TrialCurveAbout TrialCurve
TrialCurve turns published Kaplan–Meier figures into reconstructed patient-level survival data, with the checks an HTA reviewer expects. It is built for health economists, statisticians and evidence-synthesis teams who need time-to-event data from trials they do not own.
Who runs it
TrialCurve is based in London, UK. The operator and registered address are on the terms page; write to hello@trialcurve.com or use the contact form.
How the data are made
- Reading: Claude Opus 5.5, a vision model, reads the whole figure: every panel, axis, curve, numbers-at-risk table and printed statistic, zooming in where it needs to.
- Measurement: pixel tracing measures each curve independently. Where the two agree, the measured curve is used.
- Reconstruction: the Guyot (2012) algorithm rebuilds patient-level data that reproduce the curve and the at-risk table.
- Checks: hazard ratios and medians from the reconstruction are compared with those printed on the figure. A disagreement triggers a second, more thorough reading.
Results on published trials, and on simulated figures with known answers, are on the accuracy page.
The trial library
Trials in the library come from openly licensed (CC BY) publications. A curve is published only if it reproduces the statistics printed on its figure or, where none were printed, if an independent measurement confirms it. Reconstructed data approximate the published curves; the rows are not real patients.
Corrections
If a number looks wrong, tell us which page and which value. We check every report against the source figure, fix confirmed errors, and the page's "Updated" date shows when it changed.
Reconstruct IPD from your own figure. Upload a Kaplan–Meier plot and get curves, numbers at risk, pseudo-IPD, hazard ratios and a validation pack. Three figures a month are free.
Try the live demo Browse trial data API docs