This document describes the development of a risk score system to predict tumor-node-metastasis (TNM) stages in gastric cancer patients. The risk score incorporates 4 independent predictive factors identified through logistic regression analysis: serum albumin levels, tumor size, and T and N categories determined by computed tomography scans. The risk score demonstrated good discriminative ability to differentiate more advanced (stage III/IV) vs less advanced (stage I/II) disease, with a sensitivity of 79.6%, specificity of 85.2%, and accuracy of 82.4%. The risk score may help surgeons determine the appropriate extent of surgery and use of adjuvant therapies for gastric cancer patients.

1. ORIGINAL ARTICLE
A Risk Score System to Preoperatively Predict TNM Stages in
Gastric Cancer
Yixin Chen, MD,* and Linjun Mou, MD†
the likelihood of R0 resection.4,5 Thus accurate preoperative staging
Background: To improve the estimation of tumor status and facilitate the
is essential to treatment decision-making.
stage-dependent treatment planning, we developed a reliable and easy-to-use
Currently, endoscopic ultrasonography (EUS) and helical
risk score for prediction of tumor-node-metastasis stages in gastric cancer.
computerized tomography (CT) are the mainstays in estimating
Methods: Clinicopathological data were collected prospectively from 108
tumor-node-metastasis (TNM) stages of gastric cancer. The diag-
curatively resected patients with gastric cancer. The risk score was
nostic accuracy of EUS ranged between 67% and 92% for T staging,
established on the basis of independent predictive factors for tumor
and between 50% and 87% for N staging.7 Helical CT achieved
stages, and its performance was evaluated by receiver operating charac-
similar results, with accuracy of 55% to 87% for T staging and 52%
teristic (ROC) analysis.
to 71% for N staging.8 –11 Because the reported accuracy from both
Results: The following 4 independent factors were included in our score:
imaging procedures was quite variable, there is a need for a better
serum albumin levels, tumor size, T and N categories determined by helical
and more reliable method to evaluate the extent of tumor. Besides
computed tomography. Using ROC analysis, we chose a score at 7 as the
imaging findings, some other clinicopathological information is also
optimal cut point for differentiating the more advanced disease (stage III/IV)
necessary to be taken into account when assessing the stage of
from the less advanced one (stage I/II). With the defined cut point, our score
disease.12 To improve the estimation of tumor status and facilitate
allowed predicting stage III/IV with sensitivity of 79.6%, specificity of
the stage-dependent treatment planning, we established a reliable
85.2%, and accuracy of 82.4%. The discriminative ability of this score was
and easy-to-use risk score for prediction of TNM stages in gastric
good (the area under the ROC curve, 0.861– 0.965).
cancer. The risk score, which incorporated imaging findings as well
Conclusions: The risk score may be helpful to preoperative gastric cancer
as clinicopathological features, was developed from a prospective
staging. It probably assists surgeons in deciding the extent of surgery and in
cohort of 108 patients with gastric cancer.
choosing the appropriate perioperative adjuvant therapies.
Key Words: gastric cancer, TNM stages, risk score, helical computerized MATERIALS AND METHODS
tomography
(Am J Clin Oncol 2010;XX: 000 – 000) Patients
A series of 125 consecutive patients underwent curative
gastrectomy (R0 resection according to the International Union
Against Cancer UICC 13) for primary gastric cancer at the Depart-
G astric cancer is the second leading cause of cancer-related death
worldwide.1 Despite the advances in surgical techniques, the
prognosis of advanced gastric cancer (stage III/IV) remains unsat-
ment of Surgery, First Affiliated Hospital of Zhejiang University
from March 2006 to July 2007. To be eligible for participation,
patients had to have histologically confirmed adenocarcinoma of the
isfactory.2,3 A recent survey on 11,491 patients with gastric cancer stomach without previous or coexisting cancer. None had received
showed that stage-specific 5-year survival rates were 92.9% for preoperative radiation therapy or chemotherapy. Patients with con-
stage Ia, 84.2% for stage Ib, 69.3% for stage II, 45.8% for stage IIIa, comitant disease suspected of affecting hemoglobin concentration,
29.6% for stage IIIb, and 9.2% for stage IV. The recurrence rate platelet count, and serum albumin levels (ie, severe inflammatory
after curative gastrectomy (UICC R0 resection) was reported to be conditions, diabetes mellitus or metabolic syndrome) were excluded
approaching 60%.3 These poor outcomes obtained by surgery alone from the study. Overall 108 eligible patients were enrolled for
have prompted investigation of novel therapeutic strategies. construction of the score. There were 74 men and 34 women with a
R0 resection combined with perioperative adjuvant therapy median age of 59 years (ranging from 32 to 85 years). All clinico-
has become a promising treatment modality for gastric cancer.4,5 On pathological features were prospectively documented, including
the one hand, patients with gastric cancer with R0 resection had a far demographic data, tumor (T) and node (N) categories determined by
better 5-year survival rate than those with noncurative operation helical CT, preoperative hemoglobin concentration, platelet count,
(55.5% vs. 18.4%).6 R0 resection is assumed to be the primary goal serum albumin values, tumor size and location estimated by gastro-
of patient management. An ideal way of achieving R0 resection is to intestinal endoscopy, histologic differentiation and pathologic TNM
perform stage-dependent surgery for each patient, in whom the (pTNM) stages. As for the histologic classification, well and mod-
extent of surgical resection was determined by the tumor stage. On erately differentiated tubular adenocarcinoma and papillary adeno-
the other hand, perioperative adjuvant therapy was found to have a carcinoma were grouped as differentiated type; poorly differentiated
beneficial effect on locally advanced gastric cancer (stage III/IV). It adenocarcinoma, signet ring cell carcinoma and mucinous adeno-
could induce tumor downstaging and thereby potentially increase carcinoma were grouped as undifferentiated type.14 Postoperative
pTNM classification followed the 1997 UICC criteria.13
From the *Department of Surgery, Hangzhou Traditional Chinese Medicine
The institutional review board approved this protocol and all
Hospital, China; and †Department of Surgery, First Affiliated Hospital, participants gave written informed consent.
Zhejiang University School of Medicine, China.
Reprints: Linjun Mou, MD, Department of Surgery, First Affiliated Hospital, CT Imaging
Zhejiang University School of Medicine, #79 Qingchun Rd, Hangzhou, Helical CT scanning with 5 mm section thickness was per-
People s Republic of China 310003. E-mail: chenyx1999@126.com.
Copyright © 2010 by Lippincott Williams & Wilkins
formed on all patients by a Somatom Plus 4 scanner (Siemens,
ISSN: 0277-3732/10/0000-0001 Erlangen, Germany). The patients had to fast for at least 8 hours
DOI: 10.1097/COC.0b013e3181d31eeb before the examination. After receiving an intravenous infusion of
American Journal of Clinical Oncology • Volume XX, Number X, XXX 2010 www.amjclinicaloncology.com | 1

2. Chen and Mou American Journal of Clinical Oncology • Volume XX, Number X, XXX 2010
20 mg scopolamine butylbromide, each patient drank 800 mL or predictors accompanied by their regression coefficients. The
more water. An unenhanced scan was done to evaluate gastric calibration of the model, that described agreement between expected
distention. The enhanced scan was obtained following intravenous and actually observed outcomes, was measured by the Hosmer-
injection of 100 mL nonionic iodinated contrast at a rate of 3 mL/s. Lemeshow goodness-of-fit test.21 A P value of less than 0.05 was
On CT images, the gastric wall was considered thickened considered statistically significant.
when it exceeded 3 mm in the antrum, 5 mm in the body, and 7 mm Based on the multiple logistic model, an easy-to-use risk
in the fundus.9 The depth of tumor invasion visualized by CT was score was created. A partial score was computed for each indepen-
classified as follows: (1) T0, no evidence of alteration of the gastric dent variable according to the relative weight of significance in the
wall with a normal fat plane; (2) T1, focal thickening of the inner multivariate analysis. The regression coefficients pertinent to the
layer with a visible outer layer of the gastric wall and a clear fat variables retained in the final model were divided by the smallest
plane surrounding the tumor; (3) T2, focal or diffuse thickening of regression coefficient, and the results were rounded to the nearest
the gastric wall with transmural involvement and a smooth outer integer.22 The sum of partial scores produced the overall risk score
border of the wall or only a few small linear strands of soft tissue for individual patients. Receiver operating characteristic (ROC)
extending into the fat plane involving less than one-third of the
tumor extent; (4) T3, transmural tumors with obvious blurring of at
least one-third of the tumor extent or wide reticular strands sur-
rounding the outer border of the tumor; (5) T4, obliteration of fat TABLE 1. Relationship Between pTNM Stages and
plane between the gastric tumor and an adjacent organ or invasion of Clinicopathological Features in 108 Patients With Gastric
an adjacent organ.15 Cancer (Univariate Analysis)
Lymph nodes were considered to be metastatic if they were Stage Stage
larger than 10 mm in diameter.16 N1 was defined as enlarged Variables I/II III/IV P
perigastric lymph nodes within 3 cm of the primary lesion. N2 was
Age (yr) NS*
defined as enlarged lymph nodes more than 3 cm from the primary
lesion, including those located along the left gastric, common 60 28 (50.0) 27 (49.1)
hepatic, splenic or celiac arteries.15 Two experienced abdominal 60 26 (49.1) 27 (50.9)
radiologists, who were blinded to endoscopic findings, indepen- Gender NS
dently read CT images. Disagreements between readers were re- Male 41 (55.4) 33 (44.6)
solved through consensus. Female 13 (38.2) 21 (61.8)
Tumor location NS
Preoperative Laboratory Tests Low 28 (46.7) 32 (53.3)
Hemoglobin concentration, platelet count, and serum albumin Middle 20 (58.8) 14 (41.2)
values were examined in early morning samples taken before break-
Upper 5 (50.0) 5 (50.0)
fast on the second day after admission and immediately measured by
Two-third or more 1 (25.0) 3 (75.0)
an Auto-Analyzer method.17 Anemia, thrombocytosis, and hy-
poalbuminaemia were defined as hemoglobin level less than 12.0 Tumor size P 0.05
g/dL, platelet count more than 40 ( 104/ L), and serum albumin 2.0 cm 26 (89.7) 3 (10.3)
concentration less than 3.5 g/dL, respectively.18 –20 2.1–4.0 cm 16 (57.1) 12 (42.9)
4.1–6.0 cm 9 (26.5) 25 (73.5)
Surgery 6.0 cm 3 (17.6) 14 (82.4)
The following standardized operative procedures were per- T category (on CT scans) P 0.05
formed: (1) total or subtotal gastrectomy was performed, depending T0/1 87 (84.5) 1 (1.0)
on the location and macroscopic type of gastric cancer; and (2) T2 56 (35.7) 27 (17.2)
extended D2 lymphadenectomy was performed according to the
T3 20 (9.8) 71 (34.6)
rules established by the Japanese Research Society for Gastric
Cancer.14 Perigastric lymph nodes (No. 1– 6) and nodes along the T4 1 (14.3) 3 (42.9)
left gastric artery (No. 7), along the common hepatic artery (No. 8), N category (on CT scans) P 0.05
around the celiac axis (No. 9), and along the splenic artery (No. 11) N0 37 (75.5) 12 (24.5)
were resected routinely. Lymph nodes in the hepatoduodenal liga- N1 13 (43.3) 17 (56.7)
ment (No. 12a) and along the superior mesenteric vein (No. 14v) N2 4 (13.8) 25 (86.2)
were resected when the primary lesion was located in the lower third Histological type P 0.05
of the stomach. Lymph nodes at the splenic hilum (No. 10) were Differentiated 23 (67.6) 11 (32.4)
resected with the spleen when total or proximal gastrectomy was Undifferentiated 31 (41.9) 43 (58.1)
performed. A median of 26 lymph nodes (range, 15– 60 nodes) were Anemia P 0.05
retrieved for histologic examination.
No 38 (65.5) 20 (34.5)
Statistical Analysis Yes 16 (32.0) 34 (68.0)
All statistical calculations were carried out using Statistical Thrombocytosis NS
Package for the Social Sciences 10.0 for Windows (Chicago, IL). No 41 (47.7) 45 (52.3)
Univariate analysis was employed to investigate whether imaging Yes 13 (59.1) 9 (40.9)
and clinicopathological parameters correlated with tumor stages. On Hypoalbuminaemia P 0.05
multivariate analysis, a stepwise logistic regression model was No 52 (55.9) 41 (44.1)
adopted to identify independent predictive factors for tumor stages, Yes 2 (13.2) 13 (86.7)
in which the pTNM stage was chosen as a dependent variable and
significant risk factors selected by univariate analysis were chosen Values in the parenthesis are percentages.
*NS indicates not significant.
as covariates. The final model generated a set of independent
2 | www.amjclinicaloncology.com © 2010 Lippincott Williams & Wilkins

3. American Journal of Clinical Oncology • Volume XX, Number X, XXX 2010 A Score to Predict TNM Stages
analysis was used to define a cut point of the risk score for optimal
sensitivity and specificity in the prediction of stage III/IV.23 The TABLE 3. A Risk Score System for Patients With Gastric
optimal cut point was selected on the basis of Youden’s index (J), Cancer
which was calculated using the equation: J sensitivity speci- Variables Partial Score
ficity 1.24 The area under the ROC curve (AUC) reflected the
T category (on CT scans)
ability of our score to discriminate patients with stage III/IV disease
T0/1 0
from those with stage I/II disease.25
T2 2
RESULTS T3 4
T4 5
Univariate Analysis of the Relationship Between N category (on CT scans)
pTNM Stages and Clinicopathological Parameters N0 0
Table 1 lists the clinicopathological characteristics of 108 N1 2
patients with gastric cancer. Histopathologically, there were 28 N2 3
(25.9%) cases in stage I, 26 (24.1%) in stage II, 39 (36.1%) in stage III, Tumor size
and 15 (13.9%) in stage IV. The helical CT detected 94.4% (102/108)
2.0 cm 0
of the primary tumors. All 6 cases missed by helical CT were pT1
2.1–4.0 cm 1
disease. Univariate analysis demonstrated that CT category, estimated
tumor nodes (CN) category, tumor size, histologic differentiation, 4.1–6.0 cm 2
anemia and hypoalbuminaemia had a significant association with 6.0 cm 3
pTNM stages (Table 1). Hypoalbuminaemia
No 0
Multivariate Analysis of Predictive Factors for Yes 2
Tumor Stages
The 6 significant variables identified by univariate analysis
were included in a stepwise logistic regression model. The final
model revealed that CT category, CN category, tumor size and
hypoalbuminaemia were independent predictors of stage III/IV; the
corresponding regression coefficients, standard error (SE) of the
coefficients and P values were presented in Table 2. The Hosmer-
Lemeshow goodness-of-fit test was performed to assess the calibra-
tion of the logistic regression model. The Hosmer-Lemeshow sta-
tistic indicated a good fit of the final model (P 0.751).
Computation of the Score
The risk score was established on the basis of the 4 variables
retained in the final model. As shown in Table 3, a numerical score
was designated for each independent variable. The total score of a
given patient was obtained by adding his appropriate partial scores.
The mean SE value of the overall risk score was 5.7 (0.3) (range,
TABLE 2. Multivariate Analysis of Factors Related to pTNM
Stages
Independent Factors SE ( ) P
T category (on CT scans) P 0.05
T0/1
T2 2.175 1.556
T3 3.670 1.572
T4 5.132 2.060
N category (on CT scans) P 0.05
N0 FIGURE 1. The ROC curve for our risk score’s prediction of
stage III/IV.
N1 1.541 0.714
N2 2.626 0.818
Tumor size P 0.05 0 –13). When patients were stratified by pTNM stages, the mean SE
2.0 cm score values were 1.8 (0.4) in stage I, 5.3 (0.4) in stage II, 7.6 (0.3)
2.1–4.0 cm 0.969 0.907 in stage III, and 9.1 (0.6) in stage IV.
4.1–6.0 cm 1.775 0.921 Using ROC analysis, a cut point at 7 was found to be
6.0 cm 3.264 1.344 associated with optimal sensitivity and specificity of 79.6% and
Hypoalbuminaemia P 0.05 85.2%, respectively, in the prediction of stage III/IV. The discrim-
No inative ability of our score was good (stage III/IV vs. stage I/II, AUC
Yes 2.344 1.113 0.861– 0.965; Fig. 1). If the total score was less than 7, the patient
was classified as stage I/II; whether it was equal to or greater than 7,
© 2010 Lippincott Williams & Wilkins www.amjclinicaloncology.com | 3

4. Chen and Mou American Journal of Clinical Oncology • Volume XX, Number X, XXX 2010
the patient was classified as stage III/IV. Accordingly, 57 cases REFERENCES
(52.8%) were regarded in stage I/II and 51 (47.2%) in stage III/IV. 1. Boring CC, Squires TS, Tong T, et al. Cancer statistics, 1994. CA Cancer
We compared predictions from the risk score with postoperative J Clin. 1994;44:7–26.
pTNM stages. The overall accuracy was 82.4%. 2. Wanebo HJ, Kennedy BJ, Chmiel J, et al. Cancer of the stomach. A patient
care study by the American College of Surgeons. Ann Surg. 1993;218:583–
592.
3. Kim JP. Surgical results in gastric cancer. Semin Surg Oncol. 1999;17:132–
DISCUSSION 138.
So far, scoring systems have been frequently used to predict 4. Crookes P, Leichman CG, Leichman L, et al. Systemic chemotherapy for
the outcome of patients with gastric cancer.12,22,26 –30 Most of these gastric carcinoma followed by postoperative intraperitoneal therapy: a final
prognostic scoring systems took into account the postoperative report. Cancer. 1997;79:1767–1775.
pathologic properties of the tumor, so they did not work during 5. Cunningham D, Allum WH, Stenning SP, et al. Perioperative chemotherapy
preoperative decision-making. The Maruyama computer program versus surgery alone for resectable gastroesophageal cancer. N Engl J Med.
2006;355:11–20.
was one of the few systems that used preoperative variables solely
6. Doglietto GB, Pacelli F, Caprino P, et al. Surgery: independent prognostic
to calculate the incidence of lymph node metastasis as well as the factor in curable and far advanced gastric cancer. World J Surg. 2000;24:
survival time. It predicted nodal metastasis with accuracy of 80.6% 459 – 463.
to 86.1% in stations 1 to 6 and of 66.7% to 83.7% in stations 7 to 16. 7. Rosch T. Endosonographic staging of gastric cancer: a review of literature
The computer program was relatively complicated, which required results. Gastrointest Endosc Clin N Am. 1995;5:549 –557.
the input of 7 clinicopathological parameters.31 Instead, a simplified 8. Lee DH, Seo TS, Ko YT. Spiral CT of the gastric carcinoma: staging and
scoring system for patients with gastric cancer was formulated by enhancement pattern. Clin Imaging. 2001;25:32–37.
Wu et al.32 The system estimated N stages with overall accuracy of 9. D Elia F, Zingarelli A, Palli D, et al. Hydro-dynamic CT preoperative staging
of gastric cancer: correlation with pathological findings. A prospective study
65.6%. It derived from a retrospective series of patients, and con- of 107 cases. Eur Radiol. 2000;10:1877–1885.
tained no variables about imaging findings.32
10. Lee DH, Ko YT, Park SJ, et al. Comparison of hydro-US and spiral CT in the
To further improve the preoperative gastric cancer staging, staging of gastric cancer. Clin Imaging. 2001;25:181–186.
we focused on development of a risk score that combined imaging 11. Chen CY, Hsu JS, Wu DC, et al. Gastric cancer: preoperative local staging
and clinicopathological information to predict TNM stages in indi- with 3D multi-detector row CT— correlation with surgical and histopatho-
vidual cases. For this purpose, the present study prospectively logic results. Radiology. 2007;242:472– 482.
collected imaging and clinicopathological data from 108 patients 12. Kampschoer GH, Maruyama K, van de Velde CJ, et al. Computer analysis in
with gastric cancer. A stepwise logistic regression model was built making preoperative decisions: a rational approach to lymph node dissection
in gastric cancer patients. Br J Surg. 1989;76:905–908.
to investigate the independent effect of each potential predictive
13. Sobin LH, Wittekind C; International Union Against Cancer (UICC). TNM
factor. Up to now EUS is not a routine work-up at our hospital. With Classification of Malignant Tumors. 5th ed. New York, NY: Wiley-Liss;
the help of helical CT as well as other conventional tests, we 1997:59 – 62.
designed an integrated score comprising of 4 easily accessible 14. Japanese Gastric Cancer Association. Japanese classification of gastric car-
variables: CT category, CN category, tumor size, and serum albumin cinoma—2nd english edition. Gastric Cancer. 1998;1:10 –24.
levels. Unlike the Maruyama computer program which attempted to 15. Habermann CR, Weiss F, Riecken R, et al. Preoperative staging of gastric
characterize the probability of metastasis at each lymph node station, adenocarcinoma: comparison of helical CT and endoscopic US. Radiology.
this newly proposed score relied on a set of variables to evaluate the 2004;230:465– 471.
overall tumor status.12,31 16. Davies J, Chalmers AG, Sue-Ling HM, et al. Spiral computed tomography
and operative staging of gastric carcinoma: a comparison with histopatholog-
Clinically, the prognosis of patients with gastric cancer in ical staging. Gut. 1997;41:314 –319.
stage III/IV was significantly worse than that of those in stage I/II.2,3 17. Bayerdorffer E, Mannes GA, Richter WO, et al. Decreased high-density
We intended to differentiate cases with stage III/IV disease from lipoprotein cholesterol and increased low-density cholesterol levels in pa-
those with stage I/II disease, since a marked difference in survival tients with colorectal adenomas. Ann Intern Med. 1993;118:481– 487.
between them might influence the choice of treatment. The AUC for 18. Shen JG, Cheong JH, Hyung WJ, et al. Pretreatment anemia is associated with
our score indicated good discrimination between the 2 groups (stage poorer survival in patients with stage I and II gastric cancer. J Surg Oncol.
III/IV vs. stage I/II). The determination of a proper cut point always 2005;91:126 –130.
complicates a numerical scoring system.32 In this score, an optimal 19. Ikeda M, Furukawa H, Imamura H, et al. Poor prognosis associated with
thrombocytosis in patients with gastric cancer. Ann Surg Oncol. 2002;9:287–291.
cut point was set at 7 by means of ROC analysis. With the defined
20. Elahi MM, McMillan DC, McArdle CS, et al. Score based on hypoalbumin-
cut point, our risk score diagnosed stage III/IV with sensitivity of emia and elevated C-reactive protein predicts survival in patients with
79.6%, specificity of 85.2%, and accuracy of 82.4%. Because of advanced gastrointestinal cancer. Nutr Cancer. 2004;48:171–173.
high specificity, when the risk score equals to or greater than 7 for 21. Lemeshow S, Hosmer DW Jr. A review of goodness of fit statistics for use in
an individual patient with gastric cancer, he/she should be strongly the development of logistic regression models. Am J Epidemiol. 1982;115:
suspected of having stage III/IV disease. At least D2 lymphadenec- 92–106.
tomy combined with perioperative adjuvant therapy is recom- 22. Costa ML, De Cassia Braga Ribeiro K, Machado MA, et al. Prognostic score
in gastric cancer: the importance of a conjoint analysis of clinical, pathologic,
mended to increase the likelihood of R0 resection and reduce the and therapeutic factors. Ann Surg Oncol. 2006;13:843– 850.
risk of recurrence, once a patient has a score 7.
23. Chan AO, Chu KM, Lam SK, et al. Early prediction of tumor recurrence after
The next key issue needing to be addressed is the score’s curative resection of gastric carcinoma by measuring soluble E-cadherin.
generalizability across different settings. Many other scoring Cancer. 2005;104:740 –746.
systems performed well in the development sample, but poorly 24. Ishii M, Gama H, Chida N, et al. Simultaneous measurements of serum
when applied to a testing set that was not used for training.33,34 alpha-fetoprotein and protein induced by vitamin K absence for detecting
These facts highlight the urgency of validating our score on hepatocellular carcinoma. South Tohoku District study group. Am J Gastro-
enterol. 2000;95:1036 –1040.
another patient series.
25. Hanley JA, McNeil BJ. The meaning and use of the area under a receiver
In summary, the risk score may be helpful to preoperative operating characteristic (ROC) curve. Radiology. 1982;143:29 –36.
gastric cancer staging. It probably assists surgeons in deciding the 26. Victorzon M, Lundin J, Haglund C, et al. A risk score for predicting outcome
extent of surgery and in choosing the appropriate perioperative in patients with gastric cancer, based on stage, sialyl-Tn immunoreactivity
adjuvant therapies. and ploidy—a multivariate analysis. Int J Cancer. 1996;67:190 –193.
4 | www.amjclinicaloncology.com © 2010 Lippincott Williams & Wilkins

5. American Journal of Clinical Oncology • Volume XX, Number X, XXX 2010 A Score to Predict TNM Stages
27. Marrelli D, De Stefano A, de Manzoni G, et al. Prediction of recurrence after computer program accuracy for preoperative estimation of lymph node
radical surgery for gastric cancer: a scoring system obtained from a prospec- metastases from gastric cancer. World J Surg. 2000;24:1550 –1558.
tive multicenter study. Ann Surg. 2005;241:247–255. 32. Wu Y, Chen J, Yu J, et al. A practical scoring system based on ROC analysis
28. Marubini E, Bonfanti G, Bozzetti F, et al. A prognostic score for patients for evaluating potential lymph nodes metastasis during gastric surgery. J Surg
resected for gastric cancer. Eur J Cancer. 1993;29A:845– 850. Oncol. 2006;93:534 –540.
29. Kologlu M, Kama NA, Reis E, et al. A prognostic score for gastric cancer. 33. Justice AC, Covinsky KE, Berlin JA. Assessing the generalizability of
Am J Surg. 2000;179:521–526. prognostic information. Ann Intern Med. 1999;130:515–524.
30. Kattan MW, Karpeh MS, Mazumdar M, et al. Postoperative nomogram for 34. Hukkelhoven CW, Steyerberg EW, Habbema JD, et al. Predicting out-
disease-specific survival after an R0 resection for gastric carcinoma. J Clin come after traumatic brain injury: development and validation of a prognos-
Oncol. 2003;21:3647–3650. tic score based on admission characteristics. J Neurotrauma. 2005;22:1025–
31. Guadagni S, de Manzoni G, Catarci M, et al. Evaluation of the Maruyama 1039.
© 2010 Lippincott Williams & Wilkins www.amjclinicaloncology.com | 5