TrialCurveKaplan–Meier digitizer and IPD reconstruction tools compared (2026)
Summary: WebPlotDigitizer and IPDfromKM remain the standard free, manual pipeline. SurvdigitizeR, KM-GPT and KMGen automate parts of it for research use. TrialCurve automates the whole workflow, from image to validated IPD, as a supported API with a published benchmark. The table reflects each tool's public materials as of September 2026.
| Tool | Type | Digitising | At-risk table | IPD | Cost / access |
|---|---|---|---|---|---|
| WebPlotDigitizer | Web app (free) | Manual / semi-automatic | Manual | No (use another tool) | Free |
| IPDfromKM | R package + Shiny app | Click points (manual) | Manual entry | Yes (Guyot) | Free, open source |
| reconstructKM | R package | None (needs coordinates) | Manual entry | Yes | Free, open source |
| SurvdigitizeR (2024) | R package / Shiny | Automatic | Not read | Via Guyot | Free, research |
| KM-GPT (2025) | Hosted app + paper | Automatic (GPT + CV) | Automatic | Yes | Free; bring your own OpenAI key; non-commercial licence |
| KMGen (2026) | Open code (paper) | Automatic (Claude + CV) | Automatic | Yes | Free, research code |
| EasyHTA | Commercial HTA platform | You supply coordinates | Manual | Yes, plus extrapolation | Subscription (from about €959/yr) |
| TrialCurve | API + web app | Automatic: whole multi-panel figures, read by Claude with a zoom tool and confirmed by pixel measurement | Automatic | Yes, plus HR/median/RMST, checks against printed statistics and a validation pack | 3 free/month, then per figure; team plans; free for academic methods work |
How to choose
- Occasional academic use, no budget: WebPlotDigitizer plus IPDfromKM is proven, if slow.
- Research on automation methods: SurvdigitizeR, KM-GPT and KMGen publish their code and methods.
- Regular HEOR or HTA work: you need speed, reproducibility and a validation record you can put in a dossier, which is TrialCurve's focus.
Published accuracy
Studies report accuracy on different figure sets with different metrics, so compare with care. Manual digitisation plus Guyot achieved a mean HR error of 0.0094 across 118 curves (Saluja et al. 2019). KM-GPT reported a median absolute survival error of 0.005 on 540 synthetic plots. KMGen reported a mean integrated absolute error of 0.0151 on 32 plots. TrialCurve's full benchmark, with production, model-only and measurement-only results on the same figures and checks against figures from 40 published trials, is on the accuracy page.
FAQ
Which tool do HTA bodies expect?
None is mandated. Assessors expect the method (usually Guyot) to be named and the digitisation to be validated and reported, for example by comparing reconstructed and published HRs and medians.
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