NTOG Study Design & Simulation Builder

Training notice

Prototype version: v0.9. Educational use only.

This educational prototype supports early study design training. It helps users practise how to structure a research question, define variables, identify common design conflicts, and separate simulated teaching examples from real study analysis.

This tool teaches:

• PICOTS structure
• Codebook thinking
• Endpoint-variable mismatch
• Descriptive versus survival-analysis conflicts
• Time-zero alignment
• Immortal-time bias risk
• Missing-variable checks
• Model-readiness versus technical code execution
• Separation between toy simulation and real analysis

This tool is for education and early study design only. It does not replace protocol development, ethics review, data permissions, statistical review, patient-level data governance, or clinical judgement.

Do not enter patient-level, identifiable, confidential, or unpublished sensitive data.

Study concept form

Variable codebook section

Analysis plan section

Estimand & modern design considerations

Define what you are estimating before choosing how to analyse it. These are standard expectations for contemporary lung cancer study design.

Estimand framework (ICH E9(R1))

State the estimand as five attributes, so the objective and the analysis match:

• Treatment condition(s) being compared.
• Population — the target patients.
• Variable / endpoint measured per patient.
• Intercurrent events (treatment switch, subsequent therapy, death) and the strategy for handling each.
• Population-level summary (e.g. hazard ratio, difference in restricted mean survival, risk difference).

Non-proportional hazards & RMST

The hazard ratio and the log-rank test assume proportional hazards. With immunotherapy, survival curves often separate late and plateau, or cross — the assumption fails, and a single HR (and a sample size powered on it) can mislead. Consider restricted mean survival time (RMST), landmark or milestone survival, or weighted log-rank approaches, and power the study accordingly.

Registry / observational designs: target trial emulation

For non-randomised, registry-based questions, specify the protocol of the target trial you are emulating — eligibility, treatment strategies, assignment, time zero, outcome, and estimand. This is the standard way to avoid time-zero and immortal-time bias in registry studies.

Education and early design only; this does not replace formal statistical review.

Study design checks

Power scenario section

Generate outputs