a classification tree approach for pituitary adenomas

www.elsevier.com/locate/humpath

Human Pathology (2012) 43, 1627–1637

Original contribution

A classification tree approach for pituitary adenomas☆,☆☆

Alberto Righi MDa,b, Patrizia Agati BSI, PhD c, Andrea Sisto BSI d,Giorgio Frank MDe, Marco Faustini-Fustini MD f, Raffaele Agati MDg,Diego Mazzatenta MDe, Anna Farnedi BSc, PhDa, Federico Menetti MDg,Gianluca Marucci MDa, Maria P. Foschini MDa,⁎

aSection of Anatomic Pathology, Department of Haematology and Oncology “L. e A. Seragnoli” University of Bologna,Bellaria Hospital, 40139 Bologna, ItalybDepartment Biomedical Sciences and Human Oncology, University of Turin, 10126 Turin, ItalycDepartment of Statistics “P. Fortunati,” University of Bologna, 40126 Bologna, ItalydUniversity of Eastern Piedmont, Polis-Public Policy and Choice, 15121 Alessandria, ItalyeCenter of Surgery for Pituitary Tumors, Bellaria Hospital, 10126 Bologna, ItalyfEndocrine Unit, Department of Medicine, Bellaria Hospital, 10126 Bologna, ItalygDepartment of Neuroradiology, Bellaria Hospital, 40139 Bologna, Italy

Received 22 August 2011; revised 2 December 2011; accepted 8 December 2011

tuaC

B

0d

Keywords:Pituitary adenoma;Prognosis;Ki-67;p53

Summary It is difficult to evaluate the recurrence and progression potential of pituitary adenomas atpresentation. The World Health Organization classification of endocrine tumors suggests that invasionof the surrounding structures, size at presentation, an elevated mitotic index, a Ki-67 labeling indexhigher than 3%, and extensive p53 expression are indicators of aggressive behavior. Nevertheless, Ki-67and p53 labeling index evaluation is subject to interobserver variability, and their cutoff values arecontroversial. In the present study, the prognostic value of Ki-67 and p53 protein labeling indices andtheir correlation with clinical and radiologic parameters were evaluated using digital image analysis in aseries of 166 pituitary adenomas in patients having undergone a follow-up of at least 6 years to evaluatethe impact on the recurrence and progression potential of pituitary adenomas. The data were analyzedusing the receiver operating characteristic curve and classification and regression tree analysis. Theresults showed that, in the unstratified data set, the commonly used threshold of the Ki-67 index of 3%has a high specificity (89.5%) but a low sensitivity (53.8%). Unsatisfactory performance results wereobtained by performing receiver operating characteristic curve analysis on the p53 labeling index. Onthe contrary, the classification and regression tree analysis–derived tree demonstrated that each pituitaryadenoma subtype has specific prognostic factors. Specifically, the Ki-67 labeling index is a usefulprognostic factor in nonfunctioning, adrenocorticotropin, and prolactin adenomas, but with different

☆ Conflict of interest statement: We declare that we have no conflict of interest.☆☆ A. Righi is a part of and is funded by the doctoral program “Scienze Biomediche ed Oncologia Umana: Tecniche Avanzate di localizzazione dei

mori umani,” University of Turin, Italy. This manuscript was partially supported by grants from the University of Bologna, Department of Hematologynd Oncology (Fundamentally Oriented Research funds—Maria Pia Foschini). A. Farnedi received a Sandro Cavagnino and Vanda Vanini grant fromentro Interdipartimentale di Ricerche sul Cancro of the University of Bologna.⁎ Corresponding author. Section of Anatomic Pathology, Department of Haematology and Oncology “L. e A. Seragnoli,” University of Bologna, Ospedale

ellaria, via Altura 3, 40139 Bologna, Italy.

046-8177/$ – see front matter © 2012 Elsevier Inc. All rights reserved.oi:10.1016/j.humpath.2011.12.003

http://dx.doi.org/10.1016/j.humpath.2011.12.003

1628 A. Righi et al.

thresholds. In conclusion, our study emphasizes that the term pituitary adenomas includes differenttypes of tumors, each one having specific prognostic factors.© 2012 Elsevier Inc. All rights reserved.

1. Introduction with clinical and radiologic parameters in a series of 166 PAs

Most pituitary adenomas (PAs) are slow growing, benigntumors, usually cured with medical or surgical treatment. Onthe other hand, even if recurrences or disease progression islimited to a small proportion of cases, they can cause severedamage to the patient. Therefore, the management of PAs iscomplicated by the difficulty in predicting their course,particularly with respect to recurrence [1].

The 2004 edition of the World Health Organization(WHO) classification of endocrine tumors recognizes 4prognostic features (an increased mitotic index, invasivegrowth, Ki-67 labeling index [LI] N3% and extensive nuclearp53 positivity), which characterize the likelihood ofaggressive PA behavior and define this adenoma as atypical[2]. Nevertheless, the real value of these markers coupledwith appropriate thresholds for Ki-67 and p53 LIs asindicators of tumor progression is still controversial [3-12].

The reasons for which it is difficult to draw conclusionswithin subtypes are related to the relative ineffectiveness ofany single feature, histologic or otherwise, to predict tumorbehavior and to the scarcity of case series regarding recurrenttumors with a long-term follow-up. In fact, a review of 2304cases of PAs reported in the literature, which had a medianfollow-up of 4.3 years (range, 2 months to 24 years)[3,4,6,12-22] evidenced a mean recurrence or tumor progres-sion of 12.5% (289/2304 PA cases), with 2 major studiesreporting 14 (43.8%) of 32 [3] and 24 (43.6%) of 55 [12]recurrent PA cases, respectively. In these latter studies, themedian relapse latency for recurrence or progression was 6.1years (range, 5 months to 28 years). Most of the previouslymentioned studies are heterogeneous with respect to the PAsubtypes, postoperative therapies, follow-up period, andmethodologies (digital methodology or standard operatorvisual analysis) used to calculate the Ki-67 and p53 LIs.Furthermore, recent studies have suggested that each differentPA subtype requires proper prognostic indicators [1,23,24].

Therefore, additional studies involving large numbers ofPAs with a long-term follow-up are needed to validate theparameters useful for evaluating the risk of recurrence ordisease progression in each PA category [6]. Specifically,several questions remain to be answered. It is presentlyunclear which of the 4 prognostic features (an increasedmitotic index, invasive growth, Ki-67 LI N3%, or extensivenuclear p53 positivity) identified by WHO [2] relates to theaggressive behavior of PAs. Furthermore, the influence ofeach parameter on the different PA subtypes and the valuesof Ki-67 and p53 have not been clearly elucidated.

The aim of the present study was to evaluate the impact ofKi-67 and p53 protein LIs, the subtype, and the correlation

having a minimum follow-up of 6 years to predict therecurrence and progression potential of PAs.

2. Materials and methods

2.1. Cases

All PAs present in the histologic records of the Section ofAnatomic Pathology of the University of Bologna at BellariaHospital in the period between January 1989 and December2007 were retrieved.

The same neurosurgeons (G. F. and D. M.) had operatedon all cases using the transsphenoidal approach.

The cases meeting the following criteria were selected forthe present study: (1) the availability of paraffin blocks basedon a specimen size sufficient to allow morphologic andimmunohistochemical characterization; (2) no radiationtherapy before surgery; (3) the availability of clinicalinformation, including endocrinologic evaluation and neu-roimaging data; (4) and a minimum follow-up of 6 years.

Tumor invasion was defined based on 1 or more of thefollowing parameters: preoperative imaging (magneticresonance imaging [MRI] or computed tomography),intraoperative findings, and histology [2]. Dural invasionwas not considered a feature of invasion because previousstudies have demonstrated that it is not related to therecurrence rate [2]. Suprasellar growth was considered tobe an extension rather than an invasion. The MRI findingswere reviewed; PAs were classified by size into micro-adenomas (b10 mm) and macroadenomas (≥10 mm).Neuroradiologic findings were classified according to theWHO indications [2].

Follow-up monitoring included postoperative MRI scansobtained 2 to 6 months after surgery and then at variableintervals, depending on the clinical and neuroradiologicfindings and on the biochemical analysis results.

The follow-up results were classified as follows:

1. cured when no residual tumor was detected at postoperativeMRI associated with normal hormone serum levels in the casesof functional tumors;2. clinically controlled persistent tumor when minimal clinicalsymptoms persisted, confirming elevated serum hormone levelswith or without minimum tumor residue (this category includedfunctioning PAs);3. improved persistent tumor when minimum tumor residuewas visible on postoperative MRI, without clinical symptoms(this category is limited to nonfunctioning PAs);

1629Predictors of aggressiveness in pituitary adenomas

4. not cured with progression when symptoms persisted aftersurgery associated with elevated serum hormone levels in thecase of functional tumors and MRI demonstrating a growth ofthe neoplastic mass;5. recurrence of disease when a pituitary tumor was detectedin patients without evidence of residual tumor after surgicaltherapy; and6. regrowth of the PA was defined as an enlargement of thetumor remnant at postoperative imaging.

Categories 1, 2, and 3 were considered favorableoutcomes and were included in the “nonrecurrent” group,whereas categories 4, 5, and 6 were considered unfavorableoutcomes and were included in the group with “recurrent orprogression of disease.”

2.2. Immunohistochemistry

All slides were reviewed by 3 different surgicalpathologists (A. R., G. M., and M. P. F.).

The surgical specimens were fixed in 10% bufferedformalin and routinely processed with paraffin. Selectedblocks were serially cut and stained with hematoxylin andeosin, periodic acid–Schiff, and reticulin staining. A routineimmunohistochemical method was followed. An epitoperetrieval method was used as follows. Before immunostain-ing, the sections were steamed in citrate buffer for 25 minutesand then cooled for 5 minutes. The source and dilutions ofthe antisera used were as follows: thyrotropin (TSH)(Neomarkers Inc, Fremont, CA; clone TSH01 + TSH02),growth hormone (GH) (Biogenex, Fremont, CA; clone 54/92A2), prolactin (PRL) (Cell Marque, Rocklin, CA; poly-clonal), adrenocorticotropin (ACTH) (Dako, Carpintera, CA;polyclonal), follicle-stimulating hormone (Biomeda, FosterCity, CA; polyclonal), luteinizing-stimulating hormone(Biogenex; clone 3LH 5B6 YH4), Ki-67 (Dako; cloneMib-1), and p53 (Ventana, Tucson, AZ; clone DO-7). In allinstances, positive and negative controls were carried out andreacted appropriately.

Table 1 Clinical and neuroradiologic differences between not recurreexact test

Variables Not recurrent(143 cases)

Age N50 y 69/143 (48.3Sex (female/male) 65/78Size: macroadenoma 104/143 (72.7Invasive 34/143 (23.8Suprasellar extension 58/143 (40.6Partial resection or minimal residual of tumor 54/144 (37.5Mitotic index N1 per 10 HPFs 30/143 (21%Ki-67 LI N3% 7/143 (4.9%P53 LI N3% 7/143 (4.9%Recurrent time (y), mean (range)Follow-up (y), mean (range) 9.2 (6-15

2.3. Analysis of Ki-67and p53 labeling indices

The results of the analysis of Ki-67 and p53 LIs wereexpressed as the percentage of tumor cells with positive nuclei.As is commonly seen in other endocrine neoplasms, Ki-67–positive cells were heterogeneously distributed throughoutthe tumor. As in other studies on proliferative activityin PAs or in other endocrine tumors [6,12,13,21,25-27],the Ki-67–labeled and p53-labeled nuclei were evaluated inthe tumoral areas where these markers were predominantlyexpressed. Given that cell counting on numerous microscopicfields is known to be a source of interobserver andintraobserver variability, a digital camera (Olympus Q-color3, Tokyo, Japan) with area-based image analysis software(Dot-Slide 1.2 version) was used. Immunostained sectionswere screened under the microscope, and the areas containingthe greatest number of Ki-67–stained and p53-stained nucleiwere outlined with a slide pointer, allowing easy localizationduring image analysis. The Ki-67 and p53 LIs were calculatedas the ratio between the labeled and the total nuclear areas.Only nuclei with a strongly positive label were counted.The 7number of adenomatous cells per field, depending on thecellular density of the tumor, ranged from 250 to 400 cells perhigh-power field (HPF) (original magnification ×400). Ki-67and p53 LIs were evaluated in areas of the highest Ki-67 andp53 LIs (“hot spots”), on 10HPFs,when available according tothe tumor size.

2.4. Statistical analysis

All data were analyzed using Stata/SE statistical software(version 10.0; StataCorp LT, College Station, TX). P b .05was considered statistically significant.

Clinical and neuroradiologic differences between nonre-current and recurrent/uncontrolled PAs were determinedusing the Fisher exact test. Kaplan-Meier univariate analyses[28] for the clinical and pathologic variables related torecurrence/progression-free survival were used. The relative

nt and recurrent PA or PA with progression of disease and Fisher

Recurrent/Progression(23 cases)

Fisher exact test(P value)

%) 14/23 (60.9%) .36910/13 1

%) 19/23 (82.6%) .443%) 13/23 (56.5%) .002%) 13/23 (56.5%) .164%) 16/23 (69.6%) .006) 8/23 (34.8%) .18) 7/23 (30.4%) .001) 3/23 (13%) .149

5.3 (1.6-11.6).1) 9 (1.6-15.3)

Table 2 Kaplan-Meier univariate analyses for clinical and pathologic variables related to recurrence/progression-free survival

Variables 12 mo, 95% CI 24 mo, 95% CI 36 mo, 95% CI 60 mo, 95% CI 120 mo, 95% CI 180 mo, 95% CI P

Age (y)b50 0.99 (0.92-1.00) 0.99 (0.92-1.00) 0.99 (0.92-1.00) 0.97 (0.89-0.99) 0.88 (0.77-0.94) 0.74 (0.36-0.91) .141≥50 1.00 0.96 (0.89-0.99) 0.92 (0.83-0.96) 0.86 (0.76-0.92) 0.84 (0.74-0.90) –SexF 0.99 (0.91-1.00) 0.97 (0.90-0.99) 0.95 (0.86-0.98) 0.91 (0.81-0.95) 0.88 (0.76-0.94) 0.44 (0.07-0.79) .845M 1.00 0.98 (0.92-0.99) 0.96 (0.89-0.98) 0.91 (0.83-0.97) 0.85 (0.76-0.91) –Size (cm)b1 1.00 1.00 0.98 (0.85-1.00) 0.95 (0.83-0.99) 0.88 (0.70-0.96) – .293≥1 0.99 (0.94-1.00) 0.97 (0.92-0.99) 0.94 (0.88-0.97) 0.89 (0.83-0.94) 0.86 (0.78-0.91) 0.57 (0.21-0.82)Functional statusNo 1.00 0.99 (0.92-1.00) 0.98 (0.91-0.99) 0.93 (0.85-0.97) 0.90 (0.81-0.94) – .683Yes 0.99 (0.92-1.00) 0.96 (0.92-0.99) 0.93 (0.85-0.97) 0.89 (0.80-0.94) 0.83 (0.72-0.90) 0.83 (0.72-0.90)InvasiveNo 1.00 1.00 0.99 (0.94-1.00) 0.97 (0.91-0.98) 0.92 (0.84-0.96) 0.70 (0.33-0.88) b.001Yes 0.98 (0.86-1.00) 0.91 (0.79-0.97) 0.85 (0.71-0.93) 0.77 (0.62-0.86) – –Suprasellar extensionNo 1.00 1.00 0.97 (0.91-0.99) 0.95 (0.88-0.98) 0.90 (0.81-0.95) 0.65 (0.27-0.86) .065Yes 0.99 (0.90-1.00) 0.94 (0.88-0.98) 0.93 (0.84-0.97) 0.86 (0.75-0.92) 0.81 (0.70-0.89) –Partial resectionNo 1.00 1.00 0.99 (0.93-1.00) 0.97 (0.91-0.99) 0.95 (0.87-0.98) 0.95 (0.87-0.98) b.001Yes 0.99 (0.90-1.00) 0.94 (0.85-0.98) 0.90 (0.80-0.95) 0.83 (0.72-0.90) 0.75 (0.62-0.84) –Mitotic figures (per 10 HPFs)≤1 1.00 0.99 (0.95-1.00) 0.97 (0.93-0.99) 0.93 (0.87-0.92) 0.90 (0.83-0.94) 0.60 (0.21-0.84) .148N1 0.97 (0.87-1.00) 0.92 (0.77-0.97) 0.89 (0.74-0.96) 0.84 (0.68-0.93) 0.76 (0.56-0.88) –Ki-67≤3% 1.00 1.00 0.99 (0.95-1.00) 0.95 (0.90-0.97) 0.89 (0.82-0.94) 0.67 (0.33-0.87) b.001N3% 0.93 (0.61-0.99) 0.73 (0.44-0.89) 0.60 (0.32-0.79) 0.53 (0.26-0.74) – –p53≤3% 0.99 (0.96-1.00) 0.98 (0.94-0.99) 0.96 (0.91-0.98) 0.92 (0.86-0.95) 0.87 (0.80-0.91) 0.74 (0.43-0.90) .178N3% 1.00 0.90 (0.48-0.99) 0.90 (0.48-0.99) 0.80 (0.41-0.85) 0.80 (0.41-0.85) –

NOTE. P value refers to log-rank test. Abbreviations: CI, confidence interval; F, female; M, male.

Table 3 Results of multivariate Cox regression analysis

Variable z Value HR 95% CI P

Ki-67 N3% 4.89 18.69 5.78-60.49 b.001Partial resection 2.38 7.52 1.43-39.55 .017Invasion 2.10 3.52 1.09-11.39 .036

NOTE. Likelihood ratio χ2 test: 44.82; P b .0001. Abbreviation: 95%CI, 95% confidence interval.


significance of each parameter (sex, age, site, size of PA, typeof neoplasm [functioning or not], invasion, partial resection,mitotic index, Ki-67 LI, and p53 expression) included in theunivariate analysis on recurrence/progression-free survivalwas estimated using the Cox proportional hazards regressionmodel determined by the Wald test (z value). The Coxproportional hazards regression model assumption wassatisfied for all prognostic variables in our models.

The performance of the Ki-67 proliferation index as aprognostic factor for the recurrence/progression of PAs wasinvestigated by means of receiver operating characteristic(ROC) curve analysis. Successively, a Classification andRegression Tree (CART) analysis [29] was designed tocreate a tree structure that identifies classes of patients athigh risk for PA recurrence/progression based on severalprognostic factors. Ki-67 and p53 LIs, size of the adenoma,presence of invasion, and age of the patient were includedas possible predictor variables in the splitting process. TheGini impurity index was adopted as a splitting criterion.ROC curves were derived, and performance measurementswere computed for the output trees. Classification treeswere obtained by using CART Pro 6.0 (Salford System,San Diego, CA).

3. Results

3.1. Clinical and pathologic features

One hundred sixty-six PAs met the inclusion criteria andconstituted the basis of the study. Selected cases included 82nonfunctioning and 84 functioning PAs.

The nonfunctioning PAs consisted of 56 null cell adeno-mas (of these, 10 were oncocytic variants), 20 gonadotropiccell adenomas, 2 ACTH silent-type I adenomas, 3 ACTHsilent-type II adenomas, and 1 silent-subtype III adenoma.

The functioning PAs consisted of 24 PRL cell adenomas,20 ACTH adenomas, 27 GH-containing adenomas (12sparsely and 15 densely granulated), 3 TSH-secreting


adenomas, 2 plurihormonal adenomas, 2 mammosomato-troph adenomas, and 6 mixed adenomas, secreting both PRLand GH. To diagnose silent-subtype III adenomas andseparate oncocytic tumors from nononcocytic ones, electronmicroscopy was used at the time of the original diagnosis.

Patient age ranged from 14 to 87 years (mean, 48.8 years)at the time of the first surgical treatment. Ninety-two patients(55.4%) were male, and 75 (45.1%) were female.

On operative and neuroradiologic inspection, mosttumors (123/166; 74.0%) were macroadenomas, with asuprasellar extension in 71 (42.8%) of 166 patients andtumor invasion in 47 patients (28.3%). A high mitotic index(N1 per 10 HPF) was found in 38 cases (22.9%).

Partial resection or minimum tumor residue was identifiedin 70 (42.2%) of 166 patients. During follow-up, 18 (10.8%)of 166 patients showed a recurrence after an apparent grosstotal resection. Another 5 patients (3%) showed progressionof the disease after a subtotal surgical resection of theprimary PA. The mean recurrence/progression time was 5.3years (range, 1.2-11.6 years). At the last follow-up (mean,9.2 years; range, 6-15.1 years), 35 (21.1%) of 166 cases werecontrolled/improved, and 108 (65.1%) of 166 cases werecured. The clinical and neuroradiologic differences betweennonrecurrent PAs and PAs with recurrence or progression aresummarized in Table 1.

3.2. Immunohistochemical analysis

Because the distributions of the LIs were often skewed,the data are reported in terms of mean and range, evaluatedby digital methodology (Olympus Q-color 3), to emphasizethe variations observed.

Table 4 Unstratified data set and ROC analysis for Ki-67 index: ROand accuracy

NOTE. A threshold of 3% (set in bold characters) has a high specificity (89.5%)

The mean (±SD) of the Ki-67 LI was higher in PAs withrecurrence/progression (5.40% ± 9.50%; range, 0.3-43) thanin the PAs included in the nonrecurrent group (1.30% ±0.90%; range, 0-4.3).

P53 protein expression was demonstrable in 12(52.2%) of the 23 PAs with recurrence/progression duringfollow-up and in 61 (42.8%) of the 143 nonrecurrent PAs.The mean of the p53 LI was higher in PAs withrecurrence/progression (1.30% ± 2.90%; range, 0-14) thanin those without recurrence/progression (0.60% ± 1.50%;range, 0-14).

3.3. Statistical analysis

The Fisher exact test showed a correlation between the Ki-67 LI having a cutoff of 3%, tumor invasion, and partialresection or minimal residual of tumor with the probability ofrecurrence or progression of the PA (Table 1).

Kaplan-Meier univariate analysis on the determinants ofrecurrence/tumor progression demonstrated a significantlylower recurrence/progression-free survival with invasion,partial tumor resection, and a Ki-67 LI having a cutoff of3% (Table 2).

Multivariate Cox regression analysis estimated Ki-67with a cutoff of 3% to be a strong predictive factor ofrecurrence/tumor progression (P b .001; hazard ratio [HR],18.69) followed by partial tumor resection (P = .017; HR,7.52) and tumor invasion (P = .036; HR, 3.52) (Table 3).

In the unstratified data set, the commonly used thresholdof a Ki-67 index of 3% showed a high specificity (89.5%) buta low sensitivity (53.8%) (Table 4). Unsatisfactory perfor-mance results (data not shown) were also obtained by

C curve and Stata report of classification sensitivity, specificity,

but a low sensitivity (53.8%). Abbreviation: AUC, area under the curve.

Table 5 Stratified data sets and ROC analysis for Ki-67 index: ROC curves and Stata report of classification sensitivity, specificity,and accuracy

NOTE. Performance measures for a threshold of 3% (set in bold characters): nonfunctioning subgroup: specificity (96%) and sensitivity (33.3%); PRL subgroup:specificity (76.2%) and sensitivity (100%); and ACTH subgroup: specificity (82.3%) and sensitivity (0%). Abbreviation: AUC, area under the curve.


Unlabelled image

Table 6 Stratified data sets and ROC analysis for Ki-67 index: ROC curves and Stata report of classification sensitivity, specificity, andaccuracy for gonadotroph subgroup

NOTE. Performance measures for a threshold of 3% (set in bold characters): 100% specificity but 0% sensitivity. Abbreviation: AUC, area under the curve.

Fig. 1 Nonfunctioning adenomas: classification tree for recurrence/progression. Invasion of the surrounding structures is the first parameteremerging in the CART classification tree for the identification of cases with a high probability of recurrence/progression, whereas the Ki-67 LIis the second parameter. The grey bars indicate the proportion of nonrecurrent PAs.


Unlabelled image

image of Fig. 1

Table 7 Nonfunctioning adenomas: performance measure and ROC curve

Abbreviations: AUC, area under the curve; Pos., positive.


carrying out ROC curve analysis on the p53 LI; therefore,additional evaluation of this parameter was not carried out.

To further analyze the influence of functional status, thedifferent types of PAs (null cell, PRL, ACTH functioning,and gonadotroph groups) were evaluated separately byROC curve and CART analysis (Tables 5 and 6). Theremaining PA subtypes, a total of 46 patients (27.7% ofthe entire data set), were excluded from separate analyses

Fig. 2 PRL adenomas: classification

as a consequence of their small number and too few caseswith recurrence/progression.

The commonly used threshold of a Ki-67 index of 3%appeared to be ineffective in identifying patients with highrisk or recurrence/progression in almost all the subgroups.The only exception was the PRL subgroup, in which thepreviously mentioned threshold presented a good specificity(71.4%) and attained a sensitivity of 94.1%.

tree for recurrence/progression.

Unlabelled image

image of Fig. 2

Table 8 PRL adenomas: performance measure and ROC curve



3.4. CART results by PA subgroup

• Null cell adenomas: the CART algorithm selected 2relevant prognostic factors (presence of invasion and Ki-67 LI) and identified 2 subgroups with a higher risk ofrecurrence/progression (Fig. 1; Table 7): (a) invasiveadenomas and (b) noninvasive adenomas characterized byKi-67 greater than 1.95%.

Fig. 3 ACTH adenomas: classification t

• PRL adenomas: the CART procedure selected only 1 factoras a splitting variable—Ki-67 LI—and identified only 1subgroup with a high risk of recurrence/progression:adenomas with Ki-67 greater than 4.40% (Fig. 2; Table 8).

• ACTH functioning adenomas: the CART algorithm selectedKi-67 as the only splitting factor; Ki-67 greater than 1.70%identified the subgroup with a high risk of recurrence/progression (Fig. 3; Table 9).

ree for recurrence/progression.

Unlabelled image

image of Fig. 3

Table 9 ACTH adenomas: performance measure and ROC curve



• Gonadotroph adenomas: the CART results in this group didnot give valuable predictive indications.

4. Discussion

The present study reported a morphologic and immu-nohistochemical evaluation of PAs based on 4 prognosticfeatures (an increased mitotic index, invasive growth, Ki-67 LI N3%, and extensive nuclear p53 positivity), whichcharacterize the likelihood of aggressive PA behaviorrecognized by WHO as an atypical PA [2] and thecorrelation with clinical and neuroradiologic parameters.The present study added specific information to both theshortcomings and the virtues of the tests analyzed byCART analysis.

In the present study, the mean of recurrence or tumorprogression was 13.9% (23/166), with a medium follow-upof 9 years (range, 1.6-15.3 years), which is slightly higherthan the percentage of recurrence or tumor progression(12.5%, or 289/2304 PA cases) reported in the literature,with a median follow-up of 4.3 years (range, 2 months to24 years) [3,4,6,12-22]. This difference might be related tothe longer follow-up of the present series.

Among all the criteria proposed by the WHO [2] fordefining atypical, the data herein obtained indicated that aKi-67 greater than 3% and invasive growth are the mostimportant histologic features for assessing a high risk ofprogression or recurrence of PAs, whereas the expressionof p53 and the mitotic index are inadequate. Therefore,these 2 parameters (Ki-67 LI N3% and invasive growth)appeared to be the 2 most important pathologic featuresinfluencing the clinical management of the patient affectedby PAs.

The cutoff of 3% for Ki-67 has, however, a low sensitivity(53.8%), even if it shows an elevated specificity (89.5%) inall 166 cases.

These values of sensitivity and specificity change in thevarious types of PAs, as is seen from the ROC and the CARTanalyses. Furthermore, in cases of null cell PAs, invasivegrowth was the most important prognostic factor because itidentified 5 of 6 cases with recurrence/progression. In thenoninvasive subgroup of null cell PAs, the Ki-67 LI wasuseful in identifying patients at risk for recurrence/progres-sion. In PRL and ACTH functioning PAs, Ki-67 LI emergedas the most important prognostic factor, but with differentcutoff values. The small number of other subtypes (sparselyand densely granulated GH-secreting PAs, TSH-secretingPAs, plurihormonal PAs, mammosomatotroph PAs, andmixed PAs, secreting both PRL and GH) makes anysubgroup analysis relatively weak. A much larger cohort oftumors will be needed to make any determination regardingthese subtypes of PAs.

Regarding the p53 LI, even if a slightly higher level ofp53 expression was observed in PAs with progression/recurrence as compared with the other cases, this differencewas not statistically significant. The poor prognostic valueof the p53 LI was further confirmed by a CART analysis ofeach single PA type, wherein it never appeared among theuseful factors.

In the present study, a digital quantification method wasadopted to more precisely evaluate the Ki-67 LI. Specifical-ly, a standardized method for selecting the area with thehighest proliferative activity (hot spots), on 10 HPFs, in aprinted photograph was used [6,12,25]. Adopting the digitalquantification methodology in clinical practice despite thecosts, time limitations, and the need for special training couldbe useful in the assessment of Ki-67 in PAs to reduce

Unlabelled image


intraobserver and interobserver variations and to maximizeits use as a prognostic indicator [18]. Currently, digital imageanalysis is used to evaluate multiple prognostic factors ofdifferent endocrine tumors [25].

The present classification tree approach confirms thehypothesis proposed in recent articles regarding the need tofind different prognostic values for each individual PAsubtype [1,30,31]. Specifically, the present data suggest thatthe Ki-67 LI with different cutoff values, each specific for thedifferent PA subgroups, and tumor invasion are the mostreliable prognostic pathologic parameters for detecting PAswith a high risk of recurrence/progression.

Acknowledgment

The authors would like to thank Mr Gerald Goldsmith forEnglish style revision.

References

[1] Salehi F, Agur A, Scheithauer BW, Kovacs K, Lloyd RV, CusimanoM. Ki-67 in pituitary neoplasms: a review—part I. Neurosurgery2009;65:429-37.

[2] Lloyd RV, Kovacs K, Young Jr WF, et al. Tumors of the pituitary. In:DeLellis RA, Lloyd RV, Heitz PU, Eng C, editors. World HealthOrganization classification of tumors: pathology and genetics: tumorsof endocrine organs. IARC: Lyon; 2004. p. 10-47.

[3] Dubois S, Guyétant S, Menei P, et al. Relevance of Ki-67 andprognostic factors for recurrence/progression of gonadotropic adeno-mas after first surgery. Eur J Endocrinol 2007;157:141-7.

[4] Hentschel SJ, McCutcheon E, Moore W, Durity FA. P53 andMIB-1 immunohistochemistry as predictors of the clinical behaviorof nonfunctioning pituitary adenomas. Can J Neurol Sci 2003;30:215-9.

[5] Pizarro CB, Oliveira MC, Coutinho LB, Ferreira NP. Measurement ofKi-67 antigen in 159 pituitary adenomas using the MIB-1 monoclonalantibody. Braz J Med Biol Res 2004;37:235-43.

[6] Scheithauer BW, Gaffey TA, Lloyd RV, et al. Pathobiology ofpituitary adenomas and carcinomas. Neurosurgery 2006;59:341-53.

[7] Turner HE, Nagy Z, Gatter KC, Esiri MM, Wass JA, Harris AL.Proliferation, bcl-2 expression and angiogenesis in pituitary adenomas:relationship to tumor behaviour. Br J Cancer 2000;82:1441-5.

[8] Filippella M, Galland F, Kujas M, et al. Pituitary tumor transforminggene (PTTG) expression correlates with the proliferative activity andrecurrence status of pituitary adenomas: a clinical and immunohisto-chemical study. Clin Endocrinol 2006;65:536-43.

[9] Jaffrain-Rea ML, Di Stefano D, Minniti G, et al. A critical reappraisalof MIB-1 labelling index significance in a large series of pituitarytumors: secreting versus non-secreting adenomas. Endocr RelatCancer 2002;9:103-13.

[10] Nakabayashi H, Sunada I, Hara M. Immunohistochemical analyses ofcell cycle–related proteins, apoptosis, and proliferation in pituitaryadenomas. J Histochem Cytochem 2001;49:1193-4.

[11] Paek KI, Kim SH, Song SH, et al. Clinical significance of Ki-67labeling index in pituitary macroadenoma. J Korean Med Sci 2005;20:489-94.

[12] Gejman R, Swearingen B, Hedley-Whyte ET. Role of Ki-67proliferation index and p53 expression in predicting progression ofpituitary adenomas. HUM PATHOL 2008;39:758-66.

[13] Fusco A, Zatelli MC, Bianchi A, et al. Prognostic significance of theKi-67 labeling index in growth hormone–secreting pituitary adeno-mas. J Clin Endocrinol Metab 2008;93:2746-50.

[14] Meij BP, Lopes MB, Ellegala DB, Alden TD, Laws Jr ER. The long-term significance of microscopic dural invasion in 354 patients withpituitary adenomas treated with transsphenoidal surgery. J Neurosurg2002;96:195-208.

[15] Noh TW, Jeong HJ, Lee MK, Kim TS, Kim SH, Lee EJ. Predictingrecurrence of nonfunctioning pituitary adenomas. J Clin EndocrinolMetab 2009;94:4406-13.

[16] LosaM,FranzinA,Mangili F, et al. Proliferation index of nonfunctioningpituitary adenomas: correlations with clinical characteristics and long-term follow-up results. Neurosurgery 2000;47:1313-8.

[17] Ogawa Y, Ikeda H, Tominaga TJ. Clinicopathological study ofprognostic factors in patients with pituitary adenomas and Ki-67labeling index of more than 3%. Endocrinol Invest 2009;32:581-4.

[18] Mastronardi L, Guiducci A, Spera C, Puzzilli F, Liberati F, Maira G.Ki-67 labelling index and invasiveness among anterior pituitaryadenomas: analysis of 103 cases using the MIB-1 monoclonalantibody. J Clin Pathol 1999;52:107-11.

[19] Wierinckx A, Auger C, Devauchelle P, et al. A diagnostic marker setfor invasion, proliferation, and aggressiveness of prolactin pituitarytumors. Endocr Relat Cancer 2007;14:887-900.

[20] Wang EL, Qian ZR, Rahman MM, et al. Increased expression ofHMGA1 correlates with tumor invasiveness and proliferation inhuman pituitary adenomas. Histopathology 2010;56:501-9.

[21] Ferrante E, Ferraroni M, Castrignanò T, et al. Non-functioningpituitary adenoma database: a useful resource to improve the clinicalmanagement of pituitary tumors. Eur J Endocrinol 2006;155:823-9.

[22] Auer LM, Clarici G. The first 100 transsphenoidally operated pituitaryadenomas in a non-specialised centre: surgical results and tumor-recurrence. Neurol Res 1985;7:153-60.

[23] Asa SL. Practical pituitary pathology: what does the pathologist needto know? Arch Pathol Lab Med 2008;132:1231-40.

[24] Lloyd RV. Molecular pathology of pituitary adenomas. J Neurooncol2001;54:111-9.

[25] Scarpa A, Mantovani W, Capelli P, et al. Pancreatic endocrine tumors:improved TNM staging and histopathological grading permit aclinically efficient prognostic stratification of patients. Mod Pathol2010;23:824-33.

[26] Knosp E, Kitz K, Perneczky A. Proliferation activity in pituitaryadenomas: measurement by monoclonal antibody Ki-67. Neurosur-gery 1989;25:927-30.

[27] Hsu DW, Hakim F, Biller BM, et al. Significance of proliferating cellnuclear antigen index in predicting pituitary adenoma recurrence.J Neurosurg 1993;78:753-61.

[28] Kaplan E, Meier P. Nonparametric estimation from incompleteobservations. J Am Stat Assoc 1958;53:457-81.

[29] Breiman L, Friedman JH, Olshen RA, Stone C. Classification andregression trees. Belmont (Calif): Wadsworth; 1984.

[30] Al-Shraim M, Asa SL. The 2004 World Health Organization classifi-cation of pituitary tumors: what is new? Acta Neuropathol 2006;111:1-7.

[31] Kleinschmidt-DeMasters BK. Subtyping does matter in pituitaryadenomas. Acta Neuropathol 2006;111:84-5.

a classification tree approach for pituitary adenomas

Documents