Problems of Lymph Node Dissection in Intrahepatic Cholangiocarcinoma: A Multicenter Retrospective Study Using Inverse Probability of Treatment Weighting.

Lymph node dissection (LND) is considered to improve the prognosis of patients with intrahepatic cholangiocarcinoma (ICC). Although the National Comprehensive Cancer Network (NCCN) guidelines recommend routine LND in ICC, the role of LND remains controversial. This study aimed to explore the effect of LND on the prognosis of patients with ICC from two Chinese centers. Methods Patients were identied in two Chinese academic centers. Inverse probability of treatment weighting (IPTW) was used to reduce bias. Kaplan–Meier curves and Cox proportional hazards models were used to compare overall survival (OS) and disease-free survival (DFS).


Patient characteristics
Of the 251 patients, 189 (75.2%) underwent LND, and 72 (38%) had at least one LNM. A minimum of 6 lymph nodes were dissected in 67 patients (35.5%). The LND area of 116 patients (61.4%) was limited to the hilar area (hepatoduodenal ligament). There were signi cant baseline differences between the two groups (Table 1). POD time, operative time, intraoperative blood loss, and intraoperative blood transfusion were all higher in the LND group.  CEA and multi-disease were independent risk factors in multivariate analysis (P < 0.05) ( Table 2).

Effects of LND on prognosis--IPTW
There were signi cant differences in patient characteristics between the LND and nLND groups. To further determine whether LND improved ICC prognosis, we used IPTW to minimize confounders. LNM or related variables were not included because they were unclear in the LND group. The adjustments are shown in Fig. 1a. The standardized mean differences showed that IPTW effectively balanced the between-group differences. The median OS and DFS were 61 months vs. 57 months (P = 0.75) and 34 months vs. 17 months (P = 0.10), respectively (nLND vs. LND, Fig. 1b-1e).
Further, we compared nLND with patients who underwent LND and were LNM-negative using the same method (Fig. 2a); Similarly, the difference in the OS and DFS were signi cant between the two groups; these differences were eliminated by IPTW ( Fig. 2b-e). The nLND and LND ≥ 6 groups were also compared at the same time (Fig. 3a, Fig. 3b-e). The IPTW-adjusted analysis showed that in all the groups, LND had a greater effect on DFS than on OS, and the DFS of the LND ≥ 6 group was signi cantly worse than that of the nLND group.

LND subgroup analysis
A subgroup analysis of LND was conducted to investigate whether the range, LND number, and LNM number affected prognosis. LN positivity was a risk factor for both OS and DFS, with a high hazard ratio. CA 19 − 9, CEA, tumor size, T stage, positive margin, and LNM were all risk factors for OS. Tumor size, LNM, and LND numbers were signi cant for DFS.
A relationship was observed between LNM area and OS, but not between LNM area and DFS. LNM number > 3 affected OS and DFS; however, there was no statistical signi cance (Table 3).   (Table 4).

Discussion
ICC is a malignant disease with a poor prognosis. The goal of surgery is complete (R0) resection. The prevalence of LNM in ICC is as high as 17%-39.1% [15,16] , and LND is considered as part of R0 resection. Many surgeons believe that LND improves ICC survival. However, some researchers found that LND is only a staging operation and has little effect on prognosis [4,17] . In our study, the initial analysis showed a poorer prognosis in the LND group, which might be explained by the incongruity of the baseline characteristics. Variables were better adjusted using IPTW than propensity score matching (PSM); there was no signi cant difference in OS between the groups. This is not consistent with the ndings of some prior studies. Yoh et al.
compared the effect of LND on prognosis in patients with no suspected LNM before surgery [15] . This indicated that LND improved both DFS and OS in the nLND group. Only 112 patients were included, and some deviations in preoperative imaging