Evaluation of an ELISA for p16^INK4a as a Screening Test for Cervical Cancer Free

Due to limitations in cytology-based screening programs for cervical cancer (1-3), there is substantial interest in developing better biomarkers for early detection. Testing for high-risk human papillomavirus, the chief cause of cervical cancer (4), has been found to have a relatively high sensitivity and a relatively low specificity, particularly among women ages <30 years (5). High-risk HPV DNA testing is recommended as an adjunct test for women with cytology showing atypical squamous cells of undetermined significance (ASC-US; ref. 6). As a primary screening test, HPV DNA is being recommended only for women ages >30 years as a combination test with cytology (6). Aside from HPV, several host molecules are being investigated for their use in early detection (7, 8), p16^INK4a being prominent among them.

p16^INK4a is a negative regulator of cellular proliferation that influences the retinoblastoma (Rb)–controlled checkpoint of the cell cycle. By binding with cyclin-dependent kinase proteins (CDK4/CDK6), p16^INK4a activates Rb (7-10). This prevents the G₁-S-phase transition and decelerates cell proliferation. p16^INK4a is inactivated in many cancers, which results in inactivation of Rb and uncontrolled cell proliferation (10, 11). In HPV-related tumors, Rb is inactivated by the E7 oncoprotein of high-risk HPV (7, 8). The E7-induced inactivation of Rb leads not only to increased cell proliferation but also an accumulation of p16^INK4a in the cell, because the levels of p16^INK4a and Rb are maintained through a negative feedback loop (12).

Using immunostaining techniques, increased levels of p16^INK4a protein are observed in cervical cells and tissues showing cancer or precancerous changes. Strong p16^INK4a immunostaining is observed in most high-grade cervical lesions (13, 14) and in a smaller fraction of low-grade lesions. These studies suggest that p16^INK4a might be a good marker for identifying lesions that are most likely to progress (15-20). We recently reported on the positive correlation between histologically confirmed high-grade cervical intraepithelial neoplasia (CIN) and level of solubilized p16^INK4a protein detected by testing lysed cervical cell samples using a prototypic ELISA (21). p16^INK4a ELISA testing and high-risk HPV DNA testing detected similar proportions of women with high-grade CIN, but p16^INK4a ELISA testing yielded significantly fewer positive results among women with low-grade CIN or normal histologic findings. Similar findings were reported by Wentzensen et al. among women attending a colposcopy clinic in Germany (22).

Based on these preliminary findings, we undertook a study to compare the relative performances of cervical liquid-based cytology, HPV DNA testing, and p16^INK4a ELISA testing in cervical cancer screening.

Between October 2005 and November 2007, women attending two Planned Parenthood clinics for routine Papanicolaou (Pap) screening in Washington state were invited to participate in this study. Women were eligible if they were between ages 18 and 50 years, were or had previously been sexually active, were not pregnant or chronically immunocompromised, and had no history of treatment for CIN. All study procedures were approved by the Human Subjects Division at the University of Washington.

Before the clinical examination, a brief questionnaire was administered to collect demographic, behavioral, and sexual history information. The clinical examination included collection of cervical specimens for p16^INK4a ELISA, cytology, and HPV DNA testing, in that order. The p16^INK4a ELISA sample was collected using an Ayer's spatula to obtain cells from the transformation zone and a cytobrush for endocervical cells. The spatula and cytobrush were rinsed and retained in the transport medium until laboratory processing. Ectocervical and endocervical samples for cytology were collected using an Ayer's spatula and cytobrush in one clinic and a cervix broom in the other clinic. The spatula, brush, and broom were repeatedly tapped and rinsed but not retained in the transport medium. A Dacron swab was used for ectocervical and endocervical sample collection for HPV DNA testing. The swab was placed in Standard Transport Medium (Digene) and tested by the Hybrid Capture II (hc2) assay (Digene).

Samples for testing other genital infections were collected at the discretion of the clinician before collecting the p16^INK4a ELISA sample. For two periods during the study (October 2005 to July 2006 and July 2007 to October 2007), participants were asked to self-collect vaginal samples using Dacron swabs before the clinical examination using the same protocol described by our group in a different study (23). Women were eligible to participate in the present study even if they declined to provide the vaginal sample. The analyses presented in this article do not concern HPV testing of vaginal samples; information about these tests is provided only when it affects the protocol and findings of the present study.

Women were invited to return for diagnostic testing by colposcopy and biopsy if they had any of the following screening results: (a) p16^INK4a ELISA result of ≥8 pg/mL; (b) ASC-US, low-grade squamous intraepithelial lesion (LSIL), high-grade squamous intraepithelial lesion (HSIL), atypical glandular cells (AGC), or atypical squamous cells cannot rule out HSIL (ASC-H) on cervical cytology; (c) a positive hc2 test result from the cervical and/or vaginal sample. A random subset of women with normal screening test results was invited for colposcopy and biopsy to attain histologic confirmation of disease status for this group of negative controls.

At the diagnostic visit, ectocervical biopsies were obtained from areas of the cervix appearing abnormal on colposcopic examination. An ectocervical biopsy was done at the 12 o'clock position if no lesion was visible during colposcopy. An average of 1.13 ectocervical biopsies were collected per woman attending the colposcopy visit. The procedure was done by board-certified gynecologists using a Cooper Surgical Cerveillance scope digital colposcopy system Model CS2000. A digital photograph was taken at the time of colposcopy. Endocervical curettage was done if the colposcopy was inadequate, lesions extended into the endocervical canal, the screening cytology was HSIL but no lesion was visible during colposcopy, or the screening cytology result was AGC.

Women with histologically confirmed ≥CIN2 severe lesions were offered treatment by loop electrosurgical excision procedure (LEEP). There were no cases of invasive cancer. Women with HSIL on screening cytology and <CIN1 on histology, and a few other women determined by the clinician's discretion, underwent repeat biopsy. Women with ≤CIN1 were advised to return to their usual provider for follow-up care.

HPV DNA testing for 13 high-risk types (HPV types 16, 18, 31, 33, 35, 39, 45, 51, 52, 56, 68, 59, and 68) was done using Probe B of the hc2 assay following the manufacturer's specifications. Values ≥1 RLU were classified as positive.

SurePath thin-layer cytology slides were prepared and processed according to the manufacturer's specifications (TriPath Imaging), stained with the Pap stain, and reviewed at Dynacare Laboratories and Skagit Pathology. Determination of adequacy and classification of cellular changes were according to the Bethesda classification system (24, 25). A random 10% of all slides read as normal was rescreened manually as mandated by federal law (Clinical Laboratory Improvement Amendments). Testing and review of HPV DNA and cytology were done without knowledge of any other clinical or laboratory data.

H&E-stained slides of the biopsy tissue were prepared and reviewed by pathologists at Harborview Medical Center without knowledge of the p16^INK4a ELISA results. Diagnoses were assigned using both the CIN and Bethesda classification systems. Cases for which the histologic diagnosis was ambiguous were reviewed by two pathologists for a consensus diagnosis.

The ELISA test used in this study was a prototypic assay developed at mtm laboratories as described previously (21). Based on results from pilot testing of this assay (21), a value of ≥8 pg/mL was considered to be positive.

At the end of the study, all p16^INK4a ELISA samples were retested with an enhanced version of the ELISA that was developed during the course of the study. This new assay included (a) an adjusted quantity of one detergent in the lysis medium, (b) the omission of one denaturing agent, and (c) the addition of a heating step at 95°C for 10 min to the sample in the lysis medium before further processing. These changes facilitated complete lysis of cells for better detection of p16^INK4a protein. Referral to colposcopy was always based on the original ELISA test result. Testing for p16^INK4a, with both the original and enhanced assay protocols, was done without knowledge of any other test results.

This study was designed to test the hypotheses that, compared to cytology testing, the sensitivity of p16^INK4a ELISA testing would be at least 50% greater and the specificity would differ by no more than 5% for detecting cervical lesions of ≥CIN3. To test this hypothesis at the α = 0.05 level and with 80% power, at least 25 women with the disease outcome and 982 women without the disease outcome were required. ≥CIN3 was used as the primary endpoint in all analyses, but results are also presented using ≥CIN2 as an endpoint.

Estimates of sensitivity and specificity for detection of ≥CIN3 were obtained for three screening strategies, including referral to colposcopy of women with the following screening test results: (a) a value of ≥8 pg/mL from the p16^INK4a ELISA test; (b) a positive result from hc2 testing of the cervical sample; and (c) a screening thin-layer Pap test showing ASC-H, AGC, LSIL, or HSIL or a screening thin-layer Pap test showing ASC-US with positive hc2 test results on the cervical sample (called “cytologic screening with reflex hc2 testing for ASC-US”). In further analyses, we examined the performance of p16^INK4a ELISA at a lowered cut-point of ≥6 pg/mL, the performance of the enhanced ELISA at the ≥8 and ≥6 pg/mL cut-points, and the performance of a combined screening strategy based on testing by both hc2 and p16^INK4a ELISA. In the combined strategy, women would be referred to colposcopy if they had positive results from both p16^INK4a ELISA (≥8 and ≥6 pg/mL cut-points) and hc2 testing of cervical samples.

Sensitivity and specificity estimates were calculated using results from the screening visit to evaluate the outcome of each screening strategy using the presence or absence of histology-confirmed ≥CIN3 as the main disease outcome, which was determined based on the most severe histologic finding from ectocervical biopsy, endocervical curettage, and/or LEEP tissue samples. Inverse probability weighting was used to correct sensitivity and specificity estimates for verification bias using the method described in a previous study (26). Using this method, the study population was classified into five “screening strata” based on the screening test results that influenced colposcopy attendance rates: (a) negative results on all screening tests; (b) normal results on thin-layer cytology but positive results for at least one other screening test (hc2 test of cervical or vaginal sample or p16^INK4a ELISA test of cervical sample); (c) thin-layer cytology results of ASC-US; (d) thin-layer cytology results of LSIL; and (e) thin-layer cytology results of AGC, ASC-H, or HSIL. Within each screening stratum, the distribution of lesion grades among those who did and did not return for colposcopy was assumed to be the same. This is a plausible assumption because the HPV DNA and thin-layer cytology results were the only factors found to be independently associated with both colposcopy attendance and detection of ≥CIN3. Colposcopy attendance was not independently associated with p16^INK4a ELISA test result. Other potentially confounding factors, including age, race/ethnicity, education, income, number of sex partners, hormonal contraceptive use, parity, and time between screening and colposcopy visits, were not independently associated with both colposcopy attendance and detection of ≥CIN3 (data not shown).

One case of CIN3 occurred in a woman with normal screening test results and a normal colposcopic impression (12 o’clock biopsy specimen). Because a very small proportion [35 of 1,030 (3.4%)] of women with normal screening test results attended colposcopy, diagnostic results from these women were weighted heavily in corrected estimates of sensitivity, influencing both the magnitude and the precision of the estimates. To address this, we chose to present our findings using two alternate analyses. In the first analysis, which was based on the study protocol, the most severe histologic diagnosis from ectocervical biopsy, endocervical curettage, or LEEP tissue specimens was used to define disease outcome. In the alternate post hoc analysis, women with negative results on all screening tests were included in the nondiseased group regardless of their diagnostic test results. We refer to these as the “original” and “a posteriori” analyses, respectively. The a posteriori analysis provides an indication of test performance in detecting disease that would typically be diagnosed in a routine clinic setting, as it is not standard practice to perform biopsies on women with negative screening test results.

Screening samples that were inadequate or insufficient for evaluation were classified as test positive because women with inadequate Pap test results are usually asked to return for a repeat test rather than left without follow-up. This assumption affected only two women who had inadequate Pap test results. Additional sensitivity analyses used ≥CIN2 as the criterion standard for disease outcome.

Means, medians, and SDs were calculated for continuous variables. Differences in proportions were tested using the Pearson's χ² test for independent variables and the McNemar's χ² test for paired data. Multivariate logistic regression analyses were used to estimate odds ratios of association between potentially confounding factors and colposcopy attendance and high-grade histology results. Ninety-five percent confidence intervals (95% CI) were computed using the 2.5th and 97.5th percentiles of the bootstrap distribution of weighted estimates of sensitivity and specificity. The statistical software package STATA 9.0 was used for data analyses. Weights were applied using the pweights command.

Study participants were relatively young (median age, 23 years) and predominantly of White race (Table 1). They reported a median of 6 male sex partners over their lifetimes. The prevalence at screening of high-risk HPV DNA detected by hc2 testing of cervical samples, abnormal cytology findings by liquid-based cytology, and positive p16^INK4a ELISA test results (≥8 pg/mL cut-point) were 30.6%, 13.4%, and 10.4%, respectively.

Of 1,781 women screened, 490 (27.5%) underwent colposcopy and biopsy within a median of 45 days (range, 9-225 days) following the screening visit (Fig. 1). Based on histologic findings from cervical biopsy, endocervical curettage, and/or LEEP specimens, the final diagnoses for these 490 women were 258 (52.7%) normal or mildly atypical results, 2 (0.4%) atypical glandular changes, 181 (36.9%) CIN1, 19 (3.9%) CIN2, 28 (5.7%) CIN3, and 2 (0.4%) samples that were inadequate for evaluation. Thus, 28 cases of ≥CIN3 and 47 cases of ≥CIN2 were included in the original analysis. Cervical samples of the 28 cases of ≥CIN3 revealed normal cytologic findings in 10 cases, p16^INK4a ELISA results of <8 pg/mL in 14 cases, and a negative hc2 result in 1 case. The single case of CIN3 with negative hc2 results also had normal findings by cytology and p16^INK4a ELISA. This case was diagnosed in a 28-year-old woman who volunteered to attend the diagnostic visit as a negative control. Her colposcopic examination (with visualization of the entire squamo-columnar junction) was normal and the ectocervical biopsy obtained per protocol at the 12 o’clock location showed CIN2. Her LEEP tissue revealed a CIN3 lesion that extended into the endocervical canal. This case was excluded from the disease outcome groups in the a posteriori analysis, resulting in 27 cases of ≥CIN3 and 46 cases of ≥CIN2. After adjusting for verification bias, the prevalence of ≥CIN3 was 3.9% in the original analysis and 2.2% in the a posteriori analysis.

As the a posteriori analyses represent findings that could be expected in a clinical setting, those findings are presented here. Results for the original analysis are presented in Appendix A.

The sensitivity and specificity estimates for p16^INK4a ELISA testing were similar to those for cytology testing (with reflex hc2 testing of ASC-US; Table 2). Both strategies had markedly lower estimates of sensitivity and higher estimates of specificity for detection of ≥CIN3 compared to hc2 testing.

Estimates of sensitivity or specificity did not markedly change when ≥CIN2 was used as the outcome instead of ≥CIN3. There was a trend toward lower sensitivity estimates for all three screening strategies for detection of ≥CIN2 versus ≥CIN3 (Table 2). There was no notable change to the results when the assumption regarding inadequate test results was altered such that inadequate results were treated as negative (data not shown).

Findings in the original analyses were similar, but estimates of test sensitivity were lower and less precise, as indicated by wider 95% CIs (Appendix A1).

Lowering the cut-point for a positive result for the p16^INK4a ELISA test from ≥8 to ≥6 pg/mL led to an increase in the estimate of sensitivity and a substantial decrease in the estimate of specificity for detection of ≥CIN3 (Table 3A).

The enhanced p16^INK4a test (using the ≥6 pg/mL cut-point by itself or in combination with hc2 testing) was markedly more sensitive than all other p16^INK4a ELISA-based strategies. All screening strategies using the enhanced p16^INK4a ELISA were less specific than equivalent strategies using the original test. Strategies of referral to colposcopy based on positive results from both p16^INK4a ELISA and hc2 testing were markedly more specific than equivalent strategies based on p16^INK4a ELISA testing alone (Table 3A). A strategy of referral to colposcopy based on positive results from both p16^INK4a ELISA (enhanced test, ≥6 pg/mL cut-point) and hc2 testing had a sensitivity of 91.8 (79.1-100.0) and specificity of 86.0 (82.0-89.0) for detection of ≥CIN3 (Table 3A). In comparison, hc2 testing for detection of ≥CIN3 resulted in sensitivity and specificity estimates of 100.0 (95% CI, 96.2-100.0) and 69.2 (95% CI, 61.2-75.5), respectively (Table 2).

The relative rankings of sensitivity estimates of the various screening strategies that used an alternate cut-point or laboratory protocol for p16^INK4a ELISA or a combination of p16^INK4a ELISA and hc2 tests were similar when ≥CIN2 was used as the disease outcome (Table 3A and B). Sensitivity estimates of all strategies tended to be lower for detecting ≥CIN2 versus ≥CIN3, and this trend was most pronounced for strategies involving the enhanced p16^INK4a ELISA test using the ≥6 pg/mL cut-point. Specificity estimates for ≥CIN3 versus ≥CIN2 were similar for all strategies (Table 3A and B).

Findings were similar in the original analyses (Appendices A2 and A3).

Our investigation of a prototypic ELISA test for p16^INK4a protein in cervical cancer screening indicated that this novel assay performed with sensitivity and specificity estimates for detection of ≥CIN3 that were similar to estimates for cervical cytology screening (with reflex high-risk HPV testing of women with ASC-US). Compared to these two strategies, screening by hc2 testing of cervical samples for high-risk HPV DNA was more sensitive but less specific and would lead to diagnostic referrals of approximately three times as many women. An enhanced p16^INK4a ELISA based on changes in processing of the sample lysis medium and use of a lowered cut-point for referral to colposcopy led to substantially higher sensitivity and lower specificity of p16^INK4a ELISA-based screening strategies. The specificity of screening with this enhanced ELISA could be improved by incorporating HPV DNA testing into a combined strategy in which women with positive results from both p16^INK4a ELISA (enhanced assay, ≥6 pg/mL cut-point) and hc2 tests would be referred to colposcopy. These findings were consistent whether ≥CIN2 or ≥CIN3 was used as the disease outcome.

To our knowledge, this is the first published study to evaluate the p16^INK4a ELISA test in cervical cancer screening. In the proof-of-principle study that we conducted among 319 women with abnormal cytology results, we found that, when the sample for p16^INK4a ELISA was collected before any other cervical sample, as was done in the present screening study, 90% of ≥CIN3, 77% with CIN2, and 53% with <CIN2 tested positive at the ≥8 pg/mL cut-point (21). In a similar proof-of-principle study conducted in Germany among women with abnormal screening cytology results, 97% of those with ≥CIN3, 96% with CIN2, and 55% with <CIN2 tested positive for p16^INK4a ELISA using the ≥8 pg/mL cut-point (22). Compared to results from the present screening study, the p16^INK4a ELISA (≥8 pg/mL cut-point) performed with higher sensitivity in both proof-of-principle studies possibly in part because the two proof-of-principle studies were conducted in higher-risk populations of women who had abnormal cytologic findings that met standard clinical criteria for referral to colposcopy and biopsy. In this screening study, only 17 (63%) of 27 ≥CIN3 cases occurred in women with abnormal cytologic results. Moreover, colposcopists in our screening study observed that most CIN3 lesions were small. Given the independent relationship between p16^INK4a ELISA level and lesion size that we reported earlier (21), one might expect lower sensitivity of the p16^INK4a ELISA test in detecting small lesions or lesions that shed few abnormal cells.

Other studies of p16^INK4a in cervical neoplasia have applied immunostaining techniques to measure p16^INK4a protein in cervical cell or tissue samples. These studies have suggested that immunostaining for p16^INK4a can be used to clarify equivocal findings on histology (15, 18) or cytology (17, 19, 20). Detection of p16^INK4a through immunostaining of screening cytology samples shares some of the drawbacks associated with Pap staining: staining and interpreting slides is a labor-intensive, time-consuming, and subjective process. Most studies of p16^INK4a immunocytochemistry have relied on using a count of the number of p16^INK4a-positive cells to define the level of staining, but there is substantial variation in definitions used across studies. Recently, an alternative measurement based on a nuclear score has been proposed, but even this qualitative score relies on subjective interpretation of morphologic characteristics within the cell (27). In comparison, a p16^INK4a ELISA-based screening test, if further developed and validated, could have the advantages of being more objective, faster, less reliant on technical expertise, and potentially more affordable.

Based on the a posteriori analyses, our estimates of sensitivity and specificity for cytology screening with reflex hc2 testing of ASC-US and for hc2-based screening of cervical samples are in keeping with our previous findings in a similar population of young women (26). Our estimates for performance of cytology-based screening are also in the range of those reported by others (28-34). For hc2 screening of cervical samples, our sensitivity estimate of 100% is similar, but our specificity estimate of 69% is lower than that generally seen in studies from other populations in North America and western Europe (28-34). This is primarily because all the other studies were conducted among older and less sexually active women. Several studies also excluded women with a recent history of abnormal cytology results, thus resulting in study populations of lower risk profiles and with lower prevalences of HPV infection than ours (29, 30, 32, 34). These factors contribute to the higher specificity of hc2 observed in these studies. Nonetheless, our overall finding that HPV screening is more sensitive than cytology-based screening is overwhelmingly confirmed in the literature (28-30, 32-38). Further evidence for this comes from recent randomized trials of HPV and cytology-based screening strategies (33, 39-42), which have consistently shown that HPV-based screening allows for greater (33, 39, 40, 42) and earlier (41) detection of high-grade cervical lesions.

As the comparisons with other screening studies show, our findings might not be generalizable to all populations of women. A recent meta-analysis of cervical screening studies confirmed formally that the specificity of HPV screening is significantly heterogeneous across populations, being lower in younger women (5). For cytology screening, both sensitivity and specificity are influenced by age (5). Given the young average age of our study population, our findings regarding the performance of HPV, cytology, and p16^INK4a ELISA tests in screening might not be generalizable to older age groups of women. Our findings regarding the performance in screening of cytology and hc2-based strategies were in keeping with the expected results for a population of young, sexually active women based on findings from other North American and European studies (5). Another limitation of our study is that histologic diagnoses used to define disease outcome were not based on expert panel review as has been done in some other studies (32, 43). We addressed the issue of poor interobserver reproducibility of histologic findings by using a consensus diagnosis of two pathologists in cases of ambiguous histologic results. Also, we chose to use ≥CIN3 as our primary outcome rather than ≥CIN2: compared to CIN2, CIN3 is a more reproducible diagnosis (44, 45) and is also less likely to regress (46, 47). The implications of our study findings are also limited by the prototypic nature of the p16^INK4a ELISA used. Results from using an enhanced version of the p16^INK4a ELISA suggested that the test's performance could improve as it undergoes further development. Additional studies are needed to show its performance in different populations of women. Finally, our study protocol did not take into account possible effects of order of sample collection. Three separate samples were required because, at the time of this study, neither the liquid-based Pap test medium used in these clinics nor the collection medium for p16^INK4a ELISA was validated for hc2 testing. The performance of Pap and HPV tests were in agreement with other screening studies (5). Previous studies have suggested that Pap and HPV tests are influenced by the cervical sampling device: for improved sample sufficiency and disease detection collection of both endocervical and ectocervical cell samples is recommended and a combination of cervical cytobrush and spatula is considered the optimal sampling device (48). We collected both ectocervical and endocervical samples using either a combination of a spatula and cytobrush or a cervical broom device. The literature provides less consistent information on the influence of sampling order on performance of Pap and HPV tests. Previous work by our group (21) indicated that the p16^INK4a ELISA test might perform better when the cervical cell sample for this test is collected first rather than following other cervical samples.

One of the strengths of our study protocol was the correction of verification bias by histologic confirmation of disease among a subset of women with normal screening test results. This aspect of the protocol allowed for the detection of a CIN3 lesion that would otherwise have been missed. Yet, this single missed case also highlights the methodologic challenges in screening studies as well as the limitations of our current screening and diagnostic methods. When evaluating a screening test, it is important to ascertain the disease status of women with negative screening results so that false-negative cases can be identified: yet, due to ethical and practical considerations, only a fraction of screen-negative women were referred to colposcopy and biopsy and used to represent the larger baseline screen-negative population. Our study showed how the detection of a single case of CIN3 in this small group could result in relatively low and imprecise corrected estimates of sensitivity and a disease prevalence estimate that was quite possibly inflated by as much as 75% (from 2.2% to 3.9%). To explore these effects, and to account for the fact that women with negative screening test(s) are not routinely referred for diagnostic colposcopies, we conducted a posteriori analyses that treated women with all negative screening test results as disease-free. Our qualitative findings regarding the relative performance of different screening strategies were similar for the original and a posteriori analyses. In our previous study of a similar population, no cases of ≥CIN3 were detected in the negative control group (26). Most other cervical cancer screening studies that included histologic verification of disease among women with normal screening test results have reported no high-grade lesions among this group (28-31, 35). In two studies, cases of ≥CIN2 that were detected among the negative control group were either not confirmed by the expert panel of pathologists (32) or not confirmed in the LEEP tissue specimen (33). These findings collectively suggest that the extrapolated prevalence of disease in our population was likely overestimated in the original analysis. This missed CIN3 case also highlights the challenge of using an imperfect surrogate endpoint for cervical cancer, which we are compelled to do for ethical reasons. CIN3 lesions in young women tend to be small as was seen in this study. It is unknown whether such lesions are less likely than screen-detected CIN3 to progress to cancer.

Moreover, this screen-failed case of CIN3 is a reminder that, despite our best attempts at disease detection, including the use of trained nurse practitioners for specimen collection, endocervical sampling devices, and referral to colposcopy and biopsy based on any cytological abnormality, a positive hc2 test or a positive p16^INK4a ELISA test, some clinically significant lesions will be missed. This is more likely to occur with glandular and endocervical lesions, with older women whose cervical transformation zone tends to be smaller and shifted upwards into the endocervical canal (49), and with multiparous women whose cervices may have undergone mechanical damage subsequent to vaginal delivery. Indeed, such problems with sampling might partially explain why screening programs have been more successful in reducing the incidence of squamous cell cervical carcinoma than adenocarcinoma (50). In our study, the screen-failed case of CIN3 was detected in the LEEP specimen of a 28-year-old multiparous woman. Although it is not possible to know the cancerous potential of this CIN3 lesion, immunohistochemical evaluation of lesional tissue showed strong, diffuse staining for p16^INK4a (data not shown). This increases the likelihood that the lesion was precancerous and not a misclassified benign lesion.

Our screening study of young, sexually active women indicated a potential role for a new prototypic p16^INK4a ELISA test in cervical cancer screening. This test was similar to cytology-based screening in detecting high-grade cervical lesions. An enhanced version of the ELISA showed improved sensitivity in screening and a combined screening approach including the enhanced ELISA with hc2 testing showed potential for high sensitivity and specificity. p16^INK4a ELISA-based strategies, by themselves or in combination with HPV DNA testing, would have the advantages associated with molecular rather than morphologic evaluation of disease risk, including less subjectivity and greater reproducibility of test results (44, 51), a potentially better indication of molecular changes associated with carcinogenesis (7, 8), and fewer requirements of infrastructure and trained technical personnel.

R. Ridder and M. Herkert are employed by mtm laboratories and have intellectual property ownership interests in parts of the technology that is subject of the evaluation presented in this article. R. Ridder has a financial interest in mtm laboratories.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

We thank Julie Coard, Dr. Judy Jacobson, Preetma Kooner, Meagan Lehr, Laura Merrell, and Kim Tomlinson for clinical work on this project and Drs. Stephen Hawes, Tom Koepsell, Shalini Kulasingam, and Stephen Schwartz for valuable input and insights.