Prognostic discrimination of subgrouping node-positive endometrioid uterine cancer: location vs nodal extent

Background: The 2009 International Federation of Gynecologists and Obstetricians elected to substage patients with positive retroperitoneal lymph nodes as IIIC 1 (pelvic lymph node metastasis only) and IIIC 2 (paraaortic node metastasis with or with positive pelvic lymph nodes). We have investigated the discriminatory ability of subgrouping patients with retroperitoneal nodal involvement based on location, number, and ratio of positive nodes. Methods: For 1075 patients with stage IIIC endometrioid corpus cancer abstracted from the Surveillance, Epidemiology, and End Results databases for 2003–2007, Kaplan–Meier analyses, Cox proportional hazard models, and other quantitative measures were used to compare the prognostic discrimination for disease-specific survival (DSS) of nodal subgroupings. Results: In univariate analysis, the 3-year DSS were significantly different for subgroupings by location (IIIC 1 vs IIIC 2; 80.5% vs 67.0%, respectively, P=0.001), lymph node ratio (⩽23.2% vs >23.2% 80.8% vs 67.6% P<0.001), and number of positive lymph nodes (1, 2–5, >5; 79.5, 75.4, 62.9%, P=0.016). The ratio of positive nodes showed superior discriminatory substaging in Cox models. Conclusion: Subgrouping of stage IIIC patients by the ratio of positive nodes, either as a dichotomized or continuous parameter, shows the strongest ability to discriminate the survival, controlling for other confounding factors.

Uterine cancer is the most common pelvic gynaecologic malignancy in the United States. Based on a Gynaecologic Oncology Group study of surgical staging of clinical stage I endometrial cancer, 9.3% of patients had positive pelvic lymph node involvement whereas 5.5% had positive paraaortic lymph nodes, with a total of 11.3% having either pelvic and/or paraaortic retroperitoneal metastasis (Creasman et al, 1987). The most recent modification of the International Federation of Gynecologists and Obstetricians (FIGO) staging system for endometrial cancer has elected to subclassify patients with retroperitoneal lymph node involvement (without other sites of distant metastasis) into two subgroups based on the location of the metastatic lymph nodes. Patients with only pelvic lymph node involvement are staged as IIIC 1 whereas those with positive paraaortic lymph node (with or without positive pelvic lymph nodes) are stage IIIC 2 (Pecorelli, 2009). Two recent Surveillance, Epidemiology, and End Results (SEER)-based analyses have demonstrated worse outcome for patients with stage IIIC 2 vs IIIC 1 disease (Lewin et al, 2010;Cooke et al, 2011). However these studies were limited because there was no accounting for confounders such as number of positive nodes (Touboul et al, 2001;Takeshima et al, 2006;Fujimoto et al, 2009) or lymph node ratio, which have also been shown to be prognostically important (Tang et al, 1998;Mariani et al, 2001a;Yasunaga et al, 2003;.
In this current study, we investigated the prognostic significance of the new subdivision of stage IIIC disease and compared the discriminatory ability of location, number, and ratio of positive lymph nodes controlling for other confounding factors. The identification of other subgrouping based on characteristics of lymph node involvement may have therapeutic implications.

MATERIALS AND METHODS
The SEER Program database of the United States National Cancer Institute for endometrioid uterine cancer patients during the period from 1 January 2004 to 31 December 2007 was utilised (SEER, 2010, http://www.seer.cancer.gov). Patients with nonendometriod histologies were excluded. This time period was selected because in earlier periods patients with involved paraaortic lymph nodes were included with patients with stage IV disease. Of the 22 907 patients, 1235 (5.4%) had IIIC disease. A total of 160 patients who lacked information on lymph node dissection and/or lymph node distribution were excluded, leaving 1075 patients as the study cohort. All but four patients underwent some type of hysterectomy (three had no hysterectomy and for one the type of uterine surgery was not specified).
Data on demographic, clinical-pathological, and treatment parameters were abstracted. Patients were divided into nodal subgroups based on the number of positive nodes (1, 2 -5, and 45 nodes), total number of nodes examined (p10, 11 -20, 420), and ratio of positive nodes, expressed as the percentage of positive lymph nodes to the total number of nodes examined (p10%, 10 -50%, 450%), as in our previous study . In addition, the nodal parameters were dichotomized by the median number of positive nodes (1, 41) and ratio of positive nodes (p or 4average (23.2%)) to permit comparison with the new FIGO dichotomized stage grouping.
The primary endpoint of the study was the endometrial cancer disease-specific survival (DSS). Time to death was censored in patients who died from causes other than uterine cancer. Survival analyses were performed using the Kaplan -Meier method. Pearson's w 2 -and Student's t-test were employed to compare distributions of parameters between subgroups. Two-sided P-values of o0.05 were considered statistically significant. Pearson correlations were used to investigate for multiple colinearities between the subgrouping of lymph nodal involvement based on location, number of positive nodes, and ratio of positive nodes. Because of the potential correlation between the various subgroupings of lymph nodes, separate stepwise Cox regression models were employed entering only one of the three subgroupings of the lymph nodes in each model. Because preliminary analysis demonstrated a significantly higher number of positive lymph nodes for patients with IIIC 2 vs IIIC 1 disease, comparisons were also made between the three nodal subgroupings for a subset of 487 patients with only one positive lymph node. Three additional quantitative measures were used to compare the prognostic discrimination for DSS for nodal subgrouping (Gimotty et al, 2005). All statistical analyses were performed using the SPSS Statistics GradPack 17.0, Release 17.0.0 (3 August 2008, IBM, Armonk, NY, USA).

RESULTS
The demographic and clinical characteristics of the 1075 patients with stage IIIC endometrioid corpus cancers are delineated in Table 1. A total of 725 patients (67.4%) had positive pelvic nodes only (stage IIIC 1) whereas 350 (32.6%) had paraaortic node involvement with or without positive pelvic nodes (stage IIIC 2). The average number of lymph nodes examined was 17.3 (range: 1 -90). The average number of positive nodes was three (range: 1 -82) and the average lymph node ratio was 23.2% (range 0.01 -100%). Adjuvant radiation therapy was given to 638 (59.3%) of the patients. The median follow-up was 18 months (mean 19.4, range 0 -47).
Significantly poorer DSS was seen in higher grade tumours (Po0.001) and with the lack of adjuvant radiation (Po0.001). In a separate analysis (data not shown) the patients were grouped by number of lymph nodes examined (o10, 10 -20, X20). Even for the subgroups with X20 nodes examined, significantly lower DSS was seen with increasing number of positive nodes, stage III C2 vs stage III C1, and higher ratio of positive nodes (Po0.001).
On multivariate analysis, grade of tumour (Po0.001), ratio of positive nodes (P ¼ 0.005), adjuvant radiation (Po0.001), and marital status (Po0.01) were independent factors associated with DSS whereas location of the positive nodes, number of positive nodes, and number of nodes examined were not significant ( Table 4). The hazard ratio for the ratio of positive nodes as a continuous variable was 3.10.
Pearson correlation coefficients were employed to look for colinearity between the subgroupings of positive lymph nodes (location, number of positive nodes, and node ratios) and showed small correlations between these variables (0.106, 0.263, and 0.321, respectively). Repeat multivariate analyses were performed by A subset analysis was performed on those patients with one positive node (n ¼ 487); of which 394 (81.7%) had one positive pelvic node whereas 89 (18.3%) had one positive paraaortic node (Supplementary Table S1). There was no significant difference in 3-year DSS for patients with one positive pelvic node compared with those with one positive paraaortic node (80% vs 77.3%, respectively, P ¼ 0.675, Figure 2). These findings were confirmed on multivariate analysis (Supplementary Table S2).
The relative discriminatory ability of subgrouping patients with IIIC disease by location (IIIC 1 vs IIIC 2, number of positive nodes (1 vs 41), and ratio of positive nodes (p23.2% vs 423.2%) is shown in Supplementary Table S3. The hazard ratios for the ratio of positive nodes was 2.20, location (IIIC 1 vs IIIC 2) was 1.72 and 1 vs 41 positive nodes 1.15 (Table 5).

DISCUSSION
Uterine cancer is the most common gynaecologic malignancy in the United States with 43 470 new cases and 7950 deaths expected for 2010 (Jemal et al, 2010). There has been an increase in the number of deaths, particularly in those with advanced stage (III/IV) disease (Ueda et al, 2008). Although the majority have excellent prognosis with 5-year survival rates between 80 -91%, the B5 -10% of patients presenting with retroperitoneal lymph node involvement have inferior survival (Creasman et al, 2006;Lewin et al, 2010). Those with stage IIIC disease have survivals ranging from 10% to 75% . This, in part, reflects the heterogeneity in nodal and other prognostic parameters in stage IIIC cancers. The recent revision of FIGO staging of endometrial cancer has subdivided retroperitoneal node-positive patients into two subgroups based solely on the location of the positive nodes (Pecorelli, 2009).
Prior studies have shown a wide variation in survival based on lymph node location. In several series patients with involvement of pelvic lymph nodes only had nonsignificant differences in survival compared with those with positive paraaortic lymph nodes (McMeekin et al, 2001;Mariani et al, 2002;Otsuka et al, 2002;Havrilesky et al, 2005;Hoekstra et al, 2009;Lewin et al, 2010;Todo et al, 2011). In other series better survival was noted for patients with pelvic nodes (Morrow et al, 1991;Hirahatake et al, 1997;Yokoyama et al, 1997;Watari et al, 2005;Fujimoto et al, 2007;Karube et al, 2010;Lewin et al, 2010). Furthermore one recent series has shown a better survival for patients with positive paraaortic lymph nodes compared with those with positive pelvic lymph nodes (80% vs 55% at 5 years), although the difference was not statistically significant (Klopp et al, 2009). However, many of these studies did not account for the number of positive and ratio of positive nodes in addition to the location of positive nodes. Our current study analysed 1075 patients with endometrioid uterine cancer to confirm the independent prognostic significance of this new subgrouping based on node location. In addition, we compared the prognostic discrimination of nodal location with the number of positive and ratio of positive nodes.
Nodal location was found to be a significant prognostic factor both in univariate (Table 2, Figure 1A) and multivariate analysis when it was entered as the only term relating to the lymph nodes. However, the new subgrouping by nodal location was not shown to be statistically significant in multivariate analysis when the number of positive nodes and ratio of positive nodes were included (Table 4). As patients in our study with stage IIIC 2 disease had a higher number of positive lymph nodes than those with stage IIIC 1 disease (4.2 vs 2.0, respectively, Po0.001) we performed a separate analysis for those with only one positive retroperitoneal lymph node. This subgroup analysis showed no significant difference in DSS based on nodal location (Figure 2), while the ratio of positive nodes remained significant (Supplementary Table S2).
Our findings on the prognostic significance of the number of positive lymph nodes in univariate analysis ( Figure 1B) confirm previous reports (Morrow et al, 1991;Takeshima et al, 1994;Touboul et al, 2001;. Watari et al (2005) demonstrated a better 5-year survival rate for patients with one positive paraaortic lymph node group compared with those with X2 positive paraaortic lymph node groups (60.4% vs 20.0%, respectively, P ¼ 0.0319) whereas Fujimoto et al (2009) reported better 5-year relapse-free survival for patients with one positive pelvic lymph node site compared with those with X2 positive sites (81.3% vs 41.2%, respectively, P ¼ 0.04). We also showed the prognostic significance of the ratio of positive nodes to the total number of lymph nodes examined, which confirms our prior report ). In the current study, the ratio of positive nodes was significant whether it was entered as a continuous variable in multivariate analysis, dichotomized at the mean of 23.2%, or subgrouped based on p10%, 10 -50%, or 450% involvement. The results from single institutional respective studies have also demonstrated the prognostic significance of ratio of positive lymph nodes (Tang et al, 1998;Mariani et al, 2001b). Studies in other malignancies have also attempted to define the most prognostically significant subgroupings for lymph node positive patients. Various classification schemes for lymph nodes in gastric cancer have been based on the distance, number, and anatomical location of metastatic nodes as well as the site of the primary tumour (Kajitani, 1981;Hermanek and Sobin, 1992;Adachi et al, 1995;Sobin and Wittekind, 1997;de Manzoni et al, 1999). Classification of involved regional lymph nodes in gastric cancer by the ratio of positive nodes was found to represent a simple, reliable, and reproducible staging system (Yu et al, 1997;Liu et al, 2007;Marchet et al, 2007;Persiani et al, 2008;Zhang et al, 2009;Maduekwe et al, 2010;Sianesi et al, 2010).
The major shortcoming of any substaging of endometrial cancer based on measurements of nodal involvement is the lack of standardisation of the lymphadenectomy. There is wide variation in the extent of nodal dissection reflecting both surgeon's bias and patient selection. For example, this could include performing a more limited lymph node dissection following a resection of an involved bulky node, or performing a more extensive nodal dissection in patients without bulky nodes or co-morbidities (Smith et al, 2008). The issue of standardisation of lymph node dissection has been thoroughly reviewed previously (Boronow, 1980;Kilgore et al, 1995;Chang et al, 2008;Mariani et al, 2009). As discussed by Mariani et al (2009) in their commentary on the surgical staging of endometrial cancer, a standardisation of lymphadenectomy including the anatomical extent of the paraaortic lymph node dissection is lacking. The minimum requirement of lymphadenectomy, either in terms of nodal stations resected or total number of lymph nodes examined, has not been unambiguously defined in the FIGO staging system. Recommendations as to the minimum number of lymph nodes examined for adequate nodal staging have been in effect for colon cancer (12 nodes) (Nelson et al, 2001)    and gastric cancer (15 nodes) (Green et al, 2010). We are in agreement with the NCCN guidelines for the treatment of uterine cancer recommending a complete pelvic and paraaortic lymphadenectomy (unless technically unfeasible or medically contraindicated), adhering to the ACOG surgical policy (ACOG, 2005).
Two prospective randomized trials have failed to demonstrate a survival advantage from pelvic lymphadenectomy in endometrial cancer (Benedetti Panici et al, 2008;Kitchener et al, 2009). However the inclusion of low-risk patients, lack of standardisation of systemic postoperative treatments, and minimal or lack of paraaortic lymphadenectomy are limitations of these studies (Amant et al, 2009;Uccella et al, 2009;Seamon et al, 2010). A recent retrospective study in patients with stage III C endometrial cancer demonstrated the therapeutic significance of systematic lymphadenectomy including both pelvic and paraaortic node dissection (Todo et al, 2011).
Additional limitations of our study include the lack of information on other patient and treatment factors that may be of prognostic significance in patients with retroperitoneal node involvement. In particular, there is a lack of information on the extent of the pelvic and/or paraaortic lymphadenopathy, the extent of surgical staging, the surgeon's subspecialty, the extent of lymph node debulking, involvement of other pelvic extrauterine sites including the adnexa and peritoneal cytology, involvement of the uterine cervix, depth of myometrial invasion, lymph vascular space invasion, and size of the lymph nodes. Our study was limited to patients with endometrioid histology, relatively short follow-up, and there was no central pathology review. There is also a lack of information on sites of recurrence and the use of adjuvant systemic chemotherapy and hormonal therapy. However, the   Figure 2 Kaplan -Meier disease-specific survival for stage IIIC endometrioid cancer patients (n ¼ 487) with only one positive node based on lymph node location (pelvic vs paraaortic); P ¼ 0.675.
recent years of diagnosis of the patients included in this study should make them more likely to have received adjuvant treatment with chemotherapy or volume-directed radiation therapy and chemotherapy than studies including earlier cohorts of SEER patients. Other general limitations of SEER-based research including variation in data registry, underreporting of radiation therapy, lack of details on adjuvant radiation therapy (fields treated and doses), and selection bias have recently been reviewed by Yu et al (2009). The strengths of our analysis include the large number of recently diagnosed patients with node-positive endometrioid uterine cancers studied within a wide geographic distribution in the United States. In addition, our univariate and multivariate analysis of the three major subgroupings of stage IIIC patients (based on the new FIGO substaging, number of positive lymph nodes reported, and lymph node ratio) has permitted identification of the subgroupings with better abilities to discriminate DSS in this heterogeneous group of stage IIIC patients.
In summary, better classification of retroperitoneal lymph node-positive endometrioid uterine cancer patients may permit the identification of more homogenous subgroupings for prognostic purposes, stratification in clinic trials, and possible better selection for individualised adjuvant-combined modality treatments (Mariani et al, 2004). Higher risk subgroups, for example those with multiple pelvic and paraaortic nodal involvement, may require more intense chemotherapy regimens, whereas those with limited nodal disease may best be managed with volumedirected radiation therapy and less toxic systemic treatment protocols. Our study has confirmed the value of subgrouping stage IIIC patients based on nodal location, number of positive lymph nodes and ratio of positive nodes. However based on multivariate and discrimination analyses, nodal ratio was a stronger discriminator for DSS than nodal location, controlling for other confounding factors including tumor grade and the use of adjuvant radiation therapy. If our results are validated in other patient databases, these findings may permit better modifications of the substaging of retroperitoneal lymph node positive patients. However, it is stressed that standardisation of lymphadenectomy including the boundaries of resection, uniform processing of the nodal specimens, and the criteria for adequacy of lymph node resection are needed.
Supplementary Information accompanies the paper on British Journal of Cancer website (http://www.nature.com/bjc) This work is published under the standard license to publish agreement. After 12 months the work will become freely available and the license terms will switch to a Creative Commons Attribution-NonCommercial-Share Alike 3.0 Unported License.