December 4, 2024
In restricted mean survival time (RMST) analyses, tau is set at some fixed point or as stated as this paper, treated as a nonrandom, prespecified quantity, sometimes derived from the observed data (Tian, 2020). They developed tau-inflated beta regression (τ-IBR) models for censored time-to-event data to provide a better understanding of association between predictors and τ-restricted time-to-event and also allow for more efficient estimation of τ-RMST values, finally also provide estimates of event-free probabilities.
They defined their beta regression model and they also defined using EM algorithm as well as using a semi-parametric MI Algorithm for missingness for model estimation of parameters. Using their MI Algorithm for the RMST is not restricted to the τ-IBR model they proposed. In simulations, they used various measures of bias to assess how well their τ-IBR compared to standard RMST models. In general their model performed better than the standard RMST model in various situations and with less bias. On a real dataset analysis from the Azithromycin for Prevention of COPD Exacerbations Trial where patients were randomized to Azithromycin or placebo for 1 whole year where they used different tau cutoffs, their model again performed well.
In discussion, they relate that the key advantages of using their method were to obtain better understanding of predictors associated with no event in the τ-restricted period of interest, as opposed to predictors associated with shorter expected event-free time amongst those who experienced the event and also more efficient estimation of restricted means due to properly modeling the point mass of min(τ,T) event at τ. This model became useful in their real data example since certain predictors shifted
Written by,
Usha Govindarajulu, PhD
Keywords: survival, RMST, tau, beta regression, censoring
References
Wang Y and Murray S (2024) “τ-Inflated Beta Regression Model for Estimating τ-Restricted Means and Event-Free Probabilities for Censored Time-to-Event Data” Biometrical Journal.
https://doi.org/10.1002/bimj.70009
https://onlinelibrary.wiley.com/cms/asset/3350959f-9b4b-4c6c-b51b-bc5142957436/bimj70009-fig-0002-m.jpg