Portfolio · Project 07 · Clinical and economic evidence · October 2026
From survival curve to cost-effectiveness
Almost every access decision for a cancer medicine rests on three questions. How much longer do patients live? What happens after the trial stops following them? Is the extra survival worth the price? These three hands-on tools walk through each one with a hypothetical medicine, Drug Y, and simulated data you can change and re-run.
A simulated trial: Drug Y vs standard care
Change the trial
The curve
A Kaplan–Meier curve starts at 100% and steps down each time a patient dies. Patients still alive when last seen are censored: they count as “at risk” until then, and are then removed, shown as ticks. The curve’s tail rests on few patients, so it is the least reliable part.
The median
Median survival is the time at which the curve crosses 50%. If fewer than half the patients have died by the data cut-off, the median is not reached (NR). Medians describe one point on the curve, not the whole curve.
The hazard ratio
The hazard ratio compares the risk of death at any moment among patients still alive. An HR of 0.70 means about 30% lower risk on Drug Y, at every point in time, if hazards are proportional. The same reading applies to progression-free survival, where the event is progression or death.
Observed data, then 20 years of prediction
The data you have
Models are fitted to the simulated patient-level data by maximum likelihood, the way an analyst would. The NICE DSU list also includes the generalised gamma, not shown here.
Why the tails split
Each model makes a different assumption about how the risk of death changes over time: constant (exponential), steadily rising or falling (Weibull, Gompertz), or rising then falling (log-normal, log-logistic). They can all fit three years of data well and still disagree by years about the mean.
Why the mean matters
Cost-effectiveness models use mean survival, the area under the whole curve, not the median. Most of the difference between models sits in the extrapolated tail, so the model choice can drive the ICER more than the trial result does.
How analysts choose
NICE DSU guidance asks analysts to fit all the standard models and compare them with statistical fit (AIC, BIC) and diagnostic plots. Crucially, it also asks them to check the tails for clinical plausibility against external data and expert opinion, since good fit to observed data says little about the tail.
Where patients are, month by month
Country and threshold
Inputs (all hypothetical)
Survival in both arms follows exponential curves with these medians, run monthly over 20 years. Prices and costs are illustrative and in the selected country’s currency; they are not converted between countries. This is a teaching model, not a submission model.
Partitioned survival
The two survival curves split the cohort into three states. Below the PFS curve: progression-free. Between PFS and OS: progressed. Above OS: dead. Each month, the time spent in each state collects costs and quality of life.
QALYs and the ICER
A QALY is a year of life weighted by its quality (1 = full health). The incremental cost-effectiveness ratio is the extra cost divided by the extra QALYs. Agencies compare it with a threshold or reference value; above it, the price usually has to come down.
Thresholds differ
England uses a range and a severity weighting; the Netherlands sets reference values by burden of disease; Ireland, Japan and ICER in the US each use their own figures and for different purposes. The comparison panel shows the price at which Drug Y would meet each one.
Sources
- Latimer N. NICE DSU Technical Support Document 14: Survival analysis for economic evaluations alongside clinical trials, extrapolation with patient-level data. 2011, updated 2013. NICE DSU
- NICE. Changes to NICE’s cost-effectiveness thresholds confirmed (standard range £25,000 to £35,000 per QALY gained from April 2026, replacing £20,000 to £30,000; applies to technology appraisals, not highly specialised technologies). nice.org.uk; NICE in depth
- NICE. Health technology evaluations: the manual (PMG36), including the severity modifier. nice.org.uk
- Office of Health Economics. NICE’s severity modifier (QALY weights 1.2 and 1.7 and the shortfall cut-offs). ohe.org
- Zorginstituut Nederland. Ziektelast in de praktijk, 7 May 2018 (reference values €20,000, €50,000 and €80,000 per QALY by burden of disease). Parliamentary copy
- Almajed S et al. Cost-effectiveness evidence on approved cancer drugs in Ireland. Eur J Health Econ 2022;23 (€20,000 and €45,000 per QALY from the 2016 agreement). doi:10.1007/s10198-021-01365-2
- Formal implementation of cost-effectiveness evaluations in Japan. Value Health 2020;23(1) (¥5 million per QALY for post-listing price adjustment). doi:10.1016/j.jval.2019.10.005
- Institute for Clinical and Economic Review. 2019 perspectives on cost-effectiveness threshold ranges ($100,000–$150,000 per QALY for health-benefit price benchmarks). icer.org
Be curious, not judgmental.
Design inspired by Spencer Gabor and Dieter Rams’s ten principles of good design.
© 2026 Jigar Rajpura. All rights reserved. No part of this site may be reproduced, distributed or adapted without written permission, except as set out in the Terms of use.
DisclaimerPersonal project, not sponsored by or affiliated with any employer (current or former) or any university I attended; views are my own. Developed with AI assistance, so errors or outdated details may exist. All data, medicines and prices on this page are hypothetical; thresholds reflect the cited sources as of October 2026 and may change. Not medical, regulatory, reimbursement or investment advice. Full disclaimer