A consecutive series of patients with endometrial cancer who visited our facility between April 1, 2009, and March 31, 2021, were identified from the hospital`s electronic medical records. All patients were pathologically diagnosed with endometrial cancer, and patients who were preoperatively estimated with stage IA based on preoperative CT scans and MRI imaging were selected for the study. The preoperatively estimated stage IA group was divided into two groups, SLNB or PLND, excluding cases where neither SLNB nor PLND was performed (Fig. 1). The SNNS group was composed of patients who chose to omit further PLND when negative metastasis in SLN was detected intraoperatively and the PLND group was composed of patients who underwent bilateral PLND regardless of whether SLNB was performed.

Patient flow chart. SLNB: Sentinel lymph node biopsy PLND: Pelvic lymph node dissection SNNS: Sentinel node navigation surgery (Omission of bilateral pelvic lymph node dissection followed by results of intra-operative frozen section diagnosis of the bilateral SLN)

SLNB procedures for early-stage uterine malignancies have been performed with the approval of our hospital's Institutional Review Board since November 2010, while SNNS procedures began on December 1, 2012, when ultrastaging was introduced. SLNB and SNNS are not standard methods for assessing lymph node metastasis in Japan; patients themselves must make the decision to undergo PLND or SLNB after receiving an explanation of the procedures for those surgeries. If SLNB is chosen, the patient decides whether to undergo SNNS or PLND regardless of the results of the intraoperative diagnosis.

On the day before the surgery, 99mTc-phytate (PDR Pharma, Japan) was injected into the cervix at the 3 o'clock and 9 o'clock positions at a dose of 40 MBq/0.4 ml. Then, lymphoscintigraphy was performed 2 h later to estimate the location of the SLN. On the day of the surgery, 1 ml of indocyanine green (ICG) 0.025 mg/ml (Diagnogreen, Daiichi Sankyo, Japan) was administrated at the 2, 4, 8 and 10 o'clock positions of the cervix, respectively. The SLN was detected using a gamma probe (Navigator GPS, Sheeman Co. Ltd., Japan) and a fluorescent camera (Visera Elite II video system, Olympus, Japan); then it was excised and sent to a pathologist for intraoperative frozen section diagnosis [26, 27]. The diagnostic algorithm of SNNS has as a side-specific PLND: when there is an intraoperative diagnosis of SLN negativity, further removal of PLNs is omitted, and PLND is performed on the positive SLN side or on the side where there is an undetected SLN [8, 9]. Para-aortic lymph node (PAN) dissection was performed with PLND when the patient was preoperatively diagnosed as high-risk and gave consent for PAN dissection, regardless of whether SLNB was performed. An extrafascial extended hysterectomy, a standard hysterectomy in Japan, was performed. Intraoperative diagnosis of cancer metastasis was made by taking frozen section specimens at 2-mm intervals, which were perpendicularly sliced and stained with H&E. After preparing formalin-fixed paraffin embedded tissue (FFPE), the remaining specimens were ultrastaged. Immunohistochemical analysis using cytokeratin AE1/AE3 was performed at 20-μm intervals on a 3-μm FFPE section to detect micrometastasis. A routine histopathological examination was made to evaluate the FIGO stage and risk of recurrence. Low risk was defined as FIGO stage IA endometrioid carcinoma grade 1 or grade 2 with negative lympho-vascular space invasion (LVSI). Intermediate and high risk were defined as any risk other than low risk. For patients postoperatively diagnosed as intermediate or high risk, an additional six cycles of chemotherapy or radiation therapy, and three cycles of chemotherapy if isolated tumor cells (ITC) were confirmed, were proposed and those additional treatments were performed unless refused.

The data obtained were summarized using basic statistics and tested using either a Mann–Whitney test, a chi-square test, or Fisher's exact test. The propensity score method (pair-matching) was utilized to.

adjust for confounding factors in the SNNS and PLND groups. Age (under or over 55), LVSI (negative or positive), histologic subtype (either endometrioid carcinoma grade 1 or 2, or others) and upstaging (FIGO stage IA or others) were used as explanatory variables [28,29,30]. The propensity score was calculated by logistic regression analysis. The matching caliper was set to 0.2 and it was used to create a 1:1 matched pair from both groups. Then, the survival rates of the created groups were compared using the adjusted Kaplan–Meier curves with a log-rank test. The results were considered significant at p < 0.05 and all tests were two-tailed. All survival analyses were performed using EZR 1.52 (Saitama Medical Center, Jichi Medical University, Saitama, Japan), a GUI of R ver.4.0.0 (The R Foundation for Statistical Computing, Vienna, Austria) [31].

The start of observation was the date of surgery, and the end of observation was either the last hospital visit or death confirmation by September 30, 2021. Overall survival (OS) was calculated from all-cause death events, while disease-specific survival (DSS) was calculated from death events due to recurrence diagnosed by CT scan every six months or from vaginal stump cytology. Disease-free survival (DFS) is defined as the period from the start date to the earlier of either the date of recurrence or the date of death from any cause.

The nodal recurrence rate after SNNS, P(A), is 1.2% and the nodal recurrence rate after PLND, P(B), is 1.7% in a systematic review [32]. Consider a facility H where PH(A), the true recurrence rate after SNNS, is 1.2%. It is obvious that SNNS should be performed if they know the value of PH(A). However, they have no way of knowing it. The only performance index that they can use to judge whether to continue SNNS or not is thus PH(A|N), the recurrence rate obtained from N SNNSs (the number of SNNSs) performed at the facility H. However, in the case of PH(A) = 1.2% and P(B) = 1.7%, P(PH(A|N) > P(B)), the probability that PH(A|N) is greater than P(B), is 50.4% when N = 58 if recurrences occur randomly with probability PH(A). Therefore, it is difficult for the facility to make a judgement on the basis of PH(A|N) about whether they should continue performing SNNS if PH(A|N) > P(B) at the beginning of SNNS. We thus developed a new procedure to solve this problem when PH(A|N) > P(B), as shown in Fig. 2. Here, K is a sufficiently large value greater than one: K = 2 in this study. Note that it is obvious that SNNS should be continued regardless of the value of \(\tilde}}\)(PH(A|N) > P(B)), an estimate of P(PH(A|N) > P(B)), when PH(A|N) ≦ P(B).

Flow chart to compare maximum likelihood hypothesis and alternative hypothesis (*1). The maximum likelihood hypothesis means PH(A) = PH(A/N) and this procedure is done when PH(A|N) > P(B). Thus, PH(A) > P(B) and SNNS should be discontinued

Consider an example where N = 20 and nA = 1 (PH(A|N) = 5.0%, Fig. 3a). According to the procedure described above, SNNS should be continued in this case, although PH(A|N) = 5.0% is sufficiently greater than P(B) = 1.7%. This is because \(\tilde}}\)(PH(A|N) > P(B)) for the maximum likelihood hypothesis is 85.0% and is less than two times that for the alternative hypothesis (61.7%) (Table 1); that is, PH(A) > P(B) is insignificant (the alternative hypothesis is significant) according to the procedure with K = 2. However, if nA = 3 for N = 60 (see Fig. 3b), SNNS should be discontinued although PH(A|N) = 5.0%, in the same manner as in the above case with N = 20. This is because \(\tilde}}\)(PH(A|N) > P(B)) for the maximum likelihood hypothesis is 89.9% and is greater than two times that for the alternative (43.8%) (Table 1); that is, PH(A) > P(B) is significant (the maximum likelihood hypothesis is significant). We can thus evaluate the validity of performing SNNSs in an actual situation where N is less than 100 by using a procedure with a criterion that is intuitively easy to understand.

\(\tilde}}\)(PH(A|N) > P(B)) [%] (P(B) = 1.7% for maximum likelihood hypotheses and alternative hypothesis. a: at 1 event per 20 attempts, b: at 3 events per 60 attempts. P(B): Recurrence rate after PLND (1.7%) obtained from a systematic review. PH(A): True recurrence rate after SNNS in facility H, which it is impossible to know in advance. nA: The number of recurrences in N SNNSs (a random variable). ◆: A binomial distribution B (N, PH (A)) that obeys nA. PH(A|N) = nA/N: Recurrence rate after SNNS obtained from N SNNSs performed in facility H, which is a random variable because nA is a random variable. ─◆─: Probability density function for PH(A|N), which is an approximation of B (N, PH (A)) when PH (A|N) is treated as a continuous random variable. *: Area under curve (AUC) based on the probability density function of each hypothesis. For instance, 61.7 for the alternative hypothesis in 20 attempts means the percentage to the right of 0.017 when the overall area of the graph is set to 1. \(\tilde}}\)(PH(A|N) > P(B)): The areas indicated by * in the four graphs, which means an estimate of the probability that PH (A|N) is greater than P(B)