TOF points for sensitivity and 1-specificity for ≥CIN 3 for vaginal self-collected specimens lie on the endocervical specimen Cervista ROC curve rather than on the vaginal self-collected specimen Cervista ROC curve.
This importance of our study is that the high-risk HPV assay is more important than the sampling device. Combining a self-collected specimen obtained with a simple brush, with an assay having the capability of very high throughput, we achieved a similar sensitivity and only slightly lower specificity for ≥CIN 3 as can be accomplished by a direct endocervical specimen obtained by a physician. The implications for massive country-wide screening “events” in low and middle resource countries are profound. These findings open the possibility to screen thousands of women each day at a fraction of the cost and a fraction of the time of current paradigms.
Our previous study (SPOCCS III),4 suggested that a vaginal self-collected specimen tested for HR-HPV would have increased sensitivity for ≥CIN 3 if a larger specimen was collected selectively from the upper vagina or if a PCR-based assay was used to test for HR-HPV. Although the POI/NIH self-sampler brush was designed to obtain a larger specimen and the cardboard sleeve was added to encourage selective sampling of the upper vagina, this trial showed that the POI/NIH self-collection device was ineffective in achieving our goals and thereby increasing the sensitivity for ≥CIN 3 of the vaginal self-collected specimens. We speculate that the ineffectiveness of the POI/NIH self-sampler in increasing the accuracy of the vaginal self-collected specimens is secondary to women being afraid to insert the sleeve to the top of the vagina or because the women did not fully understand that the sleeve needed to be retracted before obtaining the specimen. Tampon use is very uncommon in parts of the world with the highest rates of cervical cancer. This unfamiliarity, even with our efforts to instruct the women in the proper use of the self-sampler, surely was a contributing factor. Future self-sampling trials will focus on the simplest and least expensive brush.
Although the POI/NIH self-sampler was ineffective in increasing the sensitivity of the vaginal self-collected specimen, the PCR-based MALDI-TOFHR-HPV assay was very efficacious. The sensitivity for ≥CIN 3 of the vaginal self-collected specimens obtained with either sampling device and tested for HR-HPV by MALDI-TOF (94.3%) was identical to that of the endocervical clinician-collected specimens tested by MALDI-TOF (94.3%, p = 1.0) and similar to that of endocervical clinician-collected specimens tested by Cervista (95.0%, p = 1.0). As SPOCCS III predicted, the specificity for ≥CIN 3 of the vaginal self-collected specimen tested for HR-HPV was lower than that of similarly tested clinician-collected endocervical specimens. This decrease in specificity was found despite employing the Cervista and MALDI-TOF assays which differ from HC2 in that neither cross reacts with low-risk HPV.11, 13 The decrease in specificity of the vaginal self-collected specimens is probably secondary to HR-HPV present solely in the vagina which is not associated with ≥CIN 3.4
For this trial, we used the Cervista HR-HPV assay as the non-PCR-based assay. However, for more than a decade, HC2 has been the standard HR-HPV assay to which all new assays are compared. Therefore, it is appropriate to note that in a prior publication we documented that the ROC curves for HR-HPV testing of physician-collected specimens of the Cervista and HC2 assays were virtually identical,5 and as we have shown vaginal self-collected specimens tested for HR-HPV by MALDI-TOF also lie on the endocervical physician-collected ROC for HC2.
The strengths of this trial are that the endpoint was ≥CIN 3 rather than ≥CIN 2, the population studied had a relatively high prevalence of ≥CIN 3, and there is probably little verification bias. Using ≥CIN 3 rather than ≥CIN 2 as the endpoint avoids the problems of inaccuracy of diagnosing CIN 2 and the fact that 40–50% of lesions diagnosed as CIN 2 regress without treatment. CIN 3 is quite simply a superior surrogate for cancer as opposed to less severe epithelial abnormalities.14, 15 In addition in low-resource settings it only matters if you get cancer, not whether you have CIN 1, CIN 2 or CIN 3. Once you are primary self-sample HR-HPV positive you will enter the triage/treatment system that is appropriate for your setting. Performing the trial in a population with a prevalence of ≥CIN 3 of 1.7%, having 8,556 participating women, and obtaining paired results of screening tests allows detection of relatively small differences in sensitivity and specificity. Verification bias was limited by the protocol calling for colposcopy and biopsy of women with positive HR-HPV by any of the assays on either vaginal or endocervical specimens or cytology of ≥ASC-US. As there are very few women with ≥CIN 3 who have neither a positive endocervical HR-HPV nor an abnormal cervical cytology,2 it is unlikely that significant numbers of women with ≥CIN 3 would have been missed.
We conclude that the vaginal self-collected specimen tested for HR-HPV by MALDI-TOF has similar sensitivity and lower specificity for ≥CIN 3 compared to an endocervical clinician-collected specimen. Utilizing HR-HPV testing of vaginal self-collected specimens as a primary screen for prevention of cervical cancer has significant advantages when compared to endocervical clinician-collected specimens tested for HR-HPV or cervical cytology. Its greatest advantage is the ease with which a single individual can instruct the women and gather hundreds of self-collected samples/day.16 In addition, we have previously demonstrated that women prefer the self-collection over a traditional gynecologic examination.17
In this trial, 13.3% of women tested positive for HR-HPV on their vaginal self-collected specimen. An important focus for cervical cancer-screening research today is how to manage these positives.16 None of the current options for secondary screening, which include unaided visual inspection (VIA),18 cervical cytology, genotyping for the highest risk types19 or p16 testing20 provide the perfect solution, but each may find a role depending on the rate of positive HR-HPV in the population being screened, their geographic location, and available resources for secondary screens.
- 1IARC. GLOBOCAN 2008, Section of Cancer Information. Available at: http://globocan.iarc.fr/factsheets/cancers/cervix.asp. Accessed on November 21, 2010.
- 3Shanxi Province Cervical Cancer Screening Study II: self-sampling for high-risk human papillomavirus compared to direct sampling for HPV and liquid based cervical cytology. Int J Gyn Cancer 2003; 13: 1–8., , , , , , , , , , , , et al.
- 4Prevalence of type-specific human papillomavirus in endocervical, upper and lower vaginal, perineal, and vaginal self-collected specimens; implications for vaginal self-collection. Int J Cancer 2010; 127: 1151–7., , , , , , , , , , .
- 17Patient acceptance of self-sampling for human papillomavirus in rural China. J Lower Genital Tract Dis 2003; 7: 107–16., , , , , , .