Annual hazard rates for early-stage endometrial carcinoma after postoperative radiotherapy under FIGO 2023 staging system: results from a population-based single-center cohort study

Abstract

Objective

To investigate whether early-stage endometrial carcinoma (EC) patients have survival outcomes comparable to those of the general population, and further explore the dynamic recurrence/death surveillance value of International Federation of Gynecology and Obstetrics(FIGO) 2023 staging system.

Methods

This retrospective study analyzed FIGO 2009 stage I–II patients receiving postoperative radiotherapy at Peking Union Medical College Hospital in China (2000–2020), re-staged per FIGO 2023 staging system. Overall survival (OS) and disease-free survival (DFS) were assessed using Kaplan-Meier methods; mortality was compared to age-/sex-matched Chinese controls using standardized mortality ratios (SMRs); recurrence/death trajectories were modeled through annual hazard rates; FIGO 2023 recurrence prediction accuracy was assessed via receiver operating characteristics analysis.

Results

This study of 1222 EC patients revealed upstaging from FIGO 2009 to 2023 (IA→II: 27.2%; IB→II: 26.4%). OS rates were 97.7% (3-year), 94.9% (5-year), and 91.0% (10-year), with corresponding DFS rates of 93.4%, 87.4%, and 82.8%. Under FIGO 2023 staging system, recurrence risk peaked at 1.3 years (3.5%) and declined after 3 years. Mortality risk increased within 4 years, stabilizing at about 1.0% thereafter. IC/IIC stages showed the earliest/highest risk peaks. Stage IA1–IA2 required monitoring for late recurrences (>3 years), while IIC needed vigilance for delayed mortality (>3 years). FIGO 2023 staging system was an independent prognostic factor for recurrence prediction (area under the curve: 1-year=0.644; 3-year=0.612; 5-year=0.611).

Conclusion

Early-stage EC patients had favorable prognosis. FIGO 2023 staging system exhibit excellent recurrence surveillance and prognostic prediction for these patients after postoperative radiotherapy.

INTRODUCTION

Endometrial carcinoma (EC) remains the most commonly diagnosed malignancy in females worldwide, with 417,000 new cases and 97,000 deaths in 2020 [1]. EC is the second most common gynecological malignancy in China [1]. Approximately 80% of cases are diagnosed at early stage, with a 5-year survival rate exceeding 95% [2]. Surgery serves as the initial main treatment strategy for individuals with early-stage disease [3]. Adjuvant treatments such as radiotherapy or chemotherapy are recommended based on postoperative risk stratification [3,4,5,6]. In stage I patients with high-risk features (grade 1 with deep myometrial invasion, grade 2 with any invasion, or grade 3 with superficial invasion), radiotherapy significantly reduced the locoregional recurrence rate. The 5-year locoregional recurrence rate was 4% in patients receiving radiotherapy vs. 14% in the control group [5]. In patients with high-intermediate risk disease, vaginal brachytherapy (VBT) is the standard treatment strategy [7]. For high-risk early-stage patients of all histological types, pelvic radiotherapy remains an effective adjuvant treatment [8]. Despite favorable overall survival (OS) outcomes, early-stage patients still may recur, including vaginal, locoregional, and systemic—at a rate of approximately 15% [9]. Therefore, most attention have been directed toward recurrence patterns for this favorable malignancy.

Due to advancements of molecular characterization, understanding of biological behavior among different histological types, and refined identification of pathological features, the International Federation of Gynaecology and Obstetrics (FIGO) updated its staging system in 2023 (building upon the 2009 version) [10]. This revised staging system introduced histological type, tumor grade and the substantial lymphovascular invasion (LVSI) with the aim of better defining EC patient subgroups and facilitating more tailored treatment strategies [10]. A recent study from Japan validated the prognostic accuracy of the FIGO 2023 staging system and found that it improved prognostic precision comparing with Japan Society of Gynecologic Oncology recurrence risk categories [11]. However, robust evidence regarding recurrence patterns and the prognostic predictive value of the FIGO 2023 staging system for early-stage EC patients is still limited [12]. Some previous published studies focused on the risk stratification or annual hazard recurrences rate among different risk subgroups [13,14]. For example, a French retrospective multicenter study demonstrated that the hazard rate (HR) of recurrence over time was dynamic and varied among different EC subgroups [13]. Nevertheless, the annual hazard rate (AHR) of recurrence/death under FIGO 2023 staging system is yet not been investigated. More importantly, the prognostic prediction and recurrence surveillance value of the FIGO 2023 staging system across different substages is still unclear.

Given the favorable prognosis for early-stage EC patients [15], this study aimed to analyze the standardized mortality ratio (SMR) of the whole early-stage cohort receiving postoperative radiotherapy compared with age- and sex-matched general Chinese population [16], and to further explore the dynamic recurrence or death surveillance value of the FIGO 2023 staging system among different early-stage subgroups to establish optimal surveillance and treatment strategies.

MATERIALS AND METHODS

1. Patient selection

We retrospectively enrolled patients who were diagnosed with EC from 2000 to 2020 at Peking Union Medical College Hospital in China. The inclusion criteria included: receiving surgical staging; FIGO 2009 stage I and II, receiving postoperative radiotherapy; with or without adjuvant chemotherapy. Patients with missing follow-up data were excluded. Clinical data, pathological features, treatment information, and follow-up records were collected. The study was approved by the Peking Union Medical College Hospital Review Board (Number I-24PJ0795) and complied with good clinical practice guidelines, the Declaration of Helsinki. And the requirement for informed consent was waived by IRB due to the retrospective nature of the study.

2. Treatment

The comprehensive surgical staging procedure included a total abdominal hysterectomy, bilateral salpingo-oophorectomy, and lymph node (LN) assessment. These operations could be carried out using any surgical approach—such as laparoscopic, robotic, vaginal, or abdominal—including minimally invasive techniques. Adjuvant therapy administration was determined by risk stratification and pathological factors.

Between 2000 and 2009, adjuvant radiotherapy was standard for early-stage EC with unfavorable factors, including large tumor size, deep myometrial invasion, poor differentiation, lower uterine segment invasion, and LVSI. Key trials such as PORTEC-1 [5] and GOG-99 [6] established the benefit of external beam radiation therapy (EBRT) in reducing pelvic recurrence. However, retrospective evidence also supported VBT alone for intermediate-risk patients due to comparable efficacy and lower toxicity. After 2009, following the publication of the PORTEC-2 trial [17], the use of VBT increased among intermediate-risk patients, while EBRT remained standard for those with cervical stromal invasion. Based on these principles, enrolled patients received postoperative radiotherapy, including VBT only and EBRT with or without VBT. The EBRT target encompassed the pelvic LN region, delivered using either 3-dimensional conformal radiation therapy or intensity-modulated radiation therapy. The prescribed EBRT dose was 45–50.4 Gy administered in 25–28 fractions of 1.8 Gy. VBT targeted the upper half of the vagina or the proximal 3–5 cm of the vagina using a vaginal cylinder. The D90 dose for VBT alone was 20–30 Gy delivered over 4–6 fractions. When VBT followed EBRT, the dose was 10 Gy given in 2 fractions.

Chemotherapy was as part of the adjuvant regimen. Between 2000 and 2009, it was considered for patients with aggressive histology, high tumor grade, LVSI, or positive peritoneal cytology, despite limited evidence supporting its use at the time. After 2009, results from the GOG-122 trial [18], although focused on advanced-stage (III–IV) disease, underscored the value of chemotherapy in early-stage patients at high risk of distant metastasis. Furthermore, for stage I–II patients, the PORTEC-3 trial reported that combined chemotherapy and radiotherapy did not confer a significant survival benefit (5-year OS=83.8% in combined chemotherapy radiotherapy group, 82.0% in radiotherapy alone group; p=0.500) [19]. The use of chemotherapy in our cohort was relatively low, with only 164 patients (13.4%) receiving it. Its sequence relative to radiotherapy could be sequential, or sandwiched.

3. Risk classification and re-stage

To eliminate heterogeneity from evolving staging systems, we performed rigorous re-stage for the entire cohort. First, the cohort (2000–2009) was re-staged according to FIGO 2009 staging system criteria based on original surgical-pathological reports. Subsequently, the entire cohort (2000–2020) was re-staged using the updated FIGO 2023 staging system [10]. This process was conducted independently by 2 physicians, with discrepancies resolved by a senior clinician.

Poor prognostic factors included histological type (according to the World Health Organization 2020 classification), tumor grade, myometrial invasion, and LVSI. LVSI status was categorized as negative or positive; within the positive group, the designation “focal” and “extensive” were used. However, due to the retrospective nature of this cohort study, we did not further subdivide positive LVSI cases into focal or extensive. And sub-stages were classified as: IA1–IA2, IB, IIA–IIB, IC/IIC (aggressive group). All were stratied using the European Society for Medical Oncology-European Society of Gynaecological Oncology-European Society for Radiotherapy and Oncology (ESMO-ESGO-ESTRO) 2021 risk classification [4].

4. Categories of first recurrence site

All rst reoccurring events were classied according to site as follows: local recurrence, pelvic LN recurrence, distant metastasis, local and pelvic LN recurrence, local and distant failure, pelvic and distant failure, as well as local, pelvic, and distant failure.

5. Statistical analyses

Baseline characteristics were examined using descriptive analysis. The FIGO stage migration was showed by using Sankey diagram via R. The OS was defined as the interval between the date of surgery and the date of death or the last follow-up appointment. The disease-free survival (DFS) was defined as the interval between the date of surgery and the date of progression/death/the last follow-up. Survival outcomes were analyzed using Kaplan-Meier method. OS was compared to an age- and sex-matched general Chinese population using expected survival and SMR in Poisson regression, and calculated via the survexp function in R.

To illustrate the track of the instantaneous failure rate for EC patients over time, this study employed a kernel smoothing method combined with the rstpm2 R package to predict the dynamic AHRs. The estimation of AHRs was based on flexible parametric survival models to assess and compare time-varying hazard functions for disease recurrence or death across substage groups (IA1–IA2, IB, IIA–IIB, IC/IIC). Using the stpm2 function from the rstpm2 package, we modeled the hazard of DFS and OS over time. The models were fitted separately to data from each staging group, and AHRs were predicted over a 10-year follow-up period.

To examine whether the relative risk between groups changes over different periods of follow-up time, time-dependent ratio of HR analysis within specific time intervals was conducted. The total follow-up time was divided into predefined intervals (0–1, 1–2, 2–3, 3–4, 4–5, 5–10, 10–15 years), and HRs were estimated separately within each interval, and ratio of HR was calculated as the ratio of HRs between comparison and reference groups (IB vs. IA1–IA2, IIA–IIB vs. IA1–IA2, IC/IIC vs. IA1–IA2) at multiple time points within each interval. We calculated mean, median, and distribution statistics for ratio of HRs within each interval, confidence intervals using standard error approximation (mean ± 1.96*standard error).

To evaluate whether the FIGO staging system was an independent prognostic factor, we performed Cox regression and conducted statistical comparisons among several models: model 1 (all variables), model 2 (a selected subset of variables), model 3 (only FIGO 2023 stage), and model 4 (only FIGO 2009 stage). Variable selection for model 1 was performed using a multi-method approach: univariable screening (p<0.2), stepwise selection (forward/backward/both) optimized by Akaike Information Criterion (AIC)/Bayesian Information Criterion (BIC), and LASSO regression (lambda.min/1se). Model 2, 3, and 4 were compared against model 1 using Likelihood Ratio Test (LRT). AIC, BIC, and Concordance Index (C-Index) were calculated for model evaluations.

Based on fitted Cox proportional hazards model via the coxph function, we generated continuous risk scores from the linear predictor. To evaluate the discriminative ability of these scores at specific time points (1, 3, and 5 years), time-dependent receiver operating characteristics (ROC) analysis was performed using the timeROC package [20,21]. The analysis adopted the cumulative/dynamic case definition with marginal weighting, which handles right-censored data through inverse probability weighting. Area under the curve (AUC) estimates were computed for each time horizon, and their variances were derived using the i.i.d. decomposition method, enabling the construction of 95% confidence intervals (CIs) for robust statistical inference on the model’s time-varying predictive accuracy.

All statistical analyses were performed using SPSS v. 26 (IBM Corp., Armonk, NY, USA) and R v. 3.5.1 (The R Foundation for Statistical Computing, Vienna, Austria). A 2-sided p-value <0.05 was indicative of a statistically significant difference.

RESULTS

1. Demographic and clinicopathological characteristics

A total of 1,222 patients with EC were analyzed. Table 1 presents the clinical and pathological characteristics of the cohort. The median age at diagnosis was 57 years old (range=23–85 years). Endometrioid carcinoma was the predominant histological type (94.9%), followed by mixed carcinoma (2.0%), clear cell carcinoma (1.4%), serous carcinoma (1.0%), undifferentiated carcinoma (0.4%), and mucinous carcinoma (0.3%). Tumor grade 1–2 was more common (82.0%). LVSI was negative in 78.0% of patients, positive in 11.8%, and extensive in 1.3%. According to the FIGO 2009 staging system, the stage distribution was: stage IA (53.3%), stage IB (38.8%), and stage II (7.9%). Most patients were classified into low- or intermediate-risk groups. Application of the FIGO 2023 staging system resulted in significant alterations to the stage distribution. Fig. S1 illustrates these changes from the FIGO 2009 stages. Upstage (IA to II, 27.2%; IB to II, 26.4%) were observed. The majority of patients (70.5%) received VBT alone. Only 13.4% patients received adjuvant chemotherapy.

Table 1. Clinical and pathological features of the enrolled endometrial carcinoma patient cohort (n=1,222).

Characteristics		Values
Age (yr)		57 (23–85)
	<60	765 (62.6)
	≥60	457 (37.4)
Histologic type
	Endometrioid carcinoma	1,160 (94.9)
	Mucinous carcinoma	4 (0.3)
	Serous carcinoma	12 (1.0)
	Clear cell carcinoma	17 (1.4)
	Undifferentiated carcinoma	5 (0.4)
	Mixed carcinoma	24 (2.0)
Tumor grade
	Grade 1–2	1,002 (82.0)
	Grade 3	220 (18.0)
Tumor diameter
	≤2 cm	183 (15.0)
	>2 cm	745 (61.0)
	Unknown	294 (24.0)
Cervical invasion
	None	1,056 (86.4)
	Mucosal invasion	71 (5.8)
	Stroma invasion	91 (7.4)
	Unknown	4 (0.3)
Myometrial invasion
	<50%	702 (57.4)
	≥50%	520 (42.6)
LVSI
	Negative	953 (78.0)
	Positive	144 (11.8)
	Focal	109 (8.9)
	Extensive	16 (1.3)
Peritoneal cytology
	Not tested	135 (11.0)
	Negative	922 (75.5)
	Positive	51 (4.2)
	Unknown	114 (9.3)
Adjuvant radiotherapy
	VBT only	861 (70.5)
	EBRT	42 (3.4)
	EBRT + VBT	319 (26.1)
Adjuvant chemotherapy
	Yes	164 (13.4)
	No	981 (80.3)
	Unknown	77 (6.3)
FIGO stage 2009
	IA	651 (53.3)
	IB	474 (38.8)
	II	97 (7.9)
ESMO-ESGO-ESTRO 2021 risk classification
	Low-risk	515 (42.1)
	Intermediate risk	480 (39.3)
	High intermediate risk	165 (13.5)
	High-risk	62 (5.1)
FIGO stage 2023
	IA1	8 (0.7)
	IA2	464 (38.0)
	IB	349 (28.6)
	IC	2 (0.2)
	IIA	65 (5.3)
	IIB	93 (7.6)
	IIC	241 (19.7)
First recurrence modes
	Local recurrence	11 (0.9)
	Pelvic LN recurrence	10 (0.8)
	Distant metastasis	55 (4.5)
	Local and pelvic LN recurrence	2 (0.2)
	Local and distant failure	5 (0.4)
	Pelvic and distant failure	13 (1.1)
	Local, pelvic, and distant failure	2 (0.2)

Values are presented as median (range) or number (%).

EBRT, external beam radiotherapy; ESMO-ESGO-ESTRO, European Society for Medical Oncology-European Society of Gynaecological Oncology-European Society for Radiotherapy and Oncology; FIGO, International Federation of Gynecology and Obstetrics; LN, lymph node; LVSI, lymphovascular space invasion; VBT, vaginal brachytherapy.

2. Treatment outcomes

After a median follow-up of 67.1 months (95% CI=62.6–71.6), 51 patients (4.2%) had died, and 98 (8.0%) had experienced a treatment failure, either local recurrence (n=11; 0.9% of all patients), pelvic LN recurrence (n=10; 0.8% of all patients), distant metastasis (n=55; 4.5% of all patients), local and pelvic LN recurrence (n=2; 0.2% of all patients), local and distant recurrence (n=5; 0.4% of all patients), pelvic and distant failure (n=13; 1.1% of all patients), or local, pelvic, and distant failure (n=2; 0.2% of all patients).

The 3-year, 5-year, and 10-year OS rates were 97.7%, 94.9%, and 91.0%, respectively (Fig. S2). Corresponding DFS rates were 93.4%, 87.4%, and 82.8%. And Fig. S3 shows the OS and DFS among substages in FIGO 2009 and 2023 staging system.

Fig. 1 shows the entire cohort OS outcomes and the expected survival outcomes from an age- and sex-matched general Chinese population. Table 2 presents the SMRs for the entire cohort and subgroups. For the whole cohort, the 3-year, 5-year, and 10-year SMRs were 1.13 (95% CI=0.73–1.69), 1.31 (95% CI=0.94–1.78), and 1.13 (95% CI=0.83–1.50), respectively (all p>0.050). In subgroup analyses, the 3-year and 10-year SMRs for the non-endometrioid carcinoma group were significantly elevated (p<0.050). Significantly elevated SMRs were also observed over time in the grade 3 subgroup, LVSI-positive group, high-intermediate risk group, and FIGO 2023 stage IC/IIC group (all p<0.050).

Fig. 1. Survival outcomes of the study cohort compared to an age- and sex-matched general Chinese population.

Table 2. Treatment outcomes of the enrolled endometrial carcinoma patient cohort (n=1,222).

Subgroups		No. of patients	3-yr SMR	p-value	5-yr SMR	p-value	10-yr SMR	p-value
Whole cohort		1,222 (100.0)	1.13 (0.73–1.69)		1.31 (0.94–1.78)	0.102	1.13 (0.83–1.50)	0.398
Age (yr)
	<60	765 (62.6)	1.63 (0.78–3.00)	0.150	2.14 (1.29–3.34)	0.003	1.58 (0.96–2.43)	0.048
	≥60	457 (37.4)	0.93 (0.51–1.57)	1	0.97 (0.60–1.49)	1	0.94 (0.62–1.36)	1
Histologic type
	Endometrioid carcinoma	1,160 (94.9)	1.00 (0.61–1.54)	1	1.24 (0.87–1.71)	0.194	1.05 (0.76–1.42)	0.695
	Non-endometrioid carcinoma	62 (5.1)	3.58 (0.98–9.17)	0.027	2.68 (0.73–6.87)	0.064	2.83 (0.92–6.62)	0.034
Tumor grade
	Grade 1–2	1,002 (82.0)	0.69 (0.36–1.21)	0.229	1.03 (0.67–1.50)	0.84	0.87 (0.59–1.24)	0.502
	Grade 3	220 (18.0)	3.11 (1.61–5.43)	0.001	2.68 (1.46–4.49)	0.001	2.43 (1.42–3.89)	0.001
Myometrial invasion
	<50%	702 (57.4)	1.04 (0.52–1.87)	0.877	1.16 (0.69–1.83)	0.524	1.07 (0.68–1.60)	0.746
	≥50%	520 (42.6)	1.24 (0.66–2.13)	0.436	1.48 (0.93–2.24)	0.068	1.20 (0.78–1.77)	0.325
LVSI
	Negative	953 (78.0)	0.88 (0.49–1.46)	0.717	1.19 (0.80–1.70)	0.319	1.04 (0.74–1.43)	0.741
	Positive	144 (11.8)	3.69 (1.59–7.26)	0.002	2.96 (1.35–5.62)	0.004	2.36 (1.08–4.47)	0.016
	Focal	109 (8.9)	0.58 (0.01–3.21)	1	0.51 (0.01–2.83)	1	0.51 (0.01–2.83)	1
	Extensive	16 (1.3)	NA		NA		NA
FIGO stage 2009
	IA	651 (53.3)	1.12 (0.56–2.01)	0.631	1.17 (0.68–1.88)	0.509	1.08 (0.68–1.64)	0.657
	IB	474 (38.8)	1.08 (0.54–1.93)	0.752	1.39 (0.85–2.14)	0.145	1.13 (0.72–1.70)	0.506
	II	97 (7.9)	1.69 (0.20–6.09)	0.333	1.84 (0.38–5.37)	0.226	1.62 (0.33–4.73)	0.440
ESMO-ESGO-ESTRO 2021 risk classification
	Low-risk	515 (42.1)	0.89 (0.36–1.84)	1	1.02 (0.52–1.77)	0.884	0.78 (0.41–1.33)	0.462
	Intermediate risk	480 (39.3)	0.74 (0.30–1.53)	0.517	1.18 (0.68–1.92)	0.494	1.15 (0.72–1.74)	0.491
	High intermediate risk	165 (13.5)	2.53 (1.02–5.22)	0.023	2.43 (1.11–4.61)	0.014	1.83 (0.84–3.48)	0.070
	High-risk	62 (5.1)	2.70 (0.56–7.89)	0.102	2.00 (0.41–5.85)	0.191	2.25 (0.61–5.77)	0.105
FIGO stage 2023
	IA1 + IA2	472 (38.6)	0.83 (0.30–1.81)	0.852	1.01 (0.50–1.80)	0.880	0.77 (0.40–1.34)	0.447
	IB	349 (28.6)	0.40 (0.08–1.16)	0.102	1.01 (0.51–1.81)	0.879	0.90 (0.49–1.50)	0.800
	IIA	65 (5.3)	1.25 (0.03–6.99)	0.550	1.80 (0.22–6.51)	0.305	1.52 (0.18–5.51)	0.378
	IIB	93 (7.6)	2.30 (0.47–6.71)	0.144	1.60 (0.33–4.67)	0.443	1.33 (0.27–3.88)	0.498
	IC + IIC	243 (19.9)	2.58 (1.29–4.61)	0.005	2.24 (1.19–3.83)	0.010	2.21 (1.28–3.53)	0.003

Values are presented as number (%) or % (95% condence interval).

ESMO-ESGO-ESTRO, European Society for Medical Oncology-European Society of Gynaecological Oncology-European Society for Radiotherapy and Oncology; FIGO, International Federation of Gynecology and Obstetrics; LVSI, lymphovascular space invasion; SMR, standardized mortality ratio.

3. AHR for the whole cohort

Fig. 2A displays the smoothed AHR trajectory for the whole cohort. For the entire cohort, the peak AHR for recurrence was highest for the first 3 years, peaking at 1.3 years (3.5%), and subsequently declined. The peak AHR for death increased continuously within the first 4 years. Beyond 4 years, the AHR of death remained essentially constant at approximately 1.0%.

Fig. 2. AHR of death or recurrence under FIGO 2023 staging system and time-dependent ratio of HRs. (A) AHR of entire cohort. (B) AHR of death by FIGO 2023 substage. (C) AHR of progression or death by FIGO 2023 substage. (D) A heat map to visualize ratio of HR patterns across time and comparison groups. (E) A line chart showing ratio of HR trends over time. (F) A forest plot summarizing ratio of HRs across all intervals.

AHR, annual hazard rate; FIGO, International Federation of Gynecology and Obstetrics; HR, hazard rate.

4. AHR for FIGO 2023 sub-stages

The AHR trajectories for death and recurrence across sub-stages under FIGO 2023 staging system are presented in Fig. 2B and C, respectively. When stratified based on FIGO 2023 staging system, we observed higher peak AHR and earlier time to recurrence or death in the IC/IIC group, followed by the IIA–IIB, IB, and IA1–IA2 groups. For stage IA1–IA2 group, the peak AHR for death was 0.9% at 4.6 years postoperatively, stabilizing at 1.0% thereafter. The peak AHR for recurrence was 2.6% at 1.2 years postoperatively; a rising trend was observed after 3 years. For stage IB patients, the peak AHR for death (1.7%) occurred at 4.2 years postoperatively, while the peak for recurrence (4.2%) was at 1.9 years. For stage IIA–IIB group, the peak AHRs were 2.5% for death at 2.5 years and 4.5% for recurrence at 2.2 years postoperatively. For stage IC/IIC group, the peak AHR for death was 3.3% at 1.2 years postoperatively; while a slow increase occurred after 3 years, the rate remained below 2.0%. The peak AHR for recurrence was 8.2% at 0.9 years postoperatively.

5. Time-dependent ratio of HRs for FIGO 2023 sub-stages

The time-dependent ratios of HRs are shown in Fig. 2D-F. Compared with the IA1–IA2 group, the IB group exhibited a significantly elevated ratio of HR between 1 and 5 years, peaking at 2–3 years (ratio of HR=3.09; 95% CI=3.02–3.15). The IIA–IIB group showed a significant ratio of HR from 0 to 5 years, with a peak at 2–3 years (ratio of HR=3.36; 95% CI=3.27–3.46). The IC/IIC group demonstrated a significantly increased ratio of HR from 0 to 10 years, which peaked within 0–1 year (ratio of HR=3.75; 95% CI=3.62–3.89).

6. Cox regression model comparisons among different models

Detailed comparison results are presented in Table S1. And model 2 (a selected subset of variables) was shown in Table S2. We found model 1 performed well, with the highest likelihood ratio (χ²=66.11) and higher discriminative ability (C-index=0.701 [95% CI=0.647–0.754]). And model 2 was comparable to model 1, with the LRT (χ²=60.84 vs. 66.11, p=0.810) and discriminative ability (C-index=0.692 [95% CI=0.637–0.746] vs. 0.701 [95% CI=0.647–0.754]). Therefore, variables (grade, age, peritoneal cytology, cervical invasion, LVSI, myometrial invasion, FIGO stage 2023) were independent prognostic factors in a multi-variable Cox model.

7. Validation of the FIGO 2023 staging system for predicting recurrence

Comparison analysis revealed that the univariable Cox model containing only FIGO 2023 performed significantly better than the model containing only FIGO 2009, as evidenced by the LRT (χ²=17.23 vs. 8.09; p=0.010) and higher discriminative ability (C-index=0.614 [95% CI=0.560–0.668] vs. 0.580 [95% CI=0.531–0.629]) (shown in Table S1, Fig. 3). FIGO 2023 staging system performed well in predicting recurrence rate at 1-year, 3-year, and 5-year, with AUC=0.644 (95% CI=0.133–1.156), AUC=0.612 (95% CI=0.251–0.973), and AUC=0.611 (95% CI=0.262–0.961). However, FIGO 2009 staging system showed a poor performance in predicting recurrence rate. The 1-year, 3-year, and 5-year AUC values for FIGO 2009 were 0.525 (95% CI=0.062–0.989), 0.592 (95% CI=0.246–0.938), and 0.591 (95% CI=0.253–0.929), respectively.

Fig. 3. Predictive ability for 1-year, 3-year, and 5-year disease-free survival. (A) FIGO 2023 staging system. (B) FIGO 2009 staging system.

AUC, area under the curve; FIGO, International Federation of Gynecology and Obstetrics.

DISCUSSION

In this study, we found that patients with early-stage EC had comparable survival outcomes and favorable prognosis by comparing with the age- and sex-matched general Chinese population (p>0.050). FIGO 2023 staging system performed well and was an independent factor in recurrence or death surveillance and prediction. Under FIGO 2023 staging system, peak AHR for early EC substages occurs within the first 3 years, whereas peak mortality risk emerges within the first 4 years. Notably, among all substages, stage IC/IIC exhibits the highest and earliest peak in the AHR for death or recurrence. Furthermore, stage IA1–IA2 requires vigilance for late recurrences (>3 years), while stage IIC warrants monitoring for delayed mortality patterns (>3 years).

Previous study found that older EC patients had poorer survival outcomes [22]. The median age at EC diagnosis exceeds 60 years [2,22]. To illustrate the prognosis in old individuals, we try to introduce the expected survival to this study. In this study, we found that the 3-year, 5-year, and 10-year SMRs were 1.13 (95% CI=0.73–1.69), 1.31 (95% CI=0.94–1.78), and 1.13 (95% CI=0.83–1.50), respectively (all p>0.05), which indicated that there was no significant survival difference between the early-stage EC group receiving postoperative radiotherapy and the general Chinese population. Therefore, for early-stage EC patients with favorable prognosis (the 3-year, 5-year, and 10-year OS rates of 97.7%, 94.9%, and 91.0%), refining the follow-up and surveillance is extremely important to yield a value of health economics. Nevertheless, within the non-endometrioid carcinoma, LVSI-positive, grade 3, and high-intermediate risk groups, we observed significantly lower SMRs compared with the matched population. This indicates that individuals with high-risk disease require intensive treatment. However, we also observed that the subgroup aged <60 years showed poorer 5-year and 10-year SMRs. To explore the reason for the poorer prognosis observed in this group, we conducted a chi-square test to exclude the possibility that baseline differences could account for the disparity in outcomes. As summarized in Table S3, significant differences were observed in FIGO stage 2009, ESMO-ESGO-ESTRO 2021 risk classification, and causes of death. Although the younger patients presented with earlier FIGO 2009 stage and were more frequently classified as low-risk, they exhibited a higher incidence of cardiovascular and cerebrovascular diseases. When compared with age- and sex-matched healthy controls in China, non-cancer causes of death may account for the increased mortality risk observed in this younger patient cohort. Therefore, the inferior survival observed among younger patients is likely attributed to non-radiotherapy-related complications.

Due to inconsistent LVSI assessment and challenges in molecular classification generalization, the National Comprehensive Cancer Network (NCCN) guidelines have not yet fully incorporated the FIGO 2023 staging system [23]. However, FIGO 2023 staging system does help to predict the prognosis of EC patients [24,25,26,27], and can improve the adjuvant treatment decision making [28]. Validation efforts for the FIGO 2023 staging system are valuable for its promotion and application.

Risk of recurrence or death is generally described by survival curves [29]. However, the real-time hazard function depicts the dynamic HR for recurrence or death and how it evolves over time. Previous studies have reported the AHR trajectory in various malignancies, including breast cancer, esophagus cancer, colorectal cancer, natural killer/T-cell lymphoma, and thyroid cancer [29,30,31,32,33,34]. We first explore the AHR over time for early-stage EC under FIGO 2023 staging system. In our study, by using AHR trajectory, aggressive subgroups (stage IC/IIC) exhibited the highest and earliest peak in the AHR for death or recurrence, and hence, intensive treatment and follow-up may be needed in this subgroup. Although stage IA1–IA2 group exhibits lower recurrence rate, the occurrence of late recurrence warrants attention. The relative risk among sub-groups over different periods of follow-up time was dynamic. Compared with the IA1–IA2 group, the IB group and IIA–IIB group exhibited peaking ratio of HRs within 2–3 years. IC/IIC group showed peaking ratio of HRs within 0–1 years. These results help us identify the HR in different subgroups across various time periods, thereby enabling the tailoring of follow-up strategies.

Previous study demonstrated that most recurrences were diagnosed during the rst 2 years of follow-up and an intensive follow-up in EC patients did not improve OS, even in patients with high-risk recurrence [35]. Tailored surveillance protocols are crucial in the FIGO 2023 era, as recurrence patterns exhibit distinct time-dependent variations within specific follow-up windows. Thus, based on these findings, we outline the following approaches for integration into clinical practice. First, translating the identified recurrence patterns into personalized surveillance protocols—such as intensifying imaging to every 3 months in the first 2–3 years for early-recurrence subgroups instead of every 6 months, and extending intensified monitoring beyond 5 years for late-recurrence subgroups. We also detailed the median time to locoregional and non-locoregional failure stratified by FIGO 2023 sub-stages, as presented in Table S4. For local-regional recurrence monitoring, close follow-up is recommended for all patients beginning at 1.5 years post-treatment. As for distant metastasis surveillance, particular attention should be given to patients with stage IA1–IA2 and IC/IIC within the first 6 months. Secondly, we advise customizing patient counseling based on the identified recurrence pattern. Accordingly, “early-recurrence” patients would be clearly informed about the necessity of close monitoring in the initial years, while “late-recurrence” patients would be counseled on the need for long-term vigilance. Third, our results provide a rationale for designing prospective clinical trials to evaluate intensified adjuvant treatment strategies for these high-risk subgroups. Finally, we need further conduct health economic evaluations of these strategies within China's healthcare system.

The trajectory of the AHR based on the FIGO 2009 staging system is shown in Fig. S4; however, this figure does not adequately depict the AHR of death for the stage IA, IB, and II cohorts. Additionally, we also found that AHR in the non-endometrioid carcinoma group exhibited a sharp recurrence and death peak, with a slight peak at 1 years after treatment, followed by recurrence plateau and gradually increasing death hazard (Fig. S5A and B). AHR in the high-risk and high intermediate-risk group exhibited a recurrence peak with 2 years (Fig. S5C and D).

FIGO 2023 recurrence prediction accuracy was assessed via AUC-ROC analysis. However, certain limitations exist. First, estimating AUC based on discrete FIGO stages may be imprecise, as the resulting ROC curve simply connects a limited number of fixed points, which could misrepresent the model’s true discriminative performance. Second, this approach fails to account for the continuous spectrum of disease severity and within-stage heterogeneity, thereby limiting the identification of optimal decision thresholds. To address these issues, we derived continuous risk scores from the linear predictor of a fitted Cox model to enable a more accurate estimation of the AUC. Furthermore, we validated the independent predictive value of the FIGO 2023 staging system within a multivariable Cox model. Although the FIGO 2023 system demonstrated clear superiority over the FIGO 2009 system, model 1 and model 2 continued to show the best performance in predicting recurrence. The variables—including grade, age, peritoneal cytology, cervical invasion, LVSI, myometrial invasion, and FIGO stage 2023—served as independent prognostic factors and played a crucial role in the clinical decision-making process.

Molecular classification stratifies EC patients into 4 prognostic subtypes, including POLE mutation, mismatch repair deficient, abdominal p53, and no specific molecular profile [36]. However, due to non-accessibility to the molecular classification, there exists difference in prognosis between the FIGO staging system with known and unknown molecular classification [37]. Compared to cases with known molecular classification, FIGO 2023 staging reveals more frequent upstaging when molecular status is unknown [37]. Shifts from stage I to stage II were observed in about 13%–50% of cases, mainly due to aggressive histology and LVSI [24,25,38,39,40]. In our study, upstage (IA to II, 27.2%; IB to II, 26.4%) occurred in early-stage EC patients. In the era of molecular classification, ESGO/ESTRO/ESP 2025 guidelines also updated risk stratification based on FIGO 2023 staging system. In the future, it is a trend to refining EC management based on non-anatomical parameters such as tumor type, tumor grade, LVSI, and molecular classifications. However, applying molecular-agnostic FIGO 2023 criteria for stage I/II disease requires caution in the absence of known POLE mutation status, whereas upstaging aggressive subgroups based on histology, grade, and LVSI could predict prognosis [37].

The major strength of this study is the large size of population. Another strength is that this is the first population-based study to evaluate the dynamic recurrence or modality prediction value of FIGO 2023 staging system for EC. This study has some limitations. First, this is a retrospective single-center cohort study, which limits its generality worldwide. Secondly, this study enrolled individuals at an early stage and validated the FIGO 2023 staging system in this patient cohort. Stage III–IV group should be further investigated in the future. Thirdly, the generalizability of these findings is limited by the study's design, which specifically enrolled FIGO stage I–II patients who received adjuvant radiotherapy. Consequently, the results may not be applicable to all patients, such as those managed with chemotherapy alone or no adjuvant therapy. Future studies are needed to validate these findings in broader, unselected cohorts. Forth, this study enrolled patients with unknown molecular classification, thus the recurrence patterns with known molecular classifications should be further investigated in the future. In the era of molecular classification, prospective studies are essential to derive valid longitudinal HR estimates for EC patients across stages I–IV under the FIGO 2023 staging system. With such data, individualized treatment and surveillance strategies could be tailored. Finally, over the 20-year study period, treatment strategies and survival outcomes within the study population may have evolved concurrently with updates in clinical paradigms.

Although NCCN guidelines have not yet fully incorporated the FIGO 2023 staging system, the staging system, which integrates non-anatomic parameters, retains significant strengths in prognostic prediction for EC. Our initial evaluation focused on time-dependent HRs among early-stage patients under this revised staging system. This analysis identified the peak recurrence time points and subsequent risk trends for both recurrence and treatment modality failure. Aggressive histological types, including stage IC and IIC disease, warrant particular attention. Furthermore, this dynamic understanding of HRs enables optimization of follow-up strategies and surveillance protocols, demonstrating excellent cost-effectiveness in health economics.

In conclusion, patients with early-stage EC had comparable survival outcomes to the general population, indicating a favorable prognosis. Under FIGO 2023 staging system, for the entire cohort, the peak AHR for recurrence was highest for the first 3 years, peaking at 1.3 years (3.5%), and subsequently declined. The peak AHR for death increased continuously within the first 4 years. Beyond 4 years, the AHR of death remained essentially constant at approximately 1.0%. For early EC substages, peak AHR occurs within the first 3 years, whereas peak mortality risk emerges within the first 4 years. FIGO 2023 staging system was confirmed as an independent prognostic factor for recurrence prediction and demonstrated good predictive performance, with AUC values of 0.644, 0.612, and 0.611 for 1-year, 3-year, and 5-year recurrence prediction, respectively. These findings indicate that FIGO 2023 staging system exhibit excellent recurrence surveillance and prognostic prediction for early-stage EC patients after postoperative radiotherapy. Additionally, other clinicopathological variables—including grade, age, peritoneal cytology, cervical invasion, LVSI, and myometrial invasion—also played crucial roles in clinical decision-making.

SUPPLEMENTARY MATERIALS

Table S1

Cox regression model comparisons among different models

Table S2

Multi-variables cox regression analysis results

Table S3

Comparisons of clinical and pathological features of <60 years group and ≥60 years group

Table S4

Failure patterns per International Federation of Gynecology and Obstetrics 2023 substages

Fig. S1

Sankey plot illustrating the stage migration between the FIGO 2009 and FIGO 2023 staging system.

Fig. S2

Kaplan-Meier curves for OS and DFS in the entire cohort.

Fig. S3

Kaplan-Meier survival curves by substage. (A) OS: FIGO 2009 stages IA, IB, II. (B) OS: FIGO 2023 stages IA1–IA2, IB, IIA–IIB, IC/IIC. (C) DFS: FIGO 2009 stages IA, IB, II. (D) DFS: FIGO 2023 stages IA1–IA2, IB, IIA–IIB, IC/IIC.

Fig. S4

AHR of death or recurrence under FIGO 2009 staging system. (A) Entire cohort. (B) Death by FIGO 2009 substage. (C) Progression or death by FIGO 2009 substage.

Fig. S5

AHRs. (A) Death in endometrioid vs. non-endometrioid carcinoma groups. (B) Progression or death in endometrioid vs. non-endometrioid carcinoma groups. (C) Death by risk group. (D) Progression or death by risk group.