RWE Tools · experimental
RWE Studio
Feed in a small, messy hospital dataset → profile & flag dirtiness → clean it into a tidy master file → pick a study design and map your columns → run the analysis in real R (via WebR, in your browser) → get the estimate, Table 1, and a reproducible R script.
← All RWE tools- 1 · Question & design
- 2 · Upload & profile
- 3 · Map columns
- 4 · Clean → master file
- 5 · Analyse (R)
- 6 · Export
1 · Clinical question & study design
Your question and chosen design are auto-saved to this browser · your dataset never is
Decide this first — the design determines which columns you'll map and what cleaning actually matters.
Not sure which design? Tick what applies, then get a recommendation:
2 · Upload a dataset
CSV or Excel (.xlsx), in any layout and any language. The tool works out the sheet, the header row, and what a row represents, then asks you to confirm — nothing about the shape is assumed.
What this dataset looks like (click to collapse)
Everything below is inferred, then yours to override. Skip it entirely if the guesses look right — the defaults are what the profiler and the role suggester will use.
Data profile & dictionary
| Column | Inferred type | Missing | Distinct | Range / top values | Flags |
|---|
3 · Map columns → roles
AI classify judges each column's causal role for your specific exposure→outcome — ticking genuine confounders and warning against adjusting for mediators / colliders / instruments. Only the column names + your question are sent to the AI — never any patient data. Always review and edit; you have the final clinical say.
Crude is always shown alongside.
Restricting changes which patients the effect refers to, so it is never applied on your behalf. Whatever is chosen is reported with the results.
A coded diagnosis of a condition is in claims even though the measurement behind it is not — obesity is claims, BMI is not. Anything the seeding cannot place is left unset and must be decided before the formula is computed.
4 · Clean → master file
Apply the fixes you want; each is logged and re-emitted into the exported R script. The mapped variables are coerced to the types the analysis needs.
In a new-user cohort (ACNU) each person should appear once, so a repeated ID is usually a data error to resolve. (SCCS and case-crossover repeat IDs by design, and an interrupted time series aggregates rows into periods — so this control is shown for ACNU only.)
5 · Analyse in R (WebR)
Runs the design's model on your master file in real R, in your browser. First run downloads the R runtime (~a few seconds).
6 · Export
The R script is exactly what ran — it reproduces the on-screen result in desktop R.