Objective
We sought to evaluate whether patients with endometrial cancer in the Surveillance, Epidemiology, and End Results database who underwent lymphadenectomy demonstrate improved survival.
Study Design
The study population comprised 50,969 patients. The 3-year cause-specific survival was tested by using propensity score matching (PSM) analysis.
Results
The PSM analysis generated a balanced, matched cohort in which baseline characteristics were not significantly different. The benefit of systematic lymphadenectomy appears to be significant for presumed stage I International Federation of Gynecology and Obstetrics grade 3 cancers and presumed stages II-III cancer. The omission of lymphadenectomy in stage I did not appear to show a deleterious survival consequence if the differentiation grade was moderate (grade 2) or well (grade 1).
Conclusion
Using PSM analysis, our results show no evidence of benefit in terms of survival for systematic lymphadenectomy in women with stage I endometrial cancer, except for grade 3 cancers.
Endometrial cancer (EC) is the most common gynecological malignancy in Western Europe and North America. About 81,500 women in the European Union and 40,100 women in North America are affected every year. Approximately 85% of these patients are diagnosed with stage I or II disease. Although retrospective studies have suggested a significant survival advantage following systematic lymphadenectomy in early disease stages, debate regarding its necessity led to the performance of randomized clinical trials (RCTs). Among these, A Study in the Treatment of EC (ASTEC) trial showed that pelvic lymphadenectomy had no effect on overall or recurrence-free survival. Similarly, Benedetti Panici et al reported that the risks for first event and death were similar between groups with and without lymphadenectomy. These results from RCTs should be interpreted with caution because of some inaccuracies in studies design and a lack of sufficient power. Additional data are needed to provide evidence-based treatment algorithms. RCTs are the gold standard for treatment and practice evaluation, however these are also often criticized for their limited external validity, resulting from the enrollment of highly selected patient groups. Non-RCTs or observational studies are possible alternatives for assessing treatment effects because these are expected to have greater external validity. However in retrospective observational studies, patients receive treatment according to tumor and patient (age, health status) factors, thus biasing the comparisons. Among the commonly used statistical techniques to reduce the effect of bias, multivariate analyses such as logistic regression or Cox proportional hazards models are frequently used. The technique of propensity score (PS) matching (PSM) analysis has been proposed as an alternative method to adjust for confounding factors with statistical advantages over the standard methods of confounder adjustment. The PSM method can be used to reduce the impact of treatment selection bias using observational data. The method involves generation of a PS for each subject, which is an estimate of the conditional probability of receiving a treatment given a set of known covariates. PSs are used to reduce selection bias by equating groups based on these covariates. The result of the matching process is the reduction of covariate imbalance between the treatment and control groups, reducing biases in the analysis of treatment effect. PSM has been used in several areas of biostatistical and medical research, and has been applied in oncology to assess treatment efficacies. Several studies have shown that the PSM analysis could help to determine treatment effect in populations of cancer patients. The PSM methodology has not previously been applied to evaluate the therapeutic effect of lymphadenectomy in EC. We applied this method to the Surveillance, Epidemiology, and End Results (SEER) database, which collect patient information from approximately 14% of the US population throughout the country. These data have been shown to be reliable in reporting surgical procedures. The aim of this retrospective study was to examine the survival impact of lymphadenectomy in women with EC by using PSM techniques.
Materials and Methods
Data and study population
Using SEER registry public use data tapes provided from 17 registries, we identified patients with histologically proven EC (primary site: C54.1 [endometrium]) from Jan. 1, 1988, through Dec. 31, 2007. All information was publicly available, deidentified, and exempt from institutional review board review. Baseline, demographic, clinical, and pathological data were generated using specific SEER codes. The following variables were included in the analysis: region of diagnosis (East, Central, West); period of diagnosis (1988 through 1994, 1995 through 2000, 2001 through 2007); age at diagnosis (<40, 40-65, >65 years); race (white, black, other); histologic subtype of adenocarcinoma (combination of endometrioid carcinoma and adenocarcinoma), papillary serous, and other types (combination of clear-cell, squamous, and mucinous tumors); grade differentiation (well = 1, moderate = 2, poor = 3-4); and the primary site tumoral characteristics (endometrium and <50% myometrial invasion, ≥50% of the myometrium, cervical stroma, vaginal and adnexae extent).
We included women who underwent surgery for primary EC that involved at least hysterectomy (with or without lymphadenectomy). Treatment characteristics were documented, including the use of adjuvant radiation therapy (external beam, brachytherapy only, and external beam plus brachytherapy) and whether lymphadenectomy was performed. We reported the extent of the lymphadenectomy (<10, 10-20, >20 nodes resected). Before PSM, group differences were tested using the χ 2 test.
PS and matching procedures
PSs to determine the conditional probability of receiving lymphadenectomy were generated using logistic regression model as described by Rosenbaum and Rubin regressing the surgery assignment (0 = no lymphadenectomy, 1 = lymphadenectomy) on patient demographic, pathologic, and preexisting social characteristics. The following covariates were included in the PS model: period of diagnosis, region of diagnosis, age at diagnosis, race, tumor grade differentiation, histologic type, and the primary site tumoral characteristics. The PS for each individual was obtained by the predicted probabilities from the logistic procedure. The c-statistic for this model was 0.76 (0.71-0.81). Patients were then matched (a 1:1 match) using the PS by an optimal matching algorithm by randomly selecting for each patient who underwent lymphadenectomy a corresponding nonlymphadenectomized patient with the closest PS. To find matched patients from the 2 groups, we adopted a caliper matching approach that has the ability to avoid bad matches (too large differences in PS). Once a control was matched, the control was not considered again as a match for any other case (ie, matching without replacement). The resulting matched sets with the same PS tend to have similar covariate distributions in the 2 groups.
Survival analysis
Analysis was restricted to matched patients. Three-year cause-specific survival (CSS) and other-cause survival (OCS) were calculated for the entire cohort. Then, 3-year CSS was analyzed with stratification for presumed stage using the 2009 International Federation of Gynecology and Obstetrics (FIGO) classification and according to grade differentiation. CSS was measured as the time from diagnosis to the date of death by EC or date of the last follow-up. Patients alive at the last follow-up were censored. We used the Kaplan-Meier product limit method to describe 3-year CSS and OCS and the log rank test to assess the differences between patient groups. For each group, survival is reported with 95% confidence interval (CI). A multivariate survival analysis was performed using the Cox proportional hazard model before and after matching.
All tests were 2 tailed, and P values < .05 were considered to denote significant differences. All analyses were performed using the R package with the Design, Hmisc, Matching, MatchIt, Design, and survival libraries.
Results
Study populations
Table 1 summarizes the descriptive statistics for the selected groups before and after PSM. We identified 50,969 women with EC during the study period from the SEER register. Before PSM, 28,798 (56.5%) patients had undergone lymphadenectomy procedure (lymph node examination performed group), and 22,171 (43.5%) did not (lymph node examination not performed group). Most patients were white (88%). The mean patient age was 60.5 and 62.9 years for women who underwent lymphadenectomy and those who did not, respectively ( P < .0001). There were significant imbalances in baseline characteristics between the 2 groups ( Table 1 ). After the matching based on estimated PS, 11,400 patients who underwent lymphadenectomy were successfully matched one-to-one with 11,400 nonlymphadenectomized patients. In total, 28,169 women were excluded from the matched cohort due to the lack of an appropriate match. After the matching, the mean age was 62.5 and 62.9 years for women who underwent lymphadenectomy and those who did not, respectively ( P = .7592). PSM generated a matched cohort where the baseline characteristics included in the PS were well balanced and not significantly different between the 2 groups ( Table 1 ). The median follow-up period was 78.0 months (range, 0–251 months) and 86.0 months (range, 0–251 months) before and after matching, respectively.
| Before propensity score matching, % (n) | After propensity score matching, % (n) | |||||||
|---|---|---|---|---|---|---|---|---|
| Variable | All patients N = 50,969 | Lymphadenectomy not performed n = 22,171 | Lymphadenectomy performed n = 28,798 | P value | All patients N = 22,800 | Lymphadenectomy not performed n = 11,400 | Lymphadenectomy performed n = 11,400 | P value |
| Year of diagnosis | ||||||||
| 1988-1994 | 21 (10,452) | 33 (7321) | 11 (3131) | 20 (4510) | 20 (2259) | 20 (2251) | ||
| 1995-2000 | 29 (14,834) | 38 (8330) | 23 (6504) | 35 (7897) | 34 (3909) | 35 (3988) | ||
| 2001-2007 | 50 (25,683) | 29 (6520) | 67 (19,163) | < .0001 | 45 (10,393) | 46 (5232) | 45 (5161) | .5247 |
| Region of diagnosis | ||||||||
| East | 27 (13,939) | 24 (5372) | 30 (8567) | 26 (5903) | 26 (2976) | 26 (2927) | ||
| Central | 18 (8993) | 22 (4887) | 14 (4106) | 19 (4249) | 18 (2108) | 19 (2141) | ||
| West | 55 (28,037) | 54 (11,912) | 56 (16,125) | < .0001 | 55 (12,648) | 55 (6316) | 56 (6332) | .7106 |
| Age at diagnosis, y | ||||||||
| <40 | 3 (1497) | 3 (689) | 3 (808) | 3 (662) | 3 (323) | 3 (339) | ||
| 40-65 | 55 (27,947) | 53 (11,717) | 56 (16,230) | 54 (12,292) | 54 (6136) | 54 (6156) | ||
| >65 | 42 (21,525) | 44 (9765) | 41 (11,760) | < .0001 | 43 (9846) | 43 (4941) | 43 (4905) | .7592 |
| Race | ||||||||
| White | 88 (44,873) | 90 (19,946) | 87 (24,927) | 88 (20,122) | 89 (10,101) | 88 (10,021) | ||
| Black | 5 (2492) | 4 (879) | 6 (1613) | 5 (1126) | 5 (536) | 5 (590) | ||
| Other | 7 (3604) | 6 (1346) | 8 (2258) | < .0001 | 7 (1552) | 7 (763) | 7 (789) | .1879 |
| Tumor grade | ||||||||
| 1 | 46 (23,354) | 59 (12,977) | 36 (10,377) | 48 (11,002) | 49 (5543) | 48 (5459) | ||
| 2 | 36 (18,229) | 32 (7101) | 39 (11,128) | 38 (8595) | 38 (4279) | 38 (4316) | ||
| 3-4 | 18 (9386) | 9 (2093) | 25 (7293) | < .0001 | 14 (3203) | 14 (1578) | 14 (1625) | .4746 |
| Histologic subtype | ||||||||
| Adenocarcinoma | 95 (48,309) | 97 (21,584) | 93 (26,725) | 96 (21,796) | 96 (10,905) | 95 (10,861) | ||
| Papillary serous | 2 (1128) | 1 (235) | 3 (893) | 2 (483) | 2 (241) | 2 (272) | ||
| Other | 3 (1532) | 2 (352) | 4 (1180) | < .0001 | 2 (521) | 2 (254) | 2 (267) | .3187 |
| Primary site tumoral characteristics | ||||||||
| Endometrium and <50% myometrial invasion | 71 (36,055) | 76 (16,800) | 67 (19,255) | 74 (16,880) | 75 (8458) | 73 (8422) | ||
| ≥50% of myometrium | 22 (11,294) | 20 (4452) | 24 (6842) | 21 (4848) | 21 (2412) | 21 (2436) | ||
| Cervical stroma extent | 4 (1948) | 1 (268) | 6 (1680) | 2 (509) | 2 (255) | 2 (254) | ||
| Vaginal and adnexae extent | 3 (1672) | 3 (651) | 4 (1021) | < .0001 | 3 (563) | 2 (275) | 2 (288) | .9193 |
| FIGO 2009 | ||||||||
| I | ||||||||
| IA | 70 (35,503) | 76 (16,800) | 65 (18,703) | 73 (16,685) | 74 (8458) | 72 (8227) | ||
| IB | 20 (10,423) | 20 (4452) | 21 (5971) | 20 (4639) | 21 (2412) | 19 (2227) | ||
| II | ||||||||
| II | 3 (1531) | 1 (268) | 4 (1263) | 2 (451) | 2 (255) | 2 (196) | ||
| III | ||||||||
| IIIA | 2.7 (1407) | 2.8 (605) | 2.8 (802) | 2.5 (552) | 2 (242) | 3 (310) | ||
| IIIB | 0.3 (183) | 0.2 (46) | 0.5 (137) | 0.5 (76) | 1 (33) | 1 (43) | ||
| IIIC | 4 (1922) | — | 6.7 (1922) | < .0001 | 2 (397) | — | 3 (397) | .0003 |
| Radiotherapy | ||||||||
| Not performed | 75 (38,676) | 83 (18,475) | 70 (20,201) | 79.5 (18,126) | 87 (9918) | 72 (8208) | ||
| Performed | 25 (12,293) | 17 (3696) | 30 (8597) | 20.5 (4674) | 13 (1482) | 28 (3192) | ||
| EB | 50 (6025) | 56 (2070) | 46 (3955) | 49 (2298) | 56 (830) | 46 (1468) | ||
| Intracavitary implants | 28 (3490) | 20 (739) | 32 (2751) | 28 (1286) | 20 (296) | 31 (990) | ||
| EB + intracavitary implants | 22 (2778) | 24 (887) | 22 (1891) | < .0001 | 23 (1090) | 24 (356) | 23 (734) | <.0001 |
| Extent of lymphadenectomy, nodes resected | ||||||||
| <10 | — | 43 (12,311) | — | — | 46 (5209) | |||
| 10-20 | — | 34 (9787) | — | — | — | 33 (3759) | ||
| >20 | — | 23 (6700) | — | — | 21 (2432) | — | ||
| Metastatic nodes | ||||||||
| None positive | — | 93 (26,876) | — | 90 (10,348) | ||||
| Positive | — | 7 (1922) | — | — | 10 (1052) | — | ||
| Vital status | ||||||||
| Alive | 75 (38,350) | 69 (15,290) | 80 (23,060) | 87.5 (19,950) | 88 (10,032) | 87 (9918) | ||
| Dead | 25 (12,619) | 31 (6881) | 20 (5738) | 12.5 (2850) | 12 (1368) | 13 (1482) | ||
| Because of EC | 26 (3244) | 19 (1272) | 35 (1972) | 36 (1815) | 59 (807) | 68 (1008) | ||
| Because of other cause | 74 (9375) | 81 (5609) | 65 (3766) | — | 64 (1035) | 41 (561) | 32 (474) | — |
Propensity score
The adjusted multivariate odds ratio estimates for the logistic regression model covariates are presented in Table 2 . The distribution of PS of patients who underwent lymphadenectomy and those who did not is illustrated in the Appendix , both before and after PSM. This demonstrates imbalances in the baseline distribution of PS before matching and a similar distribution afterward, which is related to good matching results.
| Variable | OR (95% CI) | P value |
|---|---|---|
| Year of diagnosis | ||
| 1988-1994 | Referent | |
| 1995-2000 | 1.96 (1.85–2.07) | |
| 2001-2007 | 7.79 (7.37–8.24) | < .001 |
| Region of diagnosis | ||
| Central | Referent | |
| West | 1.16 (1.1–1.23) | |
| East | 1.11 (1.05–1.18) | < .001 |
| Age at diagnosis, y | ||
| <40 | Referent | |
| 40-65 | 0.81 (0.78–0.85) | |
| >65 | 0.95 (0.85–1.07) | < .001 |
| Race | ||
| Black | Referent | |
| White | 0.92 (0.84–1.02) | |
| Other | 1.13 (1–1.27) | < .001 |
| Tumor grade | ||
| 1 | Referent | |
| 2 | 2.14 (1.93–2.38) | |
| 3-4 | 4.57 (4.28–4.88) | < .001 |
| Histologic subtype | ||
| Adenocarcinoma | Referent | |
| Others | 1.21 (1.05–1.39) | |
| Papillary serous | 1.37 (1.17–1.62) | < .001 |
| Primary site tumoral characteristics | ||
| Vaginal and adnexae extent | Referent | |
| Endometrium and <50% myometrial invasion | 0.95 (0.84–1.06) | |
| ≥50% of myometrium | 1.52 (1.35–1.72) | |
| Cervical stroma extent | 2.43 (2.04–2.9) | < .001 |
Survival analysis
In the matched population, the respective 3-year CSS rates according to grades 1, 2, and 3 were 99.0% (95% CI, 98.7–99.3) vs 99.1% (95% CI, 98.8–99.3), P = .062 ( Figure 1 , A) ; 97.0% (95% CI, 96.5–97.5) vs 96.9% (95% CI, 96.3–97.4), P = .098 ( Figure 1 , B); and 88.3% (95% CI, 87.1–895) vs 83.1% (95% CI, 80.7–85.6), P < .0001 ( Figure 1 , C) for women who underwent lymphadenectomy and those who did not, respectively.
When staging system stratification was performed on presumed stage I FIGO, the beneficial effect of lymphadenectomy was significantly observed for lymphadenectomized women with poorly differentiated tumors, ie, for grade 3 cases: 91.1% (95% CI, 89.9–92.2) vs 88.8% (95% CI, 86.5–91.0), P = .0202 ( Figure 2 , C) . In contrast, no beneficial effect of lymphadenectomy was observed in presumed stage I FIGO patients with moderate (grade 2) and well-differentiated (grade 1) tumors. The 3-year CSS rates were 97.3% (95% CI, 96.8–97.8) vs 97.5% (95% CI, 96.9–98.0), P = .1147 ( Figure 2 , B) and 99.2% (95% CI, 99.0–99.5) vs 99.3% (95% CI, 99.0–99.5), P = .0141 ( Figure 2 , A) for women who underwent lymphadenectomy and those who did not, respectively.
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree
