Early chemotherapy after endoscopic retrograde cholangiopancreatography and survival in advanced pancreatic cancer with biliary obstruction

Abstract

BACKGROUND

Advanced pancreatic ductal adenocarcinoma is commonly complicated by malignant distal biliary obstruction, for which endoscopic retrograde cholangiopancreatography (ERCP)-guided drainage is used to relieve jaundice and restore treatment eligibility. However, its clinical value may depend on whether drainage enables timely transition to systemic chemotherapy. Evidence evaluating early chemotherapy after ERCP while accounting for time-related bias remains limited.

AIM

To evaluate whether chemotherapy initiation within 30 days after ERCP is associated with landmark overall survival (OS).

METHODS

This single-center retrospective cohort study included patients with unresectable locally advanced or metastatic pancreatic ductal adenocarcinoma and malignant distal biliary obstruction who underwent index ERCP-guided drainage from 2016 to 2025. A 30-day landmark design was used. Inverse probability of treatment weighting-weighted Kaplan-Meier and Cox models evaluated landmark OS. Firth logistic regression and competing-risk analyses explored factors associated with chemotherapy initiation within 30 days.

RESULTS

Among 63 ERCP patients, 61 entered the landmark cohort; 36 initiated chemotherapy within 30 days and 25 did not. Chemotherapy initiation within 30 days was associated with better landmark OS (hazard ratio = 0.36, 95%CI: 0.20-0.67, P = 0.001), with a 365-day restricted mean survival time difference of 120.4 days. Higher baseline bilirubin was associated with lower odds of chemotherapy initiation (odds ratio = 0.29, 95%CI: 0.09-0.74, P = 0.008). The 30-day cumulative incidence of chemotherapy initiation was 72.7% in the low-bilirubin group and 41.4% in the high-bilirubin group; persistently high bilirubin at baseline and day 7 identified the lowest-accessibility subgroup.

CONCLUSION

Early chemotherapy after ERCP was associated with better landmark survival. Baseline and early bilirubin dynamics may help identify patients at risk of bridging failure.

Key Words: Pancreatic ductal adenocarcinoma; Malignant distal biliary obstruction; Endoscopic retrograde cholangiopancreatography; Biliary drainage; Chemotherapy accessibility; Landmark analysis; Competing risk; Overall survival

Core Tip: This study reframed endoscopic biliary drainage in advanced pancreatic cancer as a bridge-to-chemotherapy intervention rather than a purely technical procedure. Using a 30-day landmark design with inverse probability weighting, early chemotherapy after drainage was associated with better survival. Baseline total bilirubin and early post-drainage bilirubin trajectory identified patients less likely to enter systemic therapy, supporting a pathway-oriented view of biliary drainage and early risk stratification after endoscopic intervention.

INTRODUCTION

Pancreatic ductal adenocarcinoma (PDAC) is among the most lethal malignancies of the digestive system, with a 5-year survival rate of approximately 11%[1,2]. By 2030, PDAC is projected to become the second leading cause of cancer-related deaths in the United States[3]. Recent global burden analyses further suggest that the disease burden of pancreatic cancer continues to increase, with metabolic risk-related mortality and disability-adjusted life years showing rising trends over recent decades[4].

Because PDAC often presents insidiously and early symptoms are nonspecific, most patients are diagnosed at a locally advanced, unresectable, or metastatic stage. For these patients, systemic chemotherapy remains the main treatment for prolonging survival. Although systemic regimens have been progressively optimized, access to and completion of anticancer treatment in real-world practice remain substantially constrained by baseline performance status, the severity of biliary obstruction, and treatment-related complications. As a result, prognostic differences in advanced PDAC reflect not only tumor biology but also whether patients can successfully enter subsequent treatment pathways[5].

In advanced PDAC, tumors involving the pancreatic head or distal common bile duct frequently lead to malignant distal biliary obstruction (MDBO). The resulting obstructive jaundice can cause pruritus, increase the risk of cholangitis, and impair hepatic function, thereby worsening quality of life and limiting access to systemic therapy. For patients with advanced PDAC and MDBO, biliary decompression is therefore not merely a symptomatic intervention, but a critical prerequisite for subsequent anticancer treatment. Effective biliary drainage relieves cholestasis and helps create the conditions necessary for chemotherapy initiation, making it an essential component of supportive care and multidisciplinary management[6,7]. In clinical practice, systemic chemotherapy is often delayed until biliary obstruction and hyperbilirubinemia have improved sufficiently to permit safer drug delivery. In addition, cholangitis or stent-related complications may further interrupt or delay treatment, so chemotherapy is generally initiated only after infection is controlled and the patient’s overall condition is considered adequate for systemic therapy[8-11].

For distal MBO, endoscopic retrograde cholangiopancreatography (ERCP) with stent placement remains the most widely used and well-established drainage modality. Recent society guidelines have further reinforced its central role. The 2024 American Society for Gastrointestinal Endoscopy guideline suggests using self-expandable metal stents rather than plastic stents for ERCP-guided drainage of distal malignant biliary obstruction[12]. Nevertheless, ERCP is not always successful. Owing to technical difficulties such as tumor-related papillary distortion, failed cannulation, or inability to traverse the stricture, ERCP fails in approximately 5%-20% of cases[13,14]. When ERCP fails, endoscopic ultrasound-guided biliary drainage is now recommended over percutaneous drainage in appropriate patients to reduce the inconvenience and adverse consequences associated with external drainage[15]. Together, these advances suggest that the goal of contemporary biliary drainage has shifted from simply achieving drainage to establishing durable and effective decompression. In this context, the clinical value of ERCP should not be assessed solely in terms of technical success, but also in terms of whether it facilitates subsequent treatment access.

In real-world practice, however, not all patients who undergo ERCP proceed successfully to chemotherapy. Some remain unable to initiate systemic therapy because of persistent hyperbilirubinemia, uncontrolled infection, post-procedural complications, poor performance status, or high tumor burden[6,8,16]. These patients may remain in an intermediate state in which biliary drainage has technically been achieved, yet entry into anticancer treatment has not occurred. This distinction is clinically important: Successful drainage and successful treatment pathway entry are not synonymous. The former is a procedural outcome, whereas the latter reflects whether the patient has truly gained access to systemic therapy. Therefore, evaluating the real-world effectiveness of the drainage-to-chemotherapy transition, and identifying factors associated with failure of this transition, is highly relevant for treatment decision-making and pathway optimization[6,17,18]. Existing ERCP literature in MDBO has focused predominantly on stent type, stent patency, complications, and reintervention[19-21]. In contrast, quantitative evidence addressing whether ERCP can reliably bridge patients to chemotherapy, a question more directly relevant to treatment continuity in advanced PDAC, remains limited.

From a mechanistic perspective, bilirubin dynamics may play an important role in this bridging process. Persistent hyperbilirubinemia after drainage may indicate inadequate biliary decompression or ongoing disease burden and is often accompanied by infection, malnutrition, systemic inflammation, and impaired treatment tolerance. Accumulating evidence suggests that bilirubin is not merely a biochemical marker of obstruction severity, but may also contribute to clinical stratification. Observational studies in pancreatic disease-related biliary obstruction have shown that higher bilirubin levels are associated with greater short-term risk and worse perioperative outcomes[22-24]. These observations have led to increasing interest in a goal-directed view of biliary drainage, in which success is judged not only by stent deployment itself but also by whether bilirubin falls to a range compatible with subsequent treatment[25,26]. In this context, a bilirubin level of approximately 171 μmol/L (10 mg/dL) has been used in some studies as a clinically interpretable reference threshold[27,28]. Although most of this evidence derives from preoperative or perioperative settings and does not establish a universal standard for advanced PDAC, it nonetheless suggests that bilirubin level and early bilirubin decline may be relevant to subsequent treatment eligibility.

Against this background, the value of ERCP in advanced PDAC with MDBO may be more appropriately understood within a treatment continuity framework rather than a purely procedural one. Two clinically relevant questions therefore arise. First, can patients successfully initiate systemic chemotherapy within a clinically meaningful time window after ERCP? Second, which baseline characteristics and early post-drainage changes influence this transition from drainage to chemotherapy? The former addresses whether patients truly gain access to an effective anticancer pathway, whereas the latter informs early risk identification and may support more timely optimization of supportive care and treatment planning.

Because systemic chemotherapy after biliary drainage is generally considered only after improvement in cholestasis, control of infection, acceptable hepatic function, and adequate general condition, treatment initiation in routine practice usually occurs within days to weeks after ERCP rather than immediately on the day of drainage. A fixed 30-day window therefore provides a clinically interpretable short-term interval for evaluating whether drainage has successfully enabled transition to systemic therapy. This definition captures treatment accessibility during the immediate post-drainage period and also provides the temporal basis for the subsequent landmark analysis[29,30].

At the same time, observational studies examining whether chemotherapy after ERCP is associated with improved survival are particularly vulnerable to immortal time bias. Because chemotherapy is not initiated on the day of ERCP but only after an interval of survival, classifying patients according to whether they eventually received chemotherapy and then comparing survival from the ERCP date can introduce a structural bias: Patients in the chemotherapy group must survive long enough to receive treatment and thus to be classified as exposed[31,32]. This immortal period can spuriously exaggerate the apparent survival benefit of chemotherapy. Recent methodological work has emphasized the importance of aligning eligibility, treatment assignment, time zero, and estimand in observational analyses, ideally following principles of target trial emulation[33-35]. In this setting, using the ERCP date as a common time zero, predefining a 30-day exposure assessment window, and comparing subsequent outcomes within a 30-day landmark framework provide an analytic approach that is better aligned with both clinical care pathways and causal temporal ordering.

Based on this clinical and methodological design choice, the present study sought to characterize ERCP as a bridging intervention in patients with advanced PDAC and MDBO. Specifically, we first evaluated the association between chemotherapy initiation within 30 days after ERCP and overall survival (OS) using a 30-day landmark framework. Second, using chemotherapy initiation within 30 days as an operational definition of early bridging success, we identified baseline factors associated with this outcome. Third, with chemotherapy initiation treated as the primary event and death within 30 days as the competing event, we applied competing risk analysis to examine the relationships of bilirubin stratification and early post-drainage bilirubin dynamics with chemotherapy accessibility. By integrating survival analysis with treatment pathway analysis, this study aimed to provide a clinically grounded assessment of the practical value of ERCP in advanced PDAC with MDBO and to support risk stratification and pathway optimization during the transition from biliary drainage to systemic therapy.

MATERIALS AND METHODS

Study design and ethics

This was a single-center retrospective cohort study conducted at Sun Yat-sen University Cancer Center (Guangzhou, Guangdong Province, China). Patient data were extracted from the institutional electronic medical records system. The enrollment period spanned from January 1, 2016, to June 5, 2025, with June 5, 2025 serving as the database lock and follow-up cutoff. All patients were followed until death or the date of last clinical contact. Patients who were alive at the end of follow-up were censored on that date. The study was conducted in accordance with the Declaration of Helsinki (2024 revision), International Council for Harmonisation Good Clinical Practice (2016), and applicable Chinese regulatory guidelines, and was approved by the Institutional Review Board of Sun Yat-sen University Cancer Center (No. B2025-744-01). Given the retrospective nature of the study, the requirement for individual informed consent was waived.

Study population

Inclusion criteria: (1) Age 18-80 years; (2) Histologically, cytologically, or clinically diagnosed PDAC; (3) Radiologically and clinically assessed as having unresectable locally advanced or metastatic disease (American Joint Committee on Cancer 8^th edition stage III-IV); (4) Concurrent MDBO (involving the intrapancreatic common bile duct or periampullary region) requiring biliary drainage; and (5) Underwent index ERCP-guided biliary drainage with stent placement, with the date of the first ERCP clearly documented.

Exclusion criteria: (1) Age < 18 years or > 80 years; (2) Benign biliary obstruction; (3) Prior biliary drainage before the index event, including ERCP or percutaneous transhepatic cholangial drainage (PTCD); (4) Receipt of any systemic antitumor therapy before the index biliary drainage; (5) Missing or implausible ERCP date; (6) Unavailable survival or last follow-up information precluding outcome ascertainment; and (7) Substantial missing key clinical data that precluded inclusion in the primary analyses.

A separate PTCD cohort was identified for descriptive reference only and was excluded from the primary and secondary analyses.

Exposure

The primary exposure was initiation of any systemic chemotherapy regimen within 30 days of the index ERCP procedure (hereafter “chemotherapy within 30 days”). This threshold was selected to represent a clinically actionable post-drainage treatment window during which recovery of cholestasis, infection control, hepatic function, and general condition are assessed before systemic therapy. It was also consistent with the landmark analytical framework applied in the primary analysis. Patients who did not initiate chemotherapy within this window – whether they subsequently began treatment after day 30 or never received systemic therapy – constituted the comparator group.

Outcomes and covariates

Primary outcome: The primary outcome was landmark OS, defined as the time from day 30 post-ERCP to death from any cause. Patients alive at last follow-up were censored at that date. OS measured from the index ERCP date was additionally computed and reported descriptively, but was not used for formal comparative inference.

Secondary and exploratory outcomes: (1) Bridging accessibility, operationalized as initiation of systemic chemotherapy within 30 days after ERCP and analyzed as an exploratory pathway-related outcome; and (2) Post-drainage clinical complications, including biliary tract infections and biliary re-obstruction, as documented in the medical records.

Covariates

Baseline covariates were measured at or before the index ERCP and were prespecified based on clinical relevance. These included demographic variables (age and sex), tumor-related characteristics (pathological subtype, disease stage, and liver metastases), functional status (Karnofsky Performance Status score), and key laboratory parameters, including total bilirubin, direct bilirubin, albumin, C-reactive protein, white blood cell count, and neutrophil count. Post-drainage laboratory values, particularly total bilirubin reassessed on day 7 after ERCP, were used only for dynamic exploratory stratification and were not treated as baseline confounders in the primary causal analyses.

Procedural variables, including stent type, technical success, clinical success, and post-drainage complications, were summarized descriptively but were not included in the primary propensity score model because the primary model was restricted to covariates measured at or before the index ERCP.

Statistical analysis

All statistical analyses were performed in R (version 4.5.0; R Foundation for Statistical Computing, Vienna, Austria). All tests were two-sided, and P < 0.05 was considered statistically significant. The 30-day landmark OS analysis was considered the primary analysis. Analyses of bridging accessibility, bilirubin-stratified cumulative incidence, and day-7 bilirubin trajectory were exploratory and interpreted as hypothesis-generating.

Primary design (30-day landmark analysis): To mitigate immortal time bias – arising from the fact that patients must survive to their chemotherapy initiation date before being classified as exposed – the primary analysis employed a 30-day landmark design. The landmark cohort comprised all patients who were alive and under active follow-up at day 30 post-ERCP. Patients were then classified according to whether systemic chemotherapy had been initiated within the preceding 30-day window (“chemotherapy within 30 days” vs “no chemotherapy within 30 days”). Time-at-risk in the landmark analysis was measured from day 30 onward; OS from the ERCP date was reported descriptively only and was not used for formal comparative inference.

Propensity score estimation, inverse probability of treatment weighting, and balance diagnostics: To control for baseline confounding, propensity scores for chemotherapy initiation within 30 days were estimated using logistic regression. Firth's penalized likelihood correction was applied when small-sample conditions or quasi-complete separation threatened estimation stability. Stabilized inverse probability of treatment weights (IPTW) targeting the average treatment effect were derived from the estimated propensity scores and were truncated at the 1st and 99th percentiles to limit the influence of extreme weights. Covariate balance before and after weighting was assessed using absolute standardized mean difference (|SMD|), with |SMD| < 0.1 regarded as indicating adequate balance. Additional weight diagnostics included inspection of the stabilized weight distribution and the effective sample size (ESS).

Given the modest sample size, particular attention was paid to the stability of the weighting procedure. We therefore summarized the propensity score distribution, raw and truncated stabilized weights, maximum weights, percentile-based weight ranges, and ESS overall and by treatment group. These diagnostics were used to evaluate whether the weighted analysis was dominated by extreme observations.

Primary effect estimation: Within the 30-day landmark cohort, IPTW-weighted Kaplan-Meier curves were constructed to visualize the survival distributions of the two treatment groups. The primary effect estimate was obtained from an IPTW-weighted Cox proportional hazards model, using robust (sandwich) variance estimation to account for the induced weighting correlation. Results are reported as hazard ratios (HRs) with 95%CIs and two-sided P values. To provide a clinically interpretable, assumption-reduced complement to the HR, the restricted mean survival time difference between groups was estimated at a prespecified truncation horizon of τ = 365 days, with 95%CIs derived by bootstrap resampling.

Robustness and sensitivity analyses: The robustness of the primary findings was assessed through the following pre-specified sensitivity analyses: (1) Unweighted landmark Kaplan-Meier and Cox analysis; (2) Doubly robust (DR) analysis combining IPTW weighting with additional covariate adjustment for total bilirubin and age (DR1), and subsequently for all prespecified baseline covariates (DR2); (3) Common-support restriction, confining the analysis to patients with propensity scores within the overlapping range across both groups (0.223-0.815), followed by IPTW reweighting; (4) Quantitative sensitivity analysis for unmeasured confounding using the E-value framework; and (5) Time-varying exposure Cox model in which chemotherapy was coded as a binary time-dependent covariate (0 before initiation, 1 from the date of initiation onward), serving as a complementary approach to immortal time bias under an alternative modeling framework.

Exploratory accessibility and competing-risk analyses: To identify baseline factors associated with bridging accessibility, a multivariable logistic regression model was fitted with chemotherapy initiation within 30 days as the binary outcome. Predictors were prespecified on clinical grounds, and Firth’s penalized regression was applied when required by sparse data or separation. Odds ratios with 95%CIs were reported.

Because chemotherapy initiation is a time-to-event process subject to the competing risk of death before treatment initiation, the cumulative incidence of chemotherapy initiation was estimated using the cumulative incidence function, with death before chemotherapy treated as the competing event. Patients were stratified by baseline total bilirubin using a prespecified clinically interpretable threshold of 171 μmol/L. Between-group differences were assessed using Gray’s test, and Fine-Gray regression was used to estimate subdistribution HRs (sHRs) with 95%CIs.

In the competing-risk analyses, chemotherapy initiation was the event of interest. Patients were considered to have experienced the event at the date of chemotherapy initiation and were no longer at risk thereafter. Therefore, only death occurring before chemotherapy initiation was treated as the competing event.

To assess whether early post-drainage bilirubin trajectories provided additional information beyond baseline values, a dynamic exploratory stratification was performed among patients who survived to day 7 after ERCP without having initiated chemotherapy. Three bilirubin trajectory groups were defined according to baseline and day-7 total bilirubin values relative to the 171 μmol/L threshold: (1) Persistently low (both time points < 171 μmol/L); (2) Decreasing (baseline ≥ 171 μmol/L but day 7 < 171 μmol/L); and (3) Persistently high (both time points ≥ 171 μmol/L). The cumulative incidence of chemotherapy initiation during days 7-30 was then estimated within each group, and between-group differences were assessed using Gray’s test and Fine-Gray regression, with the persistently low group as the reference.

For descriptive assessment of regimen heterogeneity, first-line chemotherapy regimens among patients initiating chemotherapy within 30 days were categorized as fluoropyrimidine-based, gemcitabine-based, or other regimens. Fluoropyrimidine-based regimens included FOLFIRINOX, SOXIRI, and FOLFOX. Gemcitabine-based regimens included gemcitabine monotherapy, nab-paclitaxel plus gemcitabine, and gemcitabine plus S-1. Oral S-1 monotherapy was reported descriptively as another regimen and was not included in the two-category survival comparison.

Missing data

Complete-case analysis was used for each model. Missingness in key variables was summarized descriptively, and no imputation was performed.

RESULTS

Study population

A total of 75 potentially eligible patients with unresectable PDAC and MDBO after initial biliary drainage were identified. Twelve patients who underwent PTCD were excluded from the primary ERCP-based analyses, leaving 63 patients in the overall ERCP cohort. Of these, two died within 30 days after ERCP and were excluded from the landmark analysis. The remaining 61 patients were alive and under follow-up at day 30 after ERCP and constituted the primary 30-day landmark ERCP cohort. Total 36 patients received systemic chemotherapy within 30 days after ERCP, whereas 25 did not (Figure 1).

Figure 1 Study flowchart. Flowchart of patient screening, exclusion, and cohort construction. Patients with advanced pancreatic ductal adenocarcinoma and malignant distal biliary obstruction were screened. The chart shows the overall endoscopic retrograde cholangiopancreatography (ERCP) cohort and the final 30-day landmark ERCP cohort, which was further classified according to whether chemotherapy was initiated within 30 days after index ERCP. PDAC: Pancreatic ductal adenocarcinoma; MDBO: Malignant distal biliary obstruction; PTCD: Percutaneous transhepatic cholangial drainage; ERCP: Endoscopic retrograde cholangiopancreatography.

Stent type was summarized descriptively in the 30-day landmark ERCP cohort and is presented in Supplementary Table 1. Among the 61 landmark patients, 30 received metal stents and 29 received plastic stents; one patient had uncertain/other stent information and one patient had ERCP failure/no stent recorded. Metal stents were used in 11 of 25 patients in the no-chemotherapy-within-30-days group and in 19 of 36 patients in the chemotherapy-within-30-days group. The distribution of metal vs plastic stents did not differ significantly between the two groups (Fisher’s exact P = 0.435).

Baseline characteristics and weighting diagnostics

Before weighting, baseline differences were evident between the two landmark groups, with the largest imbalance observed for baseline total bilirubin (absolute |SMD| = 0.741) (Table 1). After stabilized inverse probability of treatment weighting (IPTW) with 1%/99% truncation, covariate balance improved across most prespecified baseline variables. Residual imbalance remained mainly for baseline total bilirubin (|SMD| = 0.208) and age (|SMD| = 0.104), whereas the remaining baseline covariates were generally well balanced. Inspection of the propensity score distributions showed improved overlap after weighting, supporting better comparability for the primary survival analysis while also motivating complementary common-support and DR sensitivity analyses (Figure 2).

Figure 2 Propensity score distributions before and after inverse probability of treatment weighting. Density plots showing the propensity score distributions of the no-chemo-by-day-30 group and the chemo-by-day-30 group before weighting and after stabilized inverse probability of treatment weighting with 1% and 99% truncation. It is presented to show the overlap of the propensity score distributions between groups. IPTW: Inverse probability of treatment weighting.

Table 1 Baseline characteristics of the 30-day landmark endoscopic retrograde cholangiopancreatography cohort, median (interquartile range)/n (%).

Baseline characteristics	Unweighted			IPTW-weighted
	No chemo by day 30	Chemo by day 30	|SMD|	No chemo by day 30	Chemo by day 30	|SMD|(IPTW)
Patients (n)	25	36		22.421	35.851
Age (years)	59.37 (10.23)	57.70 (8.86)	0.175	58.33 (8.99)	57.62 (9.23)	0.104
Female	9 (36.0)	17 (47.2)	0.112	9.6 (49.2)	16.4 (45.9)	0.030
Stage IV	14 (56.0)	24 (66.7)	0.107	13.5 (60.2)	21.7(60.4)	0.002
KPS score	86.20 (7.40)	84.44 (10.40)	0.194	84.89 (7.10)	84.28 (9.65)	0.068
Liver metastasis	10 (40.0)	14 (38.9)	0.011	10.2 (45.3)	14.3 (39.9)	0.054
TBil (μmol/L)	214.1 (158.0, 325.6)	147.25 (92.00, 193.85)	0.741	191.30 (147.60, 307.80)	162.60 (112.90, 268.80)	0.208
Albumin (g/L)	39.42 (4.54)	39.86 (4.58)	0.095	39.60 (4.26)	39.71 (4.69)	0.023
CA19-9 (U/mL)	276.00 (51.41, 3232.00)	983.40 (215.12, 3250.50)	0.179	242.50 (51.41, 3232.00)	985.90 (216.00, 4874.00)	0.034
CEA (ng/mL)	8.50 (4.83, 16.00)	3.71 (2.55, 18.03)	0.168	8.50 (4.83, 13.61)	3.65 (2.62, 18.58)	0.060
CRP (mg/L)	9.55 (6.38, 20.38)	10.14 (4.33, 22.56)	0.115	12.64 (6.38, 20.38)	11.14 (6.27, 22.29)	0.074
WBC (× 109/L)	6.75 (2.33)	6.95 (4.07)	0.060	6.61 (2.30)	6.87 (3.76)	0.079
GGT (U/L)	634.60 (367.00, 874.90)	493.25 (211.97, 768.60)	0.138	550.40 (368.20, 909.20)	596.60 (280.90, 1200.70)	0.066
ALT (U/L)	204.00 (116.60, 288.10)	200.30 (98.05, 297.45)	0.092	204.00 (115.60, 288.10)	211.40 (98.10, 296.40)	0.018

¹Weighted patient numbers represent weighted sample sizes rather than actual case counts.

Baseline variables were defined at or before index endoscopic retrograde cholangiopancreatography. Continuous variables are presented as median (interquartile range), as appropriate, and categorical variables are presented as n (%) or weighted percentages. Stabilized inverse probability of treatment weighting with 1% and 99% truncation was used. Covariate balance was assessed using the absolute standardized mean difference. IPTW: Inverse probability of treatment weights; SMD: Standardized mean difference; KPS: Karnofsky Performance Status; TBil: Total bilirubin; CRP: C-reactive protein; CA19-9: Carbohydrate antigen 19-9; CEA: Carcinoembryonic antigen; WBC: White blood cell; GGT: Gamma-glutamyl transferase; ALT: Alanine aminotransferase.

Propensity scores ranged from 0.152 to 0.815 in the no-chemotherapy-within-30-days group and from 0.223 to 0.895 in the chemotherapy-within-30-days group, yielding a common-support range of 0.223 to 0.815. Stabilized weights were truncated at the 1% and 99%; the maximum raw stabilized weight was 2.643, and the maximum trimmed weight was 2.493. The ESS after trimmed IPTW was 51.99 overall, 21.19 in the no-chemotherapy-within-30-days group, and 30.91 in the chemotherapy-within-30-days group. The maximum post-weighting absolute SMD was 0.208 for baseline total bilirubin (Supplementary Table 2).

Association between chemotherapy within 30 days and landmark OS

In the 30-day landmark cohort, initiation of chemotherapy within 30 days after ERCP was associated with better landmark OS. In the IPTW-weighted Cox proportional hazards model, the HR for death was 0.36 (95%CI: 0.20-0.67; P = 0.001) for patients who initiated chemotherapy within 30 days compared with those who did not. The weighted Kaplan-Meier curves showed separation in favor of the chemotherapy-within-30-days group, consistent with the Cox estimate (Figure 3). At 365 days after the landmark, the restricted mean survival time was 274.9 days in the chemotherapy-within-30-days group and 154.5 days in the comparator group, corresponding to an absolute between-group difference of 120.4 days.

Figure 3 Landmark overall survival Kaplan-Meier curves. Kaplan-Meier curves for landmark overall survival in the 30-day landmark endoscopic retrograde cholangiopancreatography (ERCP) cohort, stratified by whether chemotherapy was initiated within 30 days after ERCP. Landmark overall survival was calculated from day 30 after index ERCP. Shaded areas indicate 95%CIs. Numbers at risk are shown below the X-axis.

Because chemotherapy regimen intensity may also influence survival, we performed a supplementary descriptive regimen analysis among patients who initiated chemotherapy within 30 days. First-line regimens included fluoropyrimidine-based regimens in 13 patients, gemcitabine-based regimens in 22 patients, and oral S-1 monotherapy in 1 patient; an exploratory two-category comparison between the fluoropyrimidine-based and gemcitabine-based groups did not show a statistically significant separation in landmark survival curves (log-rank P = 0.21; Supplementary Table 3 and Supplementary Figure 1).

Sensitivity analyses

The primary finding remained directionally consistent across all prespecified sensitivity analyses. In the DR analysis with additional adjustment for baseline total bilirubin and age (DR1), the association remained statistically significant (HR = 0.30, 95%CI: 0.18-0.50; P < 0.001). When all prespecified baseline covariates were further included (DR2), the association persisted (HR = 0.25, 95%CI: 0.13-0.48; P < 0.001). In the common-support-restricted analysis combined with IPTW, the HR was 0.30 (95%CI: 0.16-0.56; P < 0.001), indicating that the primary finding was not materially driven by extrapolation outside the region of propensity score overlap. Quantitative sensitivity analysis using the E-value framework suggested that an unmeasured confounder associated with both chemotherapy initiation and mortality by approximately 4.98-fold each would be required to fully explain away the observed association; an association of 2.36-fold each would be required to move the lower confidence bound to the null (Figure 4).

Figure 4 Robustness and sensitivity forest plot. Forest plot showing hazard ratios and 95%CIs for the association between chemotherapy initiation within 30 days after endoscopic retrograde cholangiopancreatography and landmark overall survival across the primary weighted Cox model and prespecified sensitivity analyses, including doubly robust adjustment and common-support restriction. IPTW: Inverse probability of treatment weights; HR: Hazard ratio; SMD: Standardized mean difference; ESS: Effective sample size; DR: Doubly robust; TBil: Total bilirubin; CRP: C-reactive protein; CS: Common-support.

Bridging accessibility: Baseline determinants of chemotherapy initiation within 30 days

Not all patients in the ERCP cohort successfully transitioned to systemic chemotherapy within 30 days after drainage. To identify baseline factors associated with bridging accessibility, Firth’s penalized logistic regression was fitted with chemotherapy initiation within 30 days as the binary outcome. Among the prespecified covariates, only baseline total bilirubin was independently associated with chemotherapy initiation within 30 days (odds ratio = 0.29, 95%CI: 0.09-0.74; P = 0.008; Table 2). Higher baseline bilirubin was associated with a substantially lower probability of early bridging. The remaining prespecified covariates did not achieve statistical significance in this model.

Table 2 Multivariable model for chemotherapy initiation within 30 days.

Variable	OR (95%CI)	P value
Baseline TBil (μmol/L)	0.28 (0.09-0.73)	0.008
Baseline CRP (mg/L)	1.04 (0.56-1.97)	0.892
KPS/10	0.85 (0.43-1.57)	0.615
Baseline albumin (g/L)	0.98 (0.85-1.11)	0.707
Stage IV	1.80 (0.59-5.67)	0.297

The outcome was chemotherapy initiation within 30 days after endoscopic retrograde cholangiopancreatography. Odds ratios and 95%CIs were estimated using Firth penalized logistic regression. Baseline total bilirubin and C-reactive protein were log-transformed for modeling, and Karnofsky Performance Status score was scaled per 10-point increase. Stage III/Locally advanced disease was used as the reference category for stage. An odds ratio > 1 indicates a higher likelihood of chemotherapy initiation within 30 days, whereas an odds ratio < 1 indicates a lower likelihood. KPS: Karnofsky Performance Status; TBil: Total bilirubin; CRP: C-reactive protein; OR: Odds ratio.

Among patients who did not initiate chemotherapy within 30 days, documented barriers were multifactorial and often overlapping, including delayed biochemical or hepatic recovery, infectious or stent-related barriers, poor clinical condition, and treatment-related decision-making (Supplementary Table 4).

Competing-risk analysis according to baseline bilirubin

Because chemotherapy initiation is a time-to-event process subject to the competing risk of death before treatment initiation, chemotherapy accessibility was further evaluated using competing-risk methods in the overall ERCP cohort. Patients were stratified by baseline total bilirubin using a prespecified clinically interpretable threshold of 171 μmol/L. The 30-day cumulative incidence of chemotherapy initiation was 72.7% in the low-bilirubin group and 41.4% in the high-bilirubin group (Figure 5). In the Fine-Gray model, high baseline bilirubin was associated with a significantly lower subdistribution hazard of chemotherapy initiation (sHR = 0.36, 95%CI: 0.19-0.70; P = 0.003).

Figure 5 Cumulative incidence function of chemotherapy initiation by bilirubin group. Cumulative incidence function curves for chemotherapy initiation within 30 days after endoscopic retrograde cholangiopancreatography according to baseline total bilirubin group defined by the 171 μmol/L threshold. Death before chemotherapy initiation was treated as a competing event. The lower panel shows the day-30 cumulative incidence and the number of patients and chemotherapy events in each group. CIF: Cumulative incidence function; sHR: Subdistribution hazard ratio; ERCP: Endoscopic retrograde cholangiopancreatography; TBIL: Total bilirubin.

Day-7 dynamic bilirubin trajectory analysis

To assess whether early post-drainage bilirubin reassessment provided additional information beyond baseline values, a dynamic competing-risk analysis was performed among patients who survived to day 7 after ERCP without having initiated chemotherapy (n = 54). Three bilirubin trajectory groups were defined according to baseline and day-7 total bilirubin values relative to the 171 μmol/L threshold: (1) Persistently low (both time points < 171 μmol/L; n = 26); (2) Decreasing (baseline ≥ 171 μmol/L but day 7 < 171 μmol/L; n = 19); and (3) Persistently high (both time points ≥ 171 μmol/L; n = 9). The cumulative incidence of chemotherapy initiation during days 7-30 differed significantly across the three groups (Gray’s test, P = 0.040). The persistently high group had the lowest cumulative incidence of chemotherapy initiation over this interval (22.2%). Compared with the persistently low group, the persistently high group had a significantly lower subdistribution hazard of chemotherapy initiation (sHR = 0.22, 95%CI: 0.05-0.89; P = 0.033), whereas the decreasing group showed a nonsignificant reduction (sHR = 0.53, 95%CI: 0.25-1.13; P = 0.101; Figure 6).

Figure 6 Dynamic bilirubin trajectory and day 7-30 chemotherapy accessibility. Cumulative incidence function curves for chemotherapy initiation during days 7-30 after endoscopic retrograde cholangiopancreatography according to dynamic bilirubin trajectory groups defined by baseline and day-7 total bilirubin relative to the 171 μmol/L threshold. Death before chemotherapy initiation was treated as a competing event. The lower panel shows the day-30 cumulative incidence and the number of patients and chemotherapy events in each group. CIF: Cumulative incidence function; sHR: Subdistribution hazard ratio; ERCP: Endoscopic retrograde cholangiopancreatography.

DISCUSSION

In this single-center retrospective cohort of patients with advanced PDAC and MDBO undergoing ERCP, chemotherapy initiation within 30 days after drainage was associated with better landmark OS. This association was observed in the primary IPTW-weighted analysis and remained directionally consistent across prespecified sensitivity analyses. Given the modest sample size and single-center design, these findings should be interpreted as exploratory and hypothesis-generating rather than as definitive evidence that timing itself improves survival[36]. These findings should not be interpreted as definitive evidence that chemotherapy timing alone caused the observed survival difference. Rather, they support an association between timely transition to chemotherapy and subsequent landmark survival under an analytic design intended to reduce time-related bias.

A key interpretive point is that “chemotherapy within 30 days” served different analytic roles in different parts of the study. In the primary survival analysis, it functioned as the exposure definition within a prespecified 30-day landmark framework[31,32]. In contrast, in the accessibility analyses, it was treated as a process-related outcome reflecting whether ERCP was successfully translated into timely access to systemic therapy[17]. This distinction is methodologically important. It clarifies that the landmark survival analysis addressed the association between a clinically relevant treatment-transition strategy and subsequent survival among patients alive at day 30, whereas the logistic and competing-risk analyses addressed which patients were less likely to achieve that transition in the first place[37].

The observed association between chemotherapy initiation within 30 days and longer landmark survival is clinically plausible. In advanced PDAC with MDBO, ERCP is generally performed not as an endpoint in itself, but as a prerequisite for subsequent systemic therapy[18,38]. Patients who can initiate chemotherapy promptly after drainage are likely to represent those in whom bilirubin declines adequately, infection is controlled, hepatic function recovers sufficiently, and performance status remains adequate for oncology referral and treatment delivery[39]. Conversely, failure to initiate chemotherapy within the early post-drainage window may reflect persistent cholestatic burden, procedural or infectious complications, or rapid clinical deterioration[40]. In this sense, timely chemotherapy initiation appears to function as a pragmatic pathway marker of successful transition to systemic therapy rather than as a purely biologic marker of treatment timing alone[3].

At the same time, the findings should not be overinterpreted as definitive evidence that chemotherapy timing alone caused the observed survival difference. Because treatment initiation occurs after ERCP rather than at time zero, naive comparisons from the ERCP date are vulnerable to immortal time bias. The 30-day landmark design was therefore used to address this issue by restricting the primary comparison to patients who were alive and still under follow-up at day 30. IPTW improved balance across most measured baseline covariates, but residual imbalance remained, especially for baseline bilirubin. The consistency of results across common-support, DR, and E-value analyses strengthens confidence in the internal stability of the association, but residual and unmeasured confounding cannot be excluded in this observational setting. Accordingly, the present findings support an association under a design intended to reduce time-related bias, rather than an unqualified causal estimate of chemotherapy timing itself.

Another important contribution of this study is the pathway-oriented evaluation of bilirubin as a determinant of bridging accessibility. Across both the Firth logistic model and the Fine-Gray competing-risk analysis, higher baseline bilirubin was consistently associated with a lower probability of chemotherapy initiation within 30 days. Clinically, this is coherent. Marked hyperbilirubinemia at presentation may reflect more severe biliary obstruction, more impaired hepatic metabolic reserve, slower post-drainage recovery, and a higher likelihood of infection, malnutrition, or systemic inflammation[16,41]. Under such circumstances, even technically successful ERCP may not be sufficient to create a treatment-ready state within a short time window[42]. Baseline bilirubin therefore appears to function not only as a marker of obstruction severity, but also as a practical indicator of downstream treatment accessibility.

In the descriptive review of patients who did not initiate chemotherapy within 30 days, delayed biochemical or hepatic recovery often coexisted with other documented clinical barriers, including biliary infection, stent re-obstruction or dislodgement, poor clinical condition, treatment intolerance, and treatment preference (Supplementary Table 4). Thus, bilirubin should be viewed as a measurable pathway marker of recovery and treatment readiness, rather than as the sole mechanism explaining failure to bridge to systemic therapy.

The day-7 dynamic bilirubin analysis extends this concept by introducing an early post-ERCP reassessment point with potential clinical utility. Baseline bilirubin identifies patients who begin the post-drainage period at a relative disadvantage, but day-7 reassessment further distinguishes those whose bilirubin declines promptly from those with persistently elevated levels and persistently limited treatment accessibility. In particular, patients in the persistently high group had the lowest cumulative incidence of chemotherapy initiation during days 7-30. This suggests that day 7 may serve as a practical checkpoint for pathway management[36]. Patients with persistently high bilirubin at this stage may merit intensified supportive care, re-evaluation of drainage adequacy, closer monitoring for biliary or infectious complications, and earlier multidisciplinary discussion regarding realistic treatment options[40]. The present study does not demonstrate that such pathway interventions improve outcomes, but it does support early bilirubin trajectory as a clinically useful risk-enrichment tool for identifying patients at high risk of bridging failure.

This study has several strengths. First, the primary analysis was anchored to a clearly defined clinical question and implemented using a prespecified landmark design aligned with real-world treatment timing after ERCP. Second, confounding adjustment was performed transparently using IPTW with explicit balance diagnostics and supported by common-support, DR, and E-value analyses. Third, the study did not stop at the survival association alone, but also examined why some patients failed to transition to chemotherapy, thereby linking prognosis with pathway accessibility. This combination of landmark survival analysis and accessibility-oriented pathway analysis reflects a pragmatic view of ERCP in advanced PDAC.

Several limitations should also be acknowledged. First, this was a single-center retrospective study with a modest sample size. Although IPTW, DR analyses, and common-support restriction were used to improve comparability and assess robustness, weighted estimates in a small cohort may remain sensitive to model specification and influential observations, even after weight truncation. Therefore, the present findings should be interpreted as exploratory and hypothesis-generating rather than as definitive evidence of treatment efficacy, and they require validation in larger multicenter cohorts. Second, although the 30-day landmark design improves interpretability by reducing immortal time bias, it necessarily excludes patients who died or deteriorated before the landmark. Accordingly, the primary survival findings apply to patients who survived and remained under follow-up at day 30 after ERCP and should not be directly generalized to the sickest patients in the immediate post-drainage period. The 30-day window was selected as a clinically actionable and methodologically fixed post-drainage interval. Alternative windows, such as 14 days or 45 days, may provide complementary information. However, additional landmark analyses at multiple time points would further reduce the number of eligible patients and yield unstable estimates in this modest cohort. Future larger cohorts should assess whether the observed association is robust across different clinically meaningful treatment windows.

Third, although weighting substantially improved measured covariate balance, residual imbalance remained, particularly for baseline bilirubin. In addition, unmeasured or incompletely captured factors, including comorbidities, socioeconomic status, functional-status trajectory, complication severity, supportive care intensity, treatment preference, and oncologic decision-making, may still have influenced both chemotherapy accessibility and survival. Finally, although we added a descriptive analysis of first-line chemotherapy regimens among patients initiating chemotherapy within 30 days, the small sample size precluded robust adjustment for regimen intensity. Therefore, the observed association between chemotherapy initiation within 30 days and landmark survival should not be interpreted as being attributable to treatment timing alone; regimen selection, treatment tolerance, and downstream supportive care may also have contributed. In addition, although stent type was summarized descriptively and did not show a marked imbalance between groups, stent selection, stent patency, and subsequent stent function may still influence post-drainage recovery and chemotherapy accessibility; these procedural factors warrant further evaluation in larger cohorts.

From a clinical perspective, these findings support viewing ERCP as part of a treatment pathway rather than as a purely procedural endpoint[18,43]. The practical question in advanced PDAC with MDBO is not only whether drainage is technically achieved, but whether drainage leads to a clinically meaningful opportunity for systemic therapy[44,45]. Baseline bilirubin and early bilirubin trajectory may help identify patients who are unlikely to bridge successfully and who may therefore benefit from closer reassessment and pathway optimization. In parallel, emerging patient-derived and organoid-based tumor models may help refine individualized systemic treatment selection and support translational research in precision oncology, although such approaches are beyond the scope of the present pathway-based clinical study[46,47]. Future multicenter studies with larger samples, richer measures of post-procedural recovery, and more explicit treatment-strategy definitions are needed to validate these findings and to test whether pathway-based interventions can improve transition success and, ultimately, patient outcomes.

CONCLUSION

In patients with advanced PDAC and MDBO undergoing ERCP, initiation of chemotherapy within 30 days was associated with better landmark OS. These findings suggest that the clinical value of ERCP in this setting extends beyond biliary decompression to encompass whether drainage successfully enables timely transition to systemic therapy[6,9]. Higher baseline total bilirubin was associated with lower bridging accessibility, and persistently elevated bilirubin at day 7 post-ERCP identified a subgroup at particularly high risk of failing to initiate chemotherapy, refining risk stratification beyond baseline assessment alone. These findings support early identification of patients at high risk of bridging failure and suggest that post-ERCP pathway management should incorporate active reassessment of transition readiness. Given the single-center retrospective design and modest sample size, these findings warrant further validation in larger multicenter prospective studies.

ACKNOWLEDGEMENTS

The authors thank all members of the Department of Endoscopy, Sun Yat-sen University Cancer Center, for their support in clinical data collection and patient care.