Flexsurv Methods to Estimate Cure: An Exploration of Methods for Modelling Cure
Speaker(s)
Venkada Subramaniyan S1, Clark L2, Choudhury N1, van Hest N1
1Costello Medical, London, UK, 2Costello Medical, Manchester, UK
Presentation Documents
OBJECTIVES: Mixture cure models (MCM) estimate survival assuming a population consists of cured and uncured patients. Different MCM methods exist with no guidance on method selection, although literature recommends incorporating background mortality for cured patients to capture non-disease-specific mortality. This research aimed to determine whether results are sensitive to method choice by comparing outputs from several different MCM methods.
METHODS: Two trials (ZUMA-1; ELIANA) investigating the efficacy of chimeric antigen receptor T-cell (CAR-T) treatments with different populations (adult; paediatric) and follow-up duration (63.1 and 38.8 months) were used as case studies. Pseudo individual patient data were estimated using the Guyot algorithm on digitized survival data. MCMs were fitted using the methods in R packages flexsurvcure and cuRe. Overall, 12 MCM methods were compared by varying three link functions and exclusion/inclusion of UK-specific mortality.
RESULTS: Using a logistic link function, both packages estimated similar cure fractions to each other (range across distributions: 38–44% for ZUMA-1 and 27–60% for ELIANA). Incorporating background mortality increased the cure fractions by 2–5% across distributions in both packages for ZUMA-1 but had no impact for ELIANA. Varying the link function to probit and cloglog had negligible impact (<0.001%) on cure fractions, despite literature suggesting the link function should reflect the disease and biological mechanism of action.
CONCLUSIONS: This study exhibited some of the different available options when fitting MCMs. Estimated cure fractions differed between parametric distributions but were aligned across packages and link functions. The inclusion of background mortality only impacted the cure fraction in the ZUMA-1 adult population; this is expected given the low background mortality of the paediatric population of ELIANA. With increasing evidence demonstrating the curative potential of treatments, an effort to consolidate existing guidance and research on MCMs is required.
Code
MSR53
Topic
Methodological & Statistical Research
Disease
Genetic, Regenerative & Curative Therapies, Oncology