Benefiting Subgroup From Adjuvant Chemotherapy in Intrahepatic Cholangiocarcinoma: A Propensity Score‐Matched Study

ABSTRACT

The survival benefit of adjuvant chemotherapy (ACT) following resection of intrahepatic cholangiocarcinoma (iCCA) remains controversial; moreover, the patient group that could benefit most from this approach remains undetermined. We conducted a retrospective cohort study of 423 patients (median age 65 years; 52.2% male; 297 patients who received ACT) who underwent curative intent resection of iCCA between 2021 and 2024. To minimize selection bias, propensity score matching (PSM) at a 2:1 ratio was performed. Overall survival (OS) was the primary end point. Survival analyses were conducted using Kaplan–Meier and Cox regression methods. In the PSM‐matched cohort, ACT significantly improved OS compared to observation alone (hazard ratio [HR] = 0.52, 95% confidence interval [CI]: 0.32–0.82, p = 0.005). Subgroup analyses revealed marked heterogeneity in treatment effect. The most remarkable finding was a strong interaction between treatment and histological subtype (p for interaction = 0.041). Patients with small‐duct type iCCA derived a substantial survival benefit from ACT (HR = 0.33, 95% CI: 0.17–0.63, p = 0.001). In contrast, patients with large‐duct type iCCA did not show a significant OS benefit (HR = 0.89, 95% CI: 0.45–1.76, p = 0.737). In conclusion, postoperative ACT was associated with improved OS after resection of iCCA. Exploratory interaction analyses suggested that the magnitude of this association may differ according to histological subtype, with a more pronounced survival benefit observed in patients with small‐duct type iCCA. These findings should be considered hypothesis‐generating and require confirmation in independent prospective cohorts before histological subtype can be incorporated into treatment‐selection strategies.

1. Introduction

Intrahepatic cholangiocarcinoma (iCCA) is a malignant tumor originating from the biliary epithelial cells of the intrahepatic bile ducts, secondary to and above the level of the second‐order branches. As the second most common primary liver malignancy after hepatocellular carcinoma, the global incidence of iCCA is increasing [1, 2]. Radical surgical resection is the sole potentially curative treatment for iCCA. However, even after achieving an R0 resection, the postoperative recurrence rate remains high (50%–70%), and the 5‐year overall survival (OS) rate is unsatisfactory (25%–40%) [3, 4, 5]. This pattern of high recurrence and poor survival underscores the urgent need to explore effective postoperative adjuvant treatment strategies for iCCA.

Among various adjuvant modalities, systemic chemotherapy has been extensively studied for postoperative use in patients with iCCA because of its established survival benefit in advanced iCCA [6, 7]. Nevertheless, the value of systemic chemotherapy in the adjuvant setting for iCCA remains highly controversial. While the BILCAP and ASCOT trials demonstrated a survival benefit from adjuvant chemotherapy (ACT), iCCA patients in these two trials accounted for only 18.8% and 13.2% of the total patients, respectively [8, 9]. Influential multicenter randomized controlled trials (e.g., the French PRODIGE‐12 trial) also failed to demonstrate a substantial advantage of ACT in the overall population [10]. This controversy likely stems from differences in the study population, such as different origins of cholangiocarcinoma and variations in chemotherapy regimen selection [11].

Moreover, the population benefiting from ACT after radical surgery of iCCA showed significant heterogeneity, which can be attributed to the considerable diversity and complexity of iCCA in its biological behavior and molecular characteristics [12, 13]. iCCA can be classified into two types based on its histological features: small‐duct type and large‐duct type [14, 15]. Recent research suggests that different histological subtypes of iCCA may have distinct recurrence patterns and differential treatment responses [13, 16]. Traditional analyses treating patients as a homogeneous group may mask the true efficacy of chemotherapy for specific subpopulations. Therefore, identifying the subgroups of patients most likely to gain substantial benefits from ACT is crucial for achieving personalized, precision therapy and avoiding unnecessary toxicity for those unlikely to respond.

To systematically evaluate the impact of postoperative ACT on the long‐term survival of iCCA patients, the present study utilized a single‐center, consecutive, retrospective cohort design supported by propensity score matching (PSM), a robust statistical method that can emulate randomized controlled trials by balancing known confounding factors between treatment and control groups. More importantly, by conducting an in‐depth subgroup analysis, this study examined the interaction between various pathological features—including histological subtype, perineural invasion, and differentiation grade—and chemotherapy efficacy. The aim was to precisely delineate the profile of iCCA patients who would benefit from ACT, providing high‐quality real‐world evidence to inform clinical decision‐making.

2. Methods

2.1. Study Design and Patient Cohort

This single‐center, retrospective cohort study involved consecutive patients who underwent curative intent surgical resection for pathologically confirmed iCCA at the First Affiliated Hospital, Zhejiang University School of Medicine, between January 1, 2021 and December 31, 2024. Inclusion criteria were as follows: (a) histopathological diagnosis of iCCA; (b) underwent macroscopic complete resection (R0 or R1); and (c) availability of complete clinical, pathological, and follow‐up data. The following exclusion criteria were considered: (a) relapse or death within 90 days after surgery; (b) presence of distant metastasis at diagnosis; (c) received neoadjuvant therapy; and (d) diagnosis of concurrent or prior other malignancies.

2.2. Data Collection and Variable Definitions

Data were extracted from patients' electronic medical records and a prospectively maintained database. Patient demographic and clinicopathologic variables included the following: age; gender; Eastern Cooperative Oncology Group (ECOG) classification; hepatitis B virus (HBV) infection; cirrhosis; diabetes; hypertension; preoperative levels of carcinoembryonic antigen (CEA), carbohydrate antigen 19–9 (CA19‐9), and cancer antigen 125 (CA125); resection status (R0 [≥ 1 mm], R1 [microscopic involvement]); tumor size; satellite lesions; differentiation grade (well, moderately, or poorly differentiated); perineural invasion; vascular invasion; lymph node metastasis (N0, negative; N1, positive; Nx, not examined); and TNM stage based on the American Joint Committee on Cancer (8th edition). Histological subtype was classified as small‐duct type or large‐duct type primarily according to conventional hematoxylin and eosin (H&E) morphological features, with immunohistochemical findings used as supportive evidence, in accordance with previously established pathological classifications. Small‐duct type iCCA was characterized predominantly by small, relatively uniform tubular or acinar glands lined by cuboidal to low‐columnar tumor cells, generally with little or no intraluminal mucin. In contrast, large‐duct type iCCA was characterized by larger and more irregular glandular structures lined by tall columnar cells, frequently accompanied by conspicuous mucin production and prominent desmoplastic stroma. Immunohistochemical and histochemical findings, including mucin: alcian blue and mucicarmine; S100 calcium‐binding protein P [S100P], N‐cadherin, and neural cell adhesion molecule [NCAM]/(CD56), were used as ancillary features to support the morphological classification rather than as stand‐alone criteria [13, 17]. Cases with indeterminate histological features were excluded. Patients were assigned to the low‐risk group if they showed no (1) elevation of preoperative CA19‐9 level, (2) vascular invasion, and (3) lymph node metastasis prior to surgery. All other patients were included in the high‐risk group.

2.3. Adjuvant Chemotherapy Definition

Treatment with ACT was defined as administration of at least one cycle of curative‐intent systemic chemotherapy after surgical resection. Patients who only received chemotherapy after documented tumor recurrence were explicitly excluded from the ACT group. Chemotherapy regimens were selected based on the experience of each medical group, including capecitabine/S‐1, gemcitabine combined with capecitabine/S‐1, and gemcitabine combined with cisplatin. Systematically calculated relative dose intensity (RDI) data were not available for this retrospective cohort.

2.4. Statistical Analysis

2.4.1. Descriptive Statistics and Survival Analysis

Categorical variables are expressed as frequencies (percentages) and compared using the chi‐square or Fisher's exact test. OS was estimated with Kaplan–Meier survival analysis, and differences between the groups were compared with the log‐rank test.

Univariate Cox proportional hazards regression was performed on the entire pre‐matched cohort (n = 423) to identify variables associated with OS and recurrence‐free survival (RFS). Variables with a p‐value of < 0.05 in the univariate analysis were included in a multivariate Cox regression model by using a stepwise selection method (both forward and backward) to identify independent prognostic factors. Results are expressed as hazard ratio (HR) with 95% confidence interval (CI).

To minimize selection bias during ACT and surgery‐only group comparison, PSM at a 2:1 ratio was performed without replacement by using the nearest neighbor algorithm with a caliper width of 0.02 of the standard deviation of the logit of the propensity score. The propensity score (probability of receiving ACT) was estimated using a logistic regression model. The covariates included all clinically relevant factors and independent prognostic factors identified in the multivariate Cox regression analysis (e.g., age, gender, ECOG status, preoperative CA19‐9 level, tumor size, satellite lesions, resection status, perineural invasion, vascular invasion, and lymph node metastasis), ensuring that the matching process addressed key confounders. Histological subtype was not included in the propensity‐score model because it was pre‐specified as the primary effect‐modifier for testing treatment‐histological‐subtype interaction. Nevertheless, covariate balance for histological subtype was assessed after matching. The balance of covariates between the groups before and after matching was assessed using standardized mean difference (SMD). An SMD of < 0.10 was considered to indicate good balance. A sensitivity analysis using multivariate Cox regression for variables with a SMD > 0.1 was also performed in the matched cohort.

Survival comparison (Kaplan–Meier analysis and log‐rank test) between the ACT and Observation groups was primarily conducted in the PSM‐matched cohort. To identify potential beneficiary populations, subgroup analyses were conducted within the PSM‐matched cohort. We selected several key baseline variables with established clinical and biological importance for exploratory subgroup analyses. These included histological subtype (small‐duct type vs. large‐duct type), TNM stage (IA vs. IB‐III), risk (low risk vs. high risk), and other prespecified factors (e.g., HBV and tumor differentiation grade). A Cox proportional hazards model was fitted that included the treatment variable, the subgroup variable, and their interaction term. The p‐value of the interaction term was the primary criterion for determining whether the treatment effect differed significantly across subgroups; a p‐value for interaction < 0.05 indicated significant effect modification. The results were visually summarized using a forest plot.

All statistical tests were two‐sided, and a p‐value of < 0.05 was considered statistically significant. Statistical analyses were performed using IBM SPSS Statistics (version 31.0.1.0) and R software (version 4.5.1) with “survival” and “jstable” packages.

3. Results

3.1. Patient Characteristics and PSM Results

The study cohort comprised 423 patients with resected iCCA (Figure 1). The distribution of chemotherapy regimens among ACT‐treated patients is summarized in Supplementary Table S1. The median interval from surgery to ACT initiation was 44 days (range, 18–86 days). The median number of completed chemotherapy cycles was 6 (range, 1–8 cycles). The median age was 65 years (range: 31–87 years), and 52.1% (n = 221) of the patients were males. The median follow‐up time for survivors was 29.57 months (range: 4.37–59.47 months). During this period, 110 deaths and 217 recurrence events were recorded.

FIGURE 1.

Flow diagram of iCCA patients undergoing radical surgery. ACT, adjuvant chemotherapy; iCCA, intrahepatic cholangiocarcinoma.

Proportional‐hazards analyses were performed on this pre‐matched full cohort to identify key prognostic covariates for constructing the propensity‐score model. Full Cox results for OS and RFS are provided in Supplementary Tables [Link], [Link], [Link], [Link]. To mitigate treatment‐selection bias, 2:1 propensity‐score matching was performed. After matching, a total of 302 well‐balanced patients were retained (193 in ACT group, 109 in observation group).

Table 1 shows the baseline characteristics of the groups before and after matching, together with corresponding SMD values. Prior to matching, both ACT and Observation groups (n = 297 and 126, respectively) exhibited significant imbalances. Patients in the Observation group were significantly older (p = 0.003), had a poorer performance status (ECOG 1–2: 62.7% vs. 44.8%, p < 0.001), had a smaller tumor size (tumor size ≤ 5 cm: 77.8% vs. 67.7%), and had a lower rate of lymph node metastasis (N1: 8.7% vs. 25.3%, p < 0.001) compared to ACT group patients. These differences indicate a strong clinical selection bias where fitter patients and those with a larger tumor burden or more advanced nodal disease were more likely to be offered ACT.

TABLE 1.

Baseline characteristics of iCCA patients undergoing radical surgery.

Variables	Before matching		p	SMD	After matching		p	SMD
	ACT (n = 297)	Observation (n = 126)			ACT (n = 193)	Observation (n = 109)
Age			0.003	0.324			0.233	0.146
> 65	133 (44.8%)	77 (61.1%)			99 (53.6%)	48 (47.0%)
≤ 65	164 (55.2%)	49 (38.9%)			94 (46.4%)	61 (53.0%)
Gender			0.457	0.089			0.551	0.073
Female	138 (46.5%)	64 (50.8%)			92 (47.7%)	48 (44.0%)
Male	159 (53.5%)	62 (49.2%)			101 (52.3%)	61 (56.0%)
ECOG			< 0.001	0.366			0.189	0.162
0	164 (55.2%)	47 (37.3%)			97 (50.3%)	46 (42.2%)
1–2	133 (44.8%)	79 (62.7%)			96 (49.7%)	63 (57.8%)
CA199			0.071	0.198			0.632	0.058
< 37	148 (49.8%)	75 (59.5%)			110 (57.0%)	59 (54.1%)
≥ 37	149 (50.2%)	51 (40.5%)			83 (43.0%)	50 (45.9%)
Tumor size			0.047	0.257			0.305	0.136
≤ 5	201 (67.7%)	98 (77.8%)			156 (80.8%)	82 (75.2%)
> 5	96 (32.3%)	28 (22.2%)			37 (19.2%)	27 (24.8%)
Satellite lesions			0.224	0.147			1.000	0.014
Absent	272 (91.6%)	120 (95.2%)			183 (94.8%)	103 (94.5%)
Present	25 (8.4%)	6 (4.8%)			10 (5.2%)	6 (5.5%)
Resection status			0.353	0.096			0.706	0.044
R0	290 (97.6%)	121 (96.0%)			189 (97.9%)	106 (97.2%)
R1	7 (2.4%)	5 (4.0%)			4 (2.1%)	3 (2.8%)
Perineural invasion			0.515	0.072			0.542	0.079
Negative	175 (58.9%)	79 (62.7%)			119 (61.7%)	63 (57.8%)
Positive	122 (41.1%)	47 (37.3%)			74 (38.3%)	46 (42.2%)
Vascular invasion			0.135	0.148			0.338	0.125
Negative	199 (67.0%)	94 (74.6%)			147 (76.2%)	77 (70.6%)
Positive	98 (33.0%)	32 (25.4%)			46 (23.8%)	32 (29.4%)
N			< 0.001	0.445			0.467	0.105
0, x	222 (74.7%)	115 (91.3%)			167 (86.5%)	98 (89.9%)
1	75 (25.3%)	11 (8.7%)			26 (13.5%)	11 (10.1%)
Histological typing			0.735	0.058			0.923	0.012
Small duct‐type	182 (61.3%)	75 (59.5%)			114 (59.1%)	65 (59.6%)
Large duct‐type	115 (38.7%)	51 (40.5%)			79 (40.9%)	44 (40.4%)

Abbreviations: ACT, adjuvant chemotherapy; ECOG, eastern cooperative oncology group; iCCA, intrahepatic cholangiocarcinoma; N, lymph node metastasis.

After PSM, 302 well‐matched patients were included (ACT group: n = 193; Observation group: n = 109). Notably, histological subtype, our pre‐specified primary effect‐modifier which was not incorporated into the propensity‐score model, exhibited excellent post‐matching balance (SMD = 0.012). All previously significant imbalances were effectively eliminated. The two groups showed no significant differences in terms of age, ECOG status, tumor size, lymph node status, or any other measured clinicopathological variable (all p > 0.05). Notably, the rate of lymph node metastasis (N1), a critical prognostic factor, was balanced at approximately 12% in both groups after matching.

3.2. Survival Outcomes in the PSM‐Matched Cohorts

Survival data for the pre‐matching cohort are shown in Figure 2A,B. After matching, the mortality rate was lower in the ACT group (35 of 193, 18.1%) than in the Observation group (36 of 109, 33.0%, Figure 2C,D). At a median follow‐up period of 29.6 months, the median OS was not reached. The 2‐year OS rate was 87.9% (95% CI: 83.1%–92.9%) in the ACT group, which was significantly higher than that in the Observation group (73.1%, 95% CI: 64.8%–82.4%). The log‐rank test showed a significant difference in survival curves between the two groups (p = 0.004). Based on the HR of 0.52 (95% CI: 0.32–0.82), the risk of death was reduced by 48% in patients receiving ACT. The median RFS was 33.3 and 21.9 months in the ACT and observation groups, respectively. Although the Cox proportional hazards model indicated a trend toward reduced recurrence risk in the ACT group (HR = 0.75; 95% CI: 0.54–1.05), the difference was not significant (p = 0.097). The 2‐year RFS rates were 57.3% (95% CI: 50.3%–65.2%) and 50.5% (95% CI: 41.7%–61.2%) for the ACT and Observation groups, respectively.

FIGURE 2.

Kaplan–Meier survival curves comparing OS and RFS among patients who underwent adjuvant chemotherapy versus observation before (A, B) and after PSM (C, D). ACT, adjuvant chemotherapy; CI, confidence interval; HR, hazard ratio; OS, Overall survival; PSM, propensity score matching; RFS, Recurrence‐Free Survival.

3.3. Sensitivity Cox Regression Analysis Within PSM‐Matched Cohort

To evaluate the robustness of our primary findings, we conducted a series of sensitivity analyses to examine the association between ACT and survival outcomes across unadjusted, partially adjusted, and fully adjusted models. As shown in Table 2, the HR of RFS remained consistently below 1, shifting marginally from 0.754 (95% CI: 0.540–1.052; p = 0.097) in the unadjusted model to 0.709 (95% CI: 0.504–0.996; p = 0.048) in the fully adjusted model. Notably, while the unadjusted and partially adjusted models did not reach statistical significance, the fully adjusted model demonstrated a statistically significant reduction in recurrence risk. Similarly, OS consistently favored ACT, with the HR transitioning from 0.515 (95% CI: 0.323–0.821; p = 0.005) to 0.456 (95% CI: 0.280–0.744; p = 0.002) across the respective models. The magnitude of change in HRs between the unadjusted and fully adjusted models was minimal (−6.0% for RFS and −11.4% for OS), indicating that the observed survival benefits are not substantially driven by residual confounding. These consistent and statistically significant associations in the fully adjusted models confirm the robustness of ACT in improving both RFS and OS.

TABLE 2.

Sensitivity analysis of adjuvant chemotherapy on recurrence‐free survival and overall survival.

Analysis strategy	Recurrence‐free survival	Overall survival
Unadjusted model	HR = 0.754	HR = 0.515
	95% CI: 0.540–1.052	95% CI: 0.323–0.821
	p = 0.097	p = 0.005
Partially adjusted model (SMD > 0.1 variables only)	HR = 0.723	HR = 0.466
	95% CI: 0.515–1.016	95% CI: 0.287–0.755
	p = 0.062	p = 0.002
Fully adjusted model (All 10 matched variables)	HR = 0.709	HR = 0.456
	95% CI: 0.504–0.996	95% CI: 0.280–0.744
	p = 0.048	p = 0.002
HR change versus unadjusted	−6.0%	−11.4%

Abbreviations: CI, confidence interval; HR, hazard ratio; SMD, standardized mean difference.

3.4. Prespecified Subgroup Analyses and Interaction Testing

Prespecified subgroup analyses were performed to confirm whether ACT provided consistent benefits. The treatment effect varied across all patient subgroups (Figure 3A,B). We observed a statistically significant interaction between the histological subtype and ACT (p = 0.041), indicating that ACT shows varying effects according to the histological subtype. In subgroup analyses, patients with the small‐duct type iCCA achieved a significant survival benefit from ACT, with both improved OS (HR = 0.33, 95% CI: 0.17–0.63, p < 0.001, Figure 4A) and RFS (HR = 0.54, 95% CI: 0.35–0.86, p = 0.007, Figure 4B). In contrast, patients with large‐duct type iCCA exhibited more limited benefits without statistical significance in either OS (HR = 0.89, 95% CI: 0.45–1.76, p = 0.737, Figure 4C) or RFS (HR = 1.09, 95% CI: 0.66–1.81, p = 0.725, Figure 4D).

FIGURE 3.

Forest plot of the treatment effect on OS (A) and RFS (B) in prespecified subgroup analyses. The position of each square represents the point estimate of the treatment effect, and error bars represent 95% CI. The sizes of squares are proportional to the precision of the estimates. ACT, adjuvant chemotherapy; CA125, carbohydrate antigen125; CA199, carbohydrate antigen199; CEA, carcinoembryonic antigen; CI, confidence interval; ECOG, eastern cooperative oncology group; HBV, hepatitis B virus; OS, Overall survival; RFS, Recurrence‐Free Survival.

FIGURE 4.

Kaplan–Meier survival curves comparing OS and RFS among different histological typing subgroups of patients after PSM (small duct‐type A‐B, large duct‐type C‐D). ACT, adjuvant chemotherapy; CI, confidence interval; HR, hazard ratio; OS, Overall survival; PSM, propensity score matching; RFS, Recurrence‐Free Survival.

4. Discussion

Although postoperative ACT is accepted as a routine treatment for iCCA, its efficacy remains debatable [18, 19, 20]. To date, only BILCAP and ASCOT trials have achieved positive results for treating cholangiocarcinoma, suggesting the critical role of postoperative ACT in patients undergoing radical resection of cholangiocarcinoma. However, both studies included all types of cholangiocarcinoma and did not consider the heterogeneity of cholangiocarcinoma subtypes and the impact of surgical diversity on long‐term prognosis [8, 9].

The present study utilized the PSM approach to minimize selection bias and evaluate the role of ACT following resection of iCCA. Our analysis yielded two main findings. First, ACT significantly improved OS in the overall matched cohort. The significant OS benefit associated with ACT aligns with several large‐scale retrospective studies and meta‐analyses, offering supportive real‐world evidence for its application [21, 22]. However, improvement in RFS only showed a nonsignificant trend. Several factors may underlie this discordance. The magnitude of recurrence‐risk reduction was modest, and our matched cohort may lack sufficient statistical power to detect a statistically significant difference for the RFS end point. Moreover, overall survival can be substantially influenced by post‐recurrence salvage treatments. Then, death events tend to be more reliably ascertained in retrospective datasets compared with radiological recurrence events. This observation highlights that the survival advantage of ACT in the present cohort was mainly reflected in overall survival.

Second, and more importantly, the therapeutic benefit of ACT demonstrated marked heterogeneity, wherein patients with small‐duct type iCCA derived the maximum benefit. Given their distinct cells of origin, morphological architectures, immunophenotypes, and molecular characteristics, large‐duct type and small‐duct type iCCA exhibit diverse clinical behaviors and prognoses [16, 23, 24]. This fundamental biological heterogeneity likely underlies their differential responses to systemic therapies, such as chemotherapy, providing a crucial pathological and molecular rationale for explaining disparate treatment outcomes and for advancing personalized, precision medicine. This observation could be explained based on differences in molecular mechanisms. In small‐duct type iCCA, which is frequently characterized by IDH1/2 and BAP1 mutations and FGFR2 fusion, the resultant protein can directly or indirectly modulate DNA damage repair pathways, in which linked molecules are the favorable markers of chemotherapy in BTCs [25, 26, 27]. In contrast, TGF‐β pathway genes were frequently altered in large‐duct type iCCA [15, 16]. The TGF‐β pathway, particularly mediated through SMAD4, contributes to chemoresistance in BTC cells [13, 28]. Besides, this discrepancy may be due to the fundamental differences in the recurrence patterns of these iCCA subtypes. Similar to hepatocellular carcinoma, small‐duct type iCCA preferentially metastasizes through blood circulation because of the abundance of portal veins within the tumor [29]. Therefore, following curative intent surgery, the predominant residual threat is micrometastatic disease within the circulation. ACT, as a systemic treatment, is precisely suited to eradicate such disseminated microscopic foci, which explains the pronounced survival benefit observed in this subgroup. Conversely, large‐duct type carcinomas are more locally aggressive, with a propensity for local invasion, lymphatic spread, and longitudinal growth along bile duct walls [16, 29]. For a disease having locoregional failure as the primary mode of recurrence, the utility of systemic chemotherapy alone may be limited, which explains the lack of observed significant efficacy in this subgroup. Our results suggest that strategies combining more aggressive local therapy (e.g., wider surgical margins or adjuvant radiotherapy) with novel systemic agents may be required for treating large‐duct type iCCA.

Beyond histological subtyping, several molecular and liquid biopsy‐based predictive markers have been reported to guide ACT decisions in resected iCCA, which provide complementary perspectives to our findings. The MINT pathologic risk score, which integrates margin status, perineural invasion, pathologic nodal status, and pathologic tumor size, has emerged as a promising predictive tool for ACT benefit in biliary tract cancers [30]. It demonstrated that patients with resected cholangiocarcinoma with a high‐risk MINT score derived significant survival benefit from ACT (HR 0.60, 95% CI 0.46–0.79, p < 0.001), whereas those with a low‐risk score did not (HR 1.78, 95% CI 1.11–2.86, p = 0.016), suggesting that specific pathologic features can effectively identify candidates most likely to benefit from postoperative chemotherapy. In parallel, circulating tumor DNA (ctDNA) has garnered considerable attention as a non‐invasive tool for minimal residual disease (MRD) detection in the adjuvant setting. Postoperative detection of ctDNA has been shown to predict recurrence earlier than conventional imaging and can identify patients at high risk of relapse who may benefit from intensified adjuvant therapy. The STAMP trial (NCT03079427) evaluated the efficacy of ctDNA for detecting MRD and predicting survival outcomes [31]. The results demonstrated that ctDNA positivity at any postoperative timepoint was significantly associated with poorer clinical outcomes. Notably, patients who converted from ctDNA‐positive to ctDNA‐negative status during ACT exhibited survival outcomes comparable to those who remained persistently ctDNA‐negative. This subgroup of patients converting to ctDNA negativity appears to derive the greatest benefit from ACT.

As reported previously, ACT can improve OS in patients with T1N0M0 Stage 1 iCCA [32]. Interestingly, our data showed that chemotherapy benefit is closely associated with disease stage: a clear benefit was observed in patients with stage IB‐III disease (HR = 0.45, p = 0.006), while only a nonsignificant trend was observed in those with stage IA disease (HR = 0.56, p = 0.160). This accurately illustrates the “absolute benefit” principle in oncology. A more advanced stage implies a higher baseline risk of recurrence, wherein effective systemic therapy can translate into a greater magnitude of survival improvement with apparent statistical significance. Conversely, for patients with very early‐stage (IA) disease, their low baseline risk implies that the same relative risk reduction (HR = 0.56) corresponds to a smaller absolute benefit, the clinical significance of which requires careful consideration. A similar phenomenon was observed in patients with higher tumor burden (tumor burden score [TBS] ≥ 6) or lymph node metastasis, who derived greater survival benefit from ACT; in contrast, patients with TBS < 6 and N0/Nx showed no significant improvement [33]. This finding suggests that TNM stage has important predictive value for treatment efficacy together with its established prognostic value, aiding in the selection of patients who most need and are most likely to benefit from intensive intervention. Therefore, the clinical utility of ACT in unselected stage IA patients warrants careful consideration and should be balanced against potential treatment toxicities.

A critical concern regarding the subgroups discussed here is the lack of statistical evidence to confirm that the relative effect of chemotherapy is truly different from that in their counterpart subgroups. The observed variations are best considered to be hypothesis‐generating trends. In clinical practice, these factors, namely age, performance status, tumor size, and specific pathological risk factors, are already integral to a holistic assessment of a patient's prognosis and fitness for treatment. Our findings suggest that the benefit of ACT, while potentially varying in magnitude across these groups, is broadly applicable, with its most robust and predictable enhancement observed in patients with the small‐duct type histological profile.

This finding provides a critical foundation for developing personalized adjuvant treatment strategies in iCCA. By conducting a rigorous post‐matching subgroup analysis, our study presents novel evidence that this biological dichotomy translates directly into differential sensitivity to systemic chemotherapy. Our findings have direct translational implications for clinical practice. First, these findings suggest that small‐duct type iCCA may represent a subgroup with a greater likelihood of deriving survival benefit from postoperative ACT, even for those with early‐stage disease, as they might represent the population with the greatest potential to gain benefit from the treatment. Second, the current strategy of ACT alone for patients with large‐duct type iCCA may need re‐evaluation. Clinicians should prioritize the histological subtype mentioned in the pathology report during treatment decision‐making. Nevertheless, multidimensional consideration is required while strategizing the appropriate therapy. Although histological subtype is a promising predictive factor, other independent high‐risk features, such as lymph node metastasis, positive resection margin, or TNM stage, retain their prognostic significance. Future clinical guidelines could consider integrating histological subtype into risk stratification models for adjuvant therapy.

The present study has several limitations. First, this was a single‐center retrospective study, and the matched cohort comprised only 302 patients. While PSM effectively balanced observed covariates, the relatively small post‐matching sample size limits the statistical power of subgroup analyses, particularly for interactions. The observed interaction between histological subtype and treatment effect (p for interaction = 0.041) approached but did not robustly exceed the conventional threshold for statistical significance, and the substantial benefit of ACT confined to the small‐duct type subgroup should therefore be interpreted as hypothesis‐generating rather than confirmatory. Second, the absence of an external validation cohort represents an important limitation and it should be prospectively validated in multi‐institutional studies. Third, despite rigorous PSM, residual confounding by unmeasured factors may persist and influence the observed treatment effects, such as chemotherapy regimen selection, patient comorbidities, and molecular tumor characteristics. Finally, the definition of ACT as completion of at least one course of postoperative chemotherapy introduces heterogeneity in treatment intensity and duration, which may have diluted the precision of our effect estimates. These limitations underscore the need for larger, multi‐institutional prospective studies to validate the predictive value of histological subtyping for ACT benefit in resected iCCA.

In summary, in this single‐center retrospective cohort, ACT was associated with improved survival following resection of iCCA. Exploratory subgroup and interaction analyses suggested that histological subtype may modify the magnitude of ACT benefit, with the most pronounced association observed in patients with small‐duct type iCCA. However, these findings should be interpreted cautiously because of the retrospective study design, limited subgroup sample size, potential residual confounding, and absence of external validation. Histological subtype therefore represents a promising candidate for treatment stratification rather than an established predictive biomarker. Larger multi‐institutional and prospective studies are warranted to validate whether histological subtyping can reliably guide postoperative ACT selection in iCCA.

Author Contributions

Yeting Lu: writing – original draft, conceptualization, data curation, formal analysis, validation. Xinsuo Zhuang: writing – original draft, conceptualization, data curation, formal analysis, validation. Huan Luo: data curation, formal analysis, investigation, validation. Fan Gao: investigation, supervision, validation. Jiaqi Yang: investigation, supervision, validation. Yiwen Chen: investigation, supervision, validation. Yan Shen: investigation, supervision, validation. Min Zhang: investigation, validation, supervision. Jian Wu: investigation, validation, supervision. Shunliang Gao: investigation, validation, supervision. Jun Yu: investigation, validation, supervision. Risheng Que: investigation, validation, supervision. Xueli Bai: conceptualization, investigation, methodology, supervision, writing – review and editing.

Funding

This study was financially supported by grants from the National Science and Technology Innovation 2030, Noncommunicable Chronic Diseases‐National Science and Technology Major project (2026ZD0558202), Fundamental and Interdisciplinary Disciplines Breakthrough Plan of the Ministry of Education of China (JYB2025XDXM611), and Clinical Medicine Innovation Team Training Project, Zhejiang University School of Medicine.

Ethics Statement

The study protocol was reviewed and approved by the Research Ethics Committee of the First Affiliated Hospital of Zhejiang University (Approval number: [2025B] IIT Ethics Approval No. 1464). This approval confirms that the study design and data analysis adhere to ethical standards and guidelines. Given the retrospective nature of the research involving the analysis of existing data without compromising patient welfare and rights, the requirement for informed consent was waived by the same ethics committee. The utilization of non‐identified information ensures the protection of patient confidentiality, in accordance with ethical standards for retrospective research.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Table S1: The distribution of chemotherapy regimens among ACT‐treated patients.

Table S2: Univariate Cox regression analyses of the prognostic factors for OS.

Table S3: Multivariate Cox regression analyses of the prognostic factors for OS.

Table S4: Univariate Cox regression analyses of the prognostic factors for RFS.

Table S5: Multivariate Cox regression analyses of the prognostic factors for RFS.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.