Method chooser

Choose a method you can defend.

Answer four questions to get a sensible starting point and the checks you will need. It is a guide, not a rule: when the stakes are high, compare at least two methods.

1 · Purpose2 · Data3 · Size4 · Shape
01Decision room

Four questions

Describe your project. Get a starting point.

The recommendation updates as soon as you change an answer.

01What must the synthetic data do?
02Do you have patient-level data?
03How many patients or records?
04What does the data look like?
Start with

Also consider:

Tools:

Checks you will need

    02Workflow

    Whatever method you choose

    A six-step workflow

    The same process whatever method you choose. Step 6 is what regulators and HTA bodies will ask for.

    1. Define the purpose

      What decision or task will the data support?

    2. Prepare the data

      Clean it, set data types and write down the clinical rules.

    3. Start simple

      Fit a baseline such as a Gaussian copula.

    4. Go complex if needed

      Try a deep model only if the baseline falls short.

    5. Evaluate

      Fidelity, utility and privacy, judged against the purpose.

    6. Document

      Method, settings, checks and limitations.