Higher lymph node burden predicts greater chemotherapy benefit in resected pancreatic ductal adenocarcinoma: evidence from 22,045 patients

Abstract

Whether lymph node burden modifies chemotherapy effectiveness in pancreatic ductal adenocarcinoma (PDAC) has not been systematically investigated. We analyzed 22,045 surgically treated PDAC patients from the Surveillance, Epidemiology, and End Results database (2004–2022). Restricted cubic spline analysis examined chemotherapy hazard ratios across positive lymph node counts. Cox models estimated survival associations and interaction effects between N stage and chemotherapy on multiplicative and additive scales, with Fine-Gray competing risk models for sensitivity analyses. Additional sensitivity analyses included the exclusion of neoadjuvant therapy recipients, lymph node ratio (LNR) analysis, and stratification by lymph node examination adequacy. Chemotherapy absence increased disease-specific mortality (adjusted HR 1.72; 95% CI, 1.65–1.79), while N1 and N2 disease conferred additional risk (HR 1.54 and 2.05, respectively; both p < 0.001). Chemotherapy benefit increased progressively with nodal burden: hazard ratios for chemotherapy absence versus receipt were 1.52 (95% CI, 1.42–1.63) in N0, 1.80 (95% CI, 1.70–1.92) in N1, and 1.97 (95% CI, 1.83–2.12) in N2 disease. Adjusted 3-year disease-specific survival with versus without chemotherapy was 55.1% versus 39.3% (N0), 36.5% versus 18.2% (N1), and 26.4% versus 8.7% (N2). Notably, N1 patients receiving chemotherapy achieved better outcomes than N0 patients without chemotherapy. Adjusted interaction tests revealed significant multiplicative interactions (N1: 1.19; N2: 1.28) and additive interactions (RERI: N1: 0.65; N2: 1.29). These interaction patterns were validated in sensitivity analyses. Lymph node burden significantly modifies chemotherapy effectiveness in resected PDAC, with higher nodal burden patients deriving disproportionately greater benefit. These findings support risk-stratified treatment strategies incorporating lymph node status.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-36035-x.

Introduction

Pancreatic ductal adenocarcinoma (PDAC) remains one of the most lethal malignancies worldwide, with a 5-year survival rate of approximately 13% despite recent therapeutic advances^1,2. For patients with resectable disease, adjuvant chemotherapy has become the standard of care following pivotal trials including PRODIGE-24³ and ESPAC-4⁴, which established modern combination regimens such as modified FOLFIRINOX and gemcitabine-capecitabine as superior to monotherapy. Despite these advances, substantial heterogeneity persists in treatment response and survival outcomes^3,5, underscoring the critical need to identify factors that predict chemotherapy effectiveness and enable personalized treatment selection.

Lymph node involvement is one of the most powerful pathological prognostic factors in resected PDAC. The American Joint Committee on Cancer (AJCC) 8th edition staging system stratifies nodal disease into N0 (no positive nodes), N1 (1–3 positive nodes), and N2 (≥ 4 positive nodes), with clear prognostic discrimination⁶. While subgroup analyses from PRODIGE-24³ suggested enhanced benefit of modified FOLFIRINOX in node-positive disease, and the recent ESPAC-4⁷ long-term follow-up demonstrated that node-negative patients achieve superior outcomes with combination therapy (59% versus 53% 5-year survival), no studies have systematically investigated whether nodal burden systematically modifies chemotherapy effectiveness. This represents a critical knowledge gap limiting personalized treatment selection, particularly as real-world evidence shows that 40–55% of eligible patients never receive adjuvant therapy due to postoperative complications or disease progression^8,9, and among those who initiate treatment, only 65–68% complete the intended course^10–12.

We therefore conducted a comprehensive population-based analysis using the Surveillance, Epidemiology, and End Results (SEER) database to investigate the interaction between lymph node status and chemotherapy effectiveness in resected PDAC. Specifically, we aimed to quantify whether and to what extent lymph node burden modifies the therapeutic benefit of adjuvant chemotherapy using both multiplicative and additive interaction analyses. This investigation represents the first systematic evaluation of lymph node staging-chemotherapy interactions in a large real-world cohort, with the potential to inform risk-stratified treatment approaches and improve patient selection for adjuvant therapy in resected pancreatic cancer.

Methods

Patients and data collection

This retrospective cohort study analyzed data from the Surveillance, Epidemiology, and End Results (SEER) database (Incidence—SEER Research Data, 17 Registries, Nov 2024 Sub [2000–2022]). Patients with pancreatic cancer were identified using the International Classification of Diseases for Oncology, Third Edition (ICD-O-3) site codes C25.0-C25.9. The study cohort included patients with histologically confirmed pancreatic ductal adenocarcinoma (PDAC; ICD-O-3 morphology codes: 8140, 8500) diagnosed between 2004 and 2022. Exclusion criteria included: (1) age below 18 years; (2) T4 stage disease; (3) distant metastases (M1); (4) unknown tumor size, T stage, N stage, or M stage; (5) missing county-level data; and (6) unknown survival status or follow-up duration less than one month. The final analysis included 22,045 eligible patients (Supplementary Fig. 1). The study protocol was exempted from Institutional Review Board review due to the de-identified, publicly available nature of the SEER database.

Variable collection

Demographic, clinicopathological, and treatment variables were collected across three domains. Patient demographics included age at diagnosis (categorized as 18–44, 45–54, and ≥ 55 years), sex, race (White, Black, Other/Unknown), marital status (Married, Single, Divorced/Widowed/Separated, Unknown), household income (stratified into tertiles: low, middle, and high based on county-level median income), and county type (urban, rural). Clinicopathological characteristics encompassed year of diagnosis (2004–2010, 2011–2017, 2018–2022), primary tumor site (head, body, tail or other/overlapping sites), histological grade (well/moderately differentiated [Grade I/II], poorly differentiated/undifferentiated [Grade III/IV], or unknown), tumor size categories (≤ 20 mm, 21–40 mm, > 40 mm), and N stage classification according to the AJCC 8th edition staging system (N0: no regional lymph node metastasis; N1: 1–3 positive regional lymph nodes; N2: ≥ 4 positive regional lymph nodes). Extrapancreatic extension (EPE) was defined as tumor infiltration beyond the pancreatic parenchyma, as extracted from SEER variables "CS extension (2004–2015)" and "EOD Primary Tumor (2018 +)". Treatment variables included surgery approach (partial pancreatectomy, Whipple procedure, total/extended pancreatectomy), radiotherapy (none, external beam radiation, or other radiation modalities), neoadjuvant therapy (yes/no, identified based on the sequence of systemic therapy relative to surgery), and chemotherapy receipt (yes/no). The primary outcome was disease-specific survival (DSS), defined as the interval from diagnosis to death attributed to pancreatic cancer or last follow-up.

Statistical analysis

Univariable and multivariable Cox proportional hazards regression models were employed to estimate hazard ratios (HRs) with 95% confidence intervals (CIs) for associations between clinical variables and DSS. Multivariable models included the age, sex, race, marital status, household income, county type, primary tumor site, year of diagnosis, grade, tumor size, extrapancreatic extension, N stage, surgery type, chemotherapy, and radiotherapy.

To investigate whether lymph node burden modifies the chemotherapy treatment effect, HRs for chemotherapy across the number of positive lymph nodes as a continuous variable were obtained from the three-knot restricted cubic spline (RCS) Cox model¹³. The knots for the RCS models were strategically placed at the 10th, 50th, and 90th percentiles of the positive lymph node distribution. Subsequently, survival differences between chemotherapy groups stratified by N stage were visualized using Kaplan–Meier curves for unadjusted estimates and direct adjustment methods for covariate-adjusted estimates^14,15.

Interaction effects between N stage and chemotherapy were examined on multiplicative and additive scales^16,17. Multiplicative interaction assesses whether the combined effect differs from the product of individual effects, calculated as HR₁₁/(HR₁₀ × HR₀₁), where HR₁₁ represents patients with both a higher N stage and absence of chemotherapy. Additive interaction, measured by the Relative Excess Risk due to Interaction (RERI = HR₁₁—HR₁₀—HR₀₁ + 1), quantifies whether the combined effect exceeds the sum of individual effects. A positive RERI indicates synergistic harm exceeding additive expectations.

Multiple sensitivity analyses were conducted to test the robustness of our results. First, we repeated the survival and interaction analyses using the Fine and Gray competing risk model, considering non-cancer death as a competing risk. Second, we excluded patients who received neoadjuvant therapy to isolate the effect of adjuvant chemotherapy. Third, we conducted interaction analyses using lymph node ratio (LNR) categories (< 0.1, 0.1–0.2, > 0.2) to account for variability in lymphadenectomy extent. Fourth, we stratified analyses by lymph node examination adequacy using a threshold of 11 examined lymph nodes based on published recommendations¹⁸. Fifth, we performed sex-stratified analyses to examine whether the interaction between lymph node status and chemotherapy benefit differs between female and male patients, given the known sex differences in lymph node metastasis rates.

Baseline characteristics were compared between chemotherapy groups using x² tests for categorical variables and Mann–Whitney U tests for continuous variables. Data were presented as frequencies (percentages) for categorical variables and medians (interquartile ranges) for continuous variables. All analyses were performed using R version 4.4.3. Statistical significance was defined as a two-sided p-value < 0.05. Significance levels are denoted as follows: ***p < 0.001, **p < 0.01, and *p < 0.05.

Results

Patient characteristics

A total of 22,045 patients with resected non-metastatic PDAC were identified from the SEER database (2004–2022), with 16,682 (75.7%) receiving chemotherapy and 5,363 (24.3%) undergoing surgery alone (Table 1). The chemotherapy cohort was younger (median 67 versus 72 years), more frequently married (66.0% versus 54.5%), and had higher household income (38.5% versus 32.6% high-income category) (all p < 0.001). Pathological features revealed extrapancreatic extension in 78.0% of patients, with slightly higher rates in chemotherapy recipients (78.9% versus 75.3%, p < 0.001). Most tumors were pancreatic head lesions (73.7%) measuring 20–40 mm (61.7%). Well to moderately differentiated tumors comprised 55.4% of cases but were less common in the chemotherapy group (53.0% versus 62.8%, p < 0.001). Lymph node involvement occurred in 63.7% of patients, with the chemotherapy group demonstrating higher nodal burden: N1 disease (39.7% versus 36.7%) and N2 disease (25.3% versus 22.9%) compared to surgery alone (p < 0.001). The median number of positive lymph nodes was 1.00 (IQR 0–4) in both groups, though the chemotherapy cohort showed a broader distribution (p < 0.001). Radiation therapy was utilized in 35.4% of chemotherapy patients versus 4.0% of surgery-alone patients (p < 0.001). Neoadjuvant therapy was received by 4,635 patients (21.0%), with 27.7% of chemotherapy recipients and only 0.2% of non-chemotherapy patients having received neoadjuvant treatment (p < 0.001). Baseline characteristics stratified by N stage are presented in Supplementary Table 1. Patients with a higher N stage had larger tumors (tumors > 40 mm: N0 18.3%, N1 22.3%, N2 27.5%), more poorly differentiated tumors (Grade III/IV: N0 23.8%, N1 31.2%, N2 34.9%), and higher rates of extrapancreatic extension (N0 64.4%, N1 83.3%, N2 89.6%) (all p < 0.001). Chemotherapy receipt was higher in the advanced N stage (N0 72.9%, N1 77.1%, N2 77.5%), while neoadjuvant therapy was less common (N0 28.7%, N1 18.3%, N2 14.1%).

Table 1.

Patient characteristics stratified by chemotherapy in pancreatic ductal adenocarcinoma.

Variable	Overall N = 22,0451	Yes N = 16,6821	No N = 5,3631	p-value2
Age (year)	68.00 (60.00, 75.00)	67.00 (59.00, 73.00)	72.00 (64.00, 79.00)	< 0.001
Sex				0.018
Female	10,954 (49.7%)	8,214 (49.2%)	2,740 (51.1%)
Male	11,091 (50.3%)	8,468 (50.8%)	2,623 (48.9%)
Race				0.736
White	18,156 (82.4%)	13,753 (82.4%)	4,403 (82.1%)
Black	2,018 (9.2%)	1,527 (9.2%)	491 (9.2%)
Other/Unknown	1,871 (8.5%)	1,402 (8.4%)	469 (8.7%)
Marital status				< 0.001
Married	13,924 (63.2%)	11,003 (66.0%)	2,921 (54.5%)
Single	2,725 (12.4%)	1,978 (11.9%)	747 (13.9%)
Divorced/Widowed/Separated	4,852 (22.0%)	3,318 (19.9%)	1,534 (28.6%)
Unknown	544 (2.5%)	383 (2.3%)	161 (3.0%)
Household income				< 0.001
Low	5,334 (24.2%)	3,948 (23.7%)	1,386 (25.8%)
Middle	8,544 (38.8%)	6,318 (37.9%)	2,226 (41.5%)
High	8,167 (37.0%)	6,416 (38.5%)	1,751 (32.6%)
County type				0.808
Urban	19,641 (89.1%)	14,858 (89.1%)	4,783 (89.2%)
Rural	2,404 (10.9%)	1,824 (10.9%)	580 (10.8%)
Year of diagnosis				< 0.001
2004–2010	6,106 (27.7%)	4,010 (24.0%)	2,096 (39.1%)
2011–2017	8,497 (38.5%)	6,459 (38.7%)	2,038 (38.0%)
2018–2022	7,442 (33.8%)	6,213 (37.2%)	1,229 (22.9%)
Primary tumor site				< 0.001
Head	16,246 (73.7%)	12,447 (74.6%)	3,799 (70.8%)
Body	1,716 (7.8%)	1,294 (7.8%)	422 (7.9%)
Tail	2,285 (10.4%)	1,692 (10.1%)	593 (11.1%)
Others(overlapping/unknown)	1,798 (8.2%)	1,249 (7.5%)	549 (10.2%)
Grade				< 0.001
Grade I/II	12,214 (55.4%)	8,846 (53.0%)	3,368 (62.8%)
Grade III/IV	6,491 (29.4%)	4,713 (28.3%)	1,778 (33.2%)
Unknown	3,340 (15.2%)	3,123 (18.7%)	217 (4.0%)
Tumor size category (mm)				< 0.001
Tumor size: (0,20]	3,573 (16.2%)	2,523 (15.1%)	1,050 (19.6%)
Tumor size: (20,40]	13,599 (61.7%)	10,531 (63.1%)	3,068 (57.2%)
Tumor size: (40,90 +]	4,873 (22.1%)	3,628 (21.7%)	1,245 (23.2%)
Extension				< 0.001
No extrapancreatic extension	4,850 (22.0%)	3,524 (21.1%)	1,326 (24.7%)
Extrapancreatic extension	17,195 (78.0%)	13,158 (78.9%)	4,037 (75.3%)
Number of positive lymph nodes	1.00 (0.00, 4.00)	1.00 (0.00, 4.00)	1.00 (0.00, 3.00)	< 0.001
N stage				< 0.001
N0	7,997 (36.3%)	5,833 (35.0%)	2,164 (40.4%)
N1	8,595 (39.0%)	6,625 (39.7%)	1,970 (36.7%)
N2	5,453 (24.7%)	4,224 (25.3%)	1,229 (22.9%)
Surgery				< 0.001
Partial pancreatectomy	3,030 (13.7%)	2,233 (13.4%)	797 (14.9%)
Whipple procedure	15,578 (70.7%)	11,935 (71.5%)	3,643 (67.9%)
Total or extended pancreatectomy	3,437 (15.6%)	2,514 (15.1%)	923 (17.2%)
Radiotherapy				< 0.001
No	15,843 (71.9%)	10,702 (64.2%)	5,141 (95.9%)
Beam	6,121 (27.8%)	5,908 (35.4%)	213 (4.0%)
Other Radiation	81 (0.4%)	72 (0.4%)	9 (0.2%)
Neoadjuvant therapy				< 0.001
No	17,410 (79.0%)	12,058 (72.3%)	5,352 (99.8%)
Yes	4,635 (21.0%)	4,624 (27.7%)	11 (0.2%)
Follow-up (months)	20.00 (10.00, 37.00)	21.00 (12.00, 39.00)	13.00 (6.00, 27.00)	< 0.001
Cause of deaths				< 0.001
Alive	5,758 (26.1%)	4,984 (29.9%)	774 (14.4%)
Pancreatic cancer	14,575 (66.1%)	10,626 (63.7%)	3,949 (73.6%)
Other causes	1,712 (7.8%)	1,072 (6.4%)	640 (11.9%)

¹Median (Q1, Q3); n (%). ²Wilcoxon rank sum test; Pearson’s Chi-squared test.

Prognostic factors for disease-specific survival

Multivariable Cox regression identified lymph node status and chemotherapy as the dominant prognostic factors in resected PDAC (Fig. 1). Lymph node involvement showed a dose-dependent effect, with N1 disease increasing mortality risk by 54% (adjusted HR 1.54, 95% CI 1.48–1.61, p < 0.001) and N2 disease more than doubling the risk (adjusted HR 2.05, 95% CI 1.96–2.14, p < 0.001) compared to N0. The absence of chemotherapy increased mortality risk by 72% (adjusted HR 1.72, 95% CI 1.65–1.79, p < 0.001), equivalent to a 42% risk reduction with chemotherapy receipt. These findings remained robust in competing risk analyses accounting for non-cancer mortality, with lymph node status maintaining similar hazard ratios (N1: HR 1.54, 95% CI 1.48–1.61; N2: HR 2.05, 95% CI 1.96–2.14) and absence of chemotherapy showing increased risk (HR 1.72, 95% CI 1.65–1.79, all p < 0.001) (Supplementary Fig. 2).

Fig. 1.

Univariable and multivariable Cox regression analysis for disease-specific survival. Forest plots displaying hazard ratios with 95% confidence intervals. *p < 0.05, **p < 0.01, ***p < 0.001.

Effect of chemotherapy across number of positive lymph nodes

Restricted cubic spline analysis revealed a non-linear relationship between the number of positive lymph nodes and chemotherapy benefit (Fig. 2). The hazard ratio for patients not receiving chemotherapy compared to those receiving chemotherapy progressively increased with higher lymph node burden, demonstrating significant non-linearity (p < 0.001). In the adjusted model, the protective effect of chemotherapy was relatively modest in N0 disease (HR approximately 1.5 for no chemotherapy versus chemotherapy) but intensified substantially as nodal burden increased. At the threshold of 4 positive lymph nodes (marking the transition from N1 to N2 disease), the hazard ratio reached approximately 2.0 and plateaued thereafter, suggesting that chemotherapy benefit was maximized in patients with ≥ 4 positive nodes.

Fig. 2.

Nonlinear relationship between number of positive lymph nodes and chemotherapy benefit based on restricted cubic spline Cox models. Restricted cubic spline curves illustrating the hazard ratios for chemotherapy absence versus receipt across the continuous spectrum of positive lymph node counts. Panel (A) unadjusted hazard ratios. Panel (B) adjusted hazard ratios. The grey shaded areas represent the density distribution of positive lymph node counts.

Interaction effect between N stage and chemotherapy

Survival analysis stratified by lymph node status revealed progressively greater chemotherapy benefit with advancing nodal stage (Fig. 3). The adjusted hazard ratios for chemotherapy absence versus receipt increased with nodal burden: 1.52 (95% CI 1.42–1.63) in N0 disease, 1.80 (95% CI 1.70–1.92) in N1 disease, and 1.97 (95% CI 1.83–2.12) in N2 disease (all p < 0.001). The median disease-specific survival with chemotherapy versus without chemotherapy was 41 months versus 26 months in N0 patients, 26 months versus 13 months in N1 patients, and 21 months versus 10 months in N2 patients. This translated into substantial survival improvements across all N stages. Three-year disease-specific survival with chemotherapy versus without was 55.1% versus 39.3% in N0 patients, 36.5% versus 18.2% in N1 patients, and 26.4% versus 8.7% in N2 patients. These patterns persisted when accounting for competing risks, with cumulative incidence analysis showing three-year death probabilities of 43.3% versus 55.1% in N0 disease, 61.0% versus 74.7% in N1 disease, and 70.7% versus 84.9% in N2 disease for chemotherapy versus non-chemotherapy groups respectively, confirming the progressive magnification of chemotherapy benefit with advancing nodal stage (Supplementary Fig. 3).

Fig. 3.

Survival curves for chemotherapy groups stratified by N stage based on Cox model. Kaplan–Meier and covariate-adjusted survival curves comparing disease-specific survival between patients receiving chemotherapy (green) versus those not receiving chemotherapy (orange) across N stage categories. The left column presents unadjusted survival curves, while the right column displays adjusted survival curves. *p < 0.05, **p < 0.01, ***p < 0.001.

Formal interaction testing confirmed significant effect modification between lymph node status and chemotherapy on both multiplicative and additive scales (Table 2). The multiplicative interaction increased progressively from N1 (1.19, 95% CI 1.09–1.29, p < 0.001) to N2 (1.28, 95% CI 1.17–1.41, p < 0.001), indicating that chemotherapy’s relative benefit was 19% and 28% greater in N1 and N2 disease respectively compared to N0. More importantly, the additive interaction measured by RERI demonstrated substantial excess risk, increasing from 0.65 (95% CI 0.49–0.81) for N1 to 1.29 (95% CI 1.03–1.55) for N2, revealing that the combined adverse effect of nodal involvement and chemotherapy absence exceeded the sum of their individual effects. These interaction patterns remained robust in competing risk analyses accounting for non-cancer mortality (Supplementary Table 2). The multiplicative interactions were 1.12 (95% CI 1.02–1.23, p < 0.05) for N1 and 1.16 (95% CI 1.04–1.31, p < 0.01) for N2, while RERI values of 0.38 and 0.72 respectively confirmed the synergistic relationship between nodal burden and chemotherapy benefit.

Table 2.

Interaction analysis between different chemotherapy and N stage groups based on Cox model.

	Unadjusted interaction analysis
	N1 vs. N0			N2 vs. N0
	N stage: N0	N stage: N1	Effect of N stage within the strata of chemotherapy	N stage: N0	N stage: N1	Effect of N stage within the strata of chemotherapy
	HR (95% CI)	HR (95% CI)	HR (95% CI)	HR (95% CI)	HR (95% CI)	HR (95% CI)
Chemotherapy: Yes	1.00 (Reference)	1.60 (1.53, 1.68)***	1.60 (1.53, 1.68)***	1.00 (Reference)	2.17 (2.06, 2.28)***	2.17 (2.06, 2.28)***
Chemotherapy: No	1.47 (1.38, 1.57)***	3.08 (2.89, 3.27)***	2.09 (1.94, 2.25)***	1.47 (1.38, 1.57)***	4.53 (4.22, 4.87)***	3.08 (2.84, 3.35)***
Effect of chemotherapy within the strata of N stage	1.47 (1.38, 1.57)***	1.92 (1.81, 2.03)***		1.47 (1.38, 1.57)***	2.09 (1.95, 2.24)***
Multiplicative scale	1.30 (1.19, 1.42)***			1.42 (1.29, 1.56)***
RERI	1.00 (0.82, 1.18)***			1.89 (1.59, 2.19)***

	Adjusted interaction analysis
	N1 vs. N0			N2 vs. N0
	N stage: N0	N stage: N1	Effect of N stage within the strata of chemotherapy	N stage: N0	N stage: N1	Effect of N stage within the strata of chemotherapy
	HR (95% CI)	HR (95% CI)	HR (95% CI)	HR (95% CI)	HR (95% CI)	HR (95% CI)
Chemotherapy: Yes	1.00 (Reference)	1.45 (1.38, 1.52)***	1.45 (1.38, 1.52)***	1.00 (Reference)	1.89 (1.79, 1.99)***	1.89 (1.79, 1.99)***
Chemotherapy: No	1.52 (1.42, 1.63)***	2.62 (2.45, 2.80)***	1.72 (1.60, 1.86)***	1.52 (1.42, 1.63)***	3.71 (3.44, 4.01)***	2.42 (2.23, 2.64)***
Effect of chemotherapy within the strata of N stage	1.52 (1.42, 1.63)***	1.80 (1.70, 1.92)***		1.52 (1.42, 1.63)***	1.97 (1.83, 2.12)***
Multiplicative scale	1.19 (1.09, 1.29)***			1.28 (1.17, 1.41)***
RERI	0.65 (0.49, 0.81)***			1.29 (1.03, 1.55)***

Multivariable Cox model adjusted for age, sex, race, marital status, household income, county type, primary tumor site, year of diagnosis, grade, tumor size, extrapancreatic extension, surgery, and radiotherapy. RERI, relative excess risk due to interaction. *** p < 0.001, ** p < 0.01, * p < 0.05.

Effect of different N stage and chemotherapy groups

Combined analysis of hazard ratios for N stage and chemotherapy revealed a striking pattern where chemotherapy effectively downgraded the prognostic impact of nodal involvement (Fig. 4). Most notably, N1 patients receiving chemotherapy (adjusted HR 1.45, 95% CI 1.38–1.52) demonstrated lower mortality risk than N0 patients without chemotherapy (adjusted HR 1.52, 95% CI 1.42–1.63). Even more remarkably, N2 patients receiving chemotherapy (adjusted HR 1.89, 95% CI 1.79–1.99) achieved substantially better outcomes than N1 patients without chemotherapy (adjusted HR 2.62, 95% CI 2.45–2.80, all p < 0.001). This pattern clearly demonstrates that chemotherapy can effectively mitigate the prognostic impact of nodal involvement. The competing risk analysis confirmed these findings, with N1 patients receiving chemotherapy showing comparable risk (adjusted HR 1.42, 95% CI 1.36–1.49) to N0 patients without chemotherapy (adjusted HR 1.37, 95% CI 1.27–1.47), and N2 patients with chemotherapy (adjusted HR 1.82, 95% CI 1.73–1.91) maintaining better outcomes than N1 patients without chemotherapy (adjusted HR 2.17, 95% CI 2.02–2.34) (Supplementary Fig. 4).

Fig. 4.

Combined effects of N stage and chemotherapy on disease-specific survival based on Cox model. Forest plots displaying hazard ratios with 95% confidence intervals for six combined N stage and chemotherapy groups, with N0 patients receiving chemotherapy as the reference category. The left panel shows unadjusted hazard ratios, while the right panel presents adjusted hazard ratios from the multivariable Cox regression model. *p < 0.05, **p < 0.01, ***p < 0.001.

Sensitivity analyses

Sensitivity analyses confirmed the robustness of our findings across multiple analytical approaches. First, after excluding 4,635 neoadjuvant therapy recipients (21.0%), 17,410 patients remained for purely adjuvant analysis. The interaction patterns remained highly consistent: adjusted multiplicative interactions were 1.19 (95% CI 1.09–1.31, p < 0.001) for N1 and 1.28 (95% CI 1.16–1.41, p < 0.001) for N2, with RERI values of 0.60 (95% CI 0.44–0.76) and 1.16 (95% CI 0.91–1.41), respectively (Supplementary Table 3 and Supplementary Fig. 5).

Second, lymph node ratio (LNR) analysis demonstrated consistent interaction patterns. The adjusted hazard ratios for chemotherapy absence progressively increased across LNR strata: 1.57 (95% CI 1.48–1.66) for LNR < 0.1, 1.86 (95% CI 1.71–2.03) for LNR 0.1–0.2, and 1.87 (95% CI 1.75–2.01) for LNR > 0.2. Adjusted multiplicative interactions were 1.19 (95% CI 1.07–1.31, p < 0.001) for LNR 0.1–0.2 and 1.20 (95% CI 1.10–1.30, p < 0.001) for LNR > 0.2, with corresponding RERI values of 0.68 (95% CI 0.46–0.90) and 0.98 (95% CI 0.77–1.19), respectively (Supplementary Table 4 and Supplementary Fig. 6).

Third, stratified analyses by lymph node examination adequacy using a threshold of 11 examined lymph nodes demonstrated that interaction effects were most robustly detected in patients with adequate lymphadenectomy (≥ 11 LNs; adjusted multiplicative interactions: N1: 1.25, 95% CI 1.12–1.39; N2: 1.35, 95% CI 1.20–1.51). The interaction patterns were attenuated but remained statistically significant in patients with inadequate lymph node harvest (< 11 LNs; adjusted RERI, N1:0.49, 95% CI 0.22–0.76; N2:1.36, 95% CI 0.60–2.12) (Supplementary Tables 5–6 and Supplementary Fig. 7).

Fourth, sex-stratified analyses confirmed consistent interaction patterns in both sexes (Supplementary Tables 7–8 and Supplementary Fig. 8). In female patients (n = 10,954), the adjusted RERI values were 0.54 (95% CI 0.32–0.76) for N1 and 0.90 (95% CI 0.56–1.25) for N2 (both p < 0.001). In male patients (n = 11,091), the interaction effects were more pronounced: adjusted multiplicative interactions were 1.26 (95% CI 1.11–1.43) for N1 and 1.47 (95% CI 1.28–1.69) for N2, with corresponding RERI values of 0.80 and 1.72, respectively (all p < 0.001).

Discussion

This population-based analysis of 22,045 patients represents the first comprehensive evaluation of the interaction between chemotherapy and lymph node status in resected pancreatic ductal adenocarcinoma. Our findings demonstrate a dose-dependent relationship between lymph node burden and chemotherapy benefit, with adjusted hazard ratios for chemotherapy absence increasing from 1.52 (95% CI: 1.42–1.63) in N0 disease to 1.80 (95% CI: 1.70–1.92) in N1 disease and 1.97 (95% CI: 1.83–2.12) in N2 disease. Remarkably, chemotherapy effectively mitigated the adverse prognostic impact of nodal involvement—N1 patients receiving chemotherapy achieved superior outcomes compared to untreated N0 patients, while treated N2 patients outperformed untreated N1 patients. The significant multiplicative interactions (N1: 1.19; N2: 1.28) and additive interactions (RERI N1: 0.65; N2: 1.29) provide robust evidence that lymph node status functions as a predictive biomarker for treatment benefit rather than merely a prognostic indicator.

Our findings align with and extend recent subgroup analyses from pivotal clinical trials, providing population-level validation of differential treatment effects by nodal status. The PRODIGE-24 trial’s 5-year results¹⁹ demonstrated that modified FOLFIRINOX maintained superiority across all nodal subgroups, achieving median overall survival of 53.5 versus 35.5 months compared to gemcitabine (HR 0.64, p = 0.003), with benefit preserved even in patients with lymph node ratios exceeding 0.20. Similarly, the ESPAC-4 long-term analysis demonstrated gemcitabine plus capecitabine efficacy in both N0 and N1/N2 disease⁷. Recent studies on localized PDAC have highlighted the importance of adjuvant therapy across disease stages^20–22. These consistent findings across multiple trials support our observation that the relative benefit of systemic therapy paradoxically increases with advancing nodal stage, despite worse absolute prognosis.

The greater relative benefit of chemotherapy in N1/N2 patients likely reflects several factors rather than enhanced chemosensitivity. First, patients with a higher nodal burden have a higher baseline mortality risk, creating a greater opportunity for chemotherapy to impact survival outcomes. Second, the postoperative setting with minimal residual disease provides a favorable environment where systemic therapy can effectively target micrometastatic disease. Patients with more extensive nodal involvement presumably harbor a greater micrometastatic burden that is amenable to chemotherapy. Third, the presence of lymph node metastases indicates a propensity for systemic spread, making adjuvant chemotherapy particularly relevant for addressing occult distant micrometastases. These considerations align with our empirical findings but require validation through future translational research combining genomic profiling and tumor microenvironment analysis.

The demonstration that chemotherapy benefit increases with nodal burden reveals a critical challenge in treatment delivery. Real-world evidence shows that 40–55% of eligible patients never receive adjuvant therapy due to postoperative complications or disease progression^8,9, and among those who initiate treatment, only 65–68% complete the intended course^10–12. Modified FOLFIRINOX, despite superior efficacy, achieves only 67.2% completion versus 79.0% for gemcitabine monotherapy¹⁹. This completion crisis particularly impacts N1/N2 patients who face higher discontinuation rates due to aggressive disease biology, yet paradoxically require treatment most urgently given our finding of significant multiplicative interactions (N1: 1.19; N2: 1.28) that confirm lymph node status as a predictive rather than merely prognostic biomarker. Evidence supports prioritizing completion through adaptive dosing strategies, as maintaining even 62.5% relative dose intensity preserves survival benefit²³, while early discontinuation yields inferior outcomes with a median overall survival of 16 versus 33 months for non-completers versus completers²⁴. We emphasize that treatment decisions must ultimately incorporate randomized evidence, individual patient factors, including performance status and comorbidities, and patient preferences; aggressive chemotherapy is not appropriate for all N1/N2 patients, regardless of individual circumstances. Therefore, our results support immediate implementation of risk-stratified treatment protocols prioritizing completion in N1/N2 disease through enhanced perioperative care, proactive toxicity management, and consideration of neoadjuvant approaches for high-risk patients. Successful chemotherapy delivery can effectively neutralize the adverse prognostic impact of nodal involvement, transforming lymph node status from a marker of poor prognosis to an actionable indicator for treatment intensification.

This study’s strengths include unprecedented statistical power (22,045 patients) for interaction detection, complementary analytical approaches (Cox and Fine-Gray models), and comprehensive interaction testing on multiplicative and additive scales. Multiple sensitivity analyses including exclusion of neoadjuvant therapy recipients, lymph node ratio analysis, and stratification by lymph node examination adequacy confirmed the robustness of our findings. The SEER database enhances external validity across diverse practice settings, capturing real-world treatment patterns and outcomes. However, important limitations warrant consideration. The absence of specific chemotherapy regimen details precludes analysis of differential interactions by treatment type, though the temporal evolution from gemcitabine monotherapy to modern combination regimens likely contributes to the observed effect magnitudes. The concordant interaction patterns across both absolute node counts and LNR analyses support the validity of our conclusions, despite potential concerns regarding stage migration from variable lymphadenectomy extents. The lack of molecular markers, treatment completion data, and quality-of-life metrics limits deeper risk stratification and a more complete understanding of real-world treatment challenges. Notably, interaction analyses within the neoadjuvant therapy subgroup were infeasible, as only 11 of 4,635 neoadjuvant patients (0.2%) did not receive adjuvant chemotherapy. Additionally, unmeasured confounding factors such as performance status and comorbidities may influence treatment selection and outcomes, though our multivariable models adjusted for available demographic and clinical variables to minimize potential bias.

In conclusion, this comprehensive population-based analysis establishes lymph node burden as a critical determinant of chemotherapy effectiveness in resected pancreatic ductal adenocarcinoma, with the magnitude of benefit increasing progressively from N0 through N2 disease. The demonstration that chemotherapy can effectively mitigate or reverse the adverse prognostic impact of nodal involvement challenges current uniform treatment recommendations and supports the immediate implementation of risk-stratified therapeutic approaches that prioritize treatment completion in N1/N2 patients. Future studies should focus on developing and validating clinical decision-support tools that integrate lymph node status with other established prognostic factors to optimize individualized treatment selection for resected PDAC.

Author contributions

Jun Zhou (JZ) contributed to the conception and design of the study, performed data acquisition, and drafted the manuscript. Xiaolin Dou (XD), Wei Wei (WW), and Hongqin Jiang (HJ) contributed to data collection, analysis, and interpretation, and assisted in manuscript revisions. Ling Tang (LT) supervised the study, provided critical intellectual input, and was a major contributor in writing and revising the manuscript. JZ and LT confirm the authenticity of all the raw data. All authors read and approved the final manuscript.

Data availability

The datasets used and/or analyzed for the present study are available from the corresponding author (Ling Tang, [lingtangTL@hotmail.com](mailto:lingtangTL@hotmail.com)) on reasonable request.

Declarations

Competing interests

The authors declare no competing interests.

Ethical approval

Since the SEER database is publicly available, this study was exempt from review by the Ethics Committee of the authors’ hospital.

Patient consent

Written informed consent was waived due to the use of de-identified data from the SEER public database.