Combined vascular resection and reconstruction for locally advanced perihilar cholangiocarcinoma: A multicenter study

Abstract

BACKGROUND

Vascular resection may be required to achieve margin-negative resection in advanced perihilar cholangiocarcinoma, but the relative safety and survival impact of hepatic artery resection compared with portal vein resection remain unclear. We hypothesized that hepatic artery resection would not be associated with significantly inferior long-term survival in selected patients treated at experienced centers.

AIM

To evaluate vascular resection outcomes and compare hepatic artery resection with portal vein resection in advanced perihilar cholangiocarcinoma.

METHODS

This multicenter retrospective cohort study included 557 patients with perihilar cholangiocarcinoma treated at three tertiary centers between 2016 and 2020. Overall, 320 underwent curative-intent surgery, including 204 without vascular resection and 116 with vascular resection (51 portal vein resections and 65 hepatic artery resections). Perioperative outcomes and overall survival were compared using Kaplan-Meier analysis, Cox regression, propensity score matching, and restricted mean survival time analysis.

RESULTS

Compared with non-vascular resection, portal vein resection and hepatic artery resection required longer operative times (425/496 minutes vs 314 minutes), greater blood loss (900/900 mL vs 600 mL), and more transfusion (40%/39% vs 14%; all P < 0.001). Major morbidity was higher after vascular resection (non-vascular resection 35.8% vs portal vein resection 45.1% vs hepatic artery resection 44.6%; P = 0.013). Hepatic artery-related events were more frequent after hepatic artery resection than after portal vein resection (20.0% vs 5.9%; P = 0.032). Overall survival did not differ significantly between hepatic artery resection and portal vein resection, including after propensity score matching. Predicted 3-year survival < 0.80 identified shorter restricted mean survival time (9.7 months vs 28.4 months; P < 0.001).

CONCLUSION

Vascular resection is feasible in experienced centers. In this cohort, hepatic artery resection was not associated with worse long-term survival than portal vein resection, despite distinct vascular-specific complication profiles.

Key Words: Perihilar cholangiocarcinoma; Vascular resection and reconstruction; Hepatic artery resection; Portal vein resection; Prognosis

Core Tip: Vascular resection may be required to achieve curative-intent resection in locally advanced perihilar cholangiocarcinoma, but hepatic artery resection remains controversial. In this multicenter cohort, hepatic artery resection was not associated with significantly inferior long-term survival or higher major morbidity compared with portal vein resection, although hepatic artery-associated complications were more frequent after hepatic artery resection. These findings support the selective use of vascular resection in experienced centers and highlight the need for procedure-specific postoperative surveillance.

INTRODUCTION

Perihilar cholangiocarcinoma (pCCA) is a highly aggressive cancer with a poor prognosis due to late diagnosis and complex anatomy[1]. Curative treatment requires radical resection, including major hepatectomy and lymphadenectomy, but achieving a cancer-free margin (R0) is difficult[2].

A major challenge is locally advanced disease affecting the portal vein or hepatic artery, leading to the increased use of combined vascular resection (VR)[3]. However, the benefits of VR are debated, as earlier studies indicated higher perioperative risks without clear survival advantages, possibly due to more advanced disease stages[4-6].

Not all VR procedures are the same; portal vein resection (PVR) is common in high-volume centers, while hepatic artery resection (HAR) is rare due to its complexity and risk of complications. The effectiveness of HAR compared with PVR is debated, as existing studies are often single-center with selection biases and lack direct comparisons[7].

To address this, we conducted a multicenter retrospective study to assess the safety and benefits of VR vs non-VR and to compare PVR and HAR outcomes using propensity score matching and a composite safety-efficacy endpoint, following STROBE guidelines.

MATERIALS AND METHODS

Study design and patient cohort

This multicenter retrospective cohort study included patients with pCCA treated between 2016 and 2020 at three tertiary centers. The study was approved by the institutional ethics committee, and the requirement for informed consent was waived due to the retrospective anonymized design. Exclusions included those with recurrent cancer post-surgery, synchronous malignancies, unresectable distant metastases, or incomplete records. Patients undergoing curative surgery were categorized into non-VR or VR, with VR further divided into PVR and HAR, the latter including cases with combined vascular reconstruction. Because arterial resection may co-occur with portal venous reconstruction in locally advanced pCCA, HAR was defined a priori as ‘any HAR’ (with or without concomitant PVR). This reflects the clinical decision problem (whether arterial resection is undertaken) rather than an attempt to compare two mutually exclusive technical procedures. Baseline demographics, laboratory, imaging, operative and pathological variables were extracted. The neutrophil-to-lymphocyte ratio (NLR) and platelet-to-lymphocyte ratio (PLR) were derived from preoperative counts.

Preoperative assessment and indications for VR

Patients received standard evaluations with laboratory tests, tumor markers, and computed tomography (CT) or magnetic resonance cholangiopancreatography scans. Biliary extent was classified using the Bismuth-Corlette classification, and staging followed the Union for International Cancer Control (UICC) 7^th edition, which was used in the original multicenter pathological records and study database. To maintain consistency across this retrospective historical cohort, the UICC 7^th edition was used for the primary analyses. Preoperative biliary drainage and portal vein embolization were selectively used based on jaundice, liver function, and future liver remnant. VR was planned for suspected macroscopic involvement of the main portal vein or hepatic artery to achieve R0 resection. Contraindications included unreconstructable vascular invasion, diffuse intrahepatic portal involvement, or inadequate functional reserve, with final decisions made by a multidisciplinary team (MDT). A small proportion of patients were finally classified as M1 based on intraoperative findings or postoperative pathology (occult metastasis not identified on preoperative cross-sectional imaging). In these cases, resection was initiated with curative intent after multidisciplinary discussion; the final pathological stage was recorded for prognostic description rather than as a preoperative decision variable. No patient received neoadjuvant chemotherapy, radiotherapy, or chemoradiotherapy before surgery. Preoperative management mainly included biliary drainage, portal vein embolization when indicated, nutritional optimization, liver function optimization, and multidisciplinary assessment of resectability.

Surgical procedures

Specialist hepatobiliary surgeons followed shared protocols for operations, which included laparotomy, exploration, and extensive hilar lymphadenectomy. The type of hepatectomy was based on tumor location, biliary extension, and future liver remnant. Extrahepatic bile duct resection with Roux-en-Y hepaticojejunostomy was performed, with caudate lobectomy when feasible. Pancreatoduodenectomy was added for distal bile duct or pancreatic head involvement. PVR and/or HAR were performed when imaging or intraoperative findings showed encasement, stenosis, or direct invasion, with vascular reconstruction as necessary. Portal venous and hepatic arterial resections and reconstructions were primarily performed by the hepatobiliary-pancreatic surgical team, whose senior surgeons had experience in complex VR and reconstruction; vascular surgeons were consulted or involved only when technically necessary.

Pathologic examination and perioperative outcomes

Hepatobiliary pathologists evaluated tumor extent, differentiation, lymphovascular/perineural invasion, and margin status. R0 resection meant no microscopic tumor at ductal, vascular, and parenchymal margins. Vascular invasion included tumor thrombus or direct infiltration of the portal vein/hepatic artery wall. Complications were graded by Clavien-Dindo (major ≥ III). Post-hepatectomy liver failure and bile leak were defined by the International Study Group of Liver Surgery criteria[8,9], and pancreatic fistula by the International Study Group of Pancreatic Surgery criteria[10]. Hepatic artery events and portal vein thrombosis were identified through imaging and clinical findings. In-hospital mortality was death during the initial admission or within 90 days. Doppler ultrasonography was performed within 24-48 hours after reconstruction and repeated, with CT angiography as needed, to confirm hepatic artery-related events and portal vein thrombosis. Events were defined by imaging evidence of impaired patency/flow and were captured during the index admission or within 90 days postoperatively.

Follow-up and endpoints

Patients were monitored every 3-6 months for 2 years, then annually. The primary endpoint was overall survival (OS) from surgery to death or last follow-up (censored if alive). Secondary endpoints included perioperative outcomes, vascular-specific complications, and prognostic factors in the VR group. A binary 3-year mortality marker and an exploratory survival morbidity classification combining 3-year survival status and major morbidity were also evaluated descriptively. Postoperative adjuvant chemotherapy was recorded as a postoperative treatment variable. It was generally considered after recovery from surgery, particularly for patients with high-risk pathological features. The decision to administer adjuvant chemotherapy and the specific regimen were determined by the treating MDT according to postoperative recovery, liver function, performance status, pathological risk factors, and institutional practice.

Statistical analysis

Median follow-up duration was estimated using the reverse Kaplan-Meier method. All covariates in the prognostic model were defined as baseline information available by the time of hospital discharge (including final pathology). The model is therefore a postoperative prognostic tool rather than a preoperative decision model. Continuous variables are presented as medians (range), and categorical variables as n (%). Mann-Whitney U/Kruskal-Wallis tests were used for comparing continuous variables, and χ²/Fisher’s exact tests for categorical variables. OS was estimated using the Kaplan-Meier method and compared using log-rank tests, with significance at two-sided P < 0.05. In the VR cohort, Cox regression identified factors linked to OS; variables with P < 0.10 or clinical importance were included in a multivariable model. A risk score was created and shown as a nomogram, with performance evaluated using Harrell’s concordance index (C-index), time-dependent receiver operating characteristic curves, bootstrap calibration, and decision-curve analysis. The link between VR type (HAR vs PVR) and OS was assessed using adjusted Cox models, with landmark analyses at 12 months and 24 months and interaction terms for baseline risk effect modification. Futility was explored using restricted mean survival time (RMST) over 0-36 months (τ = 36 months), with sensitivity analyses at τ = 24 months and 48 months. Propensity score matching (1:1 nearest-neighbor, without replacement; caliper 0.2 on the logit) was performed to mitigate confounding by indication, with balance assessed using standardized mean differences. To assess the potential influence of M1 disease on the comparison between PVR and HAR, an M0-only sensitivity analysis was performed after excluding patients finally classified as M1 based on intraoperative findings or postoperative pathology.

RESULTS

Patient cohort and treatment allocation

Between 2016 and 2020, 557 patients with pCCA were assessed across three centers (Figure 1). Overall, 237 patients (43%) received non-surgical treatment owing to advanced disease or inadequate functional reserve. The remaining 320 (57%) underwent curative-intent surgery, including 204 (64%) without major VR (non-VR) and 116 (36%) with VR. Within VR, 51 patients (44%) underwent PVR and 65 (56%) HAR, including 17 with combined arterial and portal venous reconstruction.

Figure 1 Study flow diagram and patient disposition. VR: Vascular resection; Non-VR: Non-vascular resection; PVR: Portal vein resection; HAR: Hepatic artery resection; PSM: Propensity score matching; RMST: Restricted mean survival time.

Baseline clinicopathological characteristics

Baseline characteristics are summarized in Table 1. Compared with non-VR, VR patients particularly HAR more frequently had locally advanced disease, including Bismuth-Corlette type IV involvement (26% vs 45% vs 75%) and pathological T4 tumors (11% vs 73% vs 92%). In contrast, nodal status and distant metastasis distributions were similar between non-VR and VR and did not differ materially between PVR and HAR.

Table 1 Baseline clinical characteristics of patients with perihilar cholangiocarcinoma stratified by vascular resection type, n (%)/median (range).

Variables	Non-VR (n = 204)	PVR (n = 51)	HAR (n = 65)	P value (PVR vs HAR)	P value (non-VR vs VR)
Age (year)	63 (34-80)	68 (31-81)	67 (35-79)	0.896	< 0.001
Body mass index (kg/m2)	21.5 (18.0-33.0)	22.1 (18.6-28.0)	22.3 (18.0-27.0)	0.397	0.141
Sex				0.728	0.158
Female	72 (35)	21 (41)	30 (46)
Male	132 (65)	30 (59)	35 (54)
Hypertension	69 (34)	21 (41)	29 (45)	0.855	0.126
Diabetes	83 (41)	25 (49)	32 (49)	1.000	0.178
CEA (ng/mL)	3 (1-135)	3 (1-42)	4 (2-80)	0.008	< 0.001
CA19-9 (U/mL)	143 (1-12000)	135 (10-27852)	150 (11-27563)	0.343	< 0.001
ASA score				0.798	0.772
1-2	140 (69)	35 (69)	42 (65)
3-4	64 (31)	16 (31)	23 (35)
Bismuth type				0.002	< 0.001
1-3	150 (74)	28 (55)	16 (25)
4	54 (26)	23 (45)	49 (75)
Preoperative biliary drainage	149 (73)	49 (96)	63 (97)	1.000	< 0.001
Preoperative portal vein embolization	15 (7.4)	3 (5.9)	5 (7.7)	1.000	1.000
Type of hepatectomy				< 0.001	< 0.001
S1234	70 (34)	11 (22)	18 (28)
S1458	3 (1.5)	1 (2.0)	1 (1.5)
S15678	71 (35)	20 (39)	12 (18)
S123458	39 (19)	7 (14)	33 (51)
S145678	7 (3.4)	12 (24)	1 (1.5)
Limited liver resection	9 (4.4)	0	0
Bile duct resection	5 (2.5)	0	0
Combined pancreatoduodenectomy	25 (12)	4 (7.8)	5 (7.7)	1.000	0.286
Longitudinal tumor invasion length (cm)	NA			0.608	NA
≤ 3	NA	21 (41)	31 (48)
> 3	NA	30 (59)	34 (52)
Pathologic nerve invasion				0.385	< 0.001
Yes	33 (16)	24 (47)	37 (57)
No	171 (84)	27 (53)	28 (43)
Pathologic vascular invasion				0.060	< 0.001
Yes	66 (32)	35 (69)	32 (49)
No	138 (68)	16 (31)	33 (51)
T stage				0.009	< 0.001
T1, T2, T3	182 (89)	14 (27)	5 (7.7)
T4	22 (11)	37 (73)	60 (92)
N stage				0.434	0.376
N0	126 (62)	26 (51)	39 (60)
N1-2	78 (38)	25 (49)	26 (40)
Distant metastasis	5 (2.5)	3 (5.9)	6 (9.2)	0.729	0.052
Postoperative adjuvant chemotherapy	22 (11)	22 (43)	32 (49)	0.642	< 0.001

VR: Vascular resection; Non-VR: Non-vascular resection; PVR: Portal vein resection; HAR: Hepatic artery resection; CEA: Carcinoembryonic antigen; CA19-9: Carbohydrate antigen 19-9; ASA: American Society of Anesthesiologists; S: Hepatic segment; NA: Not applicable.

Surgical procedures, vascular reconstruction techniques, and procedure-related complications

Operative details, vascular reconstruction techniques, and procedure-related complications are summarized in Supplementary Table 1. In the VR cohort, 51 patients underwent PVR and 65 underwent HAR, including patients who required combined arterial and portal venous reconstruction. Segmental resection was the most common portal venous procedure, whereas end-to-end anastomosis was the predominant hepatic arterial reconstruction technique. Arterial transposition, graft interposition, two arterial reconstructions, and arterial resection without reconstruction were used less frequently. Hepatic artery-associated complications occurred in 3 of 51 patients in the PVR group and 13 of 65 patients in the HAR group. Portal vein thrombosis was observed in 4 patients after PVR and 3 patients after HAR. Major morbidity was similar between the PVR and HAR groups, whereas 90-day mortality was 2.0% and 4.6%, respectively.

Perioperative outcomes and procedure-related complications

Perioperative outcomes are shown in Table 2. Compared with non-VR, both PVR and HAR were associated with longer operative time (425 minutes and 496 minutes vs 314 minutes; P < 0.001), higher blood loss (900 mL and 900 mL vs 600 mL; P < 0.001), and greater transfusion requirements (40% and 39% vs 14%; P < 0.001). Major morbidity (Clavien-Dindo ≥ III) was higher after VR than after non-VR (35.8% vs 45.1% vs 44.6%; P = 0.013). In-hospital/90-day mortality occurred only after VR (2.0% after PVR; 4.6% after HAR), without a significant difference between PVR and HAR (P = 0.630).

Table 2 Perioperative outcomes and complications stratified by vascular resection type, n (%)/median (range).

Variables	Non-VR (n = 204)	PVR (n = 51)	HAR (n = 65)	P value (PVR vs HAR)	P value (non-VR vs VR)
Operation time (minutes)	314 (230-588)	425 (270-775)	496 (326-835)	0.002	< 0.001
Operative blood loss (mL)	600 (200-3000)	900 (220-4000)	900 (250-2400)	0.532	< 0.001
Blood transfusion	29 (14)	20 (40)	24 (39)	0.916	< 0.001
Liver failure ISGLS grade				0.939	0.005
Grade B	23 (11)	12 (24)	14 (22)
Grade C	3 (1.5)	2 (3.9)	3 (4.6)
Second operation	9 (4.4)	3 (5.9)	3 (4.6)	1.000	0.973
Clavien-Dindo				0.915	0.013
Grade 3	69 (34)	19 (37)	23 (35)
Grade 4	4 (2)	3 (5.9)	3 (4.6)
Grade 5 (death)	0 (0.0)	1 (2.0)	3 (4.6)
Bile leakage ISGLS grade ≥ B	50 (25)	13 (25)	18 (28)	0.761	0.686
Pancreatic fistula ISGPS grade ≥ B	39 (19)	10 (20)	12 (18)	0.729	0.696
Intraperitoneal abscess	15 (7.4)	3 (5.9)	4 (6.2)	1.000	0.827
Pleural effusion	17 (8.3)	5 (9.8)	9 (14)	0.707	0.374
Ascites	4 (2.0)	4 (7.8)	5 (7.7)	1.000	0.809
HA-associated complication				0.032	0.001
Thrombosis	0 (0)	0 (0)	3 (4.6)
Bleeding	3 (1.5)	1 (2.0)	2 (3.1)
Liver abscess	4 (2.0)	1 (2.0)	2 (3.1)
Liver infarction	1 (0.5)	1 (2.0)	6 (9.2)
PV thrombosis	3 (1.5)	4 (7.8)	3 (4.6)	0.698	0.040

VR: Vascular resection; Non-VR: Non-vascular resection; PVR: Portal vein resection; HAR: Hepatic artery resection; ISGLS: International Study Group of Liver Surgery; ISGPS: International Study Group of Pancreatic Surgery; HA: Hepatic artery; PV: Portal vein.

Within VR, the overall Clavien-Dindo grade distribution was comparable between PVR and HAR (P = 0.915). However, vascular-specific complication profiles differed: Hepatic artery-related events (thrombosis, bleeding/pseudoaneurysm, hepatic infarction, liver abscess) were more frequent after HAR than PVR (20.0% vs 5.9%; P = 0.032), whereas portal vein thrombosis rates were similar (7.8% vs 4.6%; P = 0.698). Among patients in the VR cohort, clinically relevant postoperative pancreatic fistula was associated with a numerically higher incidence of hepatic artery-associated complications, although the difference did not reach statistical significance (27% vs 11%, P = 0.078). No significant association was observed between pancreatic fistula and portal vein thrombosis (4.5% vs 6.4%, P = 1.000) or any vascular-specific complication (27% vs 14%, P = 0.200) (Supplementary Table 2).

Long-term survival according to resection strategy

The median follow-up duration in the VR cohort was 88.5 months [95% confidence interval (CI): 76.8-97.5], as estimated using the reverse Kaplan-Meier method. Kaplan-Meier analysis showed markedly better OS after curative-intent resection (non-VR or VR) than in unresectable disease. Within the surgical cohort, OS was shorter after VR than non-VR, consistent with greater tumor burden in VR candidates (Figure 2A). When VR was subdivided, OS did not differ between PVR and HAR (P = 0.953), and both were superior to unresectable disease (P < 0.001) (Figure 2B). Additional stratified survival patterns by M/N status and by nodal status combined with pathological vascular invasion are shown in Figure 2C and D.

Figure 2 Overall survival according to resection status and vascular resection subgroups. A: Kaplan-Meier curves comparing overall survival among patients undergoing vascular resection (VR), non-vascular resection, and unresectable cases; B: Overall survival stratified by portal vein resection, hepatic artery resection, and unresectable cases (P values shown); C: Overall survival within the VR cohort stratified by combined distant metastasis status (M)/nodal status (N) categories (VR M0 N0, VR M0 N1-2, and VR M1 N1-2); D: Overall survival of four groups defined by nodal status (N0/N1-2) and vascular invasion (VI) status [A indicates N0/VI (-), B indicates N0/VI (+), C indicates N1-2/VI (-), and D indicates N1-2/VI (+)]. VR: Vascular resection; Non-VR: Non-vascular resection; PVR: Portal vein resection; HAR: Hepatic artery resection; M: Distant metastasis status; N: Nodal status; VI: Vascular invasion.

As a sensitivity analysis, propensity score matching reduced baseline imbalance between PVR and HAR (Supplementary Figure 1A). In the matched cohort (33 pairs), OS did not differ significantly between PVR and HAR (log-rank P = 0.29) (Supplementary Figure 1B). Taken together, these findings suggest that HAR was not associated with significantly inferior OS compared with PVR in this selected VR cohort; however, they should not be interpreted as evidence of formal oncological equivalence between the two procedures.

To assess the potential influence of M1 disease, we performed an M0-only sensitivity analysis after excluding patients finally classified as M1. In the M0-only VR cohort, 48 patients underwent PVR and 59 underwent HAR. OS did not differ significantly between PVR and HAR (log-rank P = 0.776). Cox analysis also showed that HAR was not associated with significantly inferior OS compared with PVR (hazard ratio = 0.94, 95%CI: 0.60-1.47, P = 0.789) (Supplementary Table 3).

Prognostic analysis in VR patients

In the VR cohort, univariable Cox regression identified variables associated with OS (Table 3). In multivariable analysis, T4 stage, nodal metastasis, distant metastasis, poor differentiation, pathological vascular invasion, higher PLR (per 20-unit increase), and longitudinal invasion length > 3 cm were independently associated with worse OS, whereas VR type (HAR vs PVR) remained non-significant.

Table 3 Univariable and multivariable Cox proportional hazards analyses for overall survival in the vascular resection cohort.

Variable	Univariable		Multivariable
	HR (95%CI)	P value	HR (95%CI)	P value
Age
> 70	1.49 (0.95-2.33)	0.079	1.67 (1.01-2.76)	0.047
≤ 70
Sex
Male	1.18 (0.77-1.80)	0.461	NA	NA
Female
Bismuth type
4	3.37 (2.09-5.43)	< 0.001	1.23 (0.69-2.19)	0.476
1, 2, 3
Combined PD
Yes	2.03 (0.88-4.68)	0.096	NA	NA
No
CA19-9
≤ 100	3.43 (2.10-5.59)	< 0.001	0.65 (0.35-1.23)	0.189
> 100
T stage
T4	2.45 (1.31-4.59)	0.005	3.47 (1.71-7.05)	0.001
T1, T2, T3
Lymph node metastasis
N1-2	5.80 (3.64-9.25)	< 0.001	2.56 (1.19-5.52)	0.016
N0
Distant metastasis
M1	34.46 (12.30-96.90)	< 0.001	9.04 (3.11-26.32)	< 0.001
M0
Histologic grade
G1, G2	0.18 (0.09-0.35)	< 0.001	0.33 (0.15-0.73)	0.006
G3
Postoperative adjuvant chemotherapy
Yes	0.51 (0.33-0.78)	0.002	0.86 (0.53-1.40)	0.547
No
Pathologic vascular invasion
Yes	2.99 (1.90-4.70)	< 0.001	1.91 (1.14-3.20)	0.013
No
Pathologic nerve invasion
Yes	4.24 (2.68-6.69)	< 0.001	1.52 (0.85-2.72)	0.162
No
Surgery type
HAR	1.01 (0.66-1.54)	0.962	NA	NA
PVR
Longitudinal tumor invasion length (cm)	11.87 (5.94-23.70)	< 0.001	3.40 (1.41-8.19)	0.007
≤ 3
> 3
NLR	1.60 (1.37-1.87)	< 0.001	1.00 (0.81-1.28)	0.883
PLR (per 20 unit)	2.19 (1.81-2.65)	< 0.001	1.45 (1.12-1.88)	0.004
CEA	1.00 (0.98-1.02)	0.982	NA	NA
ALB	0.92 (0.86-0.99)	0.019	1.05 (0.97-1.15)	0.196

HR: Hazard ratio; CI: Confidence intervals; PD: Pancreatoduodenectomy; CA19-9: Carbohydrate antigen 19-9; PVR: Portal vein resection; HAR: Hepatic artery resection; NLR: Neutrophil-to-lymphocyte ratio; PLR: Platelet-to-lymphocyte ratio; CEA: Carcinoembryonic antigen; ALB: Albumin; NA: Not applicable.

The forest plot of Cox regression is shown in Supplementary Figure 2A. A nomogram based on the multivariable model is shown in Supplementary Figure 2B. Discrimination was higher than UICC 7^th edition tumor-node-metastasis staging (C-index 0.887 vs 0.787; Δ0.099, 95%CI: 0.057-0.141), with consistently higher time-dependent area under the curve (AUC) (Supplementary Figure 2C and Supplementary Table 4). However, these performance estimates were derived from the same retrospective cohort and should be interpreted as internally assessed model performance. Integrated discrimination improvement, net reclassification improvement, and decision-curve analyses are reported in Supplementary Table 4 and Supplementary Figure 2D.

Risk stratification and conditional survival

Landmark analyses at baseline, 12 months, and 24 months demonstrated persistent separation of model-defined risk strata (Supplementary Figure 3; all P < 0.0001). Conditional survival estimates are provided in Supplementary Table 5; conditional prognosis improved with time survived, but risk gradients persisted (conditional 2-year survival among 1-year survivors was 90.8% in the low-risk group vs 23.2% in the high-risk group).

Exploratory analyses: Futility and joint safety-efficacy outcomes

Using predicted 3-year OS, a prespecified futility subgroup (lowest quintile; cut-off 0.80) had markedly shorter 0-36 months RMST than the remainder (9.7 months vs 28.4 months; ΔRMST = 18.6 months, 95%CI: -21.6 to -15.7; P < 0.001) (Supplementary Table 6). Risk-score tertiles showed a clear OS gradient (Supplementary Figure 4A; P < 0.001). Within each risk stratum, HAR vs PVR curves overlapped (Supplementary Figure 4C-E), and stratum-specific hazard ratios were not significant (Supplementary Table 7). As an exploratory descriptive analysis, a four-level survival morbidity classification combining 3-year survival and major morbidity was used to summarize the trade-off between medium-term survival and postoperative morbidity; this classification was not considered a validated survival endpoint. Its distribution showed no difference by VR type (PVR vs HAR; Fisher’s exact P = 0.987; Monte Carlo χ² P = 0.986), and multinomial logistic regression did not identify VR type as a predictor (global likelihood ratio test P = 0.978) (Supplementary Figure 5A). Individualized estimates of 3-year OS difference between hypothetical HAR and PVR were concentrated near zero (median 0.0056; interquartile range: 0.0010-0.0214; 57.8% with |ΔS (36)| < 0.01) (Supplementary Figure 5B).

DISCUSSION

pCCA is a bile duct tumor near the hepatic hilum, known for its aggressive nature[11]. Surgical resection with clear margins (R0) is the only curative option and key to better survival. About 60% of early-stage patients undergo hepatectomy and hilar lymphadenectomy, achieving over 70% R0 resection rates. For the 20% with locally advanced disease, vascular reconstruction is crucial for R0 resection[12-14]. In this study, we assessed the role of VR in advanced pCCA, comparing outcomes of PVR and HAR.

The potential benefits of VR for surgeons are unclear due to technical issues and tumor burden. Our multicenter study indicates that hepatectomy with vascular reconstruction was associated with better observed survival for locally advanced pCCA compared with non-surgical management, despite increased complexity and VR-related complications. VR was not an independent risk factor for outcomes when adjusted for major factors, aligning with the study by Poletto et al[15], which found more complications and higher mortality in VR patients but similar survival to non-VR patients. Thus, in carefully selected patients treated at experienced centers, VR may represent a feasible oncologic surgical strategy rather than an excessively aggressive procedure. Other centers have confirmed these findings[16-18].

There are three main types of VR: Isolated PVR, isolated HAR, and combined PVR and HAR[7]. Currently, very few studies have focused on the safety of HAR owing to the technical difficulty of the procedure and high risk of ischemic liver injury. Yamanaka et al[19] described 10 patients undergoing HAR with or without PVR between 1980 and 1998, and reported one perioperative death. Shimada et al[20] reported outcomes of HAR for pCCA, including cases with concomitant PVR. Among 12 patients undergoing HAR (6 HAR alone and 6 HAR with PVR), one postoperative death occurred after HAR with PVR. The cumulative 3- and 5-year OS rates after HAR (with or without PVR) were 32% and 18%, respectively. The high mortality rate of HAR also occurred in studies at other centers[21]. Some studies suggest that the prognosis of HAR is worse than that of PVR[22,23]. In contrast, a single-center study based on a Japanese population suggested that there was no significant difference in OS between patients who underwent PVR and those who underwent HAR, despite a high risk of ascites and hepatic artery-associated complications[24]. However, most of these reports did not directly compare PVR and HAR within balanced cohorts, leaving uncertainty as to whether HAR is likely to be worthwhile for patients with early invasion into hepatic artery. As a Chinese multicenter investigation, this study identified that among patients who underwent vascular reconstruction, HAR was not associated with significantly inferior short-term outcomes or long-term survival compared with PVR in this selected VR cohort, and this finding remained consistent after propensity score matching. However, these results should not be interpreted as evidence of formal equivalence between HAR and PVR. This conclusion still holds true in the research of other high-level centers[6]. Perioperative risks vary by VR type: HAR was associated with more hepatic artery-associated events, whereas PVR showed a higher incidence of portal vein thrombosis. These differences are due to the unique characteristics of the hepatic artery and portal vein systems. Despite these variations, the predicted differences in 3-year OS and short-term complications are minimal between HAR and PVR. Therefore, for locally advanced pCCA patients, achieving R0 resection should be prioritized over the choice of VR type.

From a surgical decision-making perspective, HAR should be viewed as a selective, margin-directed procedure rather than an indiscriminate extension of resection. It may be considered when arterial involvement prevents R0 clearance, when the involved segment can be safely resected and reconstructed, and when hepatic arterial inflow to the future liver remnant can be preserved or restored. Conversely, HAR should be avoided when R0 resection remains unlikely, arterial reconstruction is unsafe or technically infeasible, future liver remnant volume or function is inadequate, or the patient’s physiological reserve and tumor biology suggest limited benefit from aggressive surgery. Therefore, locally advanced T4 disease alone should not be considered an absolute contraindication; rather, the decision should integrate technical reconstructability, liver reserve, systemic condition, and expected oncological benefit.

Univariable Cox regression in the VR cohort identified advanced histological stage, nodal metastasis, advanced T stage, and possibly elevated NLR/PLR as key predictors of poor outcomes, while VR type (HAR vs PVR) was not significant. Thus, disease extent and tumor biology, rather than VR type itself, appear to drive long-term prognosis. These factors should be considered preoperatively for VR. A prognostic nomogram was developed to estimate postoperative survival risk in patients undergoing VR. Patients with bilateral nodal disease and multifocal vascular invasion may not benefit from aggressive surgery, whereas younger, healthy patients with unilateral arterial encasement and no distant or nodal spread could be ideal for HAR to achieve R0 resection. RMST analysis quantified survival time loss for high-risk patients, providing surgeons with specific insights. This study offers an objective, user-friendly tool to assess VR’s survival benefit for pCCA patients.

Our analysis used landmark and conditional survival to evaluate prognosis changes after complex resection at baseline, 12 months and 24 months post-surgery. Unlike previous studies with static Kaplan-Meier curves[25], we acknowledge that patients surviving 6-12 months post-major hepatectomy with VR are a select group with better conditions. Our results show that prognosis improves over time, which is vital for patient counseling and follow-up planning. A patient who remains alive 2 years after HAR likely has a significantly better prognosis than suggested by baseline survival estimates. Our findings emphasize the importance of managing locally advanced pCCA, highlighting that R0 resection is essential, with VR recommended only if it ensures margin-negative resection. Once R0 is achieved, tumor biology and nodal status, rather than VR type, mainly determine long-term survival. We also created a prognostic model using conditional survival and RMST analysis as a tool for surgeons.

This study’s strengths include its basis in three specialized hepatobiliary centers with comprehensive follow-up data, allowing for an in-depth analysis of VR’s technical and oncological aspects. It is notably the first to conduct a retrospective comparative risk-benefit analysis of PVR vs HAR in a Chinese cohort. The study also improves conventional survival analysis by incorporating conditional survival and a composite futility index for a more thorough evaluation of surgical value. However, several limitations should be acknowledged. First, the sample size was modest, which may have limited the ability to detect small differences between PVR and HAR and may also reflect center-specific expertise. Data on recurrence patterns and quality of life were unavailable, although these are important endpoints in complex oncologic surgery. In addition, the inclusion of a small number of patients finally classified as M1 reflects real-world staging uncertainty but may have affected survival estimates. Although the M0-only sensitivity analysis showed consistent findings, all VR-type comparisons should still be interpreted descriptively rather than as definitive treatment-effect estimates. Because some predictors, such as pathological stage and occult metastasis, are only available postoperatively, the nomogram should not be interpreted as a preoperative selection tool. Second, this cohort did not include patients who received neoadjuvant chemotherapy or chemoradiotherapy; therefore, the potential influence of neoadjuvant therapy on R0 resection, tumor biology, vascular dissection, reconstruction difficulty, and long-term outcomes could not be evaluated. A comparable non-resected control group of patients who required VR but did not undergo resection was also unavailable. Non-surgically managed patients were included only as a descriptive reference because they often had advanced disease, inadequate functional reserve, unreconstructable vascular involvement, or other contraindications to curative-intent surgery. Thus, this study cannot establish a causal survival benefit of VR over non-operative management, and the findings should be interpreted as supporting feasibility and acceptable outcomes in carefully selected patients at experienced centers. Third, although postoperative adjuvant chemotherapy was considered according to a shared MDT-based principle across the participating centers, the specific regimen, timing, and treatment duration were individualized according to postoperative recovery, pathological risk factors, liver function, performance status, patient tolerance, and institutional practice. Therefore, this study was not designed to evaluate the independent survival benefit of adjuvant chemotherapy, and residual treatment-selection bias may exist. Finally, the nomogram was developed and assessed in the same retrospective VR cohort without external validation, and its high C-index and time-dependent AUCs may partly reflect model optimism or overfitting. It should therefore be regarded as an exploratory postoperative prognostic tool. Tumor staging was based on the UICC 7^th edition to maintain consistency with the original multicenter pathological records, and future validation using contemporary UICC 8^th edition staging is warranted.

CONCLUSION

In experienced high-volume centers, VR with reconstruction for pCCA is feasible with acceptable perioperative mortality but increased complexity and morbidity. Among patients who underwent VR, HAR was not associated with significantly higher major morbidity or inferior long-term survival compared with PVR, although hepatic artery-associated complications were more frequent after HAR. These results are descriptive and should not be used for preoperative decision-making; instead, they underscore procedure-tailored postoperative surveillance, particularly after arterial reconstruction.

ACKNOWLEDGEMENTS

We thank all surgeons, nursing staff, and data managers for their contributions to patient care, perioperative management, and data collection in this multicenter study.