European Journal of Obstetrics & Gynecology and Reproductive Biology 172 (2014) 115–119
Contents lists available at ScienceDirect
European Journal of Obstetrics & Gynecology and Reproductive Biology journal homepage: www.elsevier.com/locate/ejogrb
Evaluation of a method of predicting lymph node metastasis in endometrial cancer based on five pre-operative characteristics Martin Koskas a,b,c,d,*, Anne Sophie Genin e, Olivier Graesslin f, Emmanuel Barranger g, Bassam Haddad e, Emile Darai c,h, Roman Rouzier d,i a
Department of Obstetrics and Gynaecology, APHP Hoˆpital Bichat, Paris, France Paris Diderot University, Paris 07, France UMR S 938, CdR St Antoine UPMC University, Paris 06, France d EA 7285, Universite´ Versailles Saint-Quentin, Montigny-le-Bretonneux, France e Department of Obstetrics and Gynaecology, CHIC, Cre´teil, France f Department of Obstetrics and Gynaecology, CHU Reims, Reims, France g Department of Obstetrics and Gynaecology, APHP Hoˆpital Lariboisiere, Paris, France h Department of Obstetrics and Gynaecology, APHP Hoˆpital Tenon, Paris, France i Department of Surgery, Institut Curie, Paris, France b c
A R T I C L E I N F O
A B S T R A C T
Article history: Received 28 May 2013 Received in revised form 8 October 2013 Accepted 26 October 2013
Objective: We recently developed an algorithm based on five clinical and pathological characteristics to predict lymph node (LN) metastasis in endometrial cancer. The aim of this study was to evaluate the accuracy of using this algorithm with preoperative characteristics. Study design: In this retrospective multicenter study, we evaluated the accuracy of using an algorithm to predict LN metastasis using preoperative tumor characteristics provided by endometrial sampling pathological characteristics (histological subtype and grade) and by magnetic resonance imaging (MRI) for primary site tumor extension. Results: In total, 181 patients were included in this study, and 14 patients had pelvic LN metastasis (7.7%). Using preoperative tumor characteristics, the algorithm showed good discrimination with an area under the receiver operating characteristic curve (AUC) of 0.83 (95% confidence interval (IC95) = 0.79– 0.87) and was well calibrated (average error = 1.9% and maximal error = 8.5%). LN metastasis prediction by the algorithm using preoperative data was as accurate as that obtained using the final tumor characteristics (AUC = 0.77 (CI95 = 0.70–0.83), average error = 2.8% and maximal error = 23.2%). Conclusion: Our algorithm was accurate in predicting pelvic LN metastasis even with the use of preoperative tumor characteristics provided by endometrial sampling and MRI. These findings, however, should be verified in a larger database before our algorithm is implemented for widespread use. ß 2013 Elsevier Ireland Ltd. All rights reserved.
Keywords: Endometrial cancer Algorithm Lymph node metastasis Preoperative data
1. Introduction Endometrial cancer is the most common malignant disease of the female genital tract and the seventh most common cause of cancer death in women in Western countries [1]. Histological type, grade and depth of myometrial invasion are prognostic factors in early stage disease as well as risk factors for lymph node (LN) metastasis. Systematic lymphadenectomy has recently been questioned for stage I disease based on the results from two
* Corresponding author at: Department of Obstetrics and Gynecology, CHU Bichat Claude Bernard, 46, rue Henri-Huchard, 75018 Paris, France. Tel.: +33 01 40 25 80 80; fax: +33 01 40 25 67 57. E-mail addresses: [email protected], [email protected] (M. Koskas). 0301-2115/$ – see front matter ß 2013 Elsevier Ireland Ltd. All rights reserved. http://dx.doi.org/10.1016/j.ejogrb.2013.10.028
randomized trials [2,3], but the two trials included both low and high risk patients, and hence interpretation of the results is source of debate. Yet, in several retrospective studies, patients with highrisk endometrial cancer or with node-positive endometrial carcinoma who underwent lymphadenectomy had longer overall survival than patients who did not [4,5]. Using propensity score matching analysis, it has been suggested that lymphadenectomy is beneficial for stage I disease survival only for grade 3 tumors [6]. To provide evidence-based and individualized predictions, we previously developed an algorithm for predicting the LN status for endometrial cancer by combining selected clinical and pathological risk factors using a multivariate model [7]. The algorithm was developed using data from the surveillance, epidemiology, and end results (SEER) database and has been confirmed externally on a validation set. Only final pathological characteristics, however, were used for its development and validation. Consequently, to
116
M. Koskas et al. / European Journal of Obstetrics & Gynecology and Reproductive Biology 172 (2014) 115–119
date, it should only be used after hysterectomy has been performed but lymphadenectomy has been omitted. A large number of studies have reported the discrepancy between the preoperative and final grades, histologic subtype [8] and primary site tumor extension, especially regarding the sensitivity for detecting LN metastasis [9]. The aim of this study was to test the general applicability of our algorithm when the preoperative characteristics of endometrial cancer are used instead of the definitive pathological characteristics determined from examination of the surgical specimen. 2. Materials and methods 2.1. Study population The algorithm was applied using data from a multicenter database containing information from five institutions in France: Tenon University Hospital (Paris), Bichat University Hospital (Paris), Cre´teil Hospital (Cre´teil), Lariboisie`re University Hospital (Paris) and Reims Hospital (Reims). Patients treated for primary endometrial cancer between 2000 and 2010 were included if there were data available on their preoperative and definitive findings for all components of the algorithm (e.g., age, race, pathological characteristics and primary site tumor extension) and for the final LN status (e.g., all patients who underwent a lymphadenectomy). During the study period (2000–2010), according to the national recommendations the policy toward lymphadenectomy in the five institutions was grossly to perform pelvic lymphadenectomy in patients in good condition whereas para-aortic lymphadenectomy was not considered to be a standard [10]. The study was approved by the National Ethics Committee (CEROG 2012-GYN-10-01). The preoperative histological subtype and grade were abstracted from the pathological analysis of the endometrial biopsy. The preoperative tumor extension was assessed from the MRI examination results because MRI examination constitutes the gold standard for abdominal and pelvic radiological examination in the management of endometrial cancer according to the European [11] and French [12] guidelines. Because the algorithm was built for presumed stages I and II endometrial cancers, patients with stage IIIA, IIIB, IVA or IVB were excluded. For each patient, the following variables were recorded to calculate the LN metastatic probability according to the previously developed algorithm (web based calculator supplied): age at diagnosis; race (White, Black, or other); histological subtype (endometrioid adenocarcinoma, papillary serous, clear-cell, or carcinosarcoma); grade differentiation (well = 1, moderate = 2, or poor = 3); and uterine tumor extension (endometrium, 50% myometrial invasion, >50% myometrial invasion, or cervical stromal invasion). The formula applied for the calculation of the predicted probability of metastatic LN is provided in Fig. 1. For example, for a 66 year old white patient with a grade 2 endometrioid adenocarcinoma infiltrating less than half of myometrium, the predicted probability for LN metastasis was 3%.
Fig. 1. Formula applied for the calculation of the predicted probability of metastatic lymph node.
corresponds to the agreement between the observed outcome frequencies and the predicted probabilities and was studied using graphical representations of the relationship between these two results (calibration curves). We also evaluated the average and maximal errors between the prediction and observation, which were obtained from a calibration curve. Differences were considered significant at a level of p < 0.05. All analyses were performed using the R packages rms, Hmisc, Design, Presence/absence (http://lib.stat.cmu.edu/R/CRAN). 3. Results 3.1. Patient population The flow chart of patient selection is presented in Fig. 2. Complete preoperative and final data were obtained and analyzed
2.2. Statistical analysis The Cohen’s kappa coefficient was calculated to evaluate the inter-rater agreement for comparison between the preoperative and final histologic subtype, grade and primary site tumor extension. The categorical and numerical variables were analyzed using the Chi-square test and Student’s t-test, respectively. The predictive accuracy of the model was evaluated in terms of its discrimination and calibration. Discrimination is the ability to differentiate between women with and without LN metastasis. It is measured using the receiver operating characteristic (ROC) curve and summarized by the area under the curve (AUC). Calibration
Fig. 2. Flow chart of patient selection.
M. Koskas et al. / European Journal of Obstetrics & Gynecology and Reproductive Biology 172 (2014) 115–119 Table 1 Characteristics of the study population. Variables
Table 3 Relationship between the pre-operative and final tumor grade.
Pre-operative
Post-operative
P value
Age at diagnosis (years) Median (mean) Range
64 (63.52) 38–86
–
Race White Black Other
173 (96%) 3 (2%) 5 (3%)
–
Tumor grade 1 2 3 Histologic subtype Adenocarcinoma Clear-cell Papillary serous Carcinosarcoma
93 (51%) 65 (36%) 23 (23%)
Pre-operative tumor grade
Final tumor grade Grade 1
Grade 2
Grade 3
Grade 1 Grade 2 Grade 3
65 14 2
24 44 7
4 7 14
93 65 23
Total
81
75
25
181
Total
Kappa test = 0.47 (95% CI = 0.36–0.58) (moderate).
81 (45%) 75 (41%) 25 (14%)
0.44
175 5 0 1
(97%) (3%) (0%) (1%)
167 6 5 3
(92%) (3%) (3%) (2%)
0.10
Primary site tumor characteristics Endometrium 40 50% 68 >50% 57 Cervical stroma invasion 16
(22%) (38%) (31%) (9%)
18 82 69 12
(10%) (45%) (38%) (7%)
0.10
Pelvic lymph nodes positive Yes No
117
14 (8%) 167 (92%)
Table 4 Relationship between the pre-operative and final primary site tumor characteristics provided by MRI and pathological examination of the hysterectomy specimen, respectively. Pre-operative primary site tumor characteristics MRI
–
Periaortic lymph nodes positive Yes No (or undetermined)
8 (4.4%) 173 (95.6%)
–
Extent of lymphadenectomy Median (mean) Range
13.5 (14.6) 1–57
–
Final primary site tumor characteristics Cervical stroma invasion
Total
Endometrium
50%
>50%
Endometrium 50% >50% Cervical stroma invasion
10 5 2 1
26 45 9 2
3 15 43 8
1 3 3 5
40 68 57 16
Total
18
82
69
12
181
Kappa test = 0.37 (95% CI = 0.26–0.47) (fair).
for 181 patients. The patients’ characteristics are reported in Table 1. Among the 181 patients analyzed, 20 (11%) underwent para-aortic lymphadenectomy while all patients underwent pelvic lymphadenectomy. The agreement between the preoperative and final histological subtype was good (93% (169/181)) (Table 2), though some discrepancies between the preoperative and final data were observed concerning the grade and tumor extension (agreement was 68% (123/181) and 57% (103/181), respectively) (Tables 3 and 4). In the whole study population, however, there were no statistically significant differences between the preoperative and final tumor characteristics for these three parameters (Table 1), suggesting that there were similar levels of under- and overestimation by the pre-operative examinations. 3.2. Predictive values of the algorithm using the preoperative characteristics In the study population, the discrimination accuracy of the model was 0.83 (IC95 = 0.79–0.87) (Fig. 3). The corresponding
predicted and actual probabilities of LN metastasis in the whole study population are shown in the calibration plot (Fig. 4). There was no significant difference between the predicted probabilities and the observed rate of metastatic LNs (p value of the U index = 0.53). The average difference between the predicted and observed probabilities was 1.9%. 3.2.1. Predictive values of the algorithm using the final characteristics The discrimination accuracy of the model was 0.77 (IC95 = 0.70– 0.83) (Fig. 3). The calibration plot is shown in Fig. 4. There was no difference between the predicted probabilities and the observed rate of metastatic LNs (p value of the U index = 0.29). The average difference between the predicted and observed probabilities was 2.8%. 3.2.2. Thresholds of the algorithm We investigated whether thresholds could be established to decide in which patients LN dissection should be performed. We used the presence/absence package and focused on five methods to optimize thresholds (Table 5). In particular, we considered the negative predictive value to build a model that discriminates for nodal metastasis in a low-risk group. Because the most important goal for developing this model was to obtain the best negative predictive value (to prevent misclassifying a patient with LN metastasis), we determined that a cut-off value of 99 points would
Table 2 Relationship between the pre-operative and final histologic endometrial subtype. Pre-operative histologic subtype
Final histologic subtype Adenocarcinoma
Clear-cell
Papillary serous
Carcinosarcoma
Total
Adenocarcinoma Clear-cell Papillary serous Carcinosarcoma
165 2 0 0
3 3 0 0
5 0 0 0
2 0 0 1
175 5 0 1
Total
167
6
5
3
181
Kappa test = 0.38 (95% CI = 0.11–0.65) (fair).
118
M. Koskas et al. / European Journal of Obstetrics & Gynecology and Reproductive Biology 172 (2014) 115–119 Table 5 Methods used to optimize thresholds. Methods used to optimize thresholds
Threshold (predicted probability)
PCC
Sensitivity
Specificity
Sensitivity = specificity Max sensitivity and specificity Max kappa Max PCC Predicted prevalence = observed prevalence
105 99 118 183 132
0.73 0.62 0.83 0.92 0.89
0.71 1.00 0.57 0.00 0.21
0.73 0.59 0.86 1.00 0.93
(0.13) (0.07) (0.19) (0.70) (0.26)
PCC, percent correctly classified.
4. Comment
Fig. 3. ROC curves for predicting LN metastasis according to the algorithm using the preoperative (in red) and final tumor characteristics (in blue). Using the preoperative and final tumor characteristics, the algorithm has promising discrimination with an area under the receiver operating characteristic curve (AUC) of 0.83 (95% confidence interval [IC95 = 0.79–0.87]) (red line) and 0.77 [CI95 = 0.70–0.83]) (blue line). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.)
Fig. 4. Calibration curves for predicting LN metastasis according to the algorithm using the preoperative (in red) and final tumor characteristics (in blue). Perfect prediction would correspond to the 458 broken line. The solid lines indicate the observed (apparent) algorithm performance. Triangles correspond to the quartiles of predicted probability for LN metastasis. Using the preoperative and final tumor characteristics of the primary tumor, the algorithm was well calibrated (average error = 1.9% and maximal error = 8.5%; average error = 2.8% and maximal error = 23.2%, respectively). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.)
be relevant. With a cut-off value of 99 points for the total score in the algorithm in the low-risk group (98/181, 54% of patients), no patients had pelvic LN metastasis. With this threshold, the positive predictive value is 0.169 (95% CI 0.099–0.270), but the negative predictive value is 1.000 (95% CI 0.953–1.000).
Our study suggests that the pathological characteristics of the primary tumor can be used to estimate the risk of LN metastasis even when the characteristics are based on preoperative investigations using MRI and endometrial sampling (for tumor grade and subtype). Milam et al. recently estimated the risk for nodal metastasis in 971 women with endometrial cancer based on uterine pathology characteristics [13]. They determined that approximately 40% of patients were at low risk for nodal metastasis based on three specific criteria from final pathology reports: myoinvasion, tumor size, and differentiation. In this low risk group, the rate of nodal metastasis was only 0.8%. Although this result saves 40% of patients from undergoing lymphadenectomy, the decision to perform lymphadenectomy would remain for 60% of patients with a probability that only 62/582 patients (10.7%) have metastatic LNs. Similar to our initial approach, this strategy is based on definitive pathological results and is only applicable once hysterectomy has been performed. In the present study, we found that using a cut-off value of 99 points, we could avoid lymphadenectomy in more than half of patients without the risk of missing pelvic LN metastasis. Others have attempted to predict this risk in patients with endometrial cancers before hysterectomy. Kamura et al. developed a formula based on the tumor diameter and the depth of myometrial invasion [14]. Unfortunately, the accuracy of their formula has not been externally validated. Similarly, Lee et al. developed a preoperative prediction model (based on the histological tumor grade, preoperative CA-125 levels, disease extent, and myometrial invasion) for identifying a group that is at low risk for developing LN metastasis in endometrial cancer [15], but there is no evidence of the generalizability this model. Recently Kang et al. developed a model based on the CA-125 levels and MRI results: they report promising results with an accuracy of 0.85 with 43% of the patients defined as low-risk and had a false negative rate of 1.4% [16]. For this model, however, non-endometrioid endometrial cancers were excluded. Lymphovascular space invasion (LVSI), which is highly associated with pelvic LN metastasis in endometrial cancer, beyond other tumoral characteristics included in the present algorithm [17], was not investigated because such information is not available before hysterectomy has been performed. To determine the reliability of intraoperative frozen sections for surgical staging of endometrial cancer, Kumar et al. conducted a prospective study on 784 consecutive patients with endometrial cancer who underwent a hysterectomy [18]. The need for surgical staging was decided through intraoperative frozen sections using four variables: tumor size, histologic grade, histologic subtype, and depth of myometrial invasion. Seven hundred and one (89%) had an accurate definitive diagnosis for the four variables at final pathology, but clinically significant discordance occurred in only 1.3% of cases. In other words, according to the Mayo Clinic protocol, in only 1.3% of cases, the decision to perform lymphadenectomy would not remain after final pathology. Unfortunately, several
M. Koskas et al. / European Journal of Obstetrics & Gynecology and Reproductive Biology 172 (2014) 115–119
limitations must be discussed. First, as underlined by the authors, the technical expertise of the pathologists involved in the study is robust and the reliability of the frozen section may be less in smaller practices. Second, the result of lymphadenectomy was not detailed in the study. Third, with Mayo Clinic‘s protocol, avoiding lymphadenectomy would be possible only in a low proportion of patients with endometrial cancer (endometrioid, grades 1–2, myometrial invasion of 1–49%, tumor diameter of 2 cm, no evidence of tumor outside the corpus). For example, in our population study, the number of patients with such characteristics is 91/181 (50.3%). It is likely that such patients probably do not require lymphadenectomy but we believe that performing lymphadenectomy in 49.7% of patients with endometrial cancer is too important since the risk of metastatic LN is around 10–15% in such patients. On the contrary, our algorithm permits to calculate individually the metastatic LN probability. To build the most accurate algorithm, we initially included only women who underwent a lymphadenectomy with at least 10 regional LNs removed [7]. For external validation, we did not select patients who had minimal LNs removed because we wanted to test the algorithm accuracy in everyday practice. Although the median number of nodes removed in the validation set is low, it is similar to the median number of nodes harvested in the ASTEC lymphadenectomy group [2]. One possible limitation of the present study is that only 11% of patients underwent para-aortic LN dissection. However, in a recent study involving 847 patients with endometrial cancer who underwent both pelvic and paraaortic node removal during surgery, the incidence of isolated paraaortic nodal metastasis in patients with negative pelvic LNs was approximately 1% [19]. Consequently, it is unlikely that the low rate of patients who underwent para-aortic LN dissection in the current study would influence the results to a large extent. We are aware of the results of the SEPAL study [20] suggesting that combined pelvic and para-aortic lymphadenectomy is beneficial as treatment for patients with endometrial carcinoma of intermediate or high risk of recurrence. The purpose of the present study, however, was not to evaluate the impact of lymphadenectomy on survival in patients with endometrial cancer but to evaluate the accuracy of using our algorithm with preoperative characteristics to predict LN metastasis in endometrial cancer. The algorithm evaluated in the present study is based on pathological hysterectomy characteristics to provide an estimation of LN metastasis. Because the SEER database does not distinguish pelvic and para-aortic LN metastasis, it was not possible to develop an algorithm distinguishing pelvic and para-aortic LN metastasis, but in most endometrial cancers, when there is paraaortic LN metastasis the pelvic area is also invaded. The findings of several studies like that of Mariani et al. focusing on para-aortic dissemination in endometrial cancer [21] suggest that isolated para-aortic lymph node metastasis in endometrial cancer is a particularly rare event. We are aware that in several countries, few women with low risk tumors would get an MRI. Several studies have shown, however, that ultrasound and MRI have similar diagnostic performances in predicting myometrial invasion [22] and MRI examination constitutes the gold standard in the management of endometrial cancer according to the European [11] and French [12] guidelines. Importantly, unlike ultrasound examination, MRI can be reviewed before surgery by other radiologists during a multidisciplinary meeting which makes it a more reproducible test. It is true that the correlation of the MRI with final pathology is rather poor, but the present study suggests that it provides useful information with the use of our algorithm. It would be interesting to know the correlation between ultrasonography with final pathology and particularly if ultrasonography would provide a good predictive value for LN metastasis with the algorithm.
119
In conclusion, one serious limitation of our model is that the pathological data were obtained from the surgical component analysis, after hysterectomy had been performed. We demonstrated in this study that our algorithm is still accurate using preoperative data. Since changing a component of the previously published algorithm from pathologic tumor invasion to the MRI interpretation of invasion is a major change, the current findings need to be confirmed on a larger scale. Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.ejogrb.2013.10.028. References [1] Siegel R, Naishadham D, Jemal A. Cancer statistics, 2012. CA Cancer J Clin 2012;62:10–29. [2] Kitchener H, Swart AM, Qian Q, Amos C, Parmar MK. Efficacy of systematic pelvic lymphadenectomy in endometrial cancer (MRC ASTEC trial): a randomised study. Lancet 2009;373:125–36. [3] Benedetti Panici P, Basile S, Maneschi F, et al. Systematic pelvic lymphadenectomy vs. no lymphadenectomy in early-stage endometrial carcinoma: randomized clinical trial. J Natl Cancer Inst 2008;100:1707–16. [4] Cragun JM, Havrilesky LJ, Calingaert B, et al. Retrospective analysis of selective lymphadenectomy in apparent early-stage endometrial cancer. J Clin Oncol 2005;23:3668–75. [5] Bristow RE, Zahurak ML, Alexander CJ, Zellars RC, Montz FJ. FIGO stage IIIC endometrial carcinoma: resection of macroscopic nodal disease and other determinants of survival. Int J Gynecol Cancer 2003;13:664–72. [6] Bendifallah S, Koskas M, Ballester M, Genin AS, Darai E, Rouzier R. The survival impact of systematic lymphadenectomy in endometrial cancer with the use of propensity score matching analysis. Am J Obstet Gynecol 2012;206:500 e1–5011e. [7] Bendifallah S, Genin AS, Naoura I, et al. A nomogram for predicting lymph node metastasis of presumed stage I and II endometrial cancer. Am J Obstet Gynecol 2012;207:197 e1–1978e. [8] Larson DM, Johnson KK, Broste SK, Krawisz BR, Kresl JJ. Comparison of D&C and office endometrial biopsy in predicting final histopathologic grade in endometrial cancer. Obstet Gynecol 1995;86:38–42. [9] Wu LM, Xu JR, Gu HY, Hua J, Haacke EM, Hu J. Predictive value of T2-weighted imaging and contrast-enhanced MR imaging in assessing myometrial invasion in endometrial cancer: a pooled analysis of prospective studies. Eur Radiol 2013;23:435–49. [10] Bremond A, Bataillard A, Thomas L, et al. Standards, Options and Recommendations 2000 for the management of patients with endometrial cancer (nonmetastatic) (abridged report). Gynecol Obstet Fertil 2002;30:902–16. [11] Kinkel K, Forstner R, Danza FM, et al. Staging of endometrial cancer with MRI: guidelines of the European Society of Urogenital Imaging. Eur Radiol 2009;19:1565–74. [12] Querleu D, Planchamp F, Narducci F, et al. Clinical practice guidelines for the management of patients with endometrial cancer in France: recommendations of the Institut National du Cancer and the Societe Francaise d’Oncologie Gynecologique. Int J Gynecol Cancer 2011;21:945–50. [13] Milam MR, Java J, Walker JL, Metzinger DS, Parker LP, Coleman RL. Nodal metastasis risk in endometrioid endometrial cancer. Obstet Gynecol 2012;119: 286–92. [14] Kamura T, Yahata H, Shigematsu T, et al. Predicting pelvic lymph node metastasis in endometrial carcinoma. Gynecol Oncol 1999;72:387–91. [15] Lee JY, Jung DC, Park SH, et al. Preoperative prediction model of lymph node metastasis in endometrial cancer. Int J Gynecol Cancer 2010;20:1350–5. [16] Kang S, Kang WD, Chung HH, et al. Preoperative identification of a low-risk group for lymph node metastasis in endometrial cancer: a Korean gynecologic oncology group study. J Clin Oncol 2012;30:1329–34. [17] Koskas M, Bassot K, Graesslin O, et al. Impact of lymphovascular space invasion on a nomogram for predicting lymph node metastasis in endometrial cancer. Gynecol Oncol 2013;129:292–7. [18] Kumar S, Medeiros F, Dowdy SC, et al. A prospective assessment of the reliability of frozen section to direct intraoperative decision making in endometrial cancer. Gynecol Oncol 2012;127:525–31. [19] Abu-Rustum NR, Gomez JD, Alektiar KM, et al. The incidence of isolated paraaortic nodal metastasis in surgically staged endometrial cancer patients with negative pelvic lymph nodes. Gynecol Oncol 2009;115:236–8. [20] Todo Y, Kato H, Kaneuchi M, Watari H, Takeda M, Sakuragi N. Survival effect of para-aortic lymphadenectomy in endometrial cancer (SEPAL study): a retrospective cohort analysis. Lancet 2010;375:1165–72. [21] Mariani A, Keeney GL, Aletti G, Webb MJ, Haddock MG, Podratz KC. Endometrial carcinoma: para-aortic dissemination. Gynecol Oncol 2004;92:833–8. [22] Antonsen SL, Jensen LN, Loft A, et al. MRI, PET/CT and ultrasound in the preoperative staging of endometrial cancer: a multicenter prospective comparative study. Gynecol Oncol 2013;128:300–8.
Contents lists available at ScienceDirect
European Journal of Obstetrics & Gynecology and Reproductive Biology journal homepage: www.elsevier.com/locate/ejogrb
Evaluation of a method of predicting lymph node metastasis in endometrial cancer based on five pre-operative characteristics Martin Koskas a,b,c,d,*, Anne Sophie Genin e, Olivier Graesslin f, Emmanuel Barranger g, Bassam Haddad e, Emile Darai c,h, Roman Rouzier d,i a
Department of Obstetrics and Gynaecology, APHP Hoˆpital Bichat, Paris, France Paris Diderot University, Paris 07, France UMR S 938, CdR St Antoine UPMC University, Paris 06, France d EA 7285, Universite´ Versailles Saint-Quentin, Montigny-le-Bretonneux, France e Department of Obstetrics and Gynaecology, CHIC, Cre´teil, France f Department of Obstetrics and Gynaecology, CHU Reims, Reims, France g Department of Obstetrics and Gynaecology, APHP Hoˆpital Lariboisiere, Paris, France h Department of Obstetrics and Gynaecology, APHP Hoˆpital Tenon, Paris, France i Department of Surgery, Institut Curie, Paris, France b c
A R T I C L E I N F O
A B S T R A C T
Article history: Received 28 May 2013 Received in revised form 8 October 2013 Accepted 26 October 2013
Objective: We recently developed an algorithm based on five clinical and pathological characteristics to predict lymph node (LN) metastasis in endometrial cancer. The aim of this study was to evaluate the accuracy of using this algorithm with preoperative characteristics. Study design: In this retrospective multicenter study, we evaluated the accuracy of using an algorithm to predict LN metastasis using preoperative tumor characteristics provided by endometrial sampling pathological characteristics (histological subtype and grade) and by magnetic resonance imaging (MRI) for primary site tumor extension. Results: In total, 181 patients were included in this study, and 14 patients had pelvic LN metastasis (7.7%). Using preoperative tumor characteristics, the algorithm showed good discrimination with an area under the receiver operating characteristic curve (AUC) of 0.83 (95% confidence interval (IC95) = 0.79– 0.87) and was well calibrated (average error = 1.9% and maximal error = 8.5%). LN metastasis prediction by the algorithm using preoperative data was as accurate as that obtained using the final tumor characteristics (AUC = 0.77 (CI95 = 0.70–0.83), average error = 2.8% and maximal error = 23.2%). Conclusion: Our algorithm was accurate in predicting pelvic LN metastasis even with the use of preoperative tumor characteristics provided by endometrial sampling and MRI. These findings, however, should be verified in a larger database before our algorithm is implemented for widespread use. ß 2013 Elsevier Ireland Ltd. All rights reserved.
Keywords: Endometrial cancer Algorithm Lymph node metastasis Preoperative data
1. Introduction Endometrial cancer is the most common malignant disease of the female genital tract and the seventh most common cause of cancer death in women in Western countries [1]. Histological type, grade and depth of myometrial invasion are prognostic factors in early stage disease as well as risk factors for lymph node (LN) metastasis. Systematic lymphadenectomy has recently been questioned for stage I disease based on the results from two
* Corresponding author at: Department of Obstetrics and Gynecology, CHU Bichat Claude Bernard, 46, rue Henri-Huchard, 75018 Paris, France. Tel.: +33 01 40 25 80 80; fax: +33 01 40 25 67 57. E-mail addresses: [email protected], [email protected] (M. Koskas). 0301-2115/$ – see front matter ß 2013 Elsevier Ireland Ltd. All rights reserved. http://dx.doi.org/10.1016/j.ejogrb.2013.10.028
randomized trials [2,3], but the two trials included both low and high risk patients, and hence interpretation of the results is source of debate. Yet, in several retrospective studies, patients with highrisk endometrial cancer or with node-positive endometrial carcinoma who underwent lymphadenectomy had longer overall survival than patients who did not [4,5]. Using propensity score matching analysis, it has been suggested that lymphadenectomy is beneficial for stage I disease survival only for grade 3 tumors [6]. To provide evidence-based and individualized predictions, we previously developed an algorithm for predicting the LN status for endometrial cancer by combining selected clinical and pathological risk factors using a multivariate model [7]. The algorithm was developed using data from the surveillance, epidemiology, and end results (SEER) database and has been confirmed externally on a validation set. Only final pathological characteristics, however, were used for its development and validation. Consequently, to
116
M. Koskas et al. / European Journal of Obstetrics & Gynecology and Reproductive Biology 172 (2014) 115–119
date, it should only be used after hysterectomy has been performed but lymphadenectomy has been omitted. A large number of studies have reported the discrepancy between the preoperative and final grades, histologic subtype [8] and primary site tumor extension, especially regarding the sensitivity for detecting LN metastasis [9]. The aim of this study was to test the general applicability of our algorithm when the preoperative characteristics of endometrial cancer are used instead of the definitive pathological characteristics determined from examination of the surgical specimen. 2. Materials and methods 2.1. Study population The algorithm was applied using data from a multicenter database containing information from five institutions in France: Tenon University Hospital (Paris), Bichat University Hospital (Paris), Cre´teil Hospital (Cre´teil), Lariboisie`re University Hospital (Paris) and Reims Hospital (Reims). Patients treated for primary endometrial cancer between 2000 and 2010 were included if there were data available on their preoperative and definitive findings for all components of the algorithm (e.g., age, race, pathological characteristics and primary site tumor extension) and for the final LN status (e.g., all patients who underwent a lymphadenectomy). During the study period (2000–2010), according to the national recommendations the policy toward lymphadenectomy in the five institutions was grossly to perform pelvic lymphadenectomy in patients in good condition whereas para-aortic lymphadenectomy was not considered to be a standard [10]. The study was approved by the National Ethics Committee (CEROG 2012-GYN-10-01). The preoperative histological subtype and grade were abstracted from the pathological analysis of the endometrial biopsy. The preoperative tumor extension was assessed from the MRI examination results because MRI examination constitutes the gold standard for abdominal and pelvic radiological examination in the management of endometrial cancer according to the European [11] and French [12] guidelines. Because the algorithm was built for presumed stages I and II endometrial cancers, patients with stage IIIA, IIIB, IVA or IVB were excluded. For each patient, the following variables were recorded to calculate the LN metastatic probability according to the previously developed algorithm (web based calculator supplied): age at diagnosis; race (White, Black, or other); histological subtype (endometrioid adenocarcinoma, papillary serous, clear-cell, or carcinosarcoma); grade differentiation (well = 1, moderate = 2, or poor = 3); and uterine tumor extension (endometrium, 50% myometrial invasion, >50% myometrial invasion, or cervical stromal invasion). The formula applied for the calculation of the predicted probability of metastatic LN is provided in Fig. 1. For example, for a 66 year old white patient with a grade 2 endometrioid adenocarcinoma infiltrating less than half of myometrium, the predicted probability for LN metastasis was 3%.
Fig. 1. Formula applied for the calculation of the predicted probability of metastatic lymph node.
corresponds to the agreement between the observed outcome frequencies and the predicted probabilities and was studied using graphical representations of the relationship between these two results (calibration curves). We also evaluated the average and maximal errors between the prediction and observation, which were obtained from a calibration curve. Differences were considered significant at a level of p < 0.05. All analyses were performed using the R packages rms, Hmisc, Design, Presence/absence (http://lib.stat.cmu.edu/R/CRAN). 3. Results 3.1. Patient population The flow chart of patient selection is presented in Fig. 2. Complete preoperative and final data were obtained and analyzed
2.2. Statistical analysis The Cohen’s kappa coefficient was calculated to evaluate the inter-rater agreement for comparison between the preoperative and final histologic subtype, grade and primary site tumor extension. The categorical and numerical variables were analyzed using the Chi-square test and Student’s t-test, respectively. The predictive accuracy of the model was evaluated in terms of its discrimination and calibration. Discrimination is the ability to differentiate between women with and without LN metastasis. It is measured using the receiver operating characteristic (ROC) curve and summarized by the area under the curve (AUC). Calibration
Fig. 2. Flow chart of patient selection.
M. Koskas et al. / European Journal of Obstetrics & Gynecology and Reproductive Biology 172 (2014) 115–119 Table 1 Characteristics of the study population. Variables
Table 3 Relationship between the pre-operative and final tumor grade.
Pre-operative
Post-operative
P value
Age at diagnosis (years) Median (mean) Range
64 (63.52) 38–86
–
Race White Black Other
173 (96%) 3 (2%) 5 (3%)
–
Tumor grade 1 2 3 Histologic subtype Adenocarcinoma Clear-cell Papillary serous Carcinosarcoma
93 (51%) 65 (36%) 23 (23%)
Pre-operative tumor grade
Final tumor grade Grade 1
Grade 2
Grade 3
Grade 1 Grade 2 Grade 3
65 14 2
24 44 7
4 7 14
93 65 23
Total
81
75
25
181
Total
Kappa test = 0.47 (95% CI = 0.36–0.58) (moderate).
81 (45%) 75 (41%) 25 (14%)
0.44
175 5 0 1
(97%) (3%) (0%) (1%)
167 6 5 3
(92%) (3%) (3%) (2%)
0.10
Primary site tumor characteristics Endometrium 40 50% 68 >50% 57 Cervical stroma invasion 16
(22%) (38%) (31%) (9%)
18 82 69 12
(10%) (45%) (38%) (7%)
0.10
Pelvic lymph nodes positive Yes No
117
14 (8%) 167 (92%)
Table 4 Relationship between the pre-operative and final primary site tumor characteristics provided by MRI and pathological examination of the hysterectomy specimen, respectively. Pre-operative primary site tumor characteristics MRI
–
Periaortic lymph nodes positive Yes No (or undetermined)
8 (4.4%) 173 (95.6%)
–
Extent of lymphadenectomy Median (mean) Range
13.5 (14.6) 1–57
–
Final primary site tumor characteristics Cervical stroma invasion
Total
Endometrium
50%
>50%
Endometrium 50% >50% Cervical stroma invasion
10 5 2 1
26 45 9 2
3 15 43 8
1 3 3 5
40 68 57 16
Total
18
82
69
12
181
Kappa test = 0.37 (95% CI = 0.26–0.47) (fair).
for 181 patients. The patients’ characteristics are reported in Table 1. Among the 181 patients analyzed, 20 (11%) underwent para-aortic lymphadenectomy while all patients underwent pelvic lymphadenectomy. The agreement between the preoperative and final histological subtype was good (93% (169/181)) (Table 2), though some discrepancies between the preoperative and final data were observed concerning the grade and tumor extension (agreement was 68% (123/181) and 57% (103/181), respectively) (Tables 3 and 4). In the whole study population, however, there were no statistically significant differences between the preoperative and final tumor characteristics for these three parameters (Table 1), suggesting that there were similar levels of under- and overestimation by the pre-operative examinations. 3.2. Predictive values of the algorithm using the preoperative characteristics In the study population, the discrimination accuracy of the model was 0.83 (IC95 = 0.79–0.87) (Fig. 3). The corresponding
predicted and actual probabilities of LN metastasis in the whole study population are shown in the calibration plot (Fig. 4). There was no significant difference between the predicted probabilities and the observed rate of metastatic LNs (p value of the U index = 0.53). The average difference between the predicted and observed probabilities was 1.9%. 3.2.1. Predictive values of the algorithm using the final characteristics The discrimination accuracy of the model was 0.77 (IC95 = 0.70– 0.83) (Fig. 3). The calibration plot is shown in Fig. 4. There was no difference between the predicted probabilities and the observed rate of metastatic LNs (p value of the U index = 0.29). The average difference between the predicted and observed probabilities was 2.8%. 3.2.2. Thresholds of the algorithm We investigated whether thresholds could be established to decide in which patients LN dissection should be performed. We used the presence/absence package and focused on five methods to optimize thresholds (Table 5). In particular, we considered the negative predictive value to build a model that discriminates for nodal metastasis in a low-risk group. Because the most important goal for developing this model was to obtain the best negative predictive value (to prevent misclassifying a patient with LN metastasis), we determined that a cut-off value of 99 points would
Table 2 Relationship between the pre-operative and final histologic endometrial subtype. Pre-operative histologic subtype
Final histologic subtype Adenocarcinoma
Clear-cell
Papillary serous
Carcinosarcoma
Total
Adenocarcinoma Clear-cell Papillary serous Carcinosarcoma
165 2 0 0
3 3 0 0
5 0 0 0
2 0 0 1
175 5 0 1
Total
167
6
5
3
181
Kappa test = 0.38 (95% CI = 0.11–0.65) (fair).
118
M. Koskas et al. / European Journal of Obstetrics & Gynecology and Reproductive Biology 172 (2014) 115–119 Table 5 Methods used to optimize thresholds. Methods used to optimize thresholds
Threshold (predicted probability)
PCC
Sensitivity
Specificity
Sensitivity = specificity Max sensitivity and specificity Max kappa Max PCC Predicted prevalence = observed prevalence
105 99 118 183 132
0.73 0.62 0.83 0.92 0.89
0.71 1.00 0.57 0.00 0.21
0.73 0.59 0.86 1.00 0.93
(0.13) (0.07) (0.19) (0.70) (0.26)
PCC, percent correctly classified.
4. Comment
Fig. 3. ROC curves for predicting LN metastasis according to the algorithm using the preoperative (in red) and final tumor characteristics (in blue). Using the preoperative and final tumor characteristics, the algorithm has promising discrimination with an area under the receiver operating characteristic curve (AUC) of 0.83 (95% confidence interval [IC95 = 0.79–0.87]) (red line) and 0.77 [CI95 = 0.70–0.83]) (blue line). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.)
Fig. 4. Calibration curves for predicting LN metastasis according to the algorithm using the preoperative (in red) and final tumor characteristics (in blue). Perfect prediction would correspond to the 458 broken line. The solid lines indicate the observed (apparent) algorithm performance. Triangles correspond to the quartiles of predicted probability for LN metastasis. Using the preoperative and final tumor characteristics of the primary tumor, the algorithm was well calibrated (average error = 1.9% and maximal error = 8.5%; average error = 2.8% and maximal error = 23.2%, respectively). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.)
be relevant. With a cut-off value of 99 points for the total score in the algorithm in the low-risk group (98/181, 54% of patients), no patients had pelvic LN metastasis. With this threshold, the positive predictive value is 0.169 (95% CI 0.099–0.270), but the negative predictive value is 1.000 (95% CI 0.953–1.000).
Our study suggests that the pathological characteristics of the primary tumor can be used to estimate the risk of LN metastasis even when the characteristics are based on preoperative investigations using MRI and endometrial sampling (for tumor grade and subtype). Milam et al. recently estimated the risk for nodal metastasis in 971 women with endometrial cancer based on uterine pathology characteristics [13]. They determined that approximately 40% of patients were at low risk for nodal metastasis based on three specific criteria from final pathology reports: myoinvasion, tumor size, and differentiation. In this low risk group, the rate of nodal metastasis was only 0.8%. Although this result saves 40% of patients from undergoing lymphadenectomy, the decision to perform lymphadenectomy would remain for 60% of patients with a probability that only 62/582 patients (10.7%) have metastatic LNs. Similar to our initial approach, this strategy is based on definitive pathological results and is only applicable once hysterectomy has been performed. In the present study, we found that using a cut-off value of 99 points, we could avoid lymphadenectomy in more than half of patients without the risk of missing pelvic LN metastasis. Others have attempted to predict this risk in patients with endometrial cancers before hysterectomy. Kamura et al. developed a formula based on the tumor diameter and the depth of myometrial invasion [14]. Unfortunately, the accuracy of their formula has not been externally validated. Similarly, Lee et al. developed a preoperative prediction model (based on the histological tumor grade, preoperative CA-125 levels, disease extent, and myometrial invasion) for identifying a group that is at low risk for developing LN metastasis in endometrial cancer [15], but there is no evidence of the generalizability this model. Recently Kang et al. developed a model based on the CA-125 levels and MRI results: they report promising results with an accuracy of 0.85 with 43% of the patients defined as low-risk and had a false negative rate of 1.4% [16]. For this model, however, non-endometrioid endometrial cancers were excluded. Lymphovascular space invasion (LVSI), which is highly associated with pelvic LN metastasis in endometrial cancer, beyond other tumoral characteristics included in the present algorithm [17], was not investigated because such information is not available before hysterectomy has been performed. To determine the reliability of intraoperative frozen sections for surgical staging of endometrial cancer, Kumar et al. conducted a prospective study on 784 consecutive patients with endometrial cancer who underwent a hysterectomy [18]. The need for surgical staging was decided through intraoperative frozen sections using four variables: tumor size, histologic grade, histologic subtype, and depth of myometrial invasion. Seven hundred and one (89%) had an accurate definitive diagnosis for the four variables at final pathology, but clinically significant discordance occurred in only 1.3% of cases. In other words, according to the Mayo Clinic protocol, in only 1.3% of cases, the decision to perform lymphadenectomy would not remain after final pathology. Unfortunately, several
M. Koskas et al. / European Journal of Obstetrics & Gynecology and Reproductive Biology 172 (2014) 115–119
limitations must be discussed. First, as underlined by the authors, the technical expertise of the pathologists involved in the study is robust and the reliability of the frozen section may be less in smaller practices. Second, the result of lymphadenectomy was not detailed in the study. Third, with Mayo Clinic‘s protocol, avoiding lymphadenectomy would be possible only in a low proportion of patients with endometrial cancer (endometrioid, grades 1–2, myometrial invasion of 1–49%, tumor diameter of 2 cm, no evidence of tumor outside the corpus). For example, in our population study, the number of patients with such characteristics is 91/181 (50.3%). It is likely that such patients probably do not require lymphadenectomy but we believe that performing lymphadenectomy in 49.7% of patients with endometrial cancer is too important since the risk of metastatic LN is around 10–15% in such patients. On the contrary, our algorithm permits to calculate individually the metastatic LN probability. To build the most accurate algorithm, we initially included only women who underwent a lymphadenectomy with at least 10 regional LNs removed [7]. For external validation, we did not select patients who had minimal LNs removed because we wanted to test the algorithm accuracy in everyday practice. Although the median number of nodes removed in the validation set is low, it is similar to the median number of nodes harvested in the ASTEC lymphadenectomy group [2]. One possible limitation of the present study is that only 11% of patients underwent para-aortic LN dissection. However, in a recent study involving 847 patients with endometrial cancer who underwent both pelvic and paraaortic node removal during surgery, the incidence of isolated paraaortic nodal metastasis in patients with negative pelvic LNs was approximately 1% [19]. Consequently, it is unlikely that the low rate of patients who underwent para-aortic LN dissection in the current study would influence the results to a large extent. We are aware of the results of the SEPAL study [20] suggesting that combined pelvic and para-aortic lymphadenectomy is beneficial as treatment for patients with endometrial carcinoma of intermediate or high risk of recurrence. The purpose of the present study, however, was not to evaluate the impact of lymphadenectomy on survival in patients with endometrial cancer but to evaluate the accuracy of using our algorithm with preoperative characteristics to predict LN metastasis in endometrial cancer. The algorithm evaluated in the present study is based on pathological hysterectomy characteristics to provide an estimation of LN metastasis. Because the SEER database does not distinguish pelvic and para-aortic LN metastasis, it was not possible to develop an algorithm distinguishing pelvic and para-aortic LN metastasis, but in most endometrial cancers, when there is paraaortic LN metastasis the pelvic area is also invaded. The findings of several studies like that of Mariani et al. focusing on para-aortic dissemination in endometrial cancer [21] suggest that isolated para-aortic lymph node metastasis in endometrial cancer is a particularly rare event. We are aware that in several countries, few women with low risk tumors would get an MRI. Several studies have shown, however, that ultrasound and MRI have similar diagnostic performances in predicting myometrial invasion [22] and MRI examination constitutes the gold standard in the management of endometrial cancer according to the European [11] and French [12] guidelines. Importantly, unlike ultrasound examination, MRI can be reviewed before surgery by other radiologists during a multidisciplinary meeting which makes it a more reproducible test. It is true that the correlation of the MRI with final pathology is rather poor, but the present study suggests that it provides useful information with the use of our algorithm. It would be interesting to know the correlation between ultrasonography with final pathology and particularly if ultrasonography would provide a good predictive value for LN metastasis with the algorithm.
119
In conclusion, one serious limitation of our model is that the pathological data were obtained from the surgical component analysis, after hysterectomy had been performed. We demonstrated in this study that our algorithm is still accurate using preoperative data. Since changing a component of the previously published algorithm from pathologic tumor invasion to the MRI interpretation of invasion is a major change, the current findings need to be confirmed on a larger scale. Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.ejogrb.2013.10.028. References [1] Siegel R, Naishadham D, Jemal A. Cancer statistics, 2012. CA Cancer J Clin 2012;62:10–29. [2] Kitchener H, Swart AM, Qian Q, Amos C, Parmar MK. Efficacy of systematic pelvic lymphadenectomy in endometrial cancer (MRC ASTEC trial): a randomised study. Lancet 2009;373:125–36. [3] Benedetti Panici P, Basile S, Maneschi F, et al. Systematic pelvic lymphadenectomy vs. no lymphadenectomy in early-stage endometrial carcinoma: randomized clinical trial. J Natl Cancer Inst 2008;100:1707–16. [4] Cragun JM, Havrilesky LJ, Calingaert B, et al. Retrospective analysis of selective lymphadenectomy in apparent early-stage endometrial cancer. J Clin Oncol 2005;23:3668–75. [5] Bristow RE, Zahurak ML, Alexander CJ, Zellars RC, Montz FJ. FIGO stage IIIC endometrial carcinoma: resection of macroscopic nodal disease and other determinants of survival. Int J Gynecol Cancer 2003;13:664–72. [6] Bendifallah S, Koskas M, Ballester M, Genin AS, Darai E, Rouzier R. The survival impact of systematic lymphadenectomy in endometrial cancer with the use of propensity score matching analysis. Am J Obstet Gynecol 2012;206:500 e1–5011e. [7] Bendifallah S, Genin AS, Naoura I, et al. A nomogram for predicting lymph node metastasis of presumed stage I and II endometrial cancer. Am J Obstet Gynecol 2012;207:197 e1–1978e. [8] Larson DM, Johnson KK, Broste SK, Krawisz BR, Kresl JJ. Comparison of D&C and office endometrial biopsy in predicting final histopathologic grade in endometrial cancer. Obstet Gynecol 1995;86:38–42. [9] Wu LM, Xu JR, Gu HY, Hua J, Haacke EM, Hu J. Predictive value of T2-weighted imaging and contrast-enhanced MR imaging in assessing myometrial invasion in endometrial cancer: a pooled analysis of prospective studies. Eur Radiol 2013;23:435–49. [10] Bremond A, Bataillard A, Thomas L, et al. Standards, Options and Recommendations 2000 for the management of patients with endometrial cancer (nonmetastatic) (abridged report). Gynecol Obstet Fertil 2002;30:902–16. [11] Kinkel K, Forstner R, Danza FM, et al. Staging of endometrial cancer with MRI: guidelines of the European Society of Urogenital Imaging. Eur Radiol 2009;19:1565–74. [12] Querleu D, Planchamp F, Narducci F, et al. Clinical practice guidelines for the management of patients with endometrial cancer in France: recommendations of the Institut National du Cancer and the Societe Francaise d’Oncologie Gynecologique. Int J Gynecol Cancer 2011;21:945–50. [13] Milam MR, Java J, Walker JL, Metzinger DS, Parker LP, Coleman RL. Nodal metastasis risk in endometrioid endometrial cancer. Obstet Gynecol 2012;119: 286–92. [14] Kamura T, Yahata H, Shigematsu T, et al. Predicting pelvic lymph node metastasis in endometrial carcinoma. Gynecol Oncol 1999;72:387–91. [15] Lee JY, Jung DC, Park SH, et al. Preoperative prediction model of lymph node metastasis in endometrial cancer. Int J Gynecol Cancer 2010;20:1350–5. [16] Kang S, Kang WD, Chung HH, et al. Preoperative identification of a low-risk group for lymph node metastasis in endometrial cancer: a Korean gynecologic oncology group study. J Clin Oncol 2012;30:1329–34. [17] Koskas M, Bassot K, Graesslin O, et al. Impact of lymphovascular space invasion on a nomogram for predicting lymph node metastasis in endometrial cancer. Gynecol Oncol 2013;129:292–7. [18] Kumar S, Medeiros F, Dowdy SC, et al. A prospective assessment of the reliability of frozen section to direct intraoperative decision making in endometrial cancer. Gynecol Oncol 2012;127:525–31. [19] Abu-Rustum NR, Gomez JD, Alektiar KM, et al. The incidence of isolated paraaortic nodal metastasis in surgically staged endometrial cancer patients with negative pelvic lymph nodes. Gynecol Oncol 2009;115:236–8. [20] Todo Y, Kato H, Kaneuchi M, Watari H, Takeda M, Sakuragi N. Survival effect of para-aortic lymphadenectomy in endometrial cancer (SEPAL study): a retrospective cohort analysis. Lancet 2010;375:1165–72. [21] Mariani A, Keeney GL, Aletti G, Webb MJ, Haddock MG, Podratz KC. Endometrial carcinoma: para-aortic dissemination. Gynecol Oncol 2004;92:833–8. [22] Antonsen SL, Jensen LN, Loft A, et al. MRI, PET/CT and ultrasound in the preoperative staging of endometrial cancer: a multicenter prospective comparative study. Gynecol Oncol 2013;128:300–8.
