EAU Talent Incubator Programme · New Technology & RWE

Rotterdam · Friday 4 + Saturday 5 September 2026

Hands on research challenge

TIP26 RWE Challenge

From research question to manuscript and simulated submission

This challenge starts Friday and finishes Saturday. On Friday your team defines the question, checks the literature and plans the simplest defensible analysis before seeing the full dataset. On Saturday you test that plan against the data, build a manuscript and submit a frozen PDF to independent AI reviewers.

Your challengeUse AI throughout the research process, but understand and control each scientific decision well enough to defend it.

The mission

One research project across two days

Use one Author chat from Friday until the manuscript PDF is frozen. On Saturday, continue in that same chat for data analysis and manuscript creation. The review phase must use fresh AI conversations that see only the submitted PDF.

Do not analyse the Saturday dataset on Friday. Design the study first. The full dataset remains locked until the faculty start signal on Saturday.

The data you will receive Saturday

2,400 synthetic patients

The cohort contains patients treated for non metastatic prostate cancer across eight fictional European treatment centres. You do not see the patient level values or distributions on Friday, but you do know which variables will be available. Use that information to design a question the dataset could realistically address.

Patient characteristics: age, BMI, smoking status, Charlson comorbidity index, ECOG performance status and deprivation quintile
Disease characteristics: PSA, ISUP grade group, clinical stage, metastatic status and risk group
Treatment: radical prostatectomy or radiotherapy, plus ADT status among patients receiving radiotherapy
Baseline patient reported outcomes: urinary, bowel and sexual function and EQ-5D VAS
Early outcomes: major complication within 90 days and readmission within 30 days
Follow up outcomes: urinary, bowel and sexual function and EQ-5D VAS at 12 months
Disease outcome: disease progression within 24 months
Context: diagnosis year, follow up time, follow up reference, treatment centre and country

Dataset locked until Saturday

The faculty will give you the access code when the Saturday challenge starts.

Code not recognised. Check the code shown by the faculty.

Download the complete dataset

The complete challenge

Start Friday, finish Saturday

Keep moving. The goal is a complete research workflow, not a perfect paper.

Friday 20 minQuestion
Evidence + plan

Sat 0 to 12Data
Analysis

Sat 12 to 27Manuscript
Frozen PDF

Sat 27 to 40Independent
Reviews

Sat 40 to 55Faculty led
Plenary

Friday: Design before data

Leave Friday with one research question grounded in the available variables, one evidence check and one simple analysis plan.

20 minutes

1. Start your Author chat

Open one new AI conversation for the research team. Keep this chat for the author work on both days. Agree who leads analysis, manuscript writing, evidence checking and scientific challenge.

2. Generate and choose a preliminary question

Give AI the study context and the variables that will be available. The goal is to generate questions the actual dataset could plausibly address without pretending that you have seen the data.

Prompt: You are helping our research team develop a clinically relevant research question for a multicentre observational study of 2,400 patients treated for non metastatic prostate cancer. We do not yet have access to the patient level data, but tomorrow the dataset will contain: patient characteristics including age, BMI, smoking status, Charlson comorbidity index, ECOG performance status and deprivation quintile; disease characteristics including PSA, ISUP grade group, clinical stage, metastatic status and risk group; treatment with radical prostatectomy or radiotherapy, including ADT status among patients receiving radiotherapy; baseline urinary, bowel and sexual function and EQ-5D VAS; major complications within 90 days and readmission within 30 days; urinary, bowel and sexual function and EQ-5D VAS at 12 months; disease progression within 24 months; diagnosis year, follow up time, follow up reference, treatment centre and country. Based only on these available variables, suggest three clinically relevant research questions that could realistically be investigated with this dataset. For each question, define the population, exposure, comparison and outcome; explain briefly why it matters clinically; identify the most important potential confounders or sources of bias; and state whether the available variables appear sufficient to address the question. Do not assume anything about the values or distributions in the dataset. Do not invent variables that are not listed. Do not analyse the data yet.

3. Evidence checkpoint

Take your preferred question to Consensus.app before committing to it.

Maximum 3 minutes: What does the literature already answer, and what remains uncertain? Do not stop at the headline answer. Check the Consensus Meter and inspect at least two papers behind the answer. Ask whether the evidence addresses the same population, exposure and outcome. Refine the question if needed.

4. Plan and hand over

Decide the simplest analysis you think could answer the question. Do not analyse data yet.

Prompt: Based on our research question and the variables that will be available, propose the simplest defensible analysis strategy. Define the population, exposure, comparison, outcome, important baseline covariates, likely confounders and main missing data concerns. Separate analyses that are essential from optional sensitivity analyses. Do not invent variables and do not run an analysis. Explain the rationale for every major choice in plain language. Then create a concise handover for tomorrow containing our final research question, why it matters, what the literature already tells us, what remains uncertain, our planned primary analysis, the main threats to validity and the decisions we must revisit when we see the data.

Friday stop: Save the Author chat and the handover. Do not work on the Saturday dataset before the faculty unlock.

Saturday: Test the plan against the data

Continue in the same Author chat. The question may survive, or the data may force you to change it.

12 minutes

1. Unlock, upload and inspect

Upload the dataset to your Friday Author chat. Check the data before running the planned model.

Prompt: Visualise and inspect the actual dataset. Describe the cohort, variables, missingness, treatment mix, follow up, important baseline differences and potential data quality issues. Then assess whether the research question and analysis plan we developed yesterday are still defensible. Identify any changes we need to make before analysing. Do not analyse treatment effects until you have stated those changes.

2. Analyse and challenge

Run the simplest defensible analysis, then attack the conclusion before moving on.

Prompt: Using our agreed question and the actual data, run the minimal analysis needed to answer the question. Explain the model in plain language. Then identify overlooked variables, confounding, missing data concerns, alternative explanations and claims that would be too strong. Produce one main result and one useful figure or table.

If the analysis becomes unfamiliar

Stop before accepting an analysis that nobody in your group can explain and defend.

Optional prompt: I am not confident that I understand all the analyses you just performed. Explain each analysis step by step in plain language. For each step, tell me what question it answers, why you chose it, what assumptions it relies on, which variables were included, and how the result should and should not be interpreted. Then tell me which analyses are essential and which are optional sensitivity analyses.

Researcher rule: Do not continue with an analysis that nobody in your group can explain and defend.

At 12 minutes move on. You need one main result, one figure or table and a conclusion you can defend. Do not keep optimising the model.

Manuscript sprint

Turn the analysis into one compact submission manuscript.

15 minutes

1. Title and Methods

Describe only what was actually done. Watch for invented study details.

Prompt: Based on our research question and analysis, propose a precise scientific title. Then write a concise Methods section describing the cohort, exposure, outcome, covariates, missing data and statistical analysis. Use only information available from our dataset and analysis. Do not invent information.

2. Results

Keep observation separate from interpretation.

Prompt: Write a concise Results section based only on analyses actually performed in this conversation. Do not introduce analyses or numbers that have not been generated.

3. Discussion

Include the main finding, interpretation, limitations and clinical relevance.

Prompt: Write a short Discussion. Clearly distinguish observed associations from causal claims and identify the most important limitations.

4. Evidence check on the conclusion

Take the main scientific conclusion to Consensus.app before writing the Introduction.

Ask: Does the existing literature support, qualify or contradict our conclusion? Which parts are well established, and which remain uncertain? What does the literature make us more confident about, and what does it not resolve?

5. Introduction and references

Let AI propose the literature context, then verify it.

Prompt: Write a short Introduction placing our research question in the context of existing literature. Include relevant scientific references.

6. Verify the references

A reference can be real and still be the wrong support for a claim.

Prompt: Verify each reference in our manuscript. For each reference, check whether it exists, whether the bibliographic details are correct, and whether the paper actually supports the specific claim for which we cited it. Flag any overstatement or mismatch in population, treatment, outcome or follow up. Do not replace questionable references automatically.

7. Assemble and freeze

One manuscript, one frozen PDF, no new science during formatting.

Prompt: Combine the title, Introduction, Methods, Results, Discussion and verified references from this conversation into one coherent submission style scientific manuscript. Preserve the analyses, results and scientific decisions already made. Do not run new analyses, add new numbers, invent missing methodological details or introduce new references. If sections contain conflicting information, mark it clearly as [CHECK] rather than silently resolving it. Use only references that survived our verification. Format it as a clean single column scientific manuscript with clear section headings and page numbers. Do not imitate a published journal article. Then create a PDF of exactly this manuscript for our simulated submission. Do not revise or improve the scientific content while creating the PDF. Preserve all [CHECK] markers.

At 27 minutes move on. Your frozen PDF is now the submitted manuscript. Do not revise it during the review phase.

Submission simulation

Three independent AI perspectives on exactly the same frozen manuscript.

13 minutes

Target journal for the simulation: European Urology Open Science.
Open the official journal page and Guide for Authors

Work in parallel: Open three fresh AI conversations and upload only the same frozen manuscript PDF. One person runs each review where possible. The reviewers must not see the Author chat or each other’s feedback.

1. Methodological reviewer

Prompt: Act as a methodological reviewer of this simulated scientific manuscript. This is a teaching simulation using synthetic data designed to represent a real world observational cohort. Do not treat the synthetic nature of the data as a methodological flaw. Identify a maximum of five methodological issues that matter most for whether another researcher could understand, reproduce and trust the analysis. Prioritise substantive problems over minor reporting details. Do not create a comprehensive checklist. For each issue: 1. Explain the problem in plain language. 2. Point to one concrete example from the manuscript. 3. Explain why it matters scientifically or statistically. 4. Classify it as THREAT TO THE CONCLUSION, IMPORTANT LIMITATION, or REPORTING/REPRODUCIBILITY ISSUE. 5. State what the researchers should clarify, decide or analyse before submission. Explain any statistical concept before applying it. Assume the researchers are clinicians, not statisticians. Do not fill in missing methodological details yourself. End with only three take home lessons that a clinical researcher could apply to the next study.

2. Editorial screening

Prompt: Act as an editor for European Urology Open Science reviewing this simulated manuscript. This is a teaching simulation using synthetic data designed to represent a real world observational cohort. Do not treat the synthetic nature of the data as a reason for rejection. Decide whether you would SEND FOR PEER REVIEW or RETURN WITHOUT REVIEW. Give only the three most important reasons for your decision. For each: 1. Point to one concrete feature of the manuscript. 2. Explain in plain language why an editor cares about it. 3. State whether it can realistically be fixed before submission. Focus on scientific relevance, novelty, clarity of the research question, strength of the evidence, journal fit and whether the manuscript is ready for external review. Do not perform a detailed statistical review. End with one sentence answering: What would make this paper worth publishing? If you are uncertain about a journal specific requirement, write VERIFY rather than guessing.

3. Scientific peer reviewer

Prompt: Act as a scientific peer reviewer of this simulated manuscript. This is a teaching simulation using synthetic data designed to represent a real world observational cohort. Do not treat the synthetic nature of the data as a flaw. Your task is not to repeat a methodological reproducibility audit. Focus instead on how much confidence a reader should place in the scientific interpretation and clinical conclusion. Identify a maximum of four issues that matter most for the validity and interpretation of the conclusion. Consider confounding, missing data, treatment heterogeneity, outcome choice, causal language, clinical interpretation, generalisability and whether the evidence supports the claims. For each issue: 1. Explain the problem in plain language. 2. Point to one concrete example from the manuscript. 3. Explain how it changes our confidence in the conclusion. 4. Classify it as THREAT TO THE CONCLUSION or IMPORTANT LIMITATION. Do not spend time on minor reporting details already covered by a methodological reviewer. End with two short statements: What do the data still support? What should the authors not claim?

4. Review the reviewers

Bring the three reports together. Which concern most changed your confidence in the manuscript? Which concern can your group now explain to another researcher? Is there any recommendation you would not accept without checking it independently?

At 40 minutes stop. Do not revise the manuscript. Bring the reviewer reports and your judgement to the plenary.

STOP AFTER 40 MINUTES ON SATURDAY The remaining 15 minutes are for shared learning. The faculty will lead the discussion.

Prepare a 90 second group briefing

Bring five things to the plenary

1. Your research question and main result.
2. One decision that changed when you saw the actual data.
3. The reviewer concern that most changed your confidence.
4. Whether the references survived verification.
5. One thing the human researcher still had to understand or decide.

THE RULE

AI may draft the paper.
The researcher remains accountable for the science.