Ou

1h

Original article

Increased EZH2 protein expression is associated with invasiveurothelial carcinoma of the bladder

Hang Wang, Ph.D.a, Roula Albadine, M.D.b, Ahmed Magheli, M.D.c,Thomas J. Guzzo, M.D., M.P.H.a, Mark W. Ball, M.D.a, Stefan Hinz, M.D.c,

Mark P. Schoenberg, M.D.a, George J. Netto, M.D.b, Mark L. Gonzalgo, M.D., Ph.D.d,*a Department of Urology, James Buchanan Brady Urological Institute, Baltimore, MD 21287, USA

b Department of Pathology, The Johns Hopkins Medical Institutions, Baltimore, MD 21287, USAc Department of Urology, Universitätsmedizin Berlin, Charité Campus Benjamin Franklin, Berlin, Germany

d Department of Urology, Stanford University School of Medicine, Stanford, CA 94305, USA

Received 8 February 2010; received in revised form 7 September 2010; accepted 10 September 2010

Abstract

Objectives: Elevated polycomb group protein Enhancer of Zest Homolog 2 (EZH2) expression has been associated with progression tomore advanced disease in a variety of malignancies. We examined EZH2 protein expression levels in bladder tissue specimens from patientswith urothelial carcinoma (UC) and investigated the relationship between EZH2 protein expression and clinical outcomes.

Materials and methods: Tissue microarrays (TMAs) were constructed using bladder tissue specimens from radical cystectomiesperformed for UC at our institution between 1994 and 2002. EZH2 expression was measured by immunohistochemistry and scoring wasbased on percentage and intensity of positive nuclear staining. A receiver operating curve (ROC) was generated to differentiate cancerousfrom benign lesions using EZH2 protein scores. Recurrence-free survival was estimated using the Kaplan-Meier approach with log-rank test.A multivariate Cox proportional hazards model was used to assess independent contributions.

Results: A total of 454 TMA specimen spots from 81 patients were evaluated. EZH2 protein levels in invasive high grade UC weresignificantly elevated compared with adjacent benign urothelium, noninvasive low grade UC, and CIS. EZH2 protein levels were alsosignificantly increased in CIS and noninvasive low grade UC compared with adjacent benign urothelium. We found no association betweenEZH2 protein expression and clinical outcomes following radical cystectomy in our cohort of patients.

Conclusion: EZH2 overexpression is a common event in UC of the bladder. Elevated EZH2 protein levels are associated with moreaggressive bladder cancer, including invasive UC. EZH2 may therefore serve as a useful biomarker for UC. © 2012 Elsevier Inc.All rights reserved.

Urologic Oncology: Seminars and Original Investigations 30 (2012) 428–433

Keywords: Urinary bladder cancer; Urothelial carcinoma; Tissue microarray; EZH2; Polycomb protein

1. Introduction

Bladder cancer is the fourth most common malignancy inmen and the eighth most common cancer in women in theUnited States. The National Comprehensive Cancer Net-work estimates that 68,810 new cases and 14,100 deathswould have occurred in the United States in 2008 [1].

verall, 70% of bladder tumors present as noninvasiverothelial carcinoma (UC), and the remainder as muscle-

* Corresponding author. Tel.: �1-650-725-5544; fax: �1-650-723-0765.

E-mail address: gonzalgo@stanford.edu (M.L. Gonzalgo).

078-1439/$ – see front matter © 2012 Elsevier Inc. All rights reserved.ttp://dx.doi.org/10.1016/j.urolonc.2010.09.005

invasive disease [2]. Up to 50% of patients initially present-ing with muscle-invasive UC will relapse with metastaticdisease. Carcinoma in situ (CIS) is a high grade lesion thatis a precursor of invasive UC. The heterogeneity of thisdisease complicates management decision-making. Preciseidentification of tumors that will recur and progress vs.tumors that will take a more indolent course remains chal-lenging [3].

EZH2 is a key member of the Polycomb Group (PcG) ofproteins. PcG proteins are highly conserved from insects tomammals and play crucial roles in the maintenance ofembryonic and adult stem cells and in cell cycle control via

maintenance of long-term repressive chromatin states [4,5].

Ra

oepwtofit

mhahbehdiltuwei

2

2

dtTtaApwyBaivbw

S

R

MF

I

T

C

P

LL

P

P

429H. Wang et al. / Urologic Oncology: Seminars and Original Investigations 30 (2012) 428–433

Several PcG proteins including EZH2 have been implicatedin oncogenic activities [6]. It has been shown that ectopicoverexpression of EZH2 not only stimulates cell prolifera-tion, but also promotes anchorage-independent growth andcell invasion in vitro [4,7–9]. In contrast, small interfering

NAs (siRNAs) of EZH2 inhibit cell proliferation and inducepoptosis in prostate, breast, and colon cancer cells [10].

EZH2 is frequently overexpressed or amplified in vari-us types of human cancer [8]. Moreover, elevated EZH2xpression has been associated with increased tumor cellroliferation, progression to more advanced disease, andorse prognosis in prostate cancer, breast cancer, endome-

rial cancer, and melanoma [4,11,12]. In addition to EZH2,ther PcG proteins, such as BMI-1, RING-1, and SUZ12 arerequently over-expressed in various malignancies suggest-ng that the entire PcG protein complexes are important forumor development [13].

EZH2 has shown promise as a potential moleculararker for bladder cancer. Elevated EZH2 mRNA levels

ave been associated with increasing tumor stage, grade,nd disease-recurrence [14–16]. There are few studies thatave investigated the significance of EZH2 protein levels inladder cancer. Arisan et al. reported that EZH2 proteinxpression was more frequent in more advanced stage andigher grade bladder tumors [14]. Raman et al. found noifference in EZH2 protein expression levels between non-nvasive and invasive tumors and also no difference betweenow-grade tumors and benign urothelium, although EZH2 pro-ein levels were increased in tumors compared with benignrothelium [16]. We investigated whether EZH2 protein levelsere associated with clinicopathologic parameters of UC and

valuated the predictive value of EZH2 protein levels on clin-cal outcomes of patients with UC.

. Materials and methods

.1. Tissue microarray (TMA) construction

TMAs were constructed from specimens of patients un-ergoing radical cystectomy for UC at our institution be-ween 1994 and 2002. Tissue specimens for inclusion in theMA were chosen based on availability of suitable cystec-

omy tissue samples. These were processed by members ofgenitourinary pathology team at Johns Hopkins Hospital.retrospective medical record search of the clinical and

athologic data of 81 patients was also performed. Patientsith prior intravesical therapy were excluded from the anal-sis. All data were collected under an Internal Reviewoard-approved protocol with Health Insurance Portabilitynd Accountability Act compliance. All patients providednformed consent when indicated by the Institutional Re-iew Board. At least triplicates of tumor samples, adjacentenign urothelium, and corresponding CIS (when available)

ere used in creation of the TMA.

2.2. Staging and grading

Specimens were staged according to the American JointCommittee on Cancer (AJCC) TNM Staging System forBladder Cancer [1]. Histopathologic grading was performedaccording to the 2004 WHO/ISUP version, which was mod-ified by the members of WHO and the International Societyof Urologic Pathologists (ISUP) [1].

Table 1Clinicopathologic characteristics of patients with UC undergoing radicalcystectomy

Patients (n) 81Age (year)

Median (Range) 67 (35–89)Mean � SD 65.1 � 12.2

exMale N (%) 64 (79)Female N (%) 17 (21)

aceCaucasian N (%) 71 (87.7)African-American N (%) 6 (7.4)Asian N (%) 4 (4.9)edian year of surgery 2001 (1994–2002)

ollow up (mo)Median (range) 28 (0–137)Mean � SD 38.4 � 33.8

VT N (%)None 48 (59.3)Yes 33 (40.7)

ype of urinary diversionIleal conduit 50 (61.7)Neobladder 26 (32.1)Indiana pouch 2 (2.5)Colon conduit 1 (1.2)Other 2 (2.5)

linical stage N (%)Ta 1 (1.2)Tis 9 (1.5)T1 21 (25.9)T2 34 (42)T3 15 (18.5)T4 1 (1.2)

athological stage N (%)pTa 2 (2.5)pTis 3 (3.7)pT1 21 (25.9)pT2 23 (28.4)pT3 35 (43.2)pT4 9 (11.1)

ymph node metastasis N (%) 22 (29.3)ymphatic invasion N (%)Negative 52 (64.2)Positive 29 (35.8)

ostoperative chemo N (%)Yes 73 (90.1)No 8 (9.9)

ostoperative radiation N (%)Yes 9 (11.1)

No 72 (88.9)

Tgcu

relmb

2

fUct

ibmsfcicupuwsu

3

wffbsmoorpputpr

caFTuhsv0

430 H. Wang et al. / Urologic Oncology: Seminars and Original Investigations 30 (2012) 428–433

2.3. Immunohistochemical staining of TMA

Tissue sections that were 4 �m in thickness from theMA were deparaffinized with xylene and rehydrated in araded series of ethanol. Antigen retrieval was performed initrate buffer (10 mM, pH 6.0) by heating for 25 minutessing a steamer, followed by incubation with 3% H2O2 for

10 minutes to remove endogenous peroxidase activity. Sec-tions were incubated for 1 hour at room temperature with1:500 diluted affinity-purified rabbit polyclonal antibodiesagainst human EZH2 (Zymed, San Francisco, CA) in a 1%BSA-PBST solution (1X PBS with 0.05% of Tween-20).Detection was performed by using the EnVison kit contain-ing anti-rabbit HRP-conjugated IgG secondary antibody(DAKO, Carpinteria, CA). After immunohistochemicalstaining, TMA sections were counterstained with hematox-ylin. The EZH2 antibody has been previously validated invarious malignancies, and was also confirmed in the presentstudy by immunoblot analysis using nuclear extracts ofbladder cancer cell lines treated with or without EZH2small-hairpin RNAs (shRNAs) (data not shown) [4,11,16].

2.4. Scoring of immunostaining

Scoring of EZH2 immunostaining was based on intensityand percentage of the positive nuclear staining according toprevious studies [17]. The intensity of EZH2 staining wasassessed using a 4 point scale (0, 1�, 2�, and 3�). Thepercentage of positive nuclear staining was determined byexamining all cells within each individual TMA spot. EZH2scores [�(intensity score � percentage)] with a possibleange from 0 to 3 were then calculated. EZH2 scores werevaluated by two pathologists who were blinded to patho-ogic stage and grade of the tumor. EZH2 score was deter-ined by a third pathologist when there was discordance

etween the first 2 evaluators.

.5. Statistical analysis

Nonparametric testing was employed to determine dif-erential EZH2 protein expression levels in specimens withC and adjacent benign urothelium. A receiver operating

urve (ROC) was generated by plotting the fraction of therue positives (sensitivity) vs. the false positives (1 - spec-

Fig. 1. Immunohistochemical staining of EZH2 protein in formalin-fixed, p

of figure is available online.)

ficity) using EZH2 protein scores of invasive UC andenign urothelium. Area under the ROC (AUC) was deter-ined as previously described [18]. An AUC of 1.0 repre-

ents the ideal diagnostic test, which has no false positive oralse negative results. Disease recurrence-free survivalurves were generated by Kaplan-Meier analysis and thempact of EZH2 protein levels on disease-free survival wasompared by log-rank test with the median score of EZH2sed as a cut-off. To evaluate the predictive value of EZH2rotein levels on clinical outcomes of patients with UC,nivariate and multivariate Cox proportional hazards modelsere employed. A two-tailed P value of �0.05 was considered

tatistically significant. All statistical analyses were performedsing SPSS, ver. 15.0 (SPSS Inc., Chicago, IL).

. Results

A total of 454 TMA specimen spots from 81 patientsho underwent radical cystectomy for UC were evaluated

or EZH2 expression. The TMA was comprised of 235 spotsrom invasive high-grade tumors, 149 spots of adjacentenign appearing urothelium, 49 spots from CIS, and 21pots from noninvasive low-grade tumors. Tissue speci-ens were obtained from patients with multifocal invasive

r refractory non-muscle-invasive disease. The clinicopath-logic characteristics of the 81 patients who underwentadical cystectomy for UC are shown in Table 1. Twoatients had pTa tumors, 9 patients had pT1, 3 patients hadTis, and 67 patients had �pT2 tumors. Four (5%) patientsnderwent neoadjuvant chemotherapy and 30 (37%) pa-ients received adjuvant chemotherapy. A total of 39 (48%)atients recurred at a median time of 14 months followingadical cystectomy.

Immunohistochemical staining of EZH2 protein wasoncentrated in the nuclei of UC cells. Representative im-ges of EZH2 immunohistochemical staining are shown inig. 1. A comparison of relative EZH2 protein levels fromMA specimens (mean � standard deviation) for benignrothelium, noninvasive low grade UC, CIS, and invasiveigh grade UC is shown in Fig. 2 and Table 2. EZH2 proteincores for invasive high grade UC were significantly ele-ated compared with adjacent benign urothelium (P �.0003), non-invasive low grade UC (P � 0.003), and CIS

embedded sections of bladder tissue specimens from patients with UC. (A)

araffin-Benign urothelium; (B) noninvasive papillary low grade UC; (C) CIS; (D) muscle invasive high grade UC. Magnification for A–D, �400. (Color version

are 0.3

431H. Wang et al. / Urologic Oncology: Seminars and Original Investigations 30 (2012) 428–433

(P � 0.001), respectively. In addition, EZH2 protein scoreswere significantly increased in CIS (P � 0.0002) and non-invasive low grade UC (P � 0.026) compared with adjacentbenign urothelium. There was a trend of higher EZH2 ex-pression scores in CIS compared with non-invasive lowgrade UC (P � 0.059). A ROC was generated to evaluate ifrelative EZH2 protein levels could be used to differentiatebenign urothelium from invasive UC (see Fig. 3). The AUCwas 0.89 (SE 0.033, 95% CI 0.825– 0.954, P � 0.0001)demonstrating that EZH2 can be a potentially useful diag-nostic marker for UC.

We performed Kaplan-Meier analysis to determine ifEZH2 expression levels were associated with disease recur-rence-free survival following radical cystectomy. EZH2protein expression levels were not associated with disease

3.0

2.5

2.0

1.5

1.0

.5

0.0

-.5

EZH

2 Pr

otei

n Ex

pres

sion

Sco

re

Benign Urothelium

Non-invlow gra

Fig. 2. Relative expression levels of EZH2 protein in bladder tissue specimurothelium, noninvasive low grade UC, CIS, and invasive high grade UC

Table 2Comparison of EZH2 protein scores in bladder tissue specimens ofpatients with UC

Pathology of specimens comparison of mean � SD Pvalue

Invasive high grade UC vs.benign

1.97 � 0.55 vs 0.34 � 0.39 0.0003

Invasive high grade UC vs.non invasive low grade

1.97 � 0.55 vs 0.99 � 0.54 0.003

Invasive high grade UC vs.CIS

1.97 � 0.55 vs 1.33 � 0.61 0.001

CIS vs. benign 1.33 � 0.61 vs 0.34 � 0.39 0.0002Non invasive low grade vs.

benign0.99 � 0.34 vs 0.34 � 0.39 0.026

CIS vs. non invasive low 1.33 � 0.61 vs 0.99 � 0.54 0.059

grade

recurrence-free survival in our cohort of 81 patients (P �0.893) (see Fig. 4). We also performed univariate and mul-tivariate Cox regression analyses to determine if EZH2protein levels were associated with risk of disease recur-rence (Table 3). No statistically significant association wasobserved between EZH2 protein expression and risk of

Invasive high grade UC

CIS

m patients with UC. EZH2 scores (mean � standard deviation) for benign4 � 0.39, 0.99 � 0.54, 1.33 � 0.61, and 1.97 � 0.55, respectively.

1-Specificity

Sens

itivi

ty

1.00.75.50.250.00

1.00

.75

.50

.25

0.00

Fig. 3. EZH2 is a potential diagnostic marker for UC. ROC analysis, they-axis (sensitivity) represents the true positive rate and the x-axis repre-sents the false positive rate (1-specificity). The AUC is 0.89 (SE 0.033,

asivede UC

ens fro

95% CI 0.825–0.954, P � 0.0001).

teua

4

vwHpiAmbEbi

psaluspfDsl

Ulmdbaat

mEmdosavwevsgwcwdbos

pp

432 H. Wang et al. / Urologic Oncology: Seminars and Original Investigations 30 (2012) 428–433

disease recurrence following radical cystectomy in univar-iate or multivariable analyses (P � 0.893 and 0.726, respec-ively). Lymph node metastasis, a known predictor of dis-ase recurrence, maintained predictive value in bothnivariate and multivariable regression analyses (P � 0.001nd 0.014, respectively).

. Discussion

The role of EZH2 protein levels in UC remains contro-ersial. Raman et al. demonstrated that EZH2 protein levelsere higher in UC compared with benign urothelium [16].owever, they did not observe any difference in EZH2rotein expression between superficial and invasive UC orn low-grade tumors compared with benign urothelium.risan et al. reported that EZH2 protein expression wasore frequent in more advanced stage and higher grade

ladder tumors [14]. The present study demonstrates thatZH2 protein expression levels differ significantly betweenenign urothelium, noninvasive low grade UC, CIS, andnvasive high-grade UC. Our results suggest that EZH2

Blue curve: tumors with low EZH2 scores

Green curve: tumors with high EZH2 scores

0 25 50 75 100 125

1.0

0.8

0.6

0.4

0.2

0.0

Rec

urre

nce

free

surv

ival

Fig. 4. Kaplan-Meier analysis of disease recurrence-free survival for pa-tients with UC stratified by relative EZH2 protein levels in tumors with themedian score of invasive UC as cutting off. The x-axis represents diseaserecurrence-free survival in months after the date of diagnosis. The y-axisrepresents the proportion of patients surviving with uncensored data.(Color version of figure is available online.)

Table 3Univariate and multivariate proportional hazards regression models of clincystectomy

Variables Univariate analyses

HR 95% CI

Age 1.00 0.98–1.03Pathologic stage (non-inv. vs. inv.) 2.33 1.07–5.29Lymph node invasion 2.91 1.59–5.29Positive surgical margins 3.51 1.63–7.30

EZH2 score (high vs. low) 0.96 0.51–1.78

rotein levels are gradually increased during tumor progres-ion and that elevated EZH2 protein expression is associ-ted with more aggressive bladder cancer. The well-estab-ished semiquantitative immunostaining scoring systemsed in the present study may increase the accuracy ofcoring. Moreover, ROC analysis demonstrates that EZH2rotein scores possess statistically significant value for dif-erentiation of benign from cancerous lesions (P � 0.0001).etection of elevated EZH2 may potentially be used to

upplement currently available clinical, pathologic, and mo-ecular tests for bladder cancer.

Hinz et al. showed that EZH2 mRNAs were expressed inC and that expression levels were associated with patho-

ogic tumor features such as tumor stage and grade. EZH2RNA levels were also found to be associated with risk of

isease recurrence in muscle-invasive, but not in superficialladder cancer [15]. In the present study, we did not observecorrelation between increasing EZH2 protein expression

nd worse disease-free outcomes following radical cystec-omy.

Sample size is one limitation of the present study thatight explain why we did not observe a correlation betweenZH2 expression and clinical outcomes. Our study cohortay not be large enough to detect statistically significant

ifferences in EZH2 protein levels with respect to clinicalutcomes. Furthermore, the impact of EZH2 protein expres-ion on the outcomes of patients associated with UC may beffected because of patients who received either neoadju-ant or adjuvant chemotherapy. The number of patientsith low stage pathologic disease was also limited, how-

ver, for the majority of comparisons we stratified invasives. non-invasive disease to provide sufficient numbers fortatistical analyses. Selection bias may also be present,iven the fact that only samples from patients who under-ent radical cystectomy were used to create our bladder

ancer TMA. Another limitation is that benign urotheliumas obtained from patients undergoing cystectomy for blad-er cancer. It would have been preferable to have usedladder samples from individuals with no bladder pathol-gy, but these specimens were not available for use in thistudy.

Alternatively, our study may indeed suggest that EZH2rotein expression might be less relevant to the overallrognosis of invasive bladder cancer. EZH2 proteins may

d pathologic variables predicting recurrence following radical

Multivariate analyses

P value HR 95% CI P value

0.799 1.02 0.99–1.06 0.1690.034 1.36 0.49–3.77 0.5580.001 2.71 1.30–5.60 0.0070.007 2.41 0.93–6.26 0.070

ical an

0.893 0.93 0.46–1.86 0.827

[

[

[

[

[

[

[

[

[

433H. Wang et al. / Urologic Oncology: Seminars and Original Investigations 30 (2012) 428–433

function differently in a cancer type-specific manner; there-fore, EZH2 proteins may play different functional role(s)during tumor development and progression in UC comparedwith other cancer types.

5. Conclusion

Elevated EZH2 protein expression is associated withmore aggressive bladder cancer including invasive UC.ROC analyses demonstrated that EZH2 protein levels couldbe used to differentiate benign urothelium from invasiveUC. These data support the potential utility of EZH2 as abiomarker for diagnosis of UC. Further studies are war-ranted to determine whether EZH2 protein expression hasadditional prognostic value for predicting disease recur-rence-free outcomes following radical cystectomy.

Acknowledgments

The authors thank Dr. Shuanzeng Wei for his insightfuldiscussions regarding this work.

References

[1] National Comprehensive Cancer Network (NCCN) Clinical PracticeGuidelines in Oncology, Bladder Cancer, Including Upper TractTumors and Urothelial Carcinoma of Prostate, vol. VI 2009. NationalComprehensive Cancer Network, Fort Washington, PA.

[2] Kirkali Z, Chan T, Manoharan M, et al. Bladder cancer: Epidemiol-ogy, staging and grading, and diagnosis. Urology 2005;66:4–34.

[3] Wolff EM, Liang G, Jones PA. Mechanisms of Disease: Genetic andepigenetic alterations that drive bladder cancer. Nat Clin Pract Urol2005;2:502–10.

[4] Kleer CG, Cao Q, Varambally S, et al. EZH2 is a marker of aggres-sive breast cancer and promotes neoplastic transformation of breast

epithelial cells. Proc Natl Acad Sci U S A 2003;100:11606–11.

[5] Cao R, Zhang Y. SUZ12 is required for both the histone methyltrans-ferase activity and the silencing function of the EED-EZH2 complex.Mol Cell 2004;15:57–67.

[6] Tonini T, D’Andrilli G, Fucito A, et al. Importance of Ezh2polycomb protein in tumorigenesis process interfering with thepathway of growth suppressive key elements. J Cell Physiol 2008;214:295–300.

[7] Sellers WR, Loda M. The EZH2 polycomb transcriptional repres-sor—a marker or mover of metastatic prostate cancer? Cancer Cell2002;2:349–50.

[8] Bracken AP, Pasini D, Capra M, et al. EZH2 is downstream of thepRB-E2F pathway, essential for proliferation and amplified in cancer.EMBO J 2003;22:5323–35.

[9] Bryant RJ, Cross NA, Eaton CL, et al. EZH2 promotes proliferationand invasiveness of prostate cancer cells. Prostate 2007;67:547–56.

10] Tan J, Yang X, Zhuang L, et al. Pharmacologic disruption of Poly-comb-repressive complex 2-mediated gene repression selectively in-duces apoptosis in cancer cells. Genes Dev 2007;21:1050–63.

11] Varambally S, Dhanasekaran SM, Zhou M, et al. The polycombgroup protein EZH2 is involved in progression of prostate cancer.Nature 2002;419:624–9.

12] Bachmann IM, Halvorsen OJ, Collett K, et al. EZH2 expression isassociated with high proliferation rate and aggressive tumor sub-groups in cutaneous melanoma and cancers of the endometrium,prostate, and breast. J Clin Oncol 2006;24:268–73.

13] van Kemenade FJ, Raaphorst FM, Blokzijl T, et al. Coexpression ofBMI-1 and EZH2 polycomb-group proteins is associated with cyclingcells and degree of malignancy in B-cell non-Hodgkin lymphoma.Blood 2001;97:3896–901.

14] Arisan S, Buyuktuncer ED, Palavan-Unsal N, et al. Increased expres-sion of EZH2, a polycomb group protein, in bladder carcinoma. UrolInt 2005;75:252–7.

15] Hinz S, Kempkensteffen C, Christoph F, et al. Expression of thepolycomb group protein EZH2 and its relation to outcome in patientswith urothelial carcinoma of the bladder. J Cancer Res Clin Oncol2008;134:331–6.

16] Raman JD, Mongan NP, Tickoo SK, et al. Increased expression of thepolycomb group gene, EZH2, in transitional cell carcinoma of thebladder. Clin Cancer Res 2005;11:8570–6.

17] Netto GJ, Nakai Y, Nakayama M, et al. Global DNA hypomethy-lation in intratubular germ cell neoplasia and seminoma, but not innonseminomatous male germ cell tumors. Mod Pathol 2008;21:1337– 44.

18] Altman DG, Bland JM. Diagnostic tests 3: Receiver operating char-

acteristic plots. BMJ 1994;309:188.