p16 and MIB1 improve the sensitivity and specificity of the diagnosis of high grade squamous intraepithelial lesions: Methodological issues in a report of 447 biopsies with consensus

7 (2007) S233–S240www.elsevier.com/locate/ygyno

Gynecologic Oncology 10

p16 and MIB1 improve the sensitivity and specificity of the diagnosis of highgrade squamous intraepithelial lesions: Methodological issues in a report of

447 biopsies with consensus diagnosis and HPV HCII testing☆

Dirk Van Niekerk a,b,d, Martial Guillaud a,d, Jasenka Matisic a,b,d, John L. Benedet a,d,J. Adrian Freeberg d,e,f, Michele Follen c,d,e,f,⁎, Calum MacAulay a,d

a Department of Cancer Imaging, British Columbia Cancer Agency, Vancouver, British Columbia, Canadab Department of Pathology, British Columbia Cancer Agency and Vancouver General Hospital, Vancouver, British Columbia, Canada

c Department of Bioengineering, Rice University, Houston, TX 77005, USAd Department of Gynecology, Obstetrics and Reproductive Sciences, The University of Texas Health Science Center at Houston, Houston, TX, USA

e Department of Gynecology, The University of Texas M. D. Anderson Cancer Center, Houston, TX, USAf Center for Biomedical Engineering, The University of Texas M. D. Anderson Cancer Center, Houston, TX, USA

Received 6 July 2007Available online 7 September 2007

Abstract

Objective. Many investigators are studying the additional value of biomarkers to improve histopathologic agreement, but few are using the samemethodologies. Our objectives in this analysis to differentiate High-grade Squamous Intraepithelial lesions (HGSIL) from Low Grade SquamousIntraepithelial Lesions (LGSIL), atypia, and normal were: (1) to examine the rate of Human Papilloma Virus High-Risk positivity (HPVHR+), (2) tocompare and grade the basal, parabasal, intermediate, and superficial layer staining of each marker, (3) to determine the optimal qualitative thresholdfor markers, (4) to compare p16 and MIB1 agreement, and (5) to examine the sensitivities and specificities using each markers alone and together.

Methods. A sample of biopsies from 208 patients were chosen from a total of 1850 patients and 3735 biopsies obtained during the course ofongoing optical trials. At least two independent blinded reviews were performed for each biopsy. A third review was performed if there was adisagreement between the two reviews. Both endocervical and ectocervical samples were stained for p16 and MIB1. A grading system that isdelineated in the text ranged from 0 to 3 for both markers and each biopsy was scored by each cell layer. Frequencies, sensitivities, andspecificities were calculated using Statistica. An ANOVA was used to compare p16 and MIB1 staining in the epithelial layers. Finally thesensitivity and specificity of each marker alone and together were examined.

Results. 453 specimens from 208 patients whose final diagnoseswere normal (n=244), low-grade (LG) (n=59), and high-grade (HG) (n=144) wereselected for analysis. 447 of 453 specimens were available for staining. Most LG and HG lesions were HPV HR positive. Endocervical samples stainedpositive less often than ectocervix and often results were discordant from ectocervical results. The analysis by layers showed pronounced increases instaining of both p16 andMIB1 as lesions progressed from normal to LG to HG. The cutoff or threshold for p16 was 0 versus 1–3 while for MIB1 it was0–1 versus 2–3. Using the intermediate epithelial layermeasurement of both p16 andMIB1 inHPVHigh-Risk Positive separated the normal tissue fromLGSIL, normal fromHGSIL, and LGSIL from HGSIL by a statistically significant margin (pb0.05). Each marker had sensitivities and specificities forthe diagnosis of HGSIL versus LGSIL and normal of∼85–90% and this improved by 5% for both sensitivity and specificity when used together (p16sensitivity 90%, specificity 85%; MIB1 sensitivity 89%, specificity 87%; together sensitivity 94%, specificity 90%).

Conclusion. Several important methodological issues have been studied. Overall, p16 and MIB1 are promising markers to help pathologistsdifferentiateHG lesions from all else. The staining of the endocervix and the ectocervix do not always agree, and the ectocervixmore often stains positivewith the presence of HGSIL. Each marker is helpful and both are helpful together. In conclusion, both markers are useful for the confirmation of HGlesions.© 2007 Published by Elsevier Inc.

☆ Contract grant sponsor: National Cancer Institute, Program Project Grant 2POlCA82710-09.⁎ Corresponding author. Anderson Cancer Center, 1515 Holcombe Blvd., Box 193, Houston, TX 77030, USA. Fax: +1 713 792 7586.E-mail address: [email protected] (M. Follen).

0090-8258/$ - see front matter © 2007 Published by Elsevier Inc.doi:10.1016/j.ygyno.2007.07.064

S234 D. Van Niekerk et al. / Gynecologic Oncology 107 (2007) S233–S240

Introduction

Several biomarkers are being studied to see if their readingincreases agreement among pathologists. Readings of cervicalintraepithelial neoplasia or squamous intraepithelial lesionshave been the subject of many investigations. Kappa statisticsfor the inter- and intra-pathologist agreement range from 0 to0.80, that is, chance to excellent agreement on the Landis scale[1,2]. Recent data from the ALTS study showed a kappa of 0.41,representing only moderate agreement on the Landis scale [2,3],suggesting that biomarkers could play a useful role.

These biomarker studies are critically important for patientcare. Since patients with lesions less than HGSIL are followedconservatively, while those with HGSIL are treated, it isimportant to differentiate them. Initially it was hoped that thepresence of high risk HPV would help distinguish lesions, butwe now know that the vast majority of HG lesions are HPVHigh-Risk Positive and those that are negative are probablyfalsely negative due to sampling or measurement error.

P16 is a cyclin-dependent kinase inhibitor involved in controlof the cell cycle that has been demonstrated to correlate positivelywith HG lesions and negatively with normal cervical biopsies.Additionally, recent studies show that patients who are p16positive with normal or low-grade lesions are likely to progress tohigh-grade lesions [4,5]. P16 is thought to be a surrogate markerof the integration of HPV. Table 1 summarizes the currentliterature of p16 and MIB1 staining in the cervix. Sensitivitiesobtained in these studies are typically 70+% [4,6–8].

One of the hallmarks of neoplasia is uncontrolled prolifer-ation.MIB1 is amarker of proliferation that has been extensivelyvalidated [9–12]. MIB1 has been studied both qualitatively andquantitatively in cancers from many organ sites including thecervix [12]. Table 1 also includes a study by Kruse [10] in which121 MIB1-stained samples were algorithmically classified;

Table 1Summary of recent studies using p16 and MIB 1 for cervical diagnosis

Biomarker Reference No. of specimens Classifications

p16 Wang [4] 292 CondylomaLGSILHGSILCancer

p16 Sano [6] 139 CondylomaLGSILHGSILCancer

p16 Zhang [8] 329 NormalCIN1CIN2CIN3Cancer

p16 Wang [7] 38 NormalCIN (all grades)Squamous cell carcinoma

MIB1 Kruse [1] 186 CIN1–3Training set of 65Specimens applied to121 specimens

HGSIL was distinguished from LGSIL and atypia with asensitivity of 93%.

Few studies have evaluated the use of p16 and MIB1 stainingtogether. Table 2 summarizes three recent efforts for threedifferent organ sites: the cervix, the anal cavity, and the aero-digestive tract [13–15]. The study of these markers is laborintensive, and few studies use similar measures of these markers.There is also wide variation in the thresholds of positivity used inthese studies. Research to date, in all tumor sites, has relied onrelatively small sample sizes for the amount of variability noted instaining.

Methodology is increasingly shown to be of critical im-portance in studies. As markers are developed, new gold stan-dards may be defined that improve diagnosis. For this study, thehistopathologic assessment was the gold standard. In this report,we examined methodological issues in the staining and use ofp16 and MIB1, both alone and together, to discriminate HGSILfrom all lesser diagnoses.

Our objectives in this analysis to differentiate High-gradeSquamous Intraepithelial lesions (HGSIL) from Low GradeSquamous Intraepithelial Lesions (LGSIL), atypia, and normalwere: (1) to examine the rate of Human Papilloma Virus High-Risk positivity (HPV HR+), (2) to compare and grade the basal,parabasal, intermediate, and superficial layer staining of eachmarker, (3) to determine the optimal qualitative threshold formarkers, (4) to compare p16 and MIB1 agreement, and (5) toexamine the sensitivities and specificities using each markersalone and together.

Material and methods

Sample selection

Samples were chosen from among 1850 patients and 3735 biopsies collectedduring the course of a multi-center optical detection study for patients between

Grading Pattern Trend

Sporadic/diffuse CIN2 sporadic 97%Not explained CIN2 diffuse 87%but qualitative CIN3 sporadic 100%

CIN3 diffuse 100%Qualitative Focal/diffuse HPV 6/11 32/34 94%Weak/strong HPV 16 60/60 100%

HPVother types 28/28 100%HPV negative 12/17

Qualitative Focal/diffuse 15/21 normal (34%)Weak/ 81/157 CIN1 (52%)Moderate/ 49/70 CIN2+ (70%)Strong 52/65 CIN3+ (80%)

0/16 cancer (0%)Qualitative Not specified Normal 1/8 1350–3+ CIN15/19 79%N20% cells stained + SQ CELL CA 9/11 82%Quantitative Not specified CIN1 versus CIN2/3Computer Sensitivity=93%AssistedImage collection

Table 2Selected methodologies from studies grading both p16 and MIB1

Study

Keating [13] Walts [14] Wayne [15]

Organ site Cervix Anal canal Aerodigestive tractNo. of

specimens104 104 66

p16 dataHistopathologic

classificationsMetasplasia, atypia,LGSIL, HGSIL

Negative, condylomaAIN grades 1–3

Hyperplasia, mild dysplasia, moderate dysplasia, severe dysplasia

Grading Weak/strong N10% positive Mean staining proportion and heightPattern Focal/diffuse Spotty/band NoneTrend 9/24 LGSIL (38%) Band-like pattern in

over 50% AIN grades 2–3Decreased with dysplasia

26/37 HGSIL (70%)

MIB1 dataHistopathologic

classificationsNormal, LGSIL, HGSIL Negative, condyloma

AIN grades 1–3Hyperplasia, mild dysplasia, moderate dysplasia, severe dysplasia

Grading Upper 2/3 of epitheliumstained to determine +/−

N50% positive % cases staining positiveMean staining height (continuous)Mean staining proportion

Pattern Not specified Not specified Not specifiedTrend 15/21 LGSIL (71%) N50% staining in majority

of cases of AIN 2–390–100% stained +

18/19 HGSIL (95%) Increased staining height as lesions progress to dysplasia variablemean staining proportion2/29 normal (7%)

S235D. Van Niekerk et al. / Gynecologic Oncology 107 (2007) S233–S240

18 and 85 of age. After reviewing the literature, we conducted this as a pilotstudy. One of the objectives of a pilot study is to observe variability so that adelta, alpha, beta, and power and be calculated for future studies. Seeking adistribution of lesion types and a sample size of approximately 400 biopsies, wechose 208 patients for whom 453 biopsies were available. Each biopsy wastreated as an independent sample. These samples were chosen from BritishColumbia Cancer Agency, one of the four institutions involved in the study.

Patient samples

Non-pregnant women 18 years and older were enrolled in the study and aninformed consent was obtained from each between 1999 and 2006. TheInstitutional Review Boards at all four institutions approved the protocols. In thetrials, biopsies were taken from one to two colposcopically abnormal areas andfrom two colposcopically normal areas. Each patient had specimens collectedfor routine Papanicoloau testing on a slide, Hybrid Capture II (Gaithersburg,Maryland), and cytologic liquid based sample that was used for quantitativecytologic measurements and an endocervical curettage [16].

All biopsy specimenswere fixed in buffered formalin and embedded in paraffinblocks. Three adjacent sections were cut at 4 μm and stained with hematoxylin andeosin (H & E). These sections were used for clinical histopathologicalinterpretation. Two additional adjacent sections were stained for p16 and MIB1.

Criterion standard

For the purpose of this analysis, the consensus histopathologic clinicaldiagnosis was used as the criterion standard. A study of inter-pathologist, intra-pathologist, and inter-institutional agreement has been published [1].

Consensus diagnoses

All histopathologic specimens were read at the British Columbia CancerCentre, using both the WHO and the Bethesda criteria. The first pathologyreview was performed by one of the gynecological pathologists on clinical duty(hereafter called the “pool”). There were seven pathologists in the pool atBCCC, and the pool included the study pathologists. A second blinded reviewwas performed by one of our two study pathologists. If the first and second

readings agreed exactly in theWHO system, a third reviewwas not performed. Adiscrepancy of two grades of the WHO criteria mandated a third blinded readingby both study pathologists. The kappas for the readings of the pathology arereported in Malpica et al. [1]. In summary, for all diagnoses ranging from normalto SIL to cancer, the kappa readings ranged from 0.4 to 0.80; those in the HG andcancer range were in the high 0.70–0.85 value kappa range. The Bethesdaclassification for cervical neoplasia is reported in this analysis for simplicity.

HPV testing

Endocervical samples were collected and tested for HPV using the HybridCapture II method (Digene Corporation ®, Gaithersberg, Maryland). A nucleicacid hybridization microplate assay with signal amplification for thechemiluminescent detection of HPV DNA, the Hybrid Capture II test identifiesboth low-risk HPV types (6, 11, 42, 43, 44) and high-risk HPV types (16, 18, 31,33, 35, 39, 45, 51, 52, 56, 58, 59, 68). All samples were processed at LabCorp®

using funds from public grants.

p16 staining methodology

The p16 immunostainingwas performed on 4μm thick sections from paraffinblocks using a commercially available kit (DakoCytomation ®,Milan, Italy). Thesections were dried overnight at 37°. This kit includes the monoclonal antibodyE6H4 which identifies an epitope of p16 (between aa 134 and aa 156). The slideswere heated and immersed in the kit retrieval solution in a calibrated water bath at96° for 40 min. Nuclear and cytoplasmic activity if present was noted andconsidered positive.

Assessment for p16

Staining intensity was graded by counting cells within the epithelial layerspresent, using both nuclear and cytoplasmic staining. Fig. 1 shows examples ofstudy biopsies stained with H & E, p16, and MIB1. The pathologist scoring theslide was blinded to the diagnosis. The absence of stained cells was considerednegative. If less than 5% of cells were positively stained, it was called negativeor zero (0); if between 5% and 25% of cells were positively stained, it was called1+; if between 25% and 75% of cells were positively stained, it was called 2+,

S236 D. Van Niekerk et al. / Gynecologic Oncology 107 (2007) S233–S240

and finally if more than 75% of cells were positively stained, it was called 3+.This grading system was used to grade each epithelial layer (basal, parabasal,intermediate, and superficial layers). A sensitivity analysis of this qualitativescoring system was performed and will be the subject of a future work.

MIB1 staining methodology

Paraffin sections of 4 μmwere mounted on slides and dried overnight at 37°.Sections were deparaffinized in xylene and rehydrated in alcohol. Followingrehydration, the endogenous perioxidase activity was blocked by 3% H2O2 inphosphate- buffered saline. The sections were immersed in sodium citrate bufferand heated at 1000W for 2 min and at 160 W for 15 min in a microwave. Beforeimmunostaining, the slides were soaked in phosphate buffered saline. Sectionswere incubatedwith biotinylated swine anti-rabbit antibody (DAKO®, Glostrup,Denmark) at 1:100 for 30 min. Visualization of the complex was realized withdiaminobenxidine/H2O2 for 10 min at room temperature. Two washes inphosphate buffered saline were performed prior to counterstaining with Mayer'sHematoxylin. The sections were dehydrated using graded ethanol and mountedwith a mixture of distyrene, plasticiser, and xylene (DPX) (Nustain ®,Nottingham, UK).

Fig. 1. Histologic biopsy specimens showing different diagnosis from the consensusstaining, and (c) MIB1 staining histopathologic consensus diagnosis.

MIB1 classification

Fig. 1 shows examples of study biopsies stained with H& E, p16, andMIB1.The pathologist scoring the slide was blinded to the diagnosis. When there wasno parabasal layer staining, the specimen was classified as 0. Since the parabasallayer usually stains, these specimens were classified as 1+. Staining in theintermediate layer was classified as a 2+, and staining in the superficial or basallayers was denoted as 3+.

Statistical analyses

Data from grading each layer in each slide were entered into the researchdatabase. Primary data were used to enumerate specimens for descriptivestatistics. Statistica (Statsoft ®, Tulsa, OK) was used to calculate frequencies,sensitivities, and specificities for the number of specimens, the number of HPVHR + and HR−, and the amount of tissue stained in the various layers in theepithelium. High-grade lesions were the gold standard used for positive as theendpoint in the analysis. An ANOVAwas used to compare staining intensities ofdifferent levels of epithelium. Statistica was used to graph the histograms andlevels of immunostaining and calculate p values.

diagnosis set of specimens. (a) Hematoxylin and eosin stained samples, (b) p16

Table 4Biopsy specimens stratified by HPV High-Risk +/−, histology, and biomarker(p16 and MIB1) positivity for ectocervical samples

Biopsyhistology

HPV p16ectocervix

MIB1ectocervix

HC-II Neg Pos Neg Pos

Normal HR− 62 12 3 71Normal HR+ 106 38 2 141Total 218/222 217/222LGSIL HR− 10 4 0 13LGSIL HR+ 14 28 0 42Total 56/59 55/59HGSIL HR− 0 2 0 2HGSIL HR+ 4 122 0 125Total 128/144 127/144

402⁎ 399⁎

⁎Not all patients had HPV typing results and atypias excluded.

S237D. Van Niekerk et al. / Gynecologic Oncology 107 (2007) S233–S240

Results

Table 3 provides an overview of all the samples, theirdiagnoses, and those available for marker staining for eachanalysis in these results. Consistent with all studies using biopsies,sample sizes differ by category because there are: missingsamples, insufficient tissue on recutting, samples missing anothertest (HPV), or samples missing a cell layer (e.g. the intermediatelayer). 447 of 453 biopsies from 208 patients were available forstaining, 257 were available for the ectocervical and endocervicalstaining, and 408were available for the final analysis of sensitivityand specificity.

In Table 4, showing HPV High-Risk positivity, the specimenswere: (1) classified by diagnoses normal, LGSIL, and HGSIL, (2)stratified by HPV High-Risk + and HPV High-Risk − status, and(3) stratified by p16 and MIB 1 positivity. The correlationbetween SIL andHigh-Risk positivity suggests that the specimensused comprise a representative sample.

Table 5 shows the poor agreement of endocervical andectocervical specimens. For the p16 staining, there was anagreement in 38% of the specimens, while for the MIB1 staining,there was an agreement for only 12% of the specimens. Theendocervix samples for both markers were rarely positive. Thissuggests that neither marker is often present in the endocervixperhaps limiting the use of an endocervical curettage assessment.

A labor intensive review was carried out to look at stainingamong the basal, parabasal, intermediate, and superficial layersfor both p16 and MIB1. Each sample was quantified, blinded to

Table 3Description of patient biopsy samples

Patients 208

Total biopsies 453Consensus histologic diagnosis

Negative 222Atypia 22LGSIL 59HGSIL 144Cancer 0Insufficient tissue on recut 6Total 453

Biopsy specimens with HCII HPV testingp16 402Missing 51Total 453

MIB1 399Missing 54Total 453

Biopsy Specimens for p16 and MIB1 stainingEctocervix and endocervix 257Ectocervix only 154Endocervix only 27Unsatisfactory 9Total 447

Biopsies available for ectocervical stainingEctocervix with intermediate layer 408Ectocervix missing intermediate layer 3Total 411

Total biopsies available for final sensitivity and specificity analysis 408

diagnosis, and recorded. These data can best be seen graphicallyin Fig. 2 showing p16 and Fig. 3 showing MIB1. Each graphshows increases in the respective marker as the histologyworsens; both also confirm the reliability of separating thesediagnoses using the layers in the samples. Analyses of p16 andMIB1 stains of the layers differed by HPV HR positivity. ForHPV negative samples with p16 staining, only the basal layerwas significant for separating diagnoses. For the HPV HRpositive samples with p16 staining, the basal, parabasal, andintermediate layers all showed statistically significant separa-tion. The HPV negative samples subjected to MIB1 stainingshowed significant separation of the diagnoses at the superficiallayer only. For the HPV HR positive specimens with MIB1staining, the basal, intermediate, and superficial layers signif-icantly separated normal from LG from HG. Thus, the basal andintermediate layers appear most useful in the analysis whenexamining the overall data for both the p16 and MIB1 staining.

In Table 6, we examine the agreement of p16 and MIB1positivity. After careful study, a sensitivity analysis using twodifferent thresholds for both p16 and MIB 1 positivity wasperformed. Another manuscript will further delineate the sensitiv-ity analysis performed for deciding the cutoff or threshold for eachmarker. The highest values for sensitivity and specificity in theanalysis were noted for p16 threshold “0 staining as negative”versus “1–3+ staining as positive”. In contrast, for MIB1, thehighest sensitivity and specificitywere achieved for “0–1+ stainingas negative” and “2–3+ staining as positive”. Using thesethresholds, there is 90% agreement of specimens either negativeor positive for those lesions for the diagnosis of HGSIL.

Table 5Agreement of endocervical and ectocervical biopsy specimens

Agreement for p16 Agreement for MIB1

Agree positive 21 21/257 9% 27 27/257 11%Agree negative 74 74/257 29% 2 2/257 1%Ecto + endo − 152 152/257 59% 228 228/257 89%Endo + ecto − 10 10/257 4% 0 0 0%Total 257 257

Fig. 2. Graph of the intensity and 95% confidence intervals of p16 measurements in both HPV High-Risk Negative and HPV High-Risk Positive biopsy samples bytissue pathology type (Normal, LGSIL, HGSIL).

S238 D. Van Niekerk et al. / Gynecologic Oncology 107 (2007) S233–S240

Finally, we analyzed how using both markers individually andtogether would impact the sensitivity and specificity for HGdiagnosis. In Table 7, we show the result of the sensitivity andspecificity of p16 in the intermediate layer, using each markeralone and together. The sensitivity for either marker wasapproximately 90%, and using both p16 and MIB1 togetherincreased sensitivity by 5%. Similar gains are seen in specificitywhen both markers are used. Thus, if cost effective, both markersused together yield the highest sensitivities and specificities.

Fig. 3. Graph of the intensity and 95% confidence intervals of MIB1 measurements intissue pathology type (Normal, LGSIL, HGSIL).

Discussion

Our objectives in this analysis to differentiate HGSIL fromLGSIL, atypia, and normal were: (1) to examine the rate of HPVHR positivity, (2) to compare and grade the basal, parabasal,intermediate, and superficial layer staining of each marker, (3) todetermine the optimal qualitative threshold for markers, (4) tocompare p16 and MIB1 agreement, and (5) to examine thesensitivities and specificities using each marker alone and together.

both HPV High-Risk Negative and HPV High-Risk Positive biopsy samples by

Table 7Biomarker sensitivity and specificity in the intermediate layer for HGSIL using408 specimens from Table 6

Sensitivity Specificity

p16 + 0.90 0.85MIB1 + 0.89 0.87P16 + and MIB1 + 0.94 0.90

S239D. Van Niekerk et al. / Gynecologic Oncology 107 (2007) S233–S240

We demonstrated that most HG specimens were HPV HR+,p16 +, and MIB 1+. We also demonstrated that staining of theendocervix and ectocervix was most often discordant and that theectocervix stained positively much more often than theendocervix. Since these were endocervical curettage specimensfrom patients with squamous pathology, we acknowledge that thestudy of adenomatous lesions could yield different results.

We performed a layer by layer analysis to show that the basaland intermediate layers most often represented positivity on aglobal assessment. We performed a sensitivity analysis todetermine the threshold or cutoff for positive versus negativewithin each marker (data not shown). Using those thresholds, wegraded each sample. We showed that using both markersincreased the sensitivity and specificity of the diagnosis of HGby 5% respectively to a sensitivity of 94% and specificity of 90%.

To our knowledge, this is the largest study of both p16 andMIB1 in ectocervical and endocervical samples stratified byHPV High-Risk positivity. The simple analysis of specimensoutlined here shows that most HG specimens and many LGspecimens were HPV HR positive. As we know, HPV HRpositivity does not help in distinguishing HG lesions, yet HPVHR positivity may help in confirming a true HG lesion.

Clearly both p16 and MIB1 are often associated with HGhistology. This study confirms the work of others. From theanalysis of each marker by layer (basal, parabasal, intermediate,and superficial), we see definite increases as the lesions progressfrom normal to LG to HG. Additionally, since the data arestratified by HPV HR positivity, we can see that the presence ofHR HPV increases the detection of p16 and MIB1.

One could speculate that HG lesions which are HPV HRpositive can further confirm a “true”HGdiagnosis, while suspectedHG lesions which are HPV HR negative may be falsely HPVnegative, insufficient for diagnosis, or falsely positive HG (i.e.bHG lesions). In any case, the differences in staining suggest thatthe specimens differ. The HPV HR negative sample size is alsosmaller than that of the positives, which increases the confidenceintervals.

Another statistical issue for all studies of biomarkers is the useof these samples as independent samples when some samplescame from the same patient. We have looked for confounding(data not shown) and did not found it a problem; however, werealize the theoretical issues in the design. We plan a future studyusing only one biopsy from each patient.

Further examination of the associated figures makes it clear thatboth the intermediate and parabasal layers are significantly differentthan the other layers. This led us to examine how the sensitivity andspecificity could improve if the information in layers is used torefine the diagnosis. Not surprisingly, the intermediate layer

Table 6Agreement of biopsy specimens for MIB1 (0–1 versus 2–3) and p16 (0 versus1–3) using intermediate layer positive

Agree positive 227 227/408 56%Agree negative 138 138/408 34%P16 − and MIB1 + 18 18/408 5%P16 + and MIB1 − 25 25/408 6%Total 408

positivity provided the greatest increase in sensitivity or specificityin diagnosing HG lesions, irrespective of HPV status.

Another issue is whether both markers are needed. One mightargue intuitively yes, since both markers give different informa-tion. In our simple analysis of sensitivities and specificities whenboth markers had been tested, each resulted in sensitivities andspecificities of ~85–90% and using them together improved thediagnosis by 5% for both sensitivity and specificity. This suggeststhat the parabasal or intermediate layer is useful and may behelpful with fragmented or small specimens.

One of the important reasons to use p16 and MIB1 is todecrease the inter-observer and intra-observer readings inpathology. We believe these observations support that hypoth-esis and in fact, could improve the diagnosis of HG lesions. Oneof the goals of our research team is to introduce quantitativepathology for both routine samples (stained with feulgen) andfor immunostained markers such as these.

In a subsequent publication, we hope to show additionalvalue in quantification of these markers as well as informationfrom their quantitative analysis of nuclear texture andarchitecture [17–19]. Automating the reading of pathologycould improve the accuracy and agreement of diagnoses in thedeveloped world and perhaps decrease infrastructure require-ments in the developing world.

Conflict of interest statementWe declare that we have no conflict of interest.

Acknowledgments

Special thanks go to all the patients who participate in ourtrials, Trey Kell, Deanna Haskins, Anita Carraro, Maria TheresaArbalaez, Rosa Morales, Nan Earle, and Brian Crain.

References

[1] Malpica A, Matisic JP, Niekirk DV, Crum CP, Staerkel GA, Yamal JM,et al. Kappa statistics to measure interrater and intrarater agreement for1790 cervical biopsy specimens among twelve pathologists: qualitativehistopathologic analysis and methodologic issues. Gynecol Oncol2005;99:S38–52.

[2] Landis JR, Koch GG. The measurement of observer agreement forcategorical data. Biometrics 1977;33:159–74.

[3] Stoler MH, Schiffman M. Interobserver reproducibility of cervicalcytologic and histologic interpretations: realistic estimates from theASCUS-LSIL Triage Study. Jama 2001;285:1500–5.

[4] Wang SS, TrunkM, SchiffmanM, Herrero R, ShermanME, Burk RD, et al.Validation of p16INK4a as a marker of oncogenic human papillomavirusinfection in cervical biopsies from a population-based cohort in Costa Rica.Cancer Epidemiol Biomark Prev 2004;13:1355–60.

S240 D. Van Niekerk et al. / Gynecologic Oncology 107 (2007) S233–S240

[5] Negri G, Vittadello F, Romano F,Kasal A, Rivasi F, Girlando S,MianC, et al.p16INK4a expression and progression risk of low-grade intraepithelialneoplasia of the cervix uteri. Virchows Arch 2004;445:616–20.

[6] Sano T, Oyama T, Kashiwabara K, Fukuda T, Nakajima T. Expressionstatus of p16 protein is associated with human papillomavirus oncogenicpotential in cervical and genital lesions. Am J Pathol 1998;153:1741–8.

[7] Wang JL, Zheng BY, Li XD, Nokelainen K, Angstrom T, LindstromMS, et al. p16INK4A and p14ARF expression pattern by immunohis-tochemistry in human papillomavirus-related cervical neoplasia. ModPath 2005;18:629–37.

[8] Zhang Q, Kuhn L, Denny LA, De SouzaM, Taylor S,Wright Jr TC. Impactof utilizing p16INK4A immunohistochemistry on estimated performanceof three cervical cancer screening tests. Int J Cancer 2007;120:351–6.

[9] Bulten J, van der Laak JA, Gemmink JH, Pahlplatz MM, de Wilde PC,Hanselaar AG. MIB1, a promising marker for the classification of cervicalintraepithelial neoplasia. J Pathol 1996;178:268–73.

[10] Kruse AJ, Baak JP, de Bruin PC, Jiwa M, Snijders WP, Boodt PJ, et al. Ki-67 immunoquantitation in cervical intraepithelial neoplasia (CIN): asensitive marker for grading. J Pathol 2001;193:48–54.

[11] Popiolek D, Ventura K, Mittal K. Distinction of low-grade squamousintraepithelial lesions from high-grade squamous intraepithelial lesions basedon quantitative analysis of proliferative activity. Oncol Rep 2004;11:687–91.

[12] Follen M, Schottenfeld D. Surrogate endpoint biomarkers and theirmodulation in cervical chemoprevention trials. Cancer 2001;91:1758–76.

[13] Keating JT, Cviko A, Riethdorf S, Riethdorf L, Quade BJ, Sun D. Ki-67,cyclin E, and p16INK4 are complimentary surrogate biomarkers for humanpapilloma virus-related cervical neoplasia. Am J Surg Pathol 2001;25:884–91.

[14] Walts AE, Lechago J, Bose S. P16 and Ki67 immunostaining is a usefuladjunct in the assessment of biopsies for HPV-associated anal intraepithe-lial neoplasia. Am J Surg Pathol 2006;30:795–801.

[15] Wayne S, Robinson RA. Upper aerodigestive tract squamous dysplasia:correlation with p16, p53, pRb, and Ki-67 expression. Arch Pathol Lab Med2006;130:1309–14.

[16] Guillaud M, Benedet JL, Cantor SB, Staerkel G, Follen M, MacAulay C.DNA ploidy compared with human papilloma virus testing (HybridCapture II) and conventional cervical cytology as a primary screening testfor cervical high-grade lesions and cancer in 1555 patients with biopsyconfirmation. Cancer 2006;107:309–18.

[17] Guillaud M, Cox D, Malpica A, Staerkel G, Matisic J, Van NiekirkD, et al. Quantitative histopathological analysis of cervical intra-epithelial neoplasia sections: methodological issues. Cell Oncol2004;26:31–43.

[18] Guillaud, M., Lam, N., Lam, S. MacAulay, C. in International Conferenceon Applied Genomics 719th ESACP/16th ESDQP Meeting.

[19] Guillaud M, Matthews JB, Harrison A, MacAulay C, Skov K. A novelimage cytometric method for quantitation of immunohistochemicalstaining of cytoplasmic antigens. Anal Cell Pathol 1997;14:87–99.