Mixture Cure Models in Oncology tutorial (Felizzi et al.)
This tutorial explains how to use mixture cure models to estimate long-term survival in cancer patients, especially when some patients are expected to be 'statistically cured'. It shows two approaches: one that estimates the cure rate from trial data and another that uses real-world data to inform the cure rate. The methods are implemented in R with free, accessible code.
At a glance
Use when
Estimating long-term survival in cancers where a cure fraction is plausible (e.g., early-stage cancers, immunotherapies with durable responses); conducting economic evaluations of novel oncology therapies
Avoid when
Survival curves show no evidence of plateauing; when all patients are expected to relapse; in aggressive cancers with no curative potential
Inputs
Clinical trial survival data, external real-world data (for informed approach), patient-level or aggregate survival data
Outputs
Estimated cure fraction, long-term survival curves, model parameters (e.g., latency and cure components), survival extrapolations beyond trial follow-up
How it works
The tutorial presents mixture cure models that combine a cured fraction and a non-cured fraction of patients in survival analysis. It covers data integration, maximum likelihood estimation, model fitting, and interpretation. Two models are demonstrated: an 'uninformed' approach estimating the cure fraction from clinical trial data, and an 'informed' approach using external data (e.g., real-world evidence) as input for the cure fraction. Implementation is done in R, with reproducible code available on GitHub.
- HTA domains
- Clinical Effectiveness, Costs & Economic Evaluation, Patient and Social Aspects
- Assumptions
- A subset of patients is 'statistically cured' and experiences general population mortality rates; the survival curve reaches a plateau; the model form (e.g., parametric distribution) is correctly specified
- Strengths
- Improves long-term survival predictions in curative settings; flexible in incorporating external evidence; publicly available R code enhances reproducibility; useful for health technology assessment and economic modeling
- Limitations
- Requires careful model selection and validation; assumptions about cure may not hold in all cancers; limited applicability when no plateau is observed in survival data; depends on data quality and availability
- Also known as
- Mixture Cure Model Tutorial, Felizzi et al. tutorial on cure models
Questions this answers
- › How can long-term survival be estimated when some patients are statistically cured?
- › What is the difference between 'uninformed' and 'informed' mixture cure models?
- › How can external data be incorporated into cure models?
- › How are mixture cure models implemented in R?
- › How do mixture cure models improve survival extrapolation in oncology?
- › What are the implications of using cure models in health economic evaluations?
References & sources
Similar by meaning
- Mixture and Non-mixture Cure Models for HTA (Gibson et al. guidance)
- curesurv
- Survival extrapolation incorporating general-population mortality using excess-hazard and cure models (Sweeting et al. tutorial)
- Palmer flexible survival model selection algorithm for immunotherapies
- Survival extrapolation validation-based case study (Bullement et al.)
Beta record. Generated from the primary source via AI extraction and independent audit, pending final human review.