TrialCurveUsing reconstructed survival data in HTA submissions
In HTA, reconstructed IPD from published KM curves is used whenever you need time-to-event data for a comparator trial you don't own: to fit and extrapolate parametric survival models, test proportional hazards, run network meta-analyses of survival outcomes, or perform population-adjusted indirect comparisons. Assessors accept it when the method is stated and the digitisation is shown to be validated.
Where pseudo-IPD is used
- Survival extrapolation for cost-effectiveness models (partitioned survival, state transition), following NICE DSU TSD 14 and the flexible methods of TSD 21.
- Network meta-analysis of time-to-event outcomes, including fractional-polynomial and piecewise models that need more than a single HR.
- Population-adjusted indirect comparisons (MAIC, STC, ML-NMR; TSD 18), where the comparator's outcomes usually come from its published KM curves.
- Checking proportional hazards before pooling HRs, and computing RMST when hazards are non-proportional.
The EU Joint Clinical Assessment
Under the EU HTA Regulation (2021/2282), joint clinical assessments apply to new oncology medicines and ATMPs from 12 January 2025, orphan medicines from 13 January 2028, and all new medicines from 13 January 2030. JCA scopes often include many PICOs, several of which can only be addressed with indirect comparisons. The EU practical guideline on direct and indirect comparisons discusses reconstructing pseudo-IPD from published KM curves. Assessors have flagged dossiers where the validation of the digitisation was not reported.
What to report
- The source figure and publication for each curve, and which endpoint and population it shows.
- The reconstruction method (for example Guyot 2012) and the software and version used.
- Validation: reconstructed versus published HR, medians and event counts, and agreement between independent extractions.
- Any deviations and how they were resolved, such as a misread table cell.
AI-use disclosure for JCA dossiers
The HTA Coordination Group's principles on using AI to prepare JCA dossiers (adopted July 2026) ask developers to disclose every AI-assisted step, including data extraction. The disclosure should name the tool, version, date and developer, keep prompts available on request, and keep a human responsible for each step. Every TrialCurve result includes an ai_use_disclosure block containing these fields, the prompt's hash and a ready-to-paste statement.
Every TrialCurve result carries this in its validation pack, together with the image hash, the engine and model versions, the at-risk table exactly as read, the agreement between the model's reading and the pixel measurement, and checks against the statistics printed on the figure.
FAQ
Does NICE accept reconstructed IPD?
Reconstructed IPD is widely used in NICE technology appraisals, particularly for comparators in survival extrapolation and indirect comparisons. Reviewers expect transparency about the digitisation and its validation.
What does the EU JCA expect for indirect comparisons?
The methodological and practical guidelines expect justified methods, consideration of effect modifiers, and transparent reporting of data sources. Reconstructed pseudo-IPD is discussed as a way to assess proportional hazards and to support time-varying comparisons.
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