Possible overestimation of treatment effects of pelvic and para-aortic lymphadenectomy for early-stage ovarian clear cell carcinoma: a retrospective propensity-score weighted multi-center cohort study

Abstract

Objective

The treatment effects of lymphadenectomy in early-stage ovarian clear cell carcinoma (OCCC) reported in previous studies may have been overestimated owing to confounding factors. This study aimed to investigate the treatment effect of pelvic and para-aortic lymphadenectomy (PeNPAN) in early-stage OCCC, with careful adjustment for potential confounders.

Methods

This retrospective multi-center cohort study involved women with preoperatively suspected stage I OCCC. We included patients who underwent surgery for OCCC between 2005 and 2019 at 11 affiliated institutions. The exposure (PeNPAN) group comprised patients who underwent PeNPAN. The primary outcome was disease-free survival (DFS). Additionally, hazard ratios (HRs) of lymphadenectomy for DFS were estimated using unadjusted and propensity score-weighted Cox regression models and biased models applied in previous studies. To identify strong confounders, we further examined factors associated with recurrence that differed between the groups.

Results

We analyzed 304 women who underwent surgery for preoperatively suspected stage I OCCC. The unadjusted HR for DFS was 0.63 (95% confidence interval [CI]=0.36–1.09; p=0.10), and the propensity-score adjusted HR was 0.82 (95% CI=0.42–1.58; p=0.55). The biased model showed a statistically significant HR of 0.59 (95% CI=0.36–1.00; p=0.048). Adhesions in the Douglas’ pouch and cardiovascular disease were associated with recurrence and were more prevalent in the control group, suggesting potential confounders.

Conclusion

After adjusting for potential confounders, the observed treatment effects of lymphadenectomy in the biased models were no longer statistically significant. Future investigations should carefully account for possible confounders, including intraoperative adhesions and comorbidities.

Synopsis

The adjusted hazard ratio (HR) of lymphadenectomy for early-stage ovarian clear cell carcinoma was 0.82 [0.42–1.58; p=0.55], and was inconsistent with HR estimates from the unadjusted and biased models. This discrepancy may results from the imbalance in recurrence-related factors including pelvic adhesions and cardiovascular disease.

INTRODUCTION

Among the gynecological malignancies, ovarian cancer has a poor prognosis. In the East Asian population, ovarian clear cell carcinoma (OCCC) is the second most prevalent histological type after serous carcinoma, accounting for approximately 20% of all ovarian cancers [1,2]. OCCC is associated with endometriosis and has a distinct pathogenesis compared to serous carcinoma [3,4]. Serous carcinoma is often diagnosed at an advanced stage. However, OCCC is more frequently detected at an early stage, with stage I being the most common [2,4,5]. Early-stage OCCC is primarily treated with surgery, which includes hysterectomy, bilateral salpingo-oophorectomy, omentectomy, and pelvic and para-aortic lymphadenectomy (PeNPAN) [5]. However, the treatment effect of lymphadenectomy, which is typically performed as part of staging, remains unclear [6,7,8].

Previous studies suggesting the therapeutic effects of PeNPAN did not consider important confounding factors, such as pelvic adhesions and comorbidities [6,7,8]. A recent large database study suggested a therapeutic effect of lymphadenectomy; however, the study possesses limitations, including difficulty in considering intraoperative findings, which were not accounted for in the study [9]. Consequently, the therapeutic effect of lymphadenectomy reported in previous studies may have been overestimated due to confounding factors from the higher proportion of poor prognosis cases in the non-lymphadenectomy group.

This study aimed to investigate the therapeutic effects of PeNPAN in women with preoperatively suspected stage I OCCC, with careful adjustment for potential confounding variables.

MATERIALS AND METHODS

1. Study design and setting

This multi-center retrospective cohort study utilized electronic medical records from 11 hospitals from the Kyoto Academy of Obstetrics and Gynecology Aiming for Women’s Aid (KAMOGAWA) study group [10]. This study was approved by the Institutional Review Board of Shizuoka General Hospital (reference number: SGHIRB#2020071). We included patients with OCCC who underwent surgery between January 2005 and December 2019 with preoperative suspicion of stage I disease. Women who underwent surgery before the digitization of medical records and those who underwent cystectomy, with preoperative lymph node enlargement (short axis ≥10 mm), performance status ≥2, or who underwent fertility-sparing surgery were excluded.

2. Exposure and outcomes

The exposure (PeNPAN) group included patients who underwent PeNPAN, while those who did not undergo lymphadenectomy or underwent only pelvic lymphadenectomy comprised the control group. Cases involving only lymph node sampling were excluded from the PeNPAN group. Two-stage surgery with an interval of <3 months was considered equivalent to single-stage surgery. Disease-free survival (DFS) and overall survival (OS) were defined as the primary and secondary outcomes, respectively. The index time was established as the date of the first OCCC surgery. Follow-up data up to 10 years postoperatively or until December 2019 were analyzed.

3. Statistical analysis

Statistical analyses were conducted using R version 4.0.5 (R Foundation, Vienna, Austria). Overlap weighting was employed to adjust for baseline characteristics between the 2 groups [11]. The propensity scores for PeNPAN were estimated using a logistic regression model. The following variables were incorporated into the logistic regression model as covariates based on clinical knowledge: patient age, body mass index (BMI), comorbidities (diabetes mellitus and cardiovascular disease), use of immunosuppressants, preoperative tumor size (cyst size), preoperative cyst rupture including positive cytology, intraoperative cyst rupture, year of surgery, adhesion in the Douglas’ pouch, pelvic adhesions, and hospital characteristics. Moreover, preoperative and intraoperative cyst ruptures were considered distinct variables owing to their different staging classifications. For hospital characteristics, we ascertained whether or not the hospitals were accredited institutions of the Japan Society of Gynecologic Oncology (JSGO), which is associated with oncologic outcomes in gynecologic malignancies [12]. Further definitions of each variable are described in our previous report [10]. Given that postoperative staging and adjuvant chemotherapy status are intermediate factors between surgery and outcomes, and these are not available at the time of surgery, they were excluded from the propensity score calculation. Patients with missing covariate data were also excluded, and a complete case analysis was performed.

Kaplan–Meier curves were estimated before and after propensity score weighting. Hazard ratios (HRs) and 95% confidence intervals (CIs) of PeNPAN for DFS and OS were calculated using unweighted and overlap-weighted Cox proportional hazards models. The characteristics of the study population were summarized using proportions for categorical variables and mean and standard deviations for continuous variables. Differences in the baseline characteristics of women between the groups were assessed using standardized differences. A standardized difference exceeding 0.1 was typically interpreted as a significant imbalance between the groups [13].

Our results suggest that the treatment effect of PeNPAN on DFS was overestimated before propensity score adjustment. We explored the underlying causes by calculating the unadjusted HR for DFS in variables that exhibited an imbalance between the groups before adjustment.

Additionally, for reference, we calculated HRs from 2 biased models that are frequently used in retrospective studies. First, even in our unadjusted model, we considered potential confounders such as poor performance status and fertility-sparing surgery by excluding patients with these factors during the patient selection process. Therefore, we performed an analysis without the exclusion criteria for possible confounders, as seen in previous studies (Biased Model 1) [6,9]. Furthermore, we conducted an analysis restricted to patients who were confirmed as stage I based on postoperative pathological diagnosis (Biased Model 2) [14]. The second model, which is occasionally encountered in the literature [14], may introduce bias that overestimates the treatment effect of PeNPAN, as preoperatively undiagnosed stage III cases with lymph node metastases are excluded only from the PeNPAN group.

RESULTS

1. Characteristics of the study population

Fig. 1 illustrates the selection of study participants. Between January 2005 and December 2019, 335 women underwent surgery for OCCC, which was preoperatively suspected to be stage I disease. After excluding 28 individuals who did not meet the eligibility criteria, 191 and 116 women were assigned to the PeNPAN and control groups, respectively. Among these women, one in the PeNPAN group (0.5%) and 2 in the control group (1.7%) were excluded from the analysis because of missing data. Ultimately, 190 and 114 women from the PeNPAN and control groups, respectively, were included in the final analysis. Lymph node metastases were diagnosed postoperatively in 3 of the 190 patients (1.6%) in the PeNPAN group.

Fig. 1. Selection of study participants.

OCCC, ovarian clear cell carcinoma.

Table 1 presents the baseline characteristics of the patients in each group before and after propensity score weighting adjustment. Before adjustment, the mean ages in the control and PeNPAN groups were 59.4 and 53.6 years, respectively. The unbalanced variables between the 2 groups before adjustment (standard difference >0.1) included age, BMI, diabetes, cardiovascular disease, preoperative cyst size, adhesions in the Douglas’ pouch, intraoperative cyst rupture, and JSGO-accredited hospitals. After adjustment, all variables were evenly distributed between the groups. Postoperative staging and adjuvant chemotherapy status, which were not used in the propensity score calculation, were also well balanced after adjustment.

Table 1. Baseline characteristics before and after propensity score weighting adjustment.

Characteristics		Before adjustment			After adjustment
		Control group	PeNPAN group	Standardized difference	Control group	PeNPAN group	Standardized difference
Variables used to calculate the propensity score
	Age (yr)	59.4±11.3	53.6±9.6	0.558	57.0±11.2	57.0±9.3	<0.001
	BMI (kg/m2)	22.7±3.8	21.5±3.2	0.344	22.0±3.7	22.0±3.4	<0.001
	Diabetes mellitus	9.0 (7.9)	5.0 (2.6)	0.237	2.0 (4.2)	2.0 (4.2)	<0.001
	Cardiovascular diseases	12.0 (10.5)	13.0 (6.8)	0.131	3.8 (8.0)	3.8 (8.0)	<0.001
	Use of immunosuppressant	2.0 (1.8)	6.0 (3.2)	0.091	1.0 (2.2)	1.0 (2.2)	<0.001
	Preoperative cyst size (cm)	13.4±6.3	12.0±5.3	0.249	12.9±6.1	12.9±5.5	<0.001
	Adhesions in the Douglas’ pouch	73.0 (64.0)	104.0 (54.7)	0.190	29.1 (60.8)	29.1 (60.8)	<0.001
	Preoperative cyst rupture	31.0 (27.2)	49.0 (25.8)	0.032	14.2 (29.7)	14.2 (29.7)	<0.001
	Intraoperative cyst rupture	45.0 (39.5)	85.0 (44.7)	0.107	20.4 (42.7)	20.4 (42.7)	<0.001
	JSGO-accredited hospitals	26.0 (22.8)	136.0 (71.6)	1.120	18.3 (38.3)	18.3 (38.3)	<0.001
	Year of the surgery	2,013.0±4.2	2,012.9±4.1	0.032	2,013.3±4.1	2,013.3±4.1	<0.001
Other variables
	Stage IA	38.0 (33.3)	55.0 (28.9)	0.095	13.2 (27.6)	12.7 (26.6)	0.024
	Stage IC1	45.0 (39.5)	82.0 (43.2)	0.075	20.4 (42.7)	19.9 (41.7)	0.020
	Stage IC2 or IC3	31.0 (27.2)	49.0 (25.8)	0.032	14.2 (29.7)	14.2 (29.7)	<0.001
	Adjuvant chemotherapy	91.0 (79.8)	148.0 (77.9)	0.047	38.6 (80.7)	37.7 (79.0)	0.042

Data were presented as means ± standard deviations for continuous variables and as numbers (%) for categorical variables.

BMI, body mass index; JSGO, Japan Society of Gynecologic Oncology; PeNPAN, pelvic and para-aortic lymphadenectomy.

2. DFS

Overall, 24 (21.1%) and 27 (14.2%) women in the control and PeNPAN groups, respectively, experienced recurrence. The median follow-up duration was 52.2 months (interquartile range: 19.0–87.8), with 50.5 months (17.0–81.2) in the control group and 56.5 months (21.1–96.3) in the PeNPAN group. Fig. 2A illustrates the estimated DFS before propensity score weighting analysis. The 5-year DFS were 0.79 (95% CI=0.72–0.88) and 0.83 (95% CI=0.78–0.89) in the control and PeNPAN groups, respectively. The unadjusted HR was 0.63 (95% CI=0.36–1.09; p=0.10) before the propensity score weighting adjustment.

Fig. 2. Kaplan–Meier curves of DFS for the 2 groups (A) before propensity-score weighting and (B) after overlap weighting. The unadjusted and adjusted hazard ratios were 0.63 (95% CI=0.36–1.09; p=0.10) and 0.82 (95% CI=0.42–1.58; p=0.55), respectively.

CI, confidence interval; DFS, disease-free survival; PeNPAN, pelvic and para-aortic lymphadenectomy.

Fig. 2B shows the propensity score-weighted Kaplan–Meier curve for DFS. The 5-year DFS were 0.79 (95% CI=0.70–0.89) and 0.78 (95% CI=0.68–0.88) in the control and PeNPAN groups, respectively. The adjusted HR was 0.82 (95% CI=0.42–1.58; p=0.55).

3. OS

Mortality occurred in 16 (14.0%) and 23 (12.1%) patients in the control and PeNPAN groups, respectively. The median follow-up duration was 57.1 (interquartile range: 29.7–90.6) months, with 53.6 (29.0–84.4) months in the control group and 60.3 (31.8–96.3) months in the PeNPAN group. Fig. 3 illustrate the estimated OS before and after propensity score weighting, respectively. Before adjustment, the 5-year OS was 0.84 (95% CI=0.76–0.94) in the control group and 0.81 (95% CI=0.72–0.91) in the PeNPAN group. The unadjusted and adjusted HR for OS were 0.80 (95% CI=0.42–1.51; p=0.54) and 1.21 (95% CI=0.56–2.61; p=0.61), respectively.

Fig. 3. Kaplan–Meier curves of OS for the 2 groups (A) before propensity-score weighting and (B) after overlap weighting. Unadjusted and adjusted hazard ratio for overall survival were 0.80 (95% CI=0.42–1.51; p=0.54) and 1.21 (95% CI=0.56–2.61; p=0.61), respectively.

CI, confidence interval; OS, overall survival; PeNPAN, pelvic and para-aortic lymphadenectomy.

4. Additional analysis

Table 2 presents the results of the unadjusted Cox regression for variables with unbalanced baseline characteristics. HRs for DFS were higher in patients with older age, cardiovascular disease, larger perioperative cysts, and adhesions in Douglas’ pouch. Among these variables, the point estimates of HRs exceeded 2 for adhesion in Douglas’ pouch (HR=2.48; 95% CI=1.30–4.74) and cardiovascular disease (HR=2.70; 95% CI=1.27–5.75).

Table 2. HRs for disease-free survival in variables with baseline imbalances between the 2 groups before propensity score adjustment.

Characteristics	HRs (95% CIs)	p-values
Age	1.04 (1.01–1.06)	0.01
BMI	1.00 (0.92–1.09)	0.98
Diabetes mellitus	1.94 (0.70–5.39)	0.21
Cardiovascular diseases	2.70 (1.27–5.75)	0.01
Preoperative cyst size	1.06 (1.01–1.10)	0.01
Intraoperative cyst rupture	0.80 (0.45–1.41)	0.44
Adhesions in the Douglas’ pouch	2.48 (1.30–4.74)	0.01
JSGO-accredited hospitals	0.88 (0.51–1.53)	0.66

BMI, body mass index; CI, confidence interval; HR, hazard ratio; JSGO, Japan Society of Gynecologic Oncology.

Fig. 4 shows the results of the Biased Models 1 and 2, along with the main analysis. In the univariable Cox regression analysis, the unadjusted HR in Biased Model 1, which did not apply appropriate exclusion criteria to consider potential confounding factors, was 0.65 (95% CI=0.39–1.08; p=0.09). In Biased Model 2, which further limited the analysis to patients with postoperative stage I, the adjusted HR was 0.59 (95% CI=0.36–1.00; p=0.048).

Fig. 4. Comparison of HRs of lymphadenectomy for DFS in different analytical models. Biased Model 1 is an univariable analysis of the cohort according to postoperative staging. Biased Model 2 is a multivariable Cox regression in the population of Biased Model 1, adjusted for tumor stage (IA, IC1, IC2, or IC3) and adjuvant chemotherapy status as a covariate.

CI, confidence interval; DFS, disease-free survival; HR, hazard ratio; PS, propensity score.

DISCUSSION

To our knowledge, this is the first study that investigates the prognostic influence of lymphadenectomy with careful adjustment for key confounders, including pelvic adhesions and comorbidities [15]. A significant reduction in the HR for DFS with lymphadenectomy was observed in Biased Model 2, with similar point estimates observed in the unadjusted and Biased Model 1. However, no clear therapeutic benefit of lymphadenectomy was observed after thoroughly adjusting for potential confounding variables (Fig. 4). Among the possible confounders, adhesions in the Douglas’ pouch and cardiovascular disease were identified as risk factors for recurrence with HRs >2. However, both were less common in the PeNPAN group. These results suggest that failing to adequately consider confounders in the analyses may lead to an overestimation of the therapeutic effect of lymphadenectomy.

In this study, factors with distributional differences between the PeNPAN and control groups that were related to DFS or OS may have served as confounders. Appropriate consideration of these factors is crucial to accurately estimate the therapeutic effect of lymphadenectomy [16]. One important potential confounder identified in this study is comorbidity. Severe comorbidities or poor general health often result in the avoidance of invasive surgery [17]. Notably, among the comorbidities, cardiovascular disease was unbalanced between the groups before adjustment, and thromboembolism was incorporated in our analysis. Thromboembolism, including deep venous thromboembolism, pulmonary embolism, and cerebral infarction, is often observed in OCCC and is an important poor prognostic factor regardless of the stage [5,18]. This implies that invasive lymphadenectomy was avoided in poor prognosis patients who developed thromboembolism. Thromboembolism is sometimes not analyzed independently; instead, a comorbidity index is used to account for comorbidities [19]. Given that thromboembolism is a known risk factor for recurrence [5,18], summarizing comorbidities using an index may be inappropriate when evaluating recurrence in OCCC; as such, an index may fail to adequately capture the contribution of thromboembolism to recurrence risk.

The second important possible confounder suggested in this study is pelvic adhesions. These were assessed based on adhesions in the Douglas’ pouch. Adhesions in Douglas’ pouch were a risk factor for recurrence. However, they were more frequently observed in the control group than in the PeNPAN group (Tables 1 and 2). Therefore, the therapeutic effect of lymphadenectomy may have been overestimated in previous studies that did not adjust for pelvic adhesions. Adhesions are common in OCCC and are associated with endometriosis [20,21]. While adhesions are known to increase surgical difficulty and complications [22], they are also associated with poor prognosis in ovarian carcinoma [23]. Although one study focusing solely on OCCC did not identify adhesions as a clear risk factor for recurrence, the wide CIs prevented a definitive conclusion [24]. In contrast, we defined pelvic adhesions based on adhesions in the Douglas’ pouch, which may have led to the identification of adhesions as a significant risk factor for recurrence. Our study suggests that severe adhesions may prevent complete lymphadenectomy and inclusion in the control group, underscoring the importance of considering potential confounding factors in observational studies.

Previous reports have suggested that the therapeutic effect of lymphadenectomy in early-stage OCCC does not adequately account for comorbidities and intraperitoneal findings in their analyses [6,9,14]. A study involving patients with T1 or T2 OCCC reported that lymphadenectomy was associated with prolonged DFS, with an adjusted HR of 0.4 (95% CI=0.2–0.8). However, this study did not account for comorbidities or intraoperative findings [6]. Another study from a large U.S. database that included all histological types of ovarian cancer also demonstrated the use of lymphadenectomy. However, comorbidities and intraoperative findings were not considered in that study [9]. It is important to acknowledge that in large database studies, variables uncollected during database construction cannot be utilized, and intraperitoneal findings may serve as unmeasured confounders. Furthermore, our Biased Model 2, which targeted patients with stage I OCCC based on postoperative pathological diagnosis, was subject to bias because patients with positive lymph nodes were only excluded from the PeNPAN group. However, such data are sometimes present in subgroup analyses. Therefore, caution is warranted [14]. Additionally, it should be noted that adjusting for post-operative chemotherapy when estimating the therapeutic effect of PeNPAN will prevent the calculation of the total effect because chemotherapy is an intermediate factor rather than a confounder [16].

The therapeutic effects of lymphadenectomy have also been investigated in ovarian cancers other than early-stage OCCC. In advanced ovarian cancer, while retrospective analyses have suggested a therapeutic effect of lymphadenectomy [25,26], well-designed randomized controlled trials (RCTs) have refuted it [17]. This discrepancy may be attributed to biases in the observational studies, particularly those arising from patient age, comorbidities, and overall health status [17]. Among these potential biases, our findings highlight the importance of considering comorbidities such as thromboembolism and adhesions in the Douglas’ pouch. Moreover, the therapeutic role of lymphadenectomy was recently highlighted in ovarian endometrioid carcinoma, similar to OCCC, has an endometriosis-related origin [27]. However, pelvic adhesions were not considered in that study, potentially resulting in an overestimation of the therapeutic effect.

The necessity of adjuvant chemotherapy for early-stage OCCC remains debatable. While advanced OCCC is typically regarded as chemotherapy-resistant [4,5,28], the therapeutic benefits of postoperative chemotherapy have been reported in stage I OCCC [10,29,30]. Our previous analysis revealed that chemotherapy was effective, even when patients without complete staging were included [10]. If adjuvant chemotherapy is expected to be effective, regardless of lymph node metastasis, chemotherapy without lymphadenectomy may serve as a viable treatment strategy. We only included patients without preoperative lymph node enlargement. Consequently, the incidence of lymph node metastasis was low (1.6% in the PeNPAN group, accounting for 1% of the overall study cohort). Data from the Japan Society of Obstetrics and Gynecology indicate that isolated lymph node metastasis occurs in approximately 2%–4% of OCCC without suspected intraperitoneal or distant metastases [2,31]. As our study focused on OCCC without preoperative lymph node enlargement, we posit that our findings are reasonable. Since we did not include patients with preoperative lymph node enlargement, our findings do not refute the value of lymph node dissection in these patients. Considering that preoperative lymph node enlargement is associated with a poor prognosis [6], future research should focus on evaluating the therapeutic effects of pelvic and para-aortic lymphadenectomies in these patients. Given the ongoing advances in ovarian cancer treatment, it is essential to consider personalized treatment strategies [32].

Firstly, we controlled for potential confounders as comprehensively as possible. However, the possibility of unmeasured confounding factors cannot be ruled out, given the observational nature of this study. Despite this study not demonstrating the effectiveness of PeNPAN, the HR for DFS was 0.82 (95% CI=0.42–1.58), with a wide CI that could still potentially accommodate a clinically meaningful reduction in HR. Therefore, this study could not conclude the absence of a treatment effect for lymphadenectomy [33]. Secondly, due to the retrospective nature of this study, the details of the lymphadenectomy procedures were not standardized. Prospective studies are warranted to ensure consistency in surgical techniques.

After adjusting for possible confounders, including comorbidities and intraoperative adhesions, the observed treatment effects of lymphadenectomy in the biased models were no longer statistically significant. Therefore, we could not show that lymphadenectomy reduced recurrence risk in women with OCCC suspected as stage I preoperatively. Existing observational studies that have not considered these factors may have overestimated the therapeutic effects of lymphadenectomy. Given the limited number of patients with early-stage OCCC, conducting an RCT might be challenging. Future observational studies with larger sample sizes that thoroughly address possible confounders, including adhesions and comorbidities, are warranted.