In this international, retrospective cohort study, the cohort of Chinese enrolled patients who underwent curative surgery for AA at the National Cancer Center of China, from January 1, 1998, to December 31, 2020 (China cohort). The Ethics Committee of the hospital approved the study. Informed consent was secured from participants, authorizing the utilization for potential research. Additionally, patients with AA between January 1, 2004, and December 31, 2017, in the surveillance, epidemiology and end results program were collected as the Western cohort. In which, 12 registries were enrolled including Alaska Natives, California, Connecticut, Georgia, Hawaii, Iowa, Kentucky, Louisiana, New Jersey, New Mexico, Seattle (Puget Sound), and Utah. The median follow-up time of the China cohort and Western cohort was 35 months (ranges: 1~217 months) and 34 months (ranges: 1~188 months), respectively. Additionally, power analysis was implemented to ascertain the appropriate sample size. The study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines (13). The study was approved by the National Cancer Center of China ethics committee (NCT06435767). Written informed consent for participation was collected from all patients.

Pathologically confirmed AA patients without distant metastasis treated with curative-intent resection were included in the study. According to the exclusion criteria, patients who received neoadjuvant therapy or radiotherapy, as well as those with missing clinical pathological or follow-up information, were excluded. Variables including patient sex, age, pathological characteristics, postoperative complication, hematological data, and treatment modalities were collected from the medical records. Histologic subtype of AA was not routinely determined in the Hospital from 1998 to 2020 and the pathological report did not necessarily include immunohistochemical staining features. Lymphovascular invasion (LVI) was defined by the presence of neoplastic cells within lymphatic spaces and muscular vessels. Pathologic staging (TNM) was classified based on the American Joint Committee on Cancer Staging Manual (AJCC, 8th edition).

In our center, postoperative chemotherapy regimens and doses were planned by oncologists, with adjustments made based on drug toxicities or tumor responses. ACT regimens for participants included 8 3-week cycles of GS (1000 mg/m² of gemcitabine intravenously on day 1; 8 and 60 mg/m² of S1 orally on days 1 to 14), 4 3-week cycles of GX (1000 mg/m² of gemcitabine intravenously on day 1; 8; 15 and 830 mg/m² of capecitabine orally on days 1 to 21), 4 3-week cycles of GEMOX (1000 mg/m² of gemcitabine as a 30-minute intravenous infusion on day 1; 8 and 130 mg/m² of oxaliplatin as a 2-hour intravenous infusion on day 1), 6 3-week cycles of GP (1000 mg/m² of gemcitabine intravenously on day 1; 8 and 75 mg/m² of cisplatin intravenously on days 1 to 3), 6 3-week cycles of XELOX (130 mg/m² of oxaliplatin as a 2-hour intravenous infusion on day 1 and 1000 mg/m² of capecitabine orally on days 1 to 21), 6 3-week cycles of SOX (130 mg/m² of oxaliplatin as a 2-hour intravenous infusion on day 1 and 60 mg/m² of S1 orally twice a day on days 1 to 14), 6 2-week cycles of FOLFOX4 (85 mg/m2 of oxaliplatin intravenously on day 1, 200 mg/m2 of folinic acid as a 2-hour intravenous infusion followed by a 400-mg/m2 bolus of fluorouracil, and a 22-hour intravenous infusion of 600 mg/m2 of fluorouracil), and 6 3-week cycles of single S1 (60 mg/m² of S1 orally twice daily on days 1 to 14).

The primary outcome, overall survival (OS), was defined as the time elapsed from the date of surgery to either the date of death or the last contact. Recurrence-free survival (RFS) was defined as the duration from the surgery date to the occurrence of recurrence, last contact, or death unrelated to recurrent malignancy. Cancer-specific survival (CSS) was defined as the interval between the date of surgery and the date of death due to AA or last contact.

Descriptive statistics were provided as frequencies with percentages for categorical variables and median with interquartile range (IQR) for continuous variables. Factors associated with the use of ACT were identified by steady Poisson regression adjusted by the sandwich variance estimator. We conducted a distribution test and selected parametric and nonparametric statistical methods based on whether they followed a normal distribution. The distribution of continuous variables was examined using the t test or Wilcoxon rank sum teat, as appropriate. Pearson χ² test was used to test the distribution of categorical variables. Estimates of survival rates were calculated using the Kaplan-Meier method, and the distributions of survival were compared between groups by log-rank test. Univariate and multivariate Cox proportional hazard models were applied to find the prognostic factors of long-term survival, including OS, RFS, and CSS. Variables with a univariate P-value < 0.10 were included in the multivariate model to ensure a robust selection of prognostic factors. Correlation coefficient was used to examine the multicollinearity of factors and backward selection was performed. The LODDS was defined by formula:

N⁺ and N^total refer to positive lymph node and total examined lymph node, respectively. LODDS was chosen for its capacity to reflect the overall lymph node burden and to account for node-negative patients. For each specific value of LODDS, a hazard ratio (HR) was calculated by comparing survival between patients with LODDS greater than that number (LODDS > i) and those with fewer than or equal to that number (LODDS ≤ i). Then, the adjusted HR for benefit analysis was calculated using the formula:

cHRi and nHRi refer to the HR (when LODDS = i) in patients treated with chemotherapy and non-chemotherapy, respectively. Adjusted HR for each LODDS was summarized in Supplementary Table . For ease of understanding, taking LODDS = -1 as an example, among patients who received chemotherapy, the cHR for those with LODDS greater than -1 compared to those with LODDS less than or equal to -1 was 0.4. In contrast, among patients who did not receive chemotherapy, the nHR for those with LODDS greater than -1 compared to those with LODDS less than or equal to -1 was 1.1. Therefore, at this point, the adjusted HR is calculated as 0.4/1.1 = 0.36. LOESS regression was used to establish a smooth model and the Spearman test was performed to identify the correlation across each point. Significance levels were defined as two-sided P < 0.05. All statistical analyses were conducted using R software, version 4.2.0 (R for Windows) and IBM SPSS software, version 25.0 (IBM Corp). Data analysis was conducted from February 1, 2024, to April 18, 2024.

In the China cohort, a total of 323 patients with ampullary adenocarcinoma undergoing curative resection were screened in the studied period. After the exclusion of patients with other pathological types (n = 9), missing pathological data (n = 10), adjuvant therapy other than ACT (n = 7), missing hematological data (n = 62), and incomplete follow-up information (n = 10), 225 patients (mean [SD] age, 56.8 [10.2] years, 132 males [58.7%]) met the inclusion criteria ( Figure 1A ). Compared with patients without ACT, patients treated with ACT were more likely to be younger (<65 years, 169 [75.1%] vs 56 [24.9%]; relative risk [RR], 4.67, 3.55, and 3.36 for <45, 45-55, and 55–65 years, respectively), with higher T stage (T3-4, 122 [54.2%] vs 103 [45.8%]; RR, 2.53; 95% CI, 1.39-4.61), with positive lymph node (70 [31.1%] vs 155 [68.9%]; RR, 2.08 and 2.77 for N1 and N2, respectively), poor differentiated (90 [40.0%] vs 135 [60.0%]; RR, 2.10; 95%CI, 1.26-3.49), and with LVI (59 [26.2%] vs 166 [73.8%]; RR, 2.19; 95%CI, 1.35-3.56) ( Table 1 ).

In the Western cohort, a total of 8128 patients with ampullary adenocarcinoma were identified from 2000 to 2019. After excluding patients with other pathological types (n = 2816), non-curative surgery (n = 1069), adjuvant therapy other than ACT (n = 348), distant metastasis (n = 314), and missing pathological data and follow-up information (n = 2633), 948 patients (mean [SD] age, 65.4 [11.0] years, 546 males [57.6%]) ( Figure 1B ). The major participants in the cohort tended to be elder (≥65 years, 523 [55.2%]), staged T3 (257 [27.1%]), staged N0 (457 [48.2%]), sized 2 to 4 cm (451 [47.6%]), moderately differentiated (521 [55.0%]), the White (715 [75.4%]), receiving no chemotherapy (610 [64.3%]), and married (591 [62.3%]) ( Supplementary Table 1 ).

In the China cohort, the 5-year OS rates were 48.1% and 46.3% for the ACT group and non-ACT group, respectively (ACT HR, 0.96, 95% CI, 0.60 - 1.54, P = 0.86) ( Supplementary Figure 1A ). There was also no significant difference in RFS between the two groups (5-year RFS rate, 44.2% vs 45.8%; ACT HR: 1.05, 95% CI: 0.66 - 1.69, P = 0.84) ( Supplementary Figure 1B ). The LODDS and N classification were identified as multicollinearities, and the latter was excluded from the Cox regression model. Independent prognostic factors for decreased OS in patients with AA after curative resection were T stage (T3–4 vs T1-2, HR: 1.85, 95% CI, 1.21-2.81, P = 0.004) and increasing LODDS (HR: 1.51, 95% CI, 1.20-1.91, P < 0.001) ( Table 2 ). Risk factors of decreased RFS were also identified as T stage (T3–4 vs T1-2, HR: 1.73, 95% CI, 1.13-2.64, P = 0.011) and increasing LODDS (HR: 1.55, 95% CI, 1.22-1.96, P < 0.001) ( Supplementary Table 2 ).

In the Western cohort, no significant difference in OS was found between ACT and non-ACT groups (5-year OS rate, 40.9% vs 43.0%; ACT HR: 0.87, 95% CI: 0.73 - 1.03, P = 0.11) ( Supplementary Figure 1C ). The 5-year CSS rates were 46.0% and 53.1% for the two groups, respectively (ACT HR, 1.02, 95% CI, 0.84 - 1.24, P = 0.82) ( Supplementary Figure 1D ). The N classification was excluded from the Cox regression model due to multicollinearity. Independent predictive factors for decreased OS included ≥65 years (P = 0.014), T classification (T3–4 vs T1-2, P < 0.001), grade (III-IV vs I-II, P = 0.008), and increasing LODDS (P < 0.001), while factors for increasing OS were the White (P = 0.004) and annual income > 75,000$ (P = 0.041) ( Supplementary Table 3 ). The prognostic factors for CSS were also identified using multivariate Cox regression, as shown in Supplementary Table 4 .

To determine the optimal LODDS for identifying benefits from ACT, stratified analyses were performed at each LODDS value and the adjusted HRs were calculated (range: 1.10 - 0.30). Spearman test indicated a nonlinear correlation between LODDS and adjusted HR (R² = 0.012, P = 0.41). After a smooth trendline was applied using the LOESS regression, we recognized that as the value of LODDS increased, the HR gradually decreased. However, once LODDS reached -1.4, the trend became stable, suggesting that -1.4 was the optimal LODDS for patient stratification ( Figure 2 ). Then, the patients were divided into the high-risk group (LODDS>-1.4) and the low-risk group (LODDS≤-1.4). Pearson χ² test was used to examine the ACT regimen-related bias, indicating that no significant difference was identified in the distributions of gemcitabine-based and fluorouracil/other-based regimens in the two groups (P = 0.40, χ² = .72) ( Supplementary Table 5 ). We also demonstrated baseline comparison between the LODDS > -1.4 and LODDS ≤ -1.4 groups in Supplementary Table 6 . Additionally, the distributions of LODDS were balanced in the two regimen groups as well (fluorouracil/other-based vs gemcitabine-based, mean [standard deviation, SD], -1.60 [0.92] vs -1.54 [0.81], P = 0.81). To obtain a comprehensive view of the characteristics, we depicted the clinicopathologic heatmap of AA patients in the China cohort, which was stratified by LODDS ( Figure 3 ).

In AA patients with LODDS exceeding -1.4, the survival distribution stratified by the use of ACT revealed a significant difference for OS in the China cohort (5-year OS rate, 49.2% vs 16.2%, ACT HR, 0.40, 95% CI, 0.20 - 0.81, P = 0.01) and the Western cohort (5-year OS rate, 29.8% vs 20.8%, ACT HR, 0.61, 95% CI, 0.50 - 0.75, P < 0.001) ( Figures 4A, C ). Further analysis indicated that a potential difference was found in RFS of China cohort (P = 0.09), while a significant difference emerged in CSS of the Western cohort (P < 0.001) ( Supplementary Figures 2A, C ). Regarding patients with LODDS no more than -1.4, the analysis demonstrated that there was no significant difference between ACT and non-ACT groups for OS (P = 0.44) ( Figure 4B ) and RFS (P = 0.50) ( Supplementary Figure 2B ) in the China cohort, and OS (P = 0.19) ( Figure 4D ) and CSS (P = 0.92) ( Supplementary Figure 2D ) in the Western cohort.

Stratified analysis of LODDS based on a multivariate Cox regression model was adjusted by pathological variables including LVI, grade, and the T classification. Interaction analysis for OS indicated that the differential benefit of ACT in the two LODDS groups was significant (interaction P = 0.033), suggesting patients with LODDS exceeding -1.4 favored ACT after surgery (ACT HR, 0.37, 95% CI, 0.19 - 0.75, P = 0.006). The results indicate that the precision of ACT application could be enhanced, particularly in resource-constrained settings, by incorporating LODDS into standard pathological assessments. In the subgroup analysis of LVI and grade, there was no significant difference in prognosis between the two groups (all P >0.05). Although a significant difference in OS was found between the two groups in patients with T3–4 stages (HR, 0.55, 95% CI, 0.31 - 0.98, P = 0.042), the interaction analysis showed no significance (interaction P = 0.531) ( Figure 5 ). Similarly, the multivariate analysis of the N classification adjusted by LVI, grade, and the T classification was further performed. Compared with surgery alone, patients with N1–2 stages (P = 0.018) or T3–4 stages (P = 0.037) tended to have improved OS by using ACT. However, the interaction test demonstrated that the differential benefit was insignificant for the N classification (interaction P = 0.200) and the T classification (interaction P = 0.560) ( Supplementary Figure 3 ). Patients stratified by LODDS demonstrated a more significant survival benefit from ACT compared to those stratified by N-staging.