EAU Talent Incubator Programme · New Technology & RWE

Rotterdam · Friday 4 September 2026

Interactive research session

TIP26 RWE Lab

From research question to trustworthy real world evidence

In this interactive two hour session, you will use generative AI as a research partner throughout a structured and guided research journey.

The aimUse AI to develop and refine a clinically relevant research question, explore existing evidence with Consensus and progressively analyse synthetic real world data.

As new information becomes available, you will experience how conclusions can change when baseline characteristics, confounding, missing data, different outcome domains and centre effects are considered.

Today’s research journey

From an idea to an interpretation you can defend

Research question

→

Evidence search

→

Evidence gap

→

RWE data

→

Analysis

→

Challenge

→

Interpretation

→

Refined question

→

Evidence check

Your research question

In routine clinical practice, do apparent differences in patient reported outcomes between radical prostatectomy and radiotherapy persist after accounting for differences in the patients receiving the two treatments?

How to work: Use generative AI actively as a research partner. Work only with the dataset currently released. At each step, analyse what you know, challenge your interpretation and consider what information you would want before drawing a stronger conclusion.

Progressive data challenge

Your data journey starts here

You will not receive the complete dataset at once. New information about the same patients and research problem will become available during the session.

At every step, ask:

1. What do the data suggest now?
2. What could make that conclusion wrong?
3. What information would you want next?

Dataset 1

Your starting point · See what the data show before interpreting them

OPEN

Available data

Treatment
12 month overall health outcome

Download Dataset 1

Step 1 · Visualise

Upload the dataset to your AI research partner and simply ask: “Visualise the data.”

Do not ask for additional variables or a more advanced analysis yet. First, look at what the AI chooses to show you.

2 · Write the result

Ask your AI: “Write one sentence for the Results section of a scientific paper based only on the data currently available.”

3 · What do you see?

What did the visualisation and Results sentence actually show about the two treatment groups?

4 · How strong is the claim?

Would you be comfortable changing the descriptive result into the claim that one treatment produces a better 12-month outcome? Why or why not?

Stop here and discuss with your group. Do not ask AI what additional data you need yet. Be ready to explain what the current data show: and what they do not establish.

Dataset 2

New baseline information · Does the story change?

LOCKED

Enter the code shown by the faculty.

Code not recognised. Check the code shown on the slide.

Dataset 3

More clinical information is available. Can the analysis withstand scrutiny?

LOCKED

Enter the code shown by the faculty.

Code not recognised. Check the code shown on the slide.

Dataset 4

More outcomes and the full source cohort are now available

LOCKED

Enter the code shown by the faculty.

Code not recognised. Check the code shown on the slide.

Dataset 4 unlocked

The source cohort is larger than your initial analysis dataset.

The original cohort contains 150 patients.

Your first analyses included only patients with a completed 12 month overall health outcome.

New information

Full source cohort
Urinary function at baseline and 12 months
Bowel function at baseline and 12 months
Sexual function at baseline and 12 months
PROM follow up status
Follow up time

Download Dataset 4

Step 1 · Visualise

Upload Dataset 4 in the same AI conversation and ask: “Visualise the new data.”

Look at what the AI chooses to highlight about the full cohort, the different outcome domains and the available follow up data.

2 · Write the result

Ask: “Write one sentence for the Results section of a scientific paper based only on the data currently available.”

3 · Make it clinical

Ask: “Write a clinically relevant conclusion based on these data.”

4 · Challenge the wording

Which parts of the clinical conclusion are directly supported by the data, and which parts are interpretation? Are there any words you would change as a reviewer?

5 · Investigate missing data

Ask: “Review the patients without completed 12 month PROM data. Does missingness differ between treatment groups or patient characteristics, and could this affect the conclusions from a complete case analysis?”

6 · Reconsider the question

Ask: “Considering the domain specific outcomes and the non random missingness, can we still meaningfully ask which treatment has the better overall outcome? Why or why not?”

7 · What has changed?

Discuss how your research question has changed as new outcomes and information about missing follow up became available.

Key methodological insight: A single overall outcome can hide clinically important trade offs. Who is missing from the analysis can also change what the observed results represent. The researcher must therefore defend both the choice of outcome and the population on which the conclusion is based.

Dataset 5

Final context is available. What exactly are we comparing?

LOCKED

Enter the code shown by the faculty.

Code not recognised. Check the code shown on the slide.

Dataset 5 unlocked

Final information

Treatment centre
ADT in patients receiving radiotherapy

Download Dataset 5

Step 1 · Visualise

Upload Dataset 5 in the same AI conversation and ask: “Visualise the new data.”

Look at what the AI chooses to highlight about treatment centre, ADT, patient mix, outcomes and follow up completeness.

2 · What changed?

Ask: “What new information in Dataset 5 changes how we should think about treatment as an exposure?”

3 · Challenge the ADT interpretation

Ask: “Does the larger decline in sexual function among radiotherapy patients receiving ADT allow us to conclude that ADT caused the difference? Why or why not?”

4 · Challenge the site interpretation

Ask: “Can the observed differences between treatment centres be interpreted as differences in centre quality or treatment performance? Why or why not?”

5 · Revisit missingness

Ask: “Does the difference in PROM completion between radiotherapy patients with and without ADT change how confident we should be in the observed ADT related outcome differences?”

6 · Final synthesis

Ask: “Based on all five datasets, write three statements: What do the data show? What do we infer? What can we not conclude?”

7 · Formulate the next research question

Ask: “Based on everything we have learned from the five datasets, formulate the clinically most relevant research question that should be investigated next.”

Close the research loop with Consensus.app

Take the new research question to Consensus.app and test whether the existing literature already answers it.

Ask three questions:
1. Which parts of our research question are already well answered by the existing literature?
2. Which parts remain uncertain or insufficiently studied?
3. Does the evidence support individual treatment effect prediction, or mainly associations between baseline characteristics and later outcomes?

Final reflection: Has your research question changed from the question you started with? Why?

Key methodological insight: Treatment is not always a single homogeneous exposure. Context, co treatment, patient selection and missing follow up can all shape the observed association. AI can accelerate the analysis, but the researcher remains accountable for the reasoning, the model and the final claim.

TIP26 RWE Lab

From an idea to an interpretation you can defend

Your research question

Your data journey starts here

Dataset 1

Available data

2 · Write the result

3 · What do you see?

4 · How strong is the claim?

Dataset 2

New information

2 · Write the result

3 · Make it clinical

4 · Challenge the AI

Dataset 3

New clinical information

2 · Let AI continue

3 · Review what AI used

4 · Inspect the revised analysis

5 · Challenge the conclusion

Dataset 4

New information

2 · Write the result

3 · Make it clinical

4 · Challenge the wording

5 · Investigate missing data

6 · Reconsider the question

7 · What has changed?

Dataset 5

Final information

2 · What changed?

3 · Challenge the ADT interpretation

4 · Challenge the site interpretation

5 · Revisit missingness

6 · Final synthesis

7 · Formulate the next research question

Close the research loop with Consensus.app