Tutorial on application of survival analysis to the assessment of multiple state models in clinical data analysis.
Joachim Grevel
BAST Inc Ltd, Kington HR5 3DJ UK
Objectives:
Application of survival analysis methods to competing risks, multistate models and patient benefit.
.
Methods:
If a clinical trial is represented as a multistate model, dropout is a competing risk besides others, and only the administrative data cut-off acts as right-censoring. A multistate model can be an alternative approach to exposure-response models using Cox proportional hazards. Furthermore, sojourn times in certain states and transition intensities between states can identify patient benefit in different treatment arms.
Results:
Step-by-step guide through the methodology:
- a) Organising a clinical study into a multiple state system
- b) Counting transition frequencies between states
- c) Displaying transitions with non-parametric cumulative intensity transition functions and cumulative incidence functions
- d) Setting up parametric models for sojourn times and transition intensities and estimating model parameters [1].
- e) Selecting covariates and estimating their coefficients.
- f) Evaluating goodness-of-fit graphically.
- g) Testing hypotheses related to covariate effects including treatment effects.
Simulating sojourn times and transitions to demonstrate patient benefit.
Conclusions:
The methodology to assess clinical trials as multistate systems is not new [2]. Specifically in oncology, patient benefit should be quantitatively measured besides treatment efficacy. The proposed methodology and the published results will help patients, payers and caretakers make more informed decisions about potentially life-prolonging treatment options.
References:
[1] Asanjarani A, Liquet B, Nazarathy Y. Estimation of semi-Markov multi-state models: a comparison of the sojourn times and transition intensities approaches. Int. J. Biostat. https://doi.org/10.1515/ijb-2020-0083
[2] Weiss GH, Zelen M. A semi-Markov model for clinical trials. Journal of Applied Probability 1965; 2, No. 2: 269-285
