Preoperative mFOLFIRINOX versus PAXG for stage I-III resectable and borderline resectable pancreatic ductal adenocarcinoma (PACT-21 CASSANDRA): results of the first randomisation analysis of a randomised, open-label, 2 × 2 factorial phase 3 trial.

# Preoperative mFOLFIRINOX versus PAXG for stage I–III resectable and borderline resectable pancreatic ductal adenocarcinoma (PACT-21 CASSANDRA): results of the first randomisation analysis of a randomised, open-label, 2 × 2 factorial phase 3 trial *The Lancet 2025* Articles Preoperative mFOLFIRINOX versus PAXG for stage I–III resectable and borderline resectable pancreatic ductal adenocarcinoma (PACT-21 CASSANDRA): results of the first randomisation analysis of a randomised, open-label, 2 × 2 factorial phase 3 trial Michele Reni, Marina Macchini, Giulia Orsi, Letizia Procaccio, Giuseppe Malleo, Catia Carconi, Ilario Giovanni Rapposelli, Katia Bencardino, Mario Scartozzi, Gianpaolo Balzano, Domenico Tamburrino, Barbara Merelli, Elisa Sperti, Giulio Belfiori, Nicole Liscia, Silvia Bozzarelli, Mariacristina Di Marco, Emiliano Tamburini, Michele Milella, Sara Lonardi, Giorgio Ercolani, Michele Mazzola, Diego Palumbo, Valter Torri, Massimo Falconi Summary Background Perioperative chemotherapy is a standard option for treatment of patients with resectable and borderline Lancet 2025; 406:2945–56 resectable pancreatic ductal adenocarcinoma (PDAC). This study aimed to assess the superiority of PAXG (cisplatin, nab- Published Online paclitaxel, capecitabine, and gemcitabine) over mFOLFIRINOX (modified fluorouracil, leucovorin, irinotecan, and November 20, 2025 oxaliplatin) in this population. https://doi.org/10.1016/ S0140-6736(25)01685-X See Comment page 2868 Methods CASSANDRA is a randomised, open-label, 2 × 2 factorial phase 3 trial, involving 17 Italian academic Department of Medical hospitals. Eligible patients were aged 18–75 years with pathologically confirmed resectable or borderline resectable Oncology, Pancreas PDAC. Randomisation was performed by a central web-based system using R-code lists with a computerised Translational and Clinical algorithm. The design adopted a 1:1 randomisation, with a block stratification by centre and carbohydrate antigen 19-9. Research Center, IRCCS San Participants were first randomly assigned PAXG (total daily capecitabine dose of 1250 mg/m² in a 625 mg/m² twice Raffaele Scientific Institute, Milan, Italy (M Reni MD, a day dosage and intravenous cisplatin 30 mg/m², nab-paclitaxel 150 mg/m², and gemcitabine 800 mg/m² every M Macchini MD, G Orsi MD, 14 days) or mFOLFIRINOX (intravenous fluorouracil 2400 mg/m², leucovorin 400 mg/m², irinotecan 150 mg/m², C Carconi MD, N Liscia MD); Vita- and oxaliplatin 85 mg/m² every 14 days) for 4 months, followed by a second randomisation to 2 months of additional Salute San Raffaele University, chemotherapy either before or after surgery. The primary endpoint was event-free survival (EFS) in the intention-to- Milan, Italy (M Reni, C Carconi, D Tamburrino MD, treat population and the safety population included all patients who received at least one cycle of the assigned D Palumbo MD, therapy. The results of the first randomisation are reported here. The trial, registered on ClinicalTrials.gov Prof M Falconi MD); Oncology 1 (NCT04793932) and EudraCT (2020-003080-26 and 2024-519031-42-00), completed accrual and reached the necessary Unit, Veneto Institute of events for first randomisation primary analysis but follow-up of overall survival is ongoing. Oncology IOV—IRCCS, Padua, Italy (L Procaccio MD, S Lonardi MD); Pancreatic Findings Between Nov 3, 2020, and April 24, 2024, 132 eligible patients were assigned to PAXG and 128 to mFOLFIRINOX. Surgery Unit, Pancreas In the PAXG group, the median age was 65 years (IQR 60–70), 68 (52%) of 132 patients were female, and 64 (48%) were Institute, Azienda Ospedaliera male. In the mFOLFIRNOX group, the median age was 63 years (IQR 57–69), 62 (48%) of 128 patients were female, and Universitaria Integrata Verona, Verona, Italy (G Malleo MD); 66 (52%) were male. All 260 patients received at least one assigned chemotherapy administration. PAXG prolonged the Department of Medical median EFS compared with mFOLFIRINOX (16·0 months [95% CI 12·4–19·8] vs 10·2 months [8·6–13·5]; hazard Oncology, IRCCS Istituto ratio 0·63 [0·47–0·84]; p=0·0018). At least one grade 3 or worse adverse event was observed in 87 (66%) of 132 patients in Romagnolo per lo Studio dei the PAXG group and in 78 (61%) of 128 patients in the mFOLFIRINOX group, including one fatal event. Tumori (IRST) Dino Amadori, Meldola, Italy (I G Rapposelli MD); Niguarda Interpretation PAXG significantly improved EFS compared with mFOLFIRINOX in resectable or borderline resectable Cancer Center, Department of PDAC. Preopertive PAXG could be considered a standard option for resectable or borderline resectable PDAC. Hematology, Oncology, and Accordingly, preoperative PAXG should be considered as the standard comparator group for future trials in this Molecular Medicine, Grande Ospedale Metropolitano setting. Niguarda, Milan, Italy (K Bencardino MD); Medical Funding MyEverest and Codice Viola. Oncology Unit, University Hospital and University of Cagliari, Cagliari, Italy Copyright © 2025 Elsevier Ltd. All rights reserved, including those for text and data mining, AI training, and similar (Prof M Scartozzi MD); Division technologies. of Pancreatic and Transplant Surgery, Pancreas Translational Introduction worst 5-year survival rate among solid tumours.1 Only and Clinical Research Center, IRCCS San Raffaele Hospital, Pancreatic ductal adenocarcinoma (PDAC) accounts for 10–20% of patients are surgical candidates, and surgery Milan, Italy (G Balzano MD, about 3% of new cancer cases per year, is the third alone produces disappointing results, mainly due to early D Tamburrino, G Belfiori MD, leading cause of cancer-related death, and yields the metastatic spread. Postoperative chemotherapy has been Prof M Falconi); Oncology, Articles Department of Oncology and Research in context Hematology, ASST Papa Giovanni XXIII, Bergamo, Italy Evidence before this study significant differences between groups. Similarly, the (B Merelli MD); SCDU Oncologia Medica, AO Ordine Mauriziano From Jan 1, 2000, to June 29, 2020, we searched PubMed using SWOGS1505 phase 2 trial (perioperative nab-paclitaxel plus di Torino, Turin, Italy the terms (“pancreatic ductal adenocarcinoma”) gemcitabine vs mFOLFIRINOX) for patients with resectable (E Sperti MD); Medical Oncology AND ((“neoadjuvant chemotherapy” OR “neoadjuvant chemo- disease and the PREOPANC-2 phase 3 trial (neoadjuvant and Hematology Unit, radiotherapy”) AND (“resectable” OR “borderline resectable”)), chemoradiotherapy and mFOLFIRINOX) for patients with Humanitas Cancer Center, IRCCS Humanitas Research limited to the English language. We filtered by article type resectable or borderline resectable disease showed no difference Hospital, Milan, Italy “clinical trial” and “meta-analysis”. In this scenario few phase 2 across regimens. (S Bozzarelli MD); Department randomised trials emerged, as PACT15 (surgery upfront plus of Medical and Surgical Added value of this study adjuvant gemcitabine or PEXG [cisplatin, epirubicin, Sciences, Alma Mater Overall, cumulative evidence before and during the Studiorum, University of capecitabine, and gemcitabine] vs perioperative PEXG) and CASSANDRA trial showed that neoadjuvant therapy is a Bologna, Bologna, Italy Prep02 (surgery upfront plus adjuvant S-1 vs S-1 perioperative) (M Di Marco MD, G Ercolani MD); suitable standard approach in both resectable and borderline both suggested the benefit of a neoadjuvant approach Medical Oncology Unit, IRCSS resectable PDAC, without evident differences among Azienda Ospedaliero compared with primary surgery and adjuvant chemotherapy for mFOLFIRINOX, nab-paclitaxel plus gemcitabine, capecitabine Universitaria di Bologna, patients with resectable pancreatic ductal adenocarcinoma plus gemcitabine, or chemoradiation. To date, CASSANDRA and Bologna, Italy (M Di Marco); (PDAC). Nevertheless, at that time, the only available phase 3 Department of Oncology and PREOPANC-2 are the only phase 3 studies testing the efficacy of study, PREOPANC-1, performed in a mixed resectable and Palliative Care, Cardinale different regimens in the field of resectable and borderline Giovanni Panico Hospital, borderline resectable population, did not show a benefit of resectable PDAC. The CASSANDRA trial showed superiority of Tricase, Italy (E Tamburini MD); neoadjuvant gemcitabine plus radiotherapy compared with the PAXG regimen over mFOLFIRINOX in the setting of Section of Innovation upfront surgery. Afterwards, the long-term, mature results of Biomedicine—Oncology Area, preoperative treatment of resectable and borderline resectable the study, reported in 2022 (while CASSANDRA was already Department of Engineering for PDAC. Our findings supported not only that PAXG prolongs Innovation Medicine, ongoing), showed that preoperative chemoradiotherapy event-free survival, but also that it increases the rate of disease University of Verona, and significantly prolonged survival over surgery. Thus, control, carbohydrate antigen 19-9 response, pathological Verona University and Hospital PREOPANC-1 provided evidence endorsing a neoadjuvant Trust, Verona, Italy complete response, pathological IA and IB stages, and node- approach in patients with resectable and borderline resectable (Prof M Milella MD); negative resections, and reduces the incidence of intra or early Department of General and disease. Since then, several meta-analyses and additional postoperative metastases. According to subgroup analysis, the Oncologic Surgery, Morgagni- phase 2 randomised studies were reported: NEONAX (surgery Pierantoni Hospital, AUSL upfront followed by AG [nab-paclitaxel and gemcitabine] treatment effect consistently favoured PAXG across all the Romagna, Forlì, Italy subpopulations. Furthermore, the PAXG regimen did not adjuvant vs AG perioperative) in patients with resectable (G Ercolani); Division of increase treatment-related toxicity and caused a clinically Minimally-Invasive Surgical disease and ESPAC5 (surgery plus adjuvant gemcitabine meaningful deterioration of quality of life in a smaller number Oncology, Niguarda Cancer vs either short-course neoadjuvant mFOLFIRINOX, gemcitabine Center, ASST Grande Ospedale plus capecitabine, or chemoradiotherapy) in patients with of domains compared with mFOLFIRINOX. Metropolitano Niguarda, borderline resectable disease, confirming neoadjuvant and Implications of all the available evidence Milan, Italy (M Mazzola MD); Department of Radiology, perioperative therapy as a standard in these two settings. Neoadjuvant therapy should be indicated as standard-of-care Pancreas Translational and Conversely the NORPAC-1 trial (surgery upfront followed by for patients with non-metastatic PDAC before surgery. Clinical Research Center, IRCCS adjuvant chemotherapy with mixed regimens vs 2-month CASSANDRA sets PAXG as a new standard regimen for San Raffaele Scientific preoperative chemotherapy with mixed regimens followed by preoperative treatment of patients with resectable and Institute, Milan, Italy (D Palumbo); Clinical Oncology surgery and postoperative chemotherapy) did not show borderline resectable PDAC. Department, Mario Negri Institute—IRCCS, Milan, Italy (V Torri MD) shown to significantly prolong survival.2,3 However, the In this context, the Pancreatic Adenocarcinoma Correspondence to: detection of metastases during or after surgery, along Clinical Trials-21 (PACT-21) Chemotherapy Role Dr Michele Reni, Department of with surgical complications, limits adjuvant Assessment in Stage I-III Surgically (Borderline) Medical Oncology, Pancreas Translational and Clinical chemotherapy use to about 65% of patients with initially Resectable Pancreatic Adenocarcinoma: What Research Center, IRCCS San resectable disease,4,5 with only 65% completing the Neoadjuvant Regimen? What Duration? Refinining the Raffaele Scientific Institute, planned treatment.5,6 The role of neoadjuvant therapy Therapeutic Approach (CASSANDRA) trial was initially Milan 20132, Italy was explored in randomised phase 2 trials yielding designed as a phase 2 randomised trial aimed at reni.michele@hsr.it promising results in both resectable and borderline selecting the best candidate for phase 3 testing between resectable populations.4,7,8 However, the anatomical the PAXG (cisplatin, nab-paclitaxel, capecitabine, and resectability classification does not have prognostic gemcitabine) and the mFOLFIRINOX (modified validation9 and is barely reproducible, even among fluorouracil, leucovorin, irinotecan, and oxaliplatin) experts.10 To overcome this hindrance, The Dutch regimens. Both regimens warranted evaluation because Pancreatic Cancer Group pragmatically enrolled mFOLFIRINOX and FOLFIRINOX significantly resectable or borderline resectable patients into the prolonged survival compared with gemcitabine in phase 3 PREOPANC-1 trial, providing evidence of patients with both metastatic and resected PDAC,6,12 and neoadjuvant therapy superiority over upfront surgery.11 encouraging results were reported for PAXG against 2946 Articles nab-paclitaxel plus gemcitabine in a randomised phase 2 be resectable or borderline resectable by local study in stage II–III and IV disease.13,14 During the investigators according to National Comprehensive conduct of the CASSANDRA trial, the phase 3 Cancer Network Guidelines, and biologically defined PREOPANC-2 trial (Netherlands Trial Register according to the international consensus on definition identifier: NTR7292, 2018-06-19) showed that and criteria of borderline resectable PDAC (carbohydrate mFOLFIRINOX and gemcitabine plus gemcitabine- antigen 19-9 [CA19-9] of 500 IU/mL or more).19,20 Self- based chemoradiation therapy followed by surgery reported sex (open question) and race (open question) yielded similar results in the resectable and borderline data were collected at clinical review but no ethnicity data resectable population.15 Accordingly, mFOLFIRINOX were collected due to expected uniformity in the Italian became a standard therapeutic option in the population. A protocol amendment aimed at improving CASSANDRA trial population. Since the trial fulfilled trial external validity was introduced after 21 patients the requirements for phase 3 definition,16 the trial was were randomised, allowing the enrolment of participants reclassified accordingly after 224 (86%) of 260 patients with dihydropyrimidine-dehydrogenase (DPYD) and were randomised. We aimed to analyse event-free uridine-5ʹ-diphospho-glucuronosyltransferase (UGT) survival (EFS) of PAXG compared with mFOLFIRINOX dysfunctional polymorphism. Full eligibility criteria are for first randomisation analysis. provided in the trial protocol (appendix pp 19–93). All the patients provided written informed consent before study Methods initiation. Study design This randomised, open-label, 2 × 2 factorial phase 3 trial Randomisation (PACT-21 CASSANDRA) was designed under the Patients were randomly assigned in a 1:1 ratio to receive auspices of the Associazione Italiana Studio Pancreas, PAXG or mFOLFIRINOX. Patients without progression approved by the independent ethics committee at each or limiting toxicity after 4 months of chemotherapy were participating institution (appendix pp 2–3), and further randomised in a 1:1 ratio to receive a further See Online for appendix conducted at 17 Italian academic hospitals in accordance 2 months of the same chemotherapy either before or with the International Conference on Harmonization E6 after surgery. Results of the first randomisation are requirements for Good Clinical Practice and with the reported here. The results of the second randomisation ethical principles outlined in the Declaration of will be reported in a separate manuscript. Helsinki.17 A data and safety monitoring board was not Randomisation lists were written using R code (package planned for this study because it was originally designed randomizeR) and stratified according to centre and as a phase 2 trial. The appendix contains the study CA19-9 serum level (equal or inferior to five times the protocol (pp 19–95) and the statistical analysis plan upper limit of the normal range) for the first (pp 96–108). The trial, registered on ClinicalTrials.gov randomisation or according to previously assigned (NCT04793932) and EudraCT (2020-003080-26 and treatment for the second randomisation. Both 2024-519031-42-00), has completed accrual and reached randomisations used 1:1 ratios and blocks that were the necessary events for first randomisation primary variable from two to four. The size of blocks randomly analysis but follow-up of overall survival is ongoing. varied to guarantee balancing within strata while The trial was registered 3 months and 15 days late on preserving randomisation concealment. The list was the ClinicalTrials.gov platform. The first patient was generated at the study start by a consultant statistician randomised on Nov 3, 2020, and seven patients were who was not in charge of the analysis, integrated into the enrolled before the registration on March 10, 2021. central clinical research organisation independent Autumn, 2020, was dicult in Italy due to the second web-based platform, and associated with the relevant wave of COVID-19 outbreak and some communication centre. A randomisation button was defined in the and supervision shortages led to missing this deadline electronic case report form structure. Based on the due to good faith human mistakes. The small number of conditions associated with each randomisation button, patients who were enrolled before registration guarantees the system selected a specific layer from the list and then that no attempt at selective or biased reporting of associated the patient to the first available free position research outcomes was made. Furthermore, patients and within the randomisation layer. Patient recruitment was the public were informed of the existence of the trial by overseen by the study investigators of each centre and patients’ associations that supported it. randomisation was performed by the local clinical trial assistant who was not involved in the rest of the trial. Patients Eligible patients were aged 18–75 years and had a Procedures Karnofsky performance-status score of 70 or more, Patients were randomly assigned to receive PAXG (oral stage I–III disease according to the American Joint capecitabine total daily dose of 1250 mg/m² in 625 mg/m² Committee on Cancer (AJCC) Cancer Staging Manual,18 twice daily dosage and 30–40 min intravenous infusion pathologically confirmed PDAC that was considered to of nab-paclitaxel 150 mg/m², 60–80 min intravenous Articles infusion of gemcitabine 800 mg/m², and 60 min and a physical examination including adverse events intravenous infusion of cisplatin 30 mg/m² every 14 days) assess ment by the attending oncologist. or mFOLFIRINOX (2 h intravenous infusion of Resection was planned between 4 and 8 weeks after oxaliplatin 85 mg/m², 2 h intravenous infusion of final intravenous chemotherapy adminis tration and was leucovorin 400 mg/m², 90 min intravenous infusion of performed according to standard clinical practice. After irinotecan 150 mg/m², and continuous 46 h intravenous discontinuation of the trial intervention, a physical infusion of fluorouracil 2400 mg/m² every 14 days) examination, blood samples, and tumour markers within 7 days of randomisation. Dose modification assessment, alongside a chest and abdomen imaging according to DPYD and UGT polymorphisms followed with contrast enhanced CT or MRI, was repeated every recommendations reported in the trial protocol. Each 3 months during the first year after randomisation, every treatment administration was preceded by a blood testing 4 months during the second year, and every 6 months 261 patients assessed for eligibility 261 randomly assigned (first randomisation) 133 assigned PAXG (132* included 128 assigned mFOLFIRINOX in ITT, per protocol, and safety (128 included in ITT, per analyses) protocol, and safety analyses) 16 discontinued treatment 30 discontinued treatment 4 resected 6 resected 12 not resected 24 not resected 117 event-free and eligible for 98 event-free and eligible for second randomisation second randomisation 24 not randomised 20 not randomised 18 continued PAXG 10 continued mFOLFIRINOX 6 resected 10 resected 93 randomly assigned 78 randomly assigned (second randomisation) (second randomisation) 43 assigned surgery then 2 months 50 assigned 2 months of PAXG then 40 assigned surgery then 2 months 38 assigned 2 months of of PAXG surgery of mFOLFIRINOX mFOLFIRINOX then surgery 7 not resected 4 discontinued PAXG 4 not resected 4 discontinued mFOLFIRINOX 4 resected 2 resected 2 not resected 46 resected† 64 completed PAXG‡ 52 resected§ 44 completed mFOLFIRINOX¶ 18 discontinued postoperative 14 not resected 25 discontinued postoperative 12 not resected PAXG mFOLFIRINOX 28 completed PAXG 50 resected 27 completed mFOLFIRINOX 34 resected Figure 1: Trial profile ITT=intention-to-treat. *One patient in the PAXG group had a pathological diagnosis of ampullary carcinoma at final histology after surgery, thus was excluded from the analyses. †Includes four resected before completion of first stage and six resected at second randomisation. ‡Includes 18 patients who did not undergo second randomisation and completed 6 months of PAXG. §Includes six resected before completion of first stage and ten resected at second randomisation. ¶Includes ten patients that did not undergo second randomisation and completed 6 months of mFOLFIRINOX. 2948 Articles from the third year onwards or whenever clinically Trial pertinent data wer3 collected in electronic case- indicated until disease recurrence. report forms by the investigators and analysed by a statistician who vouches with all the authors for their Outcomes accuracy and completeness. A clinical research The primary efficacy endpoint was EFS, calculated organisation collected the electronic case-report forms from the date of randomisation to the date when the local and monitored the study. An expert radiologist, who was investigator reported the first qualifying event. Any of the masked to treatment group, centrally reviewed baseline following was considered as an event: (1) radiological progression; (2) disease recurrence; (3) two consecutive CA19-9 increases, each of 20% or more, separated by PAXG mFOLFIRINOX (n=132) (n=128) 4 weeks or more (limited to patients with basal level over the superior normal laboratory limit); (4) unresectability; Age, years 65 (60–70) 63 (57–69) (5) intra-operative metastasis; and (6) death. Patients <65 66 (50%) 76 (59%) without events at the time of analysis (July 31, 2025) were ≥65 66 (50%) 52 (41%) censored on the date of the latest informative follow-up. Sex The secondary endpoints were rates of radiological Male 64 (48%) 66 (52%) response, CA19-9 response, resection, complete Female 68 (52%) 62 (48%) pathological response, absence of pathological lymph- Race node involvement, absence of pathological resection White 132 (100%) 128 (100%) margins infiltration, detection of metastases during or Karnofsky performance status within 3 months of surgery, surgical morbidity, chemo- ≤80 9 (7%) 11 (9%) therapy toxicity, quality of life, and overall survival. The 90–100 123 (93%) 117 (91%) objective response rate was assessed every 8 weeks by the Anatomical classification local investigator based on the revised RECIST criteria.21 Resectable 63 (48%) 63 (49%) Due to insucient funding, the radiological response Borderline resectable* 69 (52%) 65 (51%) assessment deviated from the per-protocol plan of a CA19-9 central radiology review. Sequential measurements of Positive 100 (76%) 85 (66%) CA19-9 serum level were repeated at baseline and every Negative 32 (24%) 43 (34%) 4 weeks during treatment. CA19-9 response was Median (95% CI)† 261 (85–940) 226 (116–940) measured in patients with basal level over the superior ≤5 × ULN 71 (54%) 79 (62%) normal laboratory limit.22 Namely, patients were classified >5 × ULN 61 (46%) 49 (38%) as non-responders if the percentage reduction at nadir Clinical tumour staging (lowest value assessed while on treatment) with respect I 71 (54%) 57 (45%) to the basal value was inferior to 50%, as minor II 50 (38%) 58 (45%) responders if it was between 50 and 89%, and as major III 11 (8%) 13 (10%) responders if it was 90% of more.22 The local investigator Tumour location graded treatment-related adverse events according to the Head 93 (70%) 96 (75%) National Cancer Institute Common Terminology Criteria Uncinate process 11 (8%) 9 (7%) for Adverse Events version 5.0.23 Quality-of-life was Body 17 (13%) 16 (13%) assessed using the European Organization for Research Tail 10 (8%) 6 (5%) and Treatment of Cancer (EORTC) 30-item Quality-of- Diffuse 1 (1%) 1 (1%) Life Questionnaire (EORTC QLQ-C30) and the 26-item Genetic variants‡ EORTC questionnaire for patients with PDAC (EORTC Not available 36 (27%) 36 (28%) QLQ-PAN26), both completed by the patients at baseline Pathogenic 16/96 (17%) 12/92 (13%) and after 4 months of chemotherapy. The scores of each BRCA1 or BRCA2 gPV 11/96 (11%) 4/92 (4%) EORTC QLQ-C30 and EORTC QLQ-PAN26 scale were Other gPVs§ 5/96 (5%) 8/92 (9%) calculated according to the EORTC scoring manual.24 None 80/96 (83%) 80/92 (87%) Between-group comparison of treatment-baseline changes in each scale score were assessed by means of Data are as median (range), n (%), or n/N (%), unless otherwise stated. CA19-9=carbohydrate antigen 19-9. ULN=upper limit of normal. gPV=germline analysis of variance. The best health response pathogenic variant. *Patients were included in the borderline resectable (improvement, no change, or deterioration) considered population based on both surgical and biological criteria (CA19–9 of 500 IU/mL or the proportion of patients with a clinically meaningful greater). †Median values and 95% CIs are calculated for the group of patients with change (defined as a ≥10-point change from baseline). positive CA19-9 values. ‡Genetic testing results refer to germline variants only. §Includes germline pathogenic variants in ATM, MLH1, MSH3, MSH6, MUTYH, Overall survival was calculated from the date of NBN, PALB2, and SDHB. randomisation until death. Participants who were alive at Table 1: Baseline characteristics in the intention-to-treat population by the time of analysis were censored at the date of the latest treatment regimen time seen alive. Articles CT scans at the end of the study to confirm both the eect estimates and interaction test.26 The statistical anatomical resectability and the AJCC classification. analysis plan is available in the appendix (pp 94–107). Statistical analysis Role of the funding source A superiority comparison was planned. Based on a All the trial costs were covered by donations from proportion of patients alive without an event at 3 years patients’ associations MyEverest and Codice Viola. The from randomisation of 20% in the mFOLFIRINOX study drugs were provided at no charge by the Italian group (null hypothesis),11 we anticipated that the EFS rate National Health System (DL Dec 23, 1996, N 648, would be 15% higher in the PAXG group (alternative Gazzetta Uciale n° 43, Feb 2, 2005). The funders of the hypothesis), which would correspond to a hazard ratio study had no role in study design, data collection, data (HR) estimated by Cox proportional hazards model analysis, data interpretation, or writing of the report. of 0·65. We calculated that, under exponential distribution and with a dropout rate of 4%, the inclusion Results of 260 patients with 173 events would provide the Between Nov 3, 2020, and April 24, 2024, 261 patients study with 80% power to show a statistically were randomly assigned to receive PAXG (133 patients) significant difference of 15% in EFS between groups or mFOLFIRINOX (128 patients; figure 1). One patient in at two-sided 5% log-rank test. There was no plan to test the PAXG group had a pathological diagnosis of overall survival at this timepoint. Overall survival analysis ampullary adenocarcinoma at final histology after will be conducted when 173 deaths have occurred. surgery, thus was excluded from all analyses. All the other Data were analysed in the intention-to-treat population patients received at least one administration of the (all randomised patients), except for surgical mortality assigned treatment. No protocol deviations were observed. and morbidity, which was assessed in the resected Intention-to-treat, per-protocol, and safety populations population, and the EORTC-QLQ questionnaire, which overlapped (figure 1). Baseline characteristics were was evaluated in patients who had a completed form similar between groups (table 1). In 132 patients in the both at baseline and at second assessment. Qualitative PAXG group, the median age was 65 years (IQR 60–70), variables were compared by the χ² test or Fisher’s exact 64 (48%) were male, and 68 (52%) were female. In test, and quantitative variables by parametric or non- 128 patients in the mFOLFIRNOX group, the median age parametric equivalent test. Survival rate estimates were was 63 years (IQR 57–69), 66 (52%) were male, and calculated with the Kaplan–Meier method. 62 (48%) were female. All participants were White. Multivariable analysis of EFS was conducted with a Cox EFS analysis was performed after a median follow-up of proportional hazards model. The treatment heterogeneity 28·5 months (IQR 20·2–38·2), after events occurred in eects in subgroups defined by prognostic factors were 183 (70%) of 260 participants (83 [63%] of 132 patients in displayed in a forest plot,25 reporting the within-level the PAXG group and 100 [78%] of 128 patients in the 90 70 50 30 10 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 Follow-up (months) Number at risk (censored) PAXG group 14 13 9 6 5 4 1 0 ·· (36) (37) (39) (43) (44) (45) (48) (49) (··) mFOLFIRINOX group 7 5 5 3 3 3 1 0 ·· (21) (23) (23) (25) (25) (25) (27) (28) (··) Figure 2: Kaplan–Meier plot of event-free survival in the intention-to-treat population Median event-free survival in PAXG group was 16·0 months (95% CI 12·4–19·8) and in mFOLFIRINOX group was 10·2 months (8·6–13·5). HR=hazard ratio. 2950 )%( lavivrus eerf-tnevE PAXG group mFOLFIRINOX group HR 0·63 (95% CI 0·47–0·84); p=0·0018 132 131 123 115 101 93 79 71 56 48 44 39 35 33 25 19 17 (0) (0) (0) (0) (0) (0) (1) (3) (10) (12) (12) (17) (21) (21) (27) (32) (34) 128 120 112 95 83 65 57 52 44 37 30 23 18 16 12 9 9 (0) (0) (0) (0) (0) (0) (0) (0) (4) (7) (12) (15) (16) (17) (19) (21) (21) Articles mFOLFIRINOX group). Median EFS was significantly longer in the PAXG group than in the mFOLFIRINOX group (16·0 months [95% CI 12·4–19·8] vs 10·2 months [8·6–13·5]; HR 0·63 [0·47–0·84]; p=0·0018; figure 2). PAXG mFOLFIRINOX Relative response p value 1-year EFS rate was 61% (95% CI 57–65) and 3-year EFS (n=132) (n=128) increase (95% CI) rate was 33% (28–37) in the PAXG group, compared RECIST best response ·· ·· ·· ·· with 45% (41–50) at 1 year and 13% (10–17) at 3 years in the Partial response 61 (46%) 50 (39%) ·· ·· mFOLFIRINOX group. Pattern of recurrence was similar Stable disease 69 (52%) 67 (52%) ·· ·· in the two groups (table 2). Event type is summarised in Progression of disease 2 (2%) 11 (9%) ·· ·· table 2. Objective response rate ·· ·· 1·18 (0·89–1·57) 0·25 The results of a multivariable Cox regression analysis No 71 (54%) 78 (61%) ·· ·· and of univariate analyses, including patient and disease- Yes 61 (46%) 50 (39%) ·· ·· related prognostic factors, confirmed the treatment Disease control rate ·· ·· 1·08 (1·02–1·14) 0·0088 eect, and the AJCC classification had a more relevant No 2 (2%) 11 (9%) ·· ·· outcome impact than the anatomical classification Yes 130 (98%) 117 (91%) ·· ·· variables included in the multivariable analysis (appendix CA19-9 response ·· ·· 1·38 (1·15–1·65) <0·0001 p 4). In the subgroup analysis, the treatment eect was Not applicable 32 (24%) 43 (34%) ·· ·· also consistently in favour of the PAXG regimen across Missing 3 (2%) 0 ·· ·· all subgroups (appendix p 5). No 12/97 (12%) 31/85 (36%) ·· ·· In the resectable population, the median EFS was Yes 85/97 (88%) 54/85 (64%) ·· ·· 19·1 months (95% CI 14·5–37·3) in the PAXG group and Pathological stage ·· ·· ·· ·· 10·4 months (8·7–14·9) in the mFOLFIRINOX group. IA* 31 (23%) 17 (13%) ·· ·· 1-year EFS rate was 71% (66–77) and the 3-year EFS rate IB 15 (11%) 12 (9%) ·· ·· was 40% (33–47) in the PAXG group, compared with a IIA 0 0 ·· ·· 1-year EFS rate of 46% (40–52) and a 3-year EFS rate IIB 39 (30%) 32 (25%) ·· ·· of 15% (9–21) in the mFOLFIRINOX group. In the III 13 (10%) 22 (17%) ·· ·· borderline resectable population, the median EFS was 12·4 months (10·7–17·4) in the PAXG group and IV 1 (1%) 5 (4%) ·· ·· 10·2 months (6·9–15·5) in the mFOLFIRINOX group. Not resected 33 (25%) 40 (31%) ·· ·· 1-year EFS rate was 52% (46–58) and the 3-year EFS rate Pathological stage