ICC is a highly malignant tumor with poor prognosis, only a small number of ICC patients (10%−15%) are suitable for surgical treatment (Roy et al. 2021). Even with surgical treatment, ICC is still associated with high recurrence and poor survival outcomes. In our study, the ICC tumor recurrence rate reached an astonishing 64.9%, and the 5-year DFS rate was only 13.6%. The survival rate was extremely poor, 5-year survival rate was only 22.4%. Consistent with previous studies, Jeong, J. et al. demonstrated that more than 50% of ICC patients developed disease progression within 20 months after radical surgery, and the 5-year OS rate was between 30 and 35% (Jeong et al. 2022). Mazzaferro, V. et al. reported that 50–70% of the patients developed recurrence after a median of 26 months. The median OS after radical surgery was 40 months, and the 5-year OS rate was 25–40% (Mazzaferro et al. 2020). LNM is an independent risk factor for poor prognosis after surgical resection as the result 3.3 described. Consistent with previous study, up to 45–65% of ICC patients found LNM at the time of clinical diagnosis (Sposito et al. 2022). The 5-year OS of pN0 patients was 35–50%, while that of pN1 patients was only 0–20% (Navarro et al. 2020). Once LNM was confirmed, the median survival time after radical surgery decreased to 15–20 months, and the 5-year OS rate decreased to 15% (Hyder et al. 2014). ICC does have a poor prognosis, especially the patients with LNM.

However, there is currently no international consensus guideline on the lymphadenectomy, which is precisely why lymphadenectomy remains a controversial procedure. The majority of surgeons followed these general criteria: (1) preoperative imaging findings of metastatic LNs; (2) intraoperative exploration revealing enlarged or suspicious LNs. Whether lymphadenectomy is beneficial to patients has indeed been a highly controversial topic. In our retrospective study, compared with LN− group, the OS (P = 0.038) and DFS (P = 0.0027) of the LN+ group were significantly lower before PSM and IPTW. The LND group has a worse prognosis, which is contrary to various previous studies that LND could promote the long-term survival (Chen et al. 2022; Ke et al. 2021; Kim et al. 2019). After the baseline comparison, we found that those confounding factors affected the prognosis. Firstly, the LN+ group had a higher T stage (P = 0.0008). The higher the T stage, the worse the prognosis (Huang et al. 2021; Sun et al. 2021). The higher T stage means the larger tumor diameter, more multiple tumors, or higher probability of vascular invasion, (Zhang et al. 2021) which could be roughly judged by the doctor from a macro perspective during the operation. Thus, the higher the tumor stage, the more likely doctors will choose to perform LNB for accurate staging or LND to prevent metastasis, making it easier for patients with higher T stages to undergo lymphadenectomy, leading higher T stage in LN+ group. In addition, the LN+ group had higher CEA level (LN- vs. LN+: 1.07 [0.59, 1.96] vs.1.43 [0.77, 2.17], P = 0.041) and CA199 level (LN- vs. LN+: 4.29 [3.10, 6.80] vs.6.06 [4.07, 9.09], P = 0.0005), which indicate the overloading tumor cells, greater risk of LNM, greater risk of distant metastasis, and poor prognosis (Li et al. 2022). In postoperative pathology, the tumor tissue of the LN+ group had higher lymphocyte infiltration (P = 0.008), also indicates a worse prognosis (Galun et al. 2018). 30.12% LN− patients received the postoperative adjuvant therapy, lower than the LN+ group (48%, P = 0.0073). The postoperative adjuvant therapy may improve the prognosis that was another confounding factor contributing to the better survival rate of LN− group. These confounding factors together led to better OS and DFS in the LN− group than in the LN+ group.

Therefore, we urgently needed to adjust these confounding factors. After univariate and multivariate COX analysis and SMD analysis, we actively selecting indicators that theoretically have a causal relationship with prognostic indicators and including them in the PSM and IPTW analysis, the results are easier to interpret, and the problem of model failure due to too many variables is avoided, improving the accuracy and reliability of the study. PSM is a classical method to reduce the confounding effect in the retrospective study (Lee et al. 2022). PSM showed robust matching effect, greatly eliminate the influence of endogenous factors as indicated in the observational study (Benedetto et al. 2018). However, there is sample size loss caused by non-pairing in the process of “finding paired samples”. In our study, we included 7 covariates that needed to be adjusted during PSM, resulting in a drastically reduction of patients included, although it further eliminated the influence of irrelevant variables. Only 55 pairs (110 in total) of patients were involved after PSM between LN− and LNB group, the data loss ratio reached 44.4%. Only 41 pairs (82 in total) of patients were involved after PSM between LN− and LND group with the data loss ratio of 57.5%. At the same time, IPTW is a rising statistical method without causing data loss (Austin & Stuart 2015). IPTW has been used by various ICC research fields to adjust the confounding factors on the basis of maintaining the sample size (Ke et al. 2023; Sposito et al. 2023).

After univariate and multivariate COX analysis, we involved those seven confounding factors (“hepatobiliary history”, “CEA(Log2)”, “CA199(Log2)”, “T stage”, “lymphocyte invasion”, “liver capsule invasion”, “surgery approach”) as the covariates of PSM and IPTW. As discussed above, “CEA”, “CA199”, “T stage”, “lymphocyte invasion” have significant influence on the prognosis in ICC patients. Although the there was no significant difference of “hepatobiliary history” between those three groups, ICC patients with a history of hepatobiliary may have a worse prognosis since the certain damage has caused to bile duct cells, (Lurje et al. 2023) thus it is necessary to be involved into adjusted factor. In addition, we found that the P value of “liver capsule invasion (P = 0.0552)”, “surgery approach (P = 0.0675)” ​​were close to 0.05 and may have an uncertain impact on prognosis. Previous research has shown the invasion of liver capsule caused worse survival rate, (Zhou et al. 2020) and the laparoscopic surgery leads to better short-term outcomes in ICC patients, (Zhao et al. 2023). According to statistical principles of PSM and IPTW, researchers can determine the threshold of the including P value of PSM and IPTW between 0.05 and 0.1 based on actual clinical conditions. Therefore, we finally included the factors with the P value of multivariate COX analysis less than 0.1 as the covariates of PSM and IPTW. These seven confounding factors have been well adjusted after PSM and IPTW (Fig. 4).

As described in results 3.4, compared with LN− group, LNB group only increased the operation time and postoperative or total hospitalization time after PSM and IPTW, without increasing the complication or risk of bleeding or risk of transfusion. Due to the inaccuracy of preoperative imaging in predicting LNM, LNM staging was inaccurate in up to 40% of ICC patients, (Tsilimigras et al. 2021) LNB is of great significance in the diagnosis of LNM. LNB as the gold standard for pathological diagnosis of LNM, provides accurate nodal staging and enables precise pathological staging for ICC patients (Sposito et al. 2023). In our study, the NO.8 LN (the common hepatic artery LN) and NO.12 LN (the hepatoduodenal ligament LN) were firstly dissected to achieve the accurate staging during LNB. Previous researches showed that 4 or more LNs are sufficient to obtain accurate staging, (Chen et al. 2021). and the No.12 LN and No.8 LN must be included during the LNB for accurate staging since those two are the highest risk areas for LNM (Kang et al. 2021; Kim et al. 2022a). Accurate nodal staging could predict and guide the postoperative adjuvant treatment to achieve better prognosis (Ke et al. 2021). Although LNB also prolongs the operation time and hospitalization days compared with LN− group, in view of the fact that LNB can bring more accurate pathological staging guides subsequent treatment, these adverse events are acceptable. In addition, no-LND may lead to the omission of LNM and inaccurate staging of ICC, LNB can significantly make up for these two shortcomings.

To assess whether LNB is sufficient for lymphadenectomy, we evaluated the possibility of missing positive LNs in results 3.6. The results showed that 29.57% of LNB group patients were diagnosed as “pLNM+”, while 40.00% of LND group patients were diagnosed as “pLNM+”. The LNB group was 10 percentage points lower than the LND group, suggesting that there may be difference suggests a potential underestimation of pLNM+ in the LNB group, LNB may have a possible 25% LN positive missing rate. However, a Chi-square test showed there was no significant difference in the proportion of pLNM+ between those two groups (P = 0.1326). This indicates that, statistically, the difference in pLNM+ rates is not significant, and the observed difference could be due to random variation. To determine whether the underestimation of pLNM+ in the LNB group affects survival outcomes, we compared the prognosis of patients with pLNM+ following LND and LNB. The results showed no difference between OS (P = 0.51) and DFS (P = 0.14) of patients with pLNM+ following LND and LNB (Fig. 7C and D). This suggests that even if there is a 25% underestimation of pLNM+ in the LNB group, it does not significantly impact the survival outcomes. Similarly, we compared the prognosis of patients with pLNM− following LND and LNB. The results again showed no difference between OS (P = 0.78) and DFS (P = 0.7) of patients with pLNM− following LND and LNB (Fig. 7E and F). This further supports the reliability of LNB in prognostic assessment and indicates that the potential missing of positive lymph nodes in LNB is acceptable. LNB is comparable to LND, indicating that the prognostic impact of pLNM− is reliable, and the potential missing of LNMs in LNB is acceptable, suggesting that LNB can achieve the same effect as LND.

LND is still under debated due to it increases the difficulty of surgery, adverse effects on postoperative recovery, and uncertainty about prognosis (Lee et al. 2020; Zhou et al. 2019) Although some centers indicated that LND could promote the long-term outcomes and prognosis of ICC patients, (Chen et al. 2022; Ke et al. 2021; Kim et al. 2019; Yoh et al. 2019) many centers including our center did not find the benefit of LND to the therapeutic effect, the OS or DFS was not significant improved after LND (Hu et al. 2021; Li et al. 2013; Zhou et al. 2019; Zhu et al. 2023). In our center, we strictly followed the standard of LND steps and resection range defined by AJCC, (Kim et al. 2019) at least 6 LNs including the NO.12 LN and NO.8 LN were dissected. However, review of LND showed that the implementation rate of LND in major hepatobiliary surgery centers in the world ranges from 26.9 to 100%, only 10% of ICC patients receive the adequate LND (Lluís et al. 2023). In addition, many centers do not strictly follow the AJCC guidelines for LND, the mode and steps of LND vary from center to center, depending on the experience of the surgeon. Those contributes to the different or even completely opposite conclusions eventually lead to the debate of LND.

The original purpose of LND is to accurately stage and prevent suspicious LNM to reduce the risk of recurrence and achieve a better prognosis. However, due to the high complexity and variability of lymphatic system around the liver, (Morine & Shimada 2015) it is impossible for us to comprehensively dissect all the LNs around the liver. Although we dissected the most suspicious LNs that may develop LNM such as NO.12 and NO.8 LNs, (Kang et al. 2021) the other LNs still have the probability of developing LNM. In addition, LNM is a systemic disease (D. Y. Li et al. 2013). Researches have proved that LNM in ICC can directly spread to distant regional LNs through the multidirectional lymphatic pathways connected to the systemic lymphatic system (Li et al. 2013). LND performed on ICC patients who are confirmed to have LNM may still only be LN sampling in a broad sense and cannot achieve the dissection effect only the LNB effect. Therefore, it is not surprising that LND does not achieve the original expected prognosis. Furthermore, LND is associated with increased post-operative morbidity (Zhou et al. 2019). As described in Results 3.5, after adjusting the confounding factors by using PSM and IPTW, compared with LN− group, LND significantly increased the operation time, the risk of postoperative complications bleeding, transfusion, prolong the total and postoperative hospitalization days, which were consistent with previous studies (Yoh et al. 2019). Previous studies indicated that the incidence of complications increases significantly after LND in patients with cirrhosis (Bagante et al. 2018). ICC patients with cirrhosis need to be more careful to perform LND. LNB can provide almost the same information about staging as LND, but significantly reduces post-operative morbidity (Choi et al. 2009). As described in Results 3.6, compared with LND, LNB shortens surgery time with minimal impact on operative duration and avoid the increased risk of bleeding, blood transfusion, and postoperative complications. In addition, according to the AJCC staging (Chun et al. 2018; Lee and Chun 2018), among the four types of biliary system tumors (ICC, gallbladder cancer, hilar cholangiocarcinoma, and distal cholangiocarcinoma), only ICC has an N stage of N1. The other three types are divided into N1 and N2, and they are collectively classified as extrahepatic cholangiocarcinoma. Once there are positive LNs detected in ICC, the N stage is determined without emphasizing the number of positive LNs. LNB is already sufficient for pathological staging, and LND is not necessary for further accurate staging. Therefore, LNB is more recommended compared with LND for accurate pathological stage and more beneficial for patients.

For the future outlook, since LNB is superior to LND and bring more benefits to patients, we will promote LNB more preferentially than LND during radical surgery of ICC in clinical work. In addition, LNM plays such important role in the prognosis of ICC patients, if LNM can be accurately determined before surgery by imaging, then targeted lymphadenectomy can be performed. However, the radiographic LNM staging was inaccurate in up to 40% of ICC patients (Tsilimigras et al. 2021). In the future we propose to establish the muti-imaging omics to build the models to improve the accuracy of preoperative LNM judgment as well as the early recurrence in ICC patients. At the same time, sentinel lymph node biopsy (SLNB) is worth further research. SLN refers to the first station of lymphatic reflux in the organ and the first LNs where LNM occurs (Kurochkin et al. 2022). Like other cancers such as breast cancer and endometrial cancer, SLNB provides a highly reliable method to achieve the accurate staging and is a potential solution to significantly reduce the negative impact of lymphadenectomy in ICC patients (Yasukawa et al. 2021). Although few studies have been performed on SLNB due to the complexity of the hepatic lymphatic system, it will be a major breakthrough in the field of ICC once successful.

In our study, we collected all the eligible ICC patients in this center over the past ten years. As the top cancer hospital in China, the surgical procedures were strictly carried out in accordance with standard procedures, many operation variables were controlled. By using both PSM and IPTW methods, the confounding factors were also well controlled. At the same time, this study was one of the few retrospective studies on LND, and we have reached a conclusion different from enormous previous literatures, providing a solid theoretical basis for opposing the removal of LND. There are still some limitations in our study. As a retrospective study, our articles inevitably face recall bias and choice bias. Furthermore, the single-center characteristics of our study limit the universality of the results. This is a single-center retrospective study with geographical limitations in China and relatively small sample size. The control of variables in retrospective study is far inferior to that in prospective studies, and the loss to follow-up bias is still exist. This article proposes and highlights the LNB, while standardized LNB needs to be further developed in the future.