Human Papillomavirus Testing for Primary Cervical Cancer Screening

The human papillomavirus (HPV) has been clearly established as the primary cause of cervical cancer in nearly all cases.¹ Thus, it should not be surprising that testing for HPV should have a role in measures aimed at control of this disease. The ultimate goal must be eradication of HPV by vaccination, but a more immediate prospect is the detection and monitoring of the virus as part of the screening and diagnostic process. Testing for HPV could have 3 potential roles: triage of patients with atypical squamous cells of undetermined significance (ASCUS) and low-grade cervical smears; surveillance of high-grade cervical intraepithelial neoplasia and localized (micro) invasive disease after treatment; and primary screening—either alone or in combination with cytology.

This third area is the most far-reaching and 2 articles in this issue of THE JOURNAL^{2 - 3} make important contributions to its evaluation. The potential role of HPV testing in cervical screening is highly dependent on the existing infrastructure. For clinical settings in which an effective, well-organized, cytology-based program is in place, the issue is whether HPV testing adds to the existing program and questions of cost-effectiveness, quality control, and added value to current practice come to the fore. In contrast, for settings in which screening is nonexistent, or is ineffective because of poor-quality cytology or inherent limitations due to a high rate of inflammatory smears, the more basic questions of sensitivity, specificity, and simplicity of testing procedures become paramount. In a review for the National Health Service of the United Kingdom,⁴ a technology assessment panel recommended limited introduction of HPV testing for ASCUS and low-grade cytology. However, while the panel recognized the greater potential of primary screening to save both money and lives, the panel emphasized that more research was needed before the test could be introduced for primary screening. In clinical settings in which an effective cytology program does not exist, a more far-reaching approach might be more appropriate.

Evaluation of all screening tests is largely based on the achievable levels of sensitivity and specificity. The great attraction of HPV testing is its high sensitivity. Several studies have shown sensitivities of approximately 90% or greater for the second-generation HPV test^{5 - 8} and comparative studies indicate a higher sensitivity than that achieved by cytology.^{5 - 6 ,9 - 14} Specificity is now the major concern and false-positive rates of 5% to 20% have been reported.^{6 ,11 - 14}

The study reported by Schiffman et al² in this issue of JAMA focuses on these parameters and provides a complete description of the performance of HPV testing by means of receiver operating characteristic curve analysis. These curves, which plot sensitivity against the positivity rate for women without disease for different thresholds of test positivity, are an elegant way to present test performance data. In their population-based study of 8554 randomly selected women, 18 years or older, Schiffman et al performed HPV testing, as well as conventional cytology, 2 new forms of cytology (liquid-based cytology and computer-assisted), and cervicography, thereby allowing comparisons among the methods.

The receiver operating characteristic curves demonstrated clearly that a 1-pg/mL level for HPV positivity was an optimal cut point for this population; lower cut points greatly increased false-positive rates without materially improving detection rates, and higher cut points reduced sensitivity. At this cut point, the sensitivity for detecting high-grade squamous intraepithelial lesions (HSILs) or cancer was 88.4% compared with a sensitivity of 77.7% for conventional cytology. However, specificities for HSIL or cancer were lower for HPV testing (89.0%) than for cytology (94.2%), leading to almost twice the referral rate for colposcopy. This is a key issue, especially if HPV testing is performed at relatively short intervals. The overall disease prevalence (for HSIL and cancer) in this study was 1.6%, leading to a positive predictive value (PPV) of 11.6% for HPV compared with a PPV of 18.0% for cytology when a diagnostic grade of ASCUS or worse was considered as positive. Sensitivity was not affected by age, but false-positive rates were, leading to a 3-fold reduction in referral rates from the lower to upper tertile of age, and a corresponding increase in PPV.

The issues of the age at which to use HPV as a primary test and the appropriate positivity levels are crucial parameters. In a study evaluating a highly screened population, 35 years or older, in the United Kingdom,⁶ HPV testing by the second-generation HPV microplate test had a 95.5% sensitivity and all positive cases had levels above 4 pg/mL. The referral rate was 6.8% using a 1-pg/mL cut point but only 4.2% at 2 pg/mL, leading to an increase in PPV from 17.1% to 27.0%. The findings for cytology were similar to those reported by Schiffman et al, with a sensitivity of 85.7% and a PPV of 21.7% for ASCUS or worse. Choice of HPV positivity threshold may be dependent on population.

The study by Wright et al³ examines testing in an area where minimal infrastructure exists, and considers a completely unscreened population, aged 35 to 65 years, in a "peri-urban" settlement near Cape Town, South Africa. Cancer or HSIL was detected in 4.25% of the 1365 women who were adequately assessed. Five methods of screening were evaluated: self-collected vaginal swabs for HPV testing, clinician-obtained conical brush samples for HPV testing, conventional cytology, cervicography, and direct visual inspection after application of 5% acetic acid. Testing for HPV was again performed by the second-generation HPV test and the 1-pg/mL cut point was used. Sensitivities were somewhat lower than for other recent studies with this test, with 83.9% for the clinician-collected sample and 66.1% for the self-collected sample. The self-collected sample had a similar sensitivity as cytology (66.1% vs 67.9%, respectively, for ASCUS or higher) but higher false-positive rates (17.1% vs 12.3% for cytology). However, the discrepancies in HPV positivity between clinician-collected and self-collected samples (11 vs 1, respectively, for HSIL or cancer; 121 vs 127, respectively, for other diagnoses), clearly indicate that the clinician-collected sample was better. This was true even though the self-collected sample was performed in optimal conditions (in the examination room just after specific instructions for its use). Performance of sample self-collection in more realistic conditions (eg, postal sample or community distribution) still needs to be evaluated. The type of collection device used in this study (conical brush for use by a clinician vs vaginal swab for self-collection) also may be important. However, as a result of the expert team conducting the study, the cytology results probably reflect optimal conditions to an even greater degree, and circumstances in which the cytology results could be much worse are easily envisaged.

An accurate self-sampled HPV test could have enormous implications. Such a test opens up the possibility of evaluating women who are otherwise unwilling or unable to submit to pelvic examinations. Whether in underdeveloped areas this would offer an advantage over the use of a trained nurse who performs visual inspection and HPV testing in the community is not clear. In areas where organized screening is in place, self-sampling offers an additional approach for reaching women who refuse to have conventional screening and also may have a role in surveillance or the monitoring after the treatment of HPV-positive cytology-negative women, in which follow-up testing at short intervals is needed. However, this potential for greater acceptability and ease of use also could have negative effects. For example, some women who would have previously had routine testing might now choose to rely on a less-sensitive home test. A clear priority is to determine the reasons for lower sensitivity (such as use of a vaginal swab) and to find ways to make a self-sampling test as good as a test performed by a clinician.

Placing HPV in the causal chain of cervical cancer has permitted a more rational approach to screening for cervical cancer through HPV testing. As demonstrated by Schiffman et al and by Wright et al, such testing potentially can enhance the sensitivity of cervical cancer screening and also could be used to reach women who otherwise might not undergo such screening. However, whether widespread HPV testing is feasible or affordable and whether it will eventually lead to fewer cases of invasive cervical cancer or reduce the morbidity and mortality of the disease will require further study to optimally identify and treat the population of women at risk.