Checked before it is delivered
Clinwright reruns every program and its checks itself before anything is delivered, whoever wrote it.
Clinical statistical programming
From SDTM to the tables in your CSR: programmed, run and checked.
Start from your protocol or from one of 36 Phase III studies. Clinwright configures your ADaM, writes the SAS programs, runs them until every check passes, and hands each data set and output to a person to review and sign. Every step stays on record.
How it works
One study, start to finish
Each step works from the one before it, automatically, and each can be opened, changed and reviewed by your programmers and statisticians.
-
1
SDTM data
Simulate SDTM from the study design to start before database lock, or upload your own.
-
2
ADaM specification
Configured for your SDTM from a standard template or a Phase III study, or imported from define.xml or Pinnacle 21.
-
3
ADaM programs and data
SAS programs written, run and checked; data sets as XPT v5 and SAS data sets.
-
4
Display plan
Shells for the tables, listings and figures, laid out from the analysis or taken from a Phase III study.
-
5
Outputs
A program for each output, run and checked, ready for QC and approval.
Dummy data that behaves like a study
Screening, randomization, early terminations, deaths and a data cutoff, drawn from your arms, visits and periods. Every generation is checked: subjects, sequence numbers, study days, visit dates, codelists and CDISC terminology.
Bring your real data later as XPT, SAS data sets or Dataset-JSON, and work from a database release.
ADaM that writes itself down
Every derived variable carries its derivation in words, and the program follows from it: from Clinwright's derivation patterns, or from the AI agent where the patterns don't reach.
A program counts as delivered only after it has run and passed its checks. Then it waits for a person to review it.
AI
An AI that works on the pages, where you can see it
Ask in words, beside any page: start a trial, set up its SDTM design, change a derivation, lay out a shell, fix a failed run. The AI works with Clinwright's own tools, as a person would, and every change it makes shows on its page.
Step by step, or done for you
Step by step, the pages follow the AI as it works, and it waits for you before the big steps: writing programs, changing every output at once, deleting. In auto mode it carries on and opens the result when it is done.
Either way, what it changed is marked where it changed, with Keep and Undo. Approving, signing QC and delivering stay with people.
A SAS and R console beside the pages
Try code on the study's SDTM and ADaM, in SAS or R. Keep WORK, and data sets and R objects stay from one run to the next. A failed run goes to the AI in one click, and its fix comes back to run.
On the public service the AI uses your organization's own Anthropic key, and Anthropic bills you for it directly. Everything else works without it.
Templates
Start from a Phase III study
36 Phase III studies, each set up end to end from its protocol and statistical analysis plan: the study design, the settings of its SDTM dummy data, the ADaM specification, and the display plan with every output's shell and analysis.
36
studies in eight therapeutic areas, from cardiovascular outcomes to rare diseases
8,756
tables, listings and figures, 64 to 475 a study
No AI
needed: their programs are written by Clinwright's rules
The whole study, or the part you need
Choose by therapeutic area, disease, primary analysis or comparator: time to event, responders, MMRM, event rates, win ratio. A new trial from a whole study runs step by step, SDTM, then ADaM, then the outputs, with its progress shown; stop it when you like and keep what is finished.
Or take a part into your own trial: the study's dummy data settings, its ADaM specification, or its outputs by topic, their columns following your arms.
Output titles are taken from clinical study reports published by Health Canada (Public Release of Clinical Information); each template names its manufacturer as the source. The manufacturers have not endorsed these templates.
Apps
Shiny, written in SAS
Turn a question about the data into a page the whole study team can use: inputs, tables and graphs in one PROC APP step, running on the study's ADaM and SDTM.
Inputs, outputs and the SAS code between them
Selects, sliders, dates and file uploads; tables to sort and search, SGPLOT graphs to click and brush, value boxes and downloads. When an input changes, only the code that reads it runs again.
Every view has its own link, opens at once when someone has seen it before, and exports as a SAS program that runs in SAS, or as an RTF or Word document.
Try them on the CDISC pilot
Live apps on the pilot's SDTM and the ADaM that Clinwright made from it, and the lineage from one to the other. No account needed; the whole study is open too: its SDTM, ADaM, CSR tables and programs. So is everyone's work on the free service: see what others have tried.
Study dashboard
Pages of value boxes and graphs: treatment, disposition, adverse events and vital signs, between the dates chosen.
Subject explorer
Filter the subjects, plot two baseline measures, brush points to list subjects and download the list.
Baseline characteristics
The CSR's demographic table for the population and characteristics ticked, saved as RTF.
Hepatic safety (eDISH)
Peak transaminase against peak bilirubin; drag to list subjects, pick one for their liver tests over time.
Adverse event drill-down
Choose a system organ class, then its terms: a plot or a table, and a listing of the events.
ADAS-Cog change from baseline
The change from baseline by visit and arm, and the SAP's ANCOVA at the week chosen.
Lineage, SDTM to the tables
Where each ADaM variable comes from, as its program derives it (the rule and the program's lines), and what each table and app reads. Pick a variable to trace it both ways.
Results
Measured against a real submission
We test Clinwright on the public CDISC pilot study. Given only its SDTM, Clinwright configures and programs the ADaM and makes the CSR's tables, and we compare both with what was published.
98%
of ADSL's values identical to the published ADSL
18 of 29
CSR tables reproduced number for number
No AI
needed for these: derivation patterns alone, in one run
Data set by data set
| ADADAS | 99% |
|---|---|
| ADSL | 98% |
| ADVS | 98% |
| ADAE | 98% |
| ADNPIX | 96% |
| ADCIBC | 95% |
| ADTTE | 90% |
| ADLBC, ADLBH | 47%* |
* Half of the published laboratory records are “change from previous visit” parameters, which the standard template keeps in data sets of their own.
Counted over every value of the published data sets (records × the specification's variables), matched by key. Two more CSR tables agree on 97% and 99% of their rows.
Built for GxP
Work you can stand behind
Runs in a sandbox
Each run is isolated: no network, a read-only system, and limits on time and memory.
Reproducible
Every run records the engine version, the hashes of its inputs and the code itself.
Audit trail
Every change is kept: who made it, when, and the values before and after.
QC and electronic signatures
Programming, QC and review, with signatures that ask for the password again and record what they mean. A signature lapses when what it signed changes.
Your data stays yours
Each organization's data is kept apart down to the database. The AI runs with your organization's own Anthropic key, and everything else works without it.
Work together
Free to use, open to collaboration
Clinwright is an independent project, built to show what metadata-driven, AI-assisted statistical programming can do, from SDTM to the tables of a CSR. It is free, sells nothing, and is looking for people to take it further.
Use it
Create an account, invite your team and start a study on simulated data; everything there is public. Each organization makes up to 30 ADaM data sets and 300 TLFs a year, and the AI uses its own Anthropic key.
Get startedTry it on a finished study
Run a study you have already reported through Clinwright, on a server set up for it alone, and compare with your own programs: what matches, what needs a person, and how much time it saves.
Get in touchBuild it together
Statistical programming teams at sponsors and CROs, standards groups and researchers: features and templates developed together, or standards work such as CDISC's Analysis Results Standard.
Get in touchADaM data sets and TLFs count once each, when a program is first delivered having passed its checks; re-runs, new database releases and revisions are free, and so are people. A template's data sets and outputs count as usual: take part of a study to stay within a year's limits. With your own Anthropic key, Anthropic bills you for the AI directly. Public organizations nobody signs in to for a year are deleted after notice (terms).