We invite you to join the
PASS-01 Challenge, a rigorous and arms-length comparison of predictive and prognostic algorithms for treatment selection in advanced pancreatic cancer.
Participating groups will be provided with the clinical dataset from the PASS-01 trial linked with genomic and transcriptomic data to test algorithms that predict differential treatment effects and patient level outputs.
Submitted predictions will be linked to centrally held outcome data to compute performance metrics, which will be reported to contributors.
Leader board:
To be updated as the challenge progresses!
Deadline for Submission 20-Nov-2026
Challenge Team
Challenge PI: Robert Grant
PASS-01 Trial PI: Jennifer Knox
Analyst: Wei Quan
Statistician: Yangqing Deng
Study Coordinator: Mariam Hasnain
Study Coordinator: Daniela Bevacqua
Study Manager: Anna Dodd
Study Background
By the end of this decade, pancreatic ductal adenocarcinoma (PDAC) will become the second most common cause of cancer death1. Most PDAC patients present at an advanced incurable stage, and are treated with palliative-intent systemic therapy. Outcomes remain dismal, with a median survival of approximately one year2–4. Novel approaches to improving outcomes are urgently needed.
Biomarker-guided treatment selection can improve outcomes by matching patients to the most effective available treatment. In the first-line setting, consensus-based guidelines5, 6 recommend three multi-agent chemotherapy regimens for metastatic PDAC: modified 5-fluorouracil, leucovorin, irinotecan, and oxaliplatin (FFX), the same regimen with nanoliposomal irinotecan substituted for irinotecan (NALIRIFOX), or gemcitabine with nab-paclitaxel (GNP). Clinical trials and meta-analysis show these regimens yield similar outcomes on average2–4, 7. However, biomarkers have been proposed that identify subgroups that may uniquely benefit from specific treatments. For example, basal-like and classical transcriptomic subtypes may derive differential benefit from GNP and FFX, respectively8–11. Predictive biomarkers can range from hypothesis-driven single-gene tests, such as hENT1 RNA expression12, to machine-learning algorithms that can be multimodal, incorporating clinical, pathological, DNA, and RNA features. While several biomarkers show promise in PDAC, none have been validated in randomized clinical trials. Randomization is required for unbiased evaluation of differential treatment effect (DTE) predictions.
The PASS-01 trial (JCO 2025) offers an unprecedented opportunity to evaluate predictive biomarkers in PDAC3. Patients with metastatic PDAC were randomized to GNP or FFX. Clinical data, digitized whole-slide histopathology images, and whole-genome and transcriptome sequencing after laser-capture microdissection were collected for correlative analysis. Similar input data are available from the COMPASS trial of advanced PDAC13; however, in COMPASS, treatment was assigned by the physician and patient choice. Therefore, COMPASS holds value for training algorithms but cannot be used to test DTE algorithms.
We have trained multimodal machine learning algorithms to predict outcomes in PDAC patients treated with FFX or GNP in the COMPASS trial. The resulting system was named MULTIPL. Next, we designed the PASS-01 Challenge, in which biomarkers and algorithms of DTE and prognosis can be formally tested in a randomized trial. In this challenge, participating groups from other institutions will submit predictions for evaluation in the PASS-01 trial, which randomized participants to FFX or GNP. The PASS-01 Challenge presents an opportunity to rigorously evaluate predictive biomarkers for treatment selecting in advanced PDAC.
References
1. Rahib L, Smith BD, Aizenberg R, et al: Projecting cancer incidence and deaths to 2030: the unexpected burden of thyroid, liver, and pancreas cancers in the United States. Cancer Res 74:2913–2921, 2014
2. Wainberg ZA, Melisi D, Macarulla T, et al: NALIRIFOX versus nab-paclitaxel and gemcitabine in treatment-naive patients with metastatic pancreatic ductal adenocarcinoma (NAPOLI 3): a randomised, open-label, phase 3 trial. Lancet 402:1272–1281, 2023
3. Knox JJ, O’Kane G, King D, et al: PASS-01: Randomized phase II trial of modified FOLFIRINOX versus gemcitabine/nab-paclitaxel and molecular correlatives for previously untreated metastatic pancreatic cancer. J Clin Oncol JCO2500436, 2025
4. Ohba A, Ozaka M, Mizusawa J, et al: Modified fluorouracil, leucovorin, irinotecan, and oxaliplatin or S-1, irinotecan, and oxaliplatin versus nabpaclitaxel + gemcitabine in metastatic or recurrent pancreatic cancer (GENERATE, JCOG1611): A randomized, open-label, phase II/III trial. J Clin Oncol 43:3345–3354, 2025
5. Conroy T, Pfeiffer P, Vilgrain V, et al: Pancreatic cancer: ESMO Clinical Practice Guideline for diagnosis, treatment and follow-up. Ann Oncol 34:987–1002, 2023
6. Guidelines Detail [Internet]. NCCN [cited 2025 Nov 10] Available from: https://www.nccn.org/guidelines/guidelines-detail?category=1&id=1455
7. Nichetti F, Rota S, Ambrosini P, et al: NALIRIFOX, FOLFIRINOX, and gemcitabine with nab-paclitaxel as first-line chemotherapy for metastatic pancreatic cancer: A systematic review and meta-analysis: A systematic review and meta-analysis. JAMA Netw Open 7:e2350756, 2024
8. O’Kane GM, Grünwald BT, Jang G-H, et al: GATA6 expression distinguishes classical and basal-like subtypes in advanced pancreatic cancer. Clin Cancer Res 26:4901–4910, 2020
9. Wenric S, Sangli C, Guittar J, et al: Real-world validation of the purity independent subtyping of tumors classifier for informing therapy selection in pancreatic ductal adenocarcinoma. JCO Precis Oncol 9:e2500197, 2025
10. Singh H, Xiu J, Kapner KS, et al: Clinical and genomic features of classical and basal transcriptional subtypes in pancreatic cancer. Clin Cancer Res 30:4932–4942, 2024
11. Rashid NU, Peng XL, Jin C, et al: Purity Independent Subtyping of Tumors (PurIST), A clinically robust, single-sample classifier for tumor subtyping in pancreatic cancer. Clin Cancer Res 26:82–92, 2020
12. Perera S, Jang GH, Wang Y, et al: HENT1 expression predicts response to gemcitabine and nab-paclitaxel in advanced pancreatic ductal adenocarcinoma. Clin Cancer Res 28:5115–5120, 2022
13. Knox JJ, Jang GH, Grant RC, et al: Whole genome and transcriptome profiling in advanced pancreatic cancer patients on the COMPASS trial. Nat Commun 16:5919, 2025
Summary tables
Primary Objective
Evaluate differential treatment effects (DTE) on progression-free survival (PFS) in the per-protocol (PP) population.
| Population | Endpoint | Prediction | Primary Metric | Secondary Metrics |
| PP | PFS | DTE (Δ = FFX − GNP) | C-for-benefit (C4B) | Hazard Ratio |
Secondary Objectives
1. Differential Treatment Effects (PP population)
| Population | Endpoint | Prediction | Primary Metric | Secondary Metrics |
| PP | OS | DTE | C4B | Hazard Ratio |
| ORR | DTE | C4B | Hazard Ratio |
2. Prognostic Prediction (ITT population)
| Endpoint | Prediction | Primary Metric | Secondary Metrics |
| PFS | Continuous risk score | C-index | Hazard Ratio |
| OS | Continuous risk score | C-index | Hazard Ratio |
| ORR | Predicted probability | AUC |
3. Arm-level Prognostic Prediction (PP population)
| Endpoint | Prediction | Primary Metric |
| PFS | Continuous risk score | C-index |
| OS | Continuous risk score | C-index |
| ORR | Predicted probability | AUC |
Additional information
- The primary analyses will use all available samples (including those with missing modalities) with any missing predictions mean imputed.
- We will also provide analyses for cases with all modalities and in the subset without missing predictions.
Differential Treatment Effects (DTE)
Prediction formats
- The preferred format for predictions is: Δ = outcome under FFX − outcome under GNP, where positive Δ → predicted benefit from FFX and negative Δ → predicted benefit from GNP
- Any monotonic scale is acceptable
Alternative formats:
- Continuous numeric (preferred)
- Discrete numeric
- “FFX” / “GNP” labels (mapped internally)
- Not allowed: other strings or unordered categorical labels.
Metrics
- Primary evaluation metric: c-for-benefit (c4b) in the per-protocol (PP) population (https://pmc.ncbi.nlm.nih.gov/articles/PMC7448760/)
- Secondary metrics: hazard ratios (HR) (progression-free survival (PFS)/ overall survival (OS)) and objective response (ORR)
Using your submitted DTE predictions, patients in the PP population are classified as:
- Matched: predicted FFX benefit and received FFX, or predicted GNP benefit and received GNP
- Unmatched: predicted FFX benefit but received GNP, or predicted GNP benefit but received FFX
We compute:
- Hazard ratio (HR) for PFS and OS
- Values < 1 indicate better outcomes for the matched group
Prognostic Risk Prediction Submission Requirements
Prediction formats
Preferred:
- Predicted survival days for PFS and OS. Alternatives are any continuous numeric predictor where higher indicates a better outcome.
- Predicted probability of ORR for ORR. Alternatives are any continuous numeric predictor where higher indicates a better outcome.
- Preferred thresholds can be provided for binary categories.
The PASS-01 Challenge is a rigorous and arms-length comparison of predictive and prognostic algorithms for treatment selection in advanced pancreatic cancer using PASS-01 trial clinical data linked with genomic and transcriptomic records. This page outlines the steps required to participate, including data access and submission procedures. We invite collaborators to contribute their models to help advance precision treatment selection.
Data Provided
The following limited clinical data variables from patients included in the PASS-01 study will be provided along with digitized histopathology images, and accompanying tumor genomic and transcriptomic data.
- Age
- Gender
- Race
- Baseline ECOG
- Baseline CA19-9
- Treatment arm
- Site of biopsy
- Metastatic sites
Steps to Participate:
- Submit your application through the “Apply” tab.
- Sign a template DTA with UHN (Click here to download the DTA)
- Obtain IRB approval if necessary, according to your institutional guidelines (Click here to download the Protocol PDF)
- After the DTA has been executed, limited clinical data will be provided to you by the PASS-01 Challenge team through secure institutional email
- Refer to the “Whole Genome Sequencing (WGS) Data” section below for steps to request and access the PASS-01 trial genomic data
- Generate predictions using your algorithm.
- Email a CSV file to SRHPBresearchcentrer@uhn.ca . Note only one submission per Principal Investigator or Group will be accepted during the challenge. A data dictionary for which fields should be provided is here. Missing values should be NA. An example file where basal-like is recommended for GNP and classical for FFX is provided here. Files in the incorrect format will be returned. Notify us in your submission email if you wish to have your algorithm included in the challenge manuscript. If so, list the name of the algorithm, and the name and affiliations of up to two coauthors.
Whole Genome Sequencing (WGS) Data
- WGS is housed at the European Genome-phenome Archive (EGA) under accession EGAD5000000239 (Click here to open EGAD50000002309)
- A Data Access Agreement (DAA) must be completed (Click here to download the DAA)
- All investigators must be from the same institution, and institutional email addresses must be used.
- The DAA must be signed by an institutional signatory who has authority to bind the corporation.
- Once completed, forward the DTA to OICR-DAC@oicr.on.ca for review. Access to WGS data will be granted upon final execution of the agreement.
After Submission:
- We will return performance metrics and figures within four weeks.
Publication Plan:
- A manuscript will be published describing the results of the challenge within six months of completion on medrxiv, followed by a peer-reviewed journal.
Timeline
Submission phase (9 months)
Participants will have 9 months to develop and submit algorithms. Authors can publish their algorithm results once performance metrics have been received from the PASS-01 team.
Comparative analysis phase (3 months)
Following the 9-month submission window, a 3-month period will be used to compare submitted algorithms and publish results.
Contact info
Apply to the PASS-01 Challenge
- Please review the Application Guidance tab before completing this page
- Complete the fields below to submit your application to participate in the PASS-01 Challenge. You will receive a confirmation email after submission. You must complete all steps outlined under the Application Guidance before you are officially entered into the Challenge.