Risk Criteria and Prognostic Factors for Predicting Recurrences After Resection of Primary...

Gastrointestinal Oncology

Risk Criteria and Prognostic Factors for PredictingRecurrences After Resection of Primary Gastrointestinal

Stromal Tumor

Piotr Rutkowski, MD, PhD,1 Zbigniew I. Nowecki, MD, PhD,1 Wanda Michej, MD,2

Maria Debiec-Rychter, MD, PhD,3 Agnieszka Wozniak, PhD,4 Janusz Limon, PhD,4

Janusz Siedlecki, PhD,5 Urszula Grzesiakowska, MD, PhD,6 Michał Kakol, MD,7

Czesław Osuch, MD, PhD,8 Marcin Polkowski, MD, PhD,9 Stanisław Głuszek, MD, PhD,10

Zbigniew _Zurawski, MD,1 and Włodzimierz Ruka, MD, PhD1

1Department of Soft Tissue/Bone Sarcoma and Melanoma, M. Sklodowska-Curie Memorial Cancer Center and Institute ofOncology, Roentgena 5, 02-781, Warsaw, Poland

2Department of Pathology, M. Sklodowska-Curie Memorial Cancer Center and Institute of Oncology, Roentgena 5, 02-781,Warsaw, Poland

3Center for Human Genetics, University of Leuven, O&N Gasthuisberg Herestraat 49, 3000, Leuven, Belgium4Department of Biology and Genetics, Medical University of Gdansk, Debinki 1, 80-211, Gdansk, Poland

5Department of Molecular Biology, M. Sklodowska-Curie Memorial Cancer Center and Institute of Oncology, Roentgena 5, 02-781,Warsaw, Poland

6Department of Radiology, M. Sklodowska-Curie Memorial Cancer Center and Institute of Oncology, Roentgena 5, 02-781,Warsaw, Poland

7Regional Oncological Center, Marii C. Sklodowskiej 2, 80-210, Gdansk, Poland8Department of General Surgery, Jagiellonian University, Kopernika 40, 31-501, Cracow, Poland

9Department of Gastroenterology, M. Sklodowska-Curie Memorial Cancer Center and Institute of Oncology, Roentgena 5, 02-781,Warsaw, Poland

10City Hospital, Grunwaldzka 45, 25-736, Kielce, Poland

Background: The introduction of adjuvant imatinib in gastrointestinal stromal tumors(GISTs) raised debate over the accuracy of National Institutes of Health risk criteria and thesignificance of other prognostic factors in GIST.Methods: Tumor aggressiveness and other clinicopathological factors influencing disease-

free survival (DFS) were assessed in 335 patients with primary resectable CD117-immuno-positive GISTs (median follow-up, 31 months after primary tumor resection) from a pro-spectively collected tumor registry.Results: Overall median DFS was 37 months, and estimated 5-year DFS was 37.8 %. In

univariate analysis, high or intermediate risk group (P < .000001), mitotic index >5/50 high-power field (P < .00001), primary tumor size >5 cm (P < .00001), nongastric primarylocation (P = .0001), male sex (P = .01), R1 resection/tumor rupture (P = .0003), andepithelioid cell or mixed cell pathological subtype (P = .05) negatively affected DFS. Inmultivariate analysis, statistically significant factors negatively influencing DFS for model 1were mitotic index >5/50 high-power field (P = .004), primary tumor size >5 cm

Received December 15, 2006; accepted January 17, 2007;published online: May 2, 2007.Address correspondence and reprint requests to: Piotr

Rutkowski, MD, PhD; E-mail: [email protected].

Published by Springer Science+Business Media, LLC � 2007 The Society ofSurgical Oncology, Inc.

Annals of Surgical Oncology 14(7):2018–2027

DOI: 10.1245/s10434-007-9377-9

2018

(P = .001), male sex (P = .003), R1 resection/tumor rupture (P = .04), and nongastricprimary tumor location (P = .02), and for model 2 were high/intermediate risk primarytumor (P < .0001 and P = .008, respectively), male sex (P = .007), resection R1/tumorrupture (P = .01), and nongastric primary tumor location (P = .02). Five-year DFS for high,intermediate, and low/very low risk group was 20%, 54%, and 96%, respectively.

Conclusions: The risk criteria for assessing the natural course of primary GISTs were val-idated, but additional independent prognostic factors—primary tumor location and sex—werealso identified.Key Words: Gastrointestinal stromal tumor—Surgery—Prognosis.

Gastrointestinal stromal tumors (GISTs) comprisea recently defined entity of the most common mes-enchymal neoplasms of the abdominal cavity. GISTsare believed to arise from precursors shared with theinterstitial cells of Cajal, the gut pacemaker cells.1–3

Activating somatic KIT mutations occur in approxi-mately 80% of GISTs, resulting in overexpression ofthe KIT receptor protein and common CD117 im-munopositivity.3–5 However, considerable contro-versy exists regarding GISTs prognosis and outcomesafter primary surgery. GISTs are morphologicallyand clinically heterogeneous tumors, for which thebiological behavior is difficult to predict, rangingfrom clinically benign to malignant.6 A consensusconference held at the National Institutes of Health(NIH) in 20014 provided both an evidence-baseddefinition and a practical scheme for the assessmentof the risk in the clinical course of this disease (Ta-ble 1). The risk categorization is based on evaluationof the size and mitotic rate of the tumors as the mostreliable prognostic factors, and its use is stronglyadvocated. However, the design of trials assessingadjuvant treatment in GIST patients with imatinibhas raised debate about the accuracy of the NIHconsensus criteria for the risk of disease recurrenceafter surgical treatment of primary tumors.7,8

The aim of our study was to analyze the value ofthe NIH criteria and other prognostic factors, as wellas clinicopathologic features, in the relationship todisease recurrences in a large series of primary GISTpatients enrolled onto our tumor registry.

PATIENTS AND METHODS

We prospectively analyzed data from 335 patientswith primary resectable GIST and no evidence ofmetastatic disease enrolled onto Clinical GIST Reg-istry from 2002. In all patients, macroscopically rad-ical resection of the tumor was performed.Clinicopathologic data was supplemented by a reviewof all available medical and histopathological records

from the referring hospitals. The diagnosis of GISTwas confirmed by histopathological review of all cases,and by immunohistochemical staining for CD117(antibodies purchased from Dako, Carpintiera, CA),which was performed in the Department of Pathol-ogy, Maria Sklodowska-Curie Memorial CancerCenter and Institute of Oncology in Warsaw, Poland.Imatinib was not used as an adjuvant treatment in

any of these patients. Patients did not undergo anyfurther selection. Follow-up information was ob-tained during regular outpatient visits or by phonewith the patient and/or the referring physician.Postoperative follow-up consisted of physical exam-ination and standard imaging (computed tomogra-phy of the abdominal cavity and the pelvis, and chestx-ray). Thorough examinations were routinely per-formed every 3 months during the first 2 years, every4 months during the third year, and every 6 months inthe years afterward.Mutational analysis of KIT and platelet derived

growth factor receptor alpha (PDGFR-a) was per-formed in randomly selected 90 cases on the basis ofDNA isolated from paraffin-embedded or fresh frozentumor tissue. KIT exons 9, 11, 13, and 17, andPDGFR-a exons 12 and 18 were amplified by poly-merase chain reaction and prescreened with denatur-ing high-performance liquid chromatography (WAVESystem, Transgenomics). Samples with abnormal

TABLE 1. Risk of aggressive behavior in gastrointestinalstromal tumor according to the National Institutes of Health

criteria

Risk category Tumor size (cm)a MI (/50 HPF)b

Very low risk <2 £5Low risk 2–5 £5Intermediate risk £5 6–10

>5–10 £5High risk >5 >5

>10 Any MIAny size >10

MI, mitotic index; HPF, high-power field.a Size represents the single largest dimension.b Ideally, MI should be standardized according to surface area

examined (based on size of HPF).

PREDICTING RECURRENCE IN PRIMARY GIST 2019

Ann. Surg. Oncol. Vol. 14, No. 7, 2007

elution profiles were bidirectionally sequenced, aspreviously described.9 Molecular analysis study hadbeen approved by the local bioethics committeeaccording to good clinical practice guidelines.During follow-up, we analyzed the incidence and

pattern of disease recurrences. Disease-free survival(DFS) was calculated from the date of primary tumorresection to the date of recurrence or the last follow-up date. All deaths from other causes were recordedas censored. DFS was assessed with respect to thefollowing variables: demographic data (age at diag-nosis £45 years vs. >45 years, and sex), tumor size(£5 cm vs. 5 to 10 cm vs. >10 cm), mitotic rate (0 to 5vs. 6–10 vs. >10 per 50 high-power fields [HPF]),NIH consensus risk criteria4 (combining tumor sizeand mitotic rate: very low/low risk vs. intermediaterisk vs. high risk), histological subtype (spindle cellvs. epithelial cell or mixed cell), primary tumor site(gastric vs. small bowel vs. large bowel vs. others),type of surgical resection (R0—microscopically radi-cal resection vs. R1—microscopically nonradical, butmacroscopically radical resection or tumor intraop-erative rupture), molecular findings (KIT exon 11mutants vs. others). We did not perform an analysisfor overall survival because most patients had beentreated with imatinib after disease recurrence.All statistical analyses were performed by SAS

software (SAS Institute, Cary, NC), version 8.2, andStatistica software (Statsoft, Tulsa, OK). Contin-gency tables were analyzed by the v2 test for non-normal distribution of parameters. Groups werecompared for age differences by the Mann-WhitneyU-test. For the survival analysis, Kaplan-Meier esti-mates were used with generalized Wilcoxon and thelog rank tests for bivariate comparisons. In multi-variate analysis of the factors associated with DFSafter radical surgery of primary GISTs, we used theCox proportional hazard models, applying the step-wise model building procedure that included all co-variates significant at the 20% level in bivariateanalysis. Two-way interactions were then consideredin the model. The scale of numerical covariates wasexamined by the martingale residuals based methodsand fractional polynomial method, with two termsand powers in the set {)2, )1, ).5, .5, 1, 2, 3} and thelog of the variables. To detect the lack of fit of theconsidered models, Gronesby and Borgan�s test wasused.10 Differences were considered statistically sig-nificant if P values were <.05.The goals of our analysis were to validate the NIH

consensus risk group criteria and to identify otherpotential prognostic factors. To accomplish this, wefirst fitted a model containing the NIH consensus

categories for mitotic index (MI) and primary tumorsize (model 1). The next step was to include the NIHconsensus risk categories (model 2). Finally, the thirdapproach was to include the number of mitoses (/50HPF) and the tumor size (cm) as continuous covari-ates (model 3).

RESULTS

Patient Characteristics

The group analyzed consisted of 335 patients (156male [47%] and 179 female [53%]) with a median ageat diagnosis of 58 years (range, 9–89 years). Onehundred four cases with primary inoperable and/ormetastatic disease at presentation (104 [24%] of 439of all GIST cases in our database; data not shown)were not included in the present study. The detailedclinicopathological data of the entire group of pa-tients with primary GISTs are shown in Table 2.The most common symptom at presentation was a

palpable intraabdominal mass (49%; 169 cases), fol-lowed by gastrointestinal bleeding or anemia (18.5%;62 cases) and abdominal pain (18%; 60 cases). In 44patients (13%), the primary operation was performedas an emergency procedure required by perforation ofthe gastrointestinal tract, bowel obstruction, or gas-trointestinal bleeding. In 10 patients (3%), GISTswere resected during surgical procedures performedfor other reasons. The diagnosis of GIST was estab-lished preoperatively in only 20 patients (6%) (theselesions were localized in the stomach in all but onecase). Lymph node metastases were rare and occurredonly in four cases (1.2%). In two patients, primarygastric GISTs were part of the Carney triad, withcoexisting pulmonary multiple chondromas. Threepatients with primary small bowel GISTs showedclinical symptoms of neurofibromatosis type 1.We found that 29 patients (8.7%) had a history of

other malignant neoplasms (24 with metachronousand 5 synchronous neoplasms).

Patterns and Time of Recurrence of the Disease

Median follow-up time was 31 months (range, 4–292 months). During follow-up, we detected 151 casesof disease recurrences (45.1%). Ten (6.6%) were localrecurrences, 9 (6.0%) were local recurrences with livermetastases, 38 (25.2%) were liver metastases only, 1(.6%) was a lung metastases only, 50 (33.1%) wereintraperitoneal dissemination, and 43 (28.5%) wereintraperitoneal disseminations and liver metastases.

P. RUTKOWSKI ET AL.2020


Overall, liver metastases were observed in 81 cases(54%). Median time to disease recurrence was 15months (range, 2–164 months).

Comparisons of Groups with and without Disease

Recurrence

The comparison of clinicopathologic features ofpatients without recurrences during the follow-up

period (recurrent-free [RF] group) with those whohad recurrences (disease recurrent [DR] group) issummarized in Table 2.We found significantly higher median tumor size

(11 cm) and median MI (10/50 HPF) in DR patientsin comparison with the RF group (5 cm and 2/50HPF, respectively) (P < .0001). The percentage oftumors with a gastric location was far lower in theDR group than in the RF group at 30.5% and 58.2%,

TABLE 2. Characteristics of 335 patients with primary resectable gastrointestinal stromal tumors

Clinicopathological featureTotal, n (%)(n = 335)

RF, n (%)(n = 184)

DR, n (%)(n = 151)

P value(RF vs. DR)

Age (y)Median (range) 58 (9–89) 62 (22–85) 53 (9–89)Mean 57 60 54 <.001<45 52 (15.5%) 20 (10.9%) 32 (21.1%)>45 283 (84.5%) 164 (89.1%) 119 (88.9%) .009

SexFemale 179 (53.4%) 106 (57.6%) 73 (48.3%)Male 156 (46.6%) 78 (42.4%) 78 (51.7%) .09

Primary tumor siteStomach 153 (45.7%) 107 (58.2%) 46 (30.5%)Duodenum 16 (4.8%) 12 (6.5%) 4 (2.6%)Small bowel 142 (42.4%) 57 (30.9 %) 85 (56.4%)Large bowel/rectum 13 (3.9%) 4 (2.2%) 9 (5.9%)Other or intraperitoneally withunknown primary origin

11 (3.2%) 4 (2.2%) 7 (4.6%) <.0001

Primary tumor size (cm)Median (range) 8 (.7–40) 5 (.7–40) 11 (2–35)£5 99 (29.6%) 92 (50%) 7 (4.6%)>5–10 122 (36.4%) 51 (27.7%) 71 (47.0%)>10 100 (29.8%) 38 (20.7%) 62 (41.1%)Missing data 14 (4.2%) 3 (1.6%) 11 (7.3%) <.0001

MI (/50 HPF)Median (range) 4 (0–100) 2 (0–100) 10 (0–100)£5 171 (51.1%) 136 (73.9%) 35 (23.2%)6–10 47 (14%) 20 (10.9%) 27 (17.9%)>10 76 (22.7%) 12 (6.5%) 64 (42.4%) <.0001Missing data 41 (12.2%) 16 (8.7%) 25 (16.5%)

Risk groupVery/low 78 (23.3%) 76 (41.3%) 2 (1.3%)Intermediate 68 (20.3%) 47 (25.5%) 21 (13.9%)High 177 (52.8%) 57 (40.0%) 120 (79.5%) <.0001Missing data 12 (3.6%) 4 (2.2%) 8 (5.3%)

Pathological subtypeSpindle cell 143(42.7%) 81 (44%) 42 (27.9%)Epithelial cell or mixed 64 (19.1%) 28 (15.2%) 36 (23.8%) .004Missing data 128 (38.2%) 75(40.8%) 73 (48.3%)

Type of surgical resectionR0 253 (75.5%) 151 (82.1%) 102 (67.5%)R1 75 (22.4%) 29 (15.8%) 46 (30.5%) .001Missing data 7 (2.1%) 4 (2.1%) 3 (2.0%)

Mutational statusa

KIT exon 11 mutation 55 (61.1%) 24 (61.5%) 31 (60.8%)Other genetic changes 35 (38.9%) 15 (38.5%) 20 (39.2%) NS

Emergency operation of primary tumorYes 44 (13.1%) 16 (8.7%) 28 (18.5%)No 244 (72.8%) 135 (73.4%) 109 (72.2%) .02Missing data 47 (14.1%) 33 (17.9%) 14 (9.3%)

RF, patients without disease recurrences during the follow-up period (recurrent-free); DR, patients with disease recurrence; R0, micro-scopically radical resection; R1, microscopically nonradical, but macroscopically radical resection or intraoperative tumor rupture; HPF,high-power field.

a Mutation status was available in 90 cases.



respectively. On the contrary, the percentage of GISTlocated in the bowel (duodenum, small and largeintestine together) was significantly higher in the DRgroup in comparison with the RF group (64.9% and39.6%, respectively; P < .0001).Most patients in the DR group constituted NIH

high risk cases (79.5%), followed by intermediate riskpatients (13.9%). Disease recurrences were observed inonly 2 (2.6%) of 78 patients from the low risk group.Mutations were detected in 71 tumors (78%), of

which 55 were in KIT exon 11 (39 deletions, 6duplications, and 10 missense mutations). We did notobserve any difference with regard to the presence ofdetectable KIT and KIT exon 11 mutations in pri-mary tumors between groups with versus withoutrecurrent disease (65.8% vs. 75% and 60.5% vs.61.5%, respectively).

Factors Influencing Disease Free Survival After

Primary Tumor Resection

The median DFS time after resection of primaryGIST was 37 months, and the estimated 5-year DFSrate was 37.8% (95% confidence interval [95% CI],29.7–45.9) (Fig. 1).

Univariate AnalysisIn bivariate analysis, the following factors dem-

onstrated negative impact on DFS: being in the highor intermediate risk group (P < .000001), MI >5/50HPF (P < .00001), primary tumor size >5 cm(P < .00001), primary nongastric tumor location(P = .0001), male sex (P = .01), R1 resection /tu-mor rupture (P = .0003), and epithelioid cell ormixed cell pathological subtype (P = .05).The categorization of risk groups according to

NIH criteria showed an excellent correlation with

DFS (Fig. 2). DFS according to risk group indicatesthat the 5-year estimated DFS rate was 96% (95% CI,87.3–100.0) for patients with very low/low risk tu-mors, 54% (95% CI, 32.1–75.9) for patients withintermediate risk tumors, and 20% (95% CI, 11.8–27.4) for patients with high risk tumors. The statis-tical difference between these rates is highly signifi-cant. The 5-year DFS rate for patients with primarytumors sized £ 5 cm, >5 to 10 cm, and >10 cm were76.5% (95% CI, 52.1–100), 31.2% (16.6–42.8), and22.5% (11.5–33.6), respectively. The 5-year DFS rateis based on the categorization of primary tumorsaccording to MI £ 5/50 HPF, 6 to 10/50 HPF, and>10/50 HPF was 62.9% (95% CI, 49.5–76.2), 23.7%(4.1–43.3), and 8.0% (.4–15.7), respectively. Patientswith primary gastric tumors had a significantly better5-year DFS rate (59.7% [95% CI, 48.1–71.4]) thanpatients with nongastric tumors (23.5%; 95% CI,13.9–33.1) (Fig. 3). Male patients had 5-year DFSrate of 33.0% (95% CI, 21.5–44.5) compared with42.0% for female patients (95% CI, 30.7–53.2)(Fig. 4). Patients with microscopically nonradical(R1) resection of the primary tumor had a higherprobability of disease recurrence than patients aftermicroscopic radical resection (R0) (5-year DFS rates:29.5% [95% CI, 15.7–43.4] vs. 40.5% [95% CI, 30.8–50.3], respectively). Patients with spindle cell tumorshad longer DFS than patients with other histologicalsubtypes (5-year DFS rates: 50.5% [95% CI, 36.6–64.5] vs. 31.6% [95% CI, 15.1–48.1], respectively).Patients with any type of KIT mutation had a

similar DFS compared with the remaining patients.No statistically significant differences were observedfor DFS in groups of patients with tumors harboringKIT exon 11 mutations versus others (also for sub-groups of KIT exon 11 mutations: missense muta-tions vs. deletions/insertions).

FIG. 1. Disease-free survival (DFS) after primary tumor resection(N = 335). Survival distribution function and 95% confidenceinterval.

FIG. 2. Disease-free survival (DFS) after primary tumor resectionaccording to National Institutes of Health risk groups.



Multivariate AnalysisBy multivariate analysis, we identified five inde-

pendent factors that negatively influence DFS (Ta-ble 3): for model 1, MI >5/50 HPF (P = .004),primary tumor size >5 cm (P = .001), male sex(P = .003), R1 resection/tumor rupture (P = .04)and primary tumor location outside the stomach(P = .02); and for model 2, high/intermediate riskprimary tumor (P < .0001 and P = .008, respec-tively), male sex (P = .007), R1 resection/tumorrupture (P = .01), and nongastric primary tumorlocation (P = .02).The initial model contained the following variables

that were significant in the bivariate analysis: sex, MI(according to the NIH consensus categories), tumorsize (according to the consensus categories), locali-zation (gastric vs. other), and extent of surgery. Ofthe two-way interactions considered, the interactionof the categorized number of mitoses and the cate-gorized tumor size was significant by partial likeli-hood test (P = .0115). The model is summarized inTable 3 (model 1). In model 2 (Table 3), the risk

categories were entered instead of MI and tumor sizein any form.Figure 5 shows the graphical assessment of the

scale of numerical covariates of tumor size andnumber of mitoses. Curves provided by two residual-based methods are displayed along with the linesfitted by categorizing MI and tumor size. The sig-nificance of the deviation from the linear trend wasassessed by modeling the curves with fractionalpolynomials. Both the residual based methods andthe fractional polynomials method indicate that thetumor size (cm) can be entered into the model as alinear covariate without transformation. On the otherhand, the examination of the functional form of thenumber of mitoses resulted in several transforma-tions, producing similar curves that fit the data muchbetter than the linear trend (the best transformationwas a combination of 1/�IM, and ln(IM))/�IM)(model 3, data not shown). According to this analy-sis, an increase of 1 cm in tumor size is associatedwith recurrence risk increase of 7% (95% CI, 3–10).The curve for mitotic is consistent with the NIHcategorization. In practice, both factors contributesimilar risks for values of <10. However, consideringthe range of 10 to 20, the MI confers a greater riskincrease. In contrast, for the numbers >20, risk isrelatively stable for the MI, but still increases fortumor size. This suggests that large tumors with smallnumbers of mitoses have a worse prognosis thansmall tumors with a larger number of mitoses. Thus,it is worthwhile to emphasize that in common prac-tical situations, the NIH risk categories (especially thehigh risk category) seem to underestimate the prog-nostic value of the MI.In all models, there are two additional risk factors

that are consistently associated with worse prognosis:male sex and nongastric localization. Extent of sur-gery was significant in the models, including tumorsize as a categorized variable. However, this is mostprobably the result of residual confounding associ-ated with tumor size (a strong correlation between theR1 and the tumor size) because this variable becomesinsignificant when tumor size is included as anumerical covariate in model 3.Although all three models have comparable R2

values for goodness of fit (data not shown), the testfor lack of fit is significant for model 2, suggestingthat this model might not provide an appropriate fit.The histological type of the tumor was available

for 207 patients. Adjusted hazard ratio for recurrencefor spindle cell as opposed to other tumor types was.90 (P = .7209) and 1.02 (P = .9473) for extendedmodels 1 and 3, respectively, thus not supporting the

FIG. 3. Disease-free survival (DFS) after primary tumor resectionaccording to primary tumor location.

FIG. 4. Disease-free survival (DFS) after primary tumor resectionaccording to patient sex.



hypothesis that the tumor type impacts the DFSprognosis.

DISCUSSION

Gastrointestinal stromal tumors (GISTs) have re-cently beenmorphologically, immunohistochemically,

and molecularly distinguished from other gastroin-testinal tract spindle cell neoplasms of mesenchymalorigin. Radical surgery is the treatment of choice inprimary resectable gastrointestinal stromal tumors,but virtually all GISTs are associated with a risk ofrecurrence, and approximately half of patients withpotentially curative resections develop recurrent ormetastatic disease.11–15 Wide, multivisceral resections

TABLE 3. Summary of multivariate analysis of factors associated with disease-free survival

Variable

Model 1 (n = 274) Model 2 (n = 308)

P value HR HR 95% CI P value HR HR 95% CI

Sex Male vs. female .0034 1.78 1.21 2.61 .0066 1.61 1.14 2.27MI (/50 HPF) and TS (cm) MI 6–10, TS £5 vs. MI £5, TS £5 .0040 12.11 2.21 66.32 – – – –

MI>10, TS £5 vs. MI £5, TS £5 .0073 14.83 2.07 106.24 – – – –MI £5, TS 6–10 vs. MI £5, TS £5 .0384 4.86 1.09 21.73 – – – –MI £5, TS>10 vs. MI £5, TS £5 .0010 11.98 2.74 52.47 – – – –MI 6–10, TS 6–10 vs. MI £5, TS £5 <.0001 20.64 4.67 91.27 – – –MI >10, TS 6–10 vs. MI £5, TS £5 <.0001 29.17 6.86 124.13 – – – –MI 6–10, TS >10 vs. MI £5, TS £5 .0071 9.40 1.84 48.14 – – – –MI >10, TS >10 vs. MI £5, TS £5 .0001 30.74 7.13 132.45 – – – –

Risk group HiR vs. LR – – – – <.0001 21.01 5.14 85.93IR vs. LR – – – – .0083 7.19 1.66 31.11

Localization Gastric vs. other .0269 .62 .41 .95 .0212 .64 .43 .93Extent of surgery R1 vs. R0 .0391 1.53 1.02 2.30 .0109 1.62 1.12 2.35

HR, hazard ratio; 95% CI, 95% confidence interval; MI, mitotic index; TS, tumor size; HiR, high risk, IR, intermediate risk; LR, low risk;R0, microscopically radical resection vs. R1, microscopically nonradical, but macroscopically radical resection or tumor intraoperativerupture.

FIG. 5. Plots for assessing the scale of the mitotic index (MI) and tumor size as numerical variables. (A) Martingale residuals and lowestsmoothed residuals from model. (B) f = log (smoothed censor/smoothed expected) + b · VARIABLE, where b is the estimated coefficientfor the variable of interest (i.e., tumor size on the upper panel and MI on the lower panel). (C) Estimated coefficients for the grouped tumorsize and grouped MI (£5, 6–10, 11–20, >20).



with lymph node dissections are usually not necessarybecause of the low incidence of lymphnodemetastases,which was also observed in our series. Nevertheless,avoiding tumor rupture is imperative. The frequencyof GIST diagnoses has greatly increased since 1992,along with a better understanding of the pathogenesisand biology of GIST and the routine availability ofCD117 antigen immunostaining.16 The survival inadvanced cases has improved since 2000, when imati-nib mesylate was introduced into clinical practice.3,17

The objective of our study was to validate the impor-tance of the NIH risk criteria and other prognosticfactors for the outcome of treatment for primaryGISTin a large, prospectively collected, nonselected groupofpatients in different locations that were included in thecontemporary tumor registry, who had confirmedpositive CD117 immunostaining.Most of the essential studies that determined the

basis for the NIH risk criteria were published beforethe introduction of CD117 immunohistochemistry (acrucial test to confirm the contemporary diagnosisof GIST), and included both primary resectable andovertly malignant tumors,11,12,14,18–21 or were basedon a small number of cases.22,23 To our knowledge,the value of the NIH criteria for primary tumors hasnot been confirmed in a large homogenous series ofprimary resectable GISTs. Miettinen et al.24–26

analyzed the value of risk factors according to theanatomical site of the primary tumor and createdtheir own classifications for risk assessment in gas-tric, duodenal, and intestinal GISTs. On the otherhand, adjuvant trials after resection of primaryGISTs (e.g., European Organization for Researchand Treatment of Cancer trial 62024 or AmericanCollege of the Surgeons Oncology Group trialZ90017) and the Scandinavian study8,27–29 on retro-spective population data called into question thepoorer prognosis in the intermediate risk group,thus raising debate regarding the accuracy of therisk criteria in primary GISTs.Overall, DFS in the group of nonselected GIST

patients in this study was 38% at 5 years, which issimilar to results observed in other published ser-ies.11,20,30,31 However, this disease represents a broadspectrum of different biological behaviors. Completeresection is curative for most low and very low riskpatients, and this group is characterized by anexcellent prognosis. In contrast, the prognosis for thehigh risk group is poor, and surgery alone is insuffi-cient for high-grade tumors. Our study confirms theimportance of stratification of the intermediate riskgroup, the existence of which was criticized in a re-cent population-based study.8 In our opinion, pa-

tients with an intermediate risk of recurrence shouldalso be recommended for adjuvant trials with smallmolecule inhibitors, such as imatinib, because the riskof recurrence is relatively high after long-term follow-up. Therefore, the stratification of risk categories(intermediate and high) is warranted in such trials.The impact of classical prognostic factors on DFS

in the current study is consistent with the results ofother studies.11,14,30,32 Thus, we confirm the role ofMI and tumor size as the most valuable ‘‘classic’’prognostic factors for primary GISTs, as included inthe NIH risk criteria.4 However, our data suggestthat the value of the MI is underestimated for tumorsof <20 cm in the largest dimension, and should beconsidered as the most important variable. On theother hand, tumor size >20 cm predicts the greatestrisk. Moreover, the estimation of tumor size gives thebest prognostic prediction when considered as acontinuous variable, and in the case of larger tumors,the value of microscopic radicalism of surgicaltreatment diminishes.We also confirm the effect of the anatomic site of

the primary lesion on the prognosis in GIST, whichwas previously suggested by other studies.3,30,33 Pri-mary tumors in the gastric area have a far betterprognosis those that in a nongastric location, inde-pendent of tumor size and MI. It is unknown, how-ever, the degree to which this may be because tumorsoutside the stomach can growth asymptomatically fora longer time. Nevertheless, the stomach was themost common primary site of tumors in our series,whereas the small intestine was the most frequentprimary location in the group of patients who expe-rienced recurrence.No statistically significant correlation between

mutation status and the disease recurrence after sur-gical treatment of primary GISTs was found, even insubgroups with common exon 11 KIT mutations,which were present in two-thirds of the cases in ourseries, or exon 11 missense mutations versus dele-tions/insertions, as suggested by some research-ers31,34–36 despite some disclaimers by others.37–39 Wecannot neglect the bias related to the small number ofcases with confirmed mutational status; however,KIT mutations are acquired early in GIST develop-ment, so they can occur in low and very low risktumors.We did not analyze the role of DNA ploidy,

immunohistochemical staining for cell proliferationantigens, or other factors involved in regulation ofthe cell cycle (p53, Ki-67, p21, or BCL-2) indicated byother authors as important prognostic factors.40–45

This was because our goal was to establish a set of



factors that could be used in routine practice. Otherauthors also indicated the significance of factors suchmucosal invasion or tumor necrosis as indicators forrisk of malignant behavior, but these factors arepoorly reproducible.GIST recurrence can be observed late after treat-

ment for the primary tumor, which indicates thenecessity for a longer follow-up than the 5 yearsusually set as the cutoff point for most patients withsarcoma. Low malignant potential GIST should befollowed up on a yearly basis because recurrence israre in this group of patients. However, patientsshould be informed about the likelihood of late re-lapse, because no form of GIST can be labeled trulybenign. In higher risk patients, recurrences will occurcontinuously for years after surgical resection(Fig. 2). Because symptomatology does not provideearly diagnostic clues, follow-up with computedtomography imaging of the abdomen and pelvis(because recurrences typically occur first in the liveror peritoneum2,3) is recommended every 3 to 6months for at least 5 years in high and intermediaterisk groups.In summary, although the NIH criteria provide an

excellent estimation of tumor behavior, physiciansshould also be familiar with other factors that influ-ence a GIST patient�s prognosis. Our data from alarge series of patients undergoing resection of GISTtumors confirm that the NIH risk criteria (which, asshown in Fig. 2, perform quite well to categorize riskof recurrence), with slightly more emphasis on MI asa discrete variable and on tumor size as a continuousvariable. Additional prognostic factors identified herewere tumor location (gastric vs. other) and patientsex. All identified or confirmed prognostic variablesare simple and easily applicable in the everydaypractices of physicians treating patients with GIST.

ACKNOWLEDGMENTS

Supported by the Polish State Committee for Sci-entific Research, grant 3P05C05925. We thank alldoctors devoted to GIST treatment and diagnosis whocollaborated in the Polish Clinical GIST Registry, M.Rosinska for statistical advice, and J. Lasota (ArmedForces Institute of Pathology, Washington, DC) forcritical remarks and mutational analyses data.

REFERENCES

1. Kindblom LG, Remotti HE, Aldenborg F, Meis-Kinblom JM.Gastrointestinal pacemaker cell tumor (GIPACT): gastroin-

testinal stromal tumors show phenotypic characteristic of theinterstitial cell of Cajal. Am J Pathol 1998; 152:1259–69.

2. Miettinen M, Lasota J. Gastrointestinal stromaltumors—definition, clinical, histological, immunohistochemi-cal, and molecular genetic features and differential diagnosis.Virchows Arch 2001; 438:1–12.

3. Trent JC, Benjamin RS. New developments in gastrointestinalstromal tumor. Curr Opin Oncol 2006; 18:386–95.

4. Fletcher CDM, Berman JJ, Corless C, et al. Diagnosis ofgastrointestinal stromal tumors: a consensus approach. HumPathol 2002; 33:459–65.

5. Miettinen M, Majidi M, Lasota J. Pathology and diagnosticcriteria of gastrointestinal stromal tumors (GISTs): a review.Eur J Cancer 2002; 38(Suppl 5):S39–51.

6. Miettinen M, El-Rifai W, Sobin LH, Lasota J. Evaluation ofmalignancy and prognosis of gastrointestinal tumors: a review.Hum Pathol 2002; 33:478–83.

7. Eisenberg BL, Judson I. Surgery, imatinib in the managementof GIST: emerging approaches to adjuvant and neoadjuvanttherapy. Ann Surg Oncol 2004; 11:465–75.

8. Nilsson B, Bumming P, Meis-Kindblom JM, et al. Gastroin-testinal stromal tumors: the incidence, prevalence, clinicalcourse, and prognostication in pre-imatinib mesylate era.Cancer 2005; 103:821–9.

9. Debiec-Rychter M, Dumez H, et al. EORTC Soft Tissue andBone Sarcoma Group. Use of c-KIT/PDGFRA mutationalanalysis to predict the clinical response to imatinib in patientswith advanced gastrointestinal stromal tumours entered onphase I and II studies of the EORTC Soft Tissue and BoneSarcoma Group. Eur J Cancer 2004; 40:689–95.

10. Hosmer DW, Lemeshow S. Applied Survival Analysis:Regression Modeling of Time to Event Data New York: JWiley and Sons; 1999.

11. DeMatteo RP, Lewis JJ, Leung D, Mudan SS, Woodruff JM,Brennan MF. Two hundred gastrointestinal stromal tumors:recurrence patterns and prognostic factors for survival. AnnSurg 2000; 231:51–8.

12. Kim CJ, Day S, Yeh KA. Gastrointestinal stromal tumors:analysis of clinical and pathologic factors. Am Surg 2001;67:135–7.

13. Hinz S, Pauser U, Egberts JH, Schafmayer C, Tepel J, Fand-rich F. Audit of a series of 40 gastrointestinal stromal tumourcases. Eur J Surg Oncol 2006; 32:1125–9.

14. Pierie JP, Choudry U, Muzikansky A, et al. The effect ofsurgery and grade on outcome of gastrointestinal stromaltumors. Arch Surg 2001; 136:383–9.

15. Ng E, Pollock RE, Munsell MF, et al. Prognostic factorsinfluencing survival in gastrointestinal leiomyosarcomas. AnnSurg 1992; 215:68–77.

16. Perez EA, Livingstone AS, Franceschi D, et al. Current inci-dence and outcomes of gastrointestinal mesenchymal tumorsincluding gastrointestinal stromal tumors. J Am Coll Surg2006; 202:623–9.

17. Demetri GD, von Mehren M, Blanke CD, et al. Efficacy andsafety of imatinib mesylate in advanced gastrointestinal stro-mal tumors. N Engl J Med 2002; 347:472–80.

18. Bucher P, Egger JF, Gervaz P, et al. An audit of surgicalmanagement of gastrointestinal stromal tumours (GIST). Eur JSurg Oncol 2006; 32:310–4.

19. Clary BM, Dematteo RP, Lewis JJ, et al. Gastrointestinalstromal tumors and leiomyosarcomas of the abdomen andretroperitoneum: a clinical comparison. Ann Surg Oncol 2001;8:290–99.

20. Crosby JA, Catton CN, Davis A, et al. Malignant gastroin-testinal stromal tumors of the small intestine: a review of 50cases from a prospective database. Ann Surg Oncol 2001; 8:50–9.

21. Miettinen M, Furlong M, Sarlomo-Rikala M, Burke A, SobinLH, Lasota J. Gastrointestinal stromal tumors, intramuralleiomyomas, and leiomyosarcomas in the rectum and anus. A



clinicopathologic, immunohistochemical, and molecular ge-netic study of 144 cases. Am J Surg Pathol 2001; 25:1121–33.

22. Aparicio T, Boige V, Sabourin JC, et al. Prognostic factorsafter surgery of primary resectable gastrointestinal stromaltumours. Eur J Surg Oncol . 2004; 30:1098–103.

23. Bucher P, Taylor S, Villiger P, Morel P, Brundler MA. Arethere any prognostic factors for small intestinal stromal tu-mors?. Am J Surg 2004; 187:761–6.

24. Miettinen M, Sobin LH, Lasota J. Gastrointestinal stromaltumors of the stomach: a clinicopathologic, immunohisto-chemical, and molecular genetic study of 1765 cases with long-term follow-up. Am J Surg Pathol 2005; 29:52–68.

25. Miettinen M, Kopczynski J, Makhlouf HR, et al. Gastroin-testinal stromal tumors, intramural leiomyomas, and leio-myosarcomas in the duodenum: a clinicopathologic,immunohistochemical, and molecular genetic study of 167cases. Am J Surg Pathol . 2003; 27:625–41.

26. Miettinen M, Makhlouf H, Sobin LH, Lasota J. Gastrointes-tinal stromal tumors of the jejunum and ileum: a clinicopath-ologic, immunohistochemical, and molecular genetic study of906 cases before imatinib with long-term follow-up. Am J SurgPathol 2006; 30:477–89.

27. Kindblom LG, Meis KindBlom J, Bumming P, et al. Incidence,prevalence, phenotype and biologic spectrum of gastrointesti-nal stromal tumors (GIST): a population based study of 600cases. Ann Oncol 2003; 13:157.

28. Buemming P, Meis-Kindblom JM, Kindblom LG, et al. Isthere an indication for adjuvant treatment with imatinib mes-ylate in patients with aggressive gastrointestinal stromal tu-mors (GISTs)? (abstract 3289). Proc Am Soc Clin Oncol 2003;22:818.

29. Bumming P, Ahlman H, Andersson J, Meis-Kindblom JM,Kindblom LG, Nilsson B. Population-based study of thediagnosis and treatment of gastrointestinal stromal tumours.Br J Surg 2006; 93:836–43.

30. Emory TS, Sobin LH, Lukes L, Lee DH, O�Leary TJ. Prog-nosis of gastrointestinal smooth-muscle (stromal) tumors:dependence on anatomic site. Am J Surg Pathol 1999; 23:82–7.

31. Singer S, Rubin BP, Lux ML, et al. Prognostic value of KITmutation type, mitotic activity and histologic subtype in gas-trointestinal stromal tumors. J Clin Oncol 2002; 20:3898–905.

32. Fujimoto Y, Nakanishi Y, Yoshimura K, Shimoda T. Clini-copathologic study of primary malignant gastrointestinalstromal tumor of the stomach, with special reference to prog-nostic factors: analysis of results in 140 surgically resectedpatients. Gastric Cancer 2003; 6:39–48.

33. Ueyama T, Guo KJ, Hashimoto H, et al. A clinicopathologicand immunohistochemical study of gastrointestinal stromaltumors. Cancer 1992; 69:947–55.

34. Andersson J, Bumming P, Meis-Kindblom JM, et al. Gastro-intestinal stromal tumors with KIT exon 11 deletions areassociated with poor prognosis. Gastroenterology 2006;130:1573–81.

35. Kim TW, Lee H, Kang YK, et al. Prognostic significance of c-kit mutation in localized gastrointestinal stromal tumors. ClinCancer Res 2004; 10:3076–81.

36. Taniguchi M, Nishida T, Hirota S, et al. Effect of c-kitmutation on prognosis of gastrointestinal stromal tumors.Cancer Res 1999; 59:4297–300.

37. Corless CL, McGreevey L, Haley A, Twon A, Heinrich MC.KIT mutations are common in incidental gastrointestinalstromal tumors one centimeter or less in size. Am J Pathol2002; 160:1567–72.

38. Corless CL, Fletcher JA, Heinrich MC. Biology of gastroin-testinal stromal tumors. J Clin Oncol 2004; 22:3813–25.

39. Lasota J, Miettinen M. KIT and PDGFRA mutations ingastrointestinal stromal tumors (GISTs). Sem Diagn Pathol2006; 23:91–102.

40. Carillo R, Candia A, Rodriguez-Peralto JL, Caz V. Prognosticsignificance of DNA ploidy and proliferative index (MIB-1index) in gastrointestinal stromal tumors. Hum Pathol 1997;28:160–5.

41. Ozguc H, Yilmazlar T, Yerci O, et al. Analysis of prognosticand immunohistochemical factors in gastrointestinal stromaltumors with malignant potential. J Gastrointest Surg 2005;9:418–29.

42. Cunningham RE, Federspiel BH, McCarthy WF, Sobin LH,O�Leary TJ. Predicting prognosis of gastrointestinal smoothmuscle tumors. Role of clinical and histologic evaluation, flowcytometry, and image cytometry. Am J Surg Pathol 1993;17:588–94.

43. Liu FY, Qi JP, Xu FL, Wu AP. Clinicopathological andimmunohistochemical analysis of gastrointestinal stromal tu-mor. World J Gastroenterol 2006; 12:4161–5.

44. Hu TH, Chuah SK, Lin JW, et al. Expression and prognosticrole of molecular markers in 99 KIT-positive gastric stromaltumors in Taiwanese. World J Gastroenterol 2006; 12:595–602.

45. Wong NA, Young R, Malcomson RD, et al. Prognostic indi-cators for gastrointestinal stromal tumours: a clinicopatho-logical and immunohistochemical study of 108 resected casesof the stomach. Histopathology 2003; 43:118–26.



Risk Criteria and Prognostic Factors for Predicting Recurrences After Resection of Primary...

Documents

Transcript of Risk Criteria and Prognostic Factors for Predicting Recurrences After Resection of Primary...