Skip to main content

Table 9 Estimands and performance measures for each of the article’s simulation studies

From: A scoping methodological review of simulation studies comparing statistical and machine learning approaches to risk prediction for time-to-event data

  

Estimands

Performance measures

Selection for time t

S(t|x)

h(t|x)

Linear predictor: η(x)

Restricted Mean Survival Time (RMST)

MSPE

C-Index*

Integrated Brier Score

Geng et al. (2014) [26]

1/5th,…,5/6th quantiles of training survival times

✓

   

✓

  

Golmakani et al. (2020) [31]

N/A

  

✓

  

✓

 

Gong et al. (2018) [27]

Unclear

✓

    

✓

 

Hu and Steingrimsson (2018) [28]

25th, 50th and 70th quantile of training marginal survival times

✓

   

✓

  

Katzman et al. (2018) [29]

N/A for linear predictor; Unclear for restricted mean survival and C-index

  

✓

✓

✓

✓

 

Lowsky et al. (2012) [25]

T = 5 with step size of 0.25

✓

     

✓ **

Omurlu et al. (2009)*** [24]

      

✓

 

Steingrimsson and Morrison (2020) [32]

S(t): t = median marginal failure time; RMST: Ï„ = 85th quantile of marginal observed times

✓

  

✓

✓

  

Wang and Li (2019) [30]

Unclear

✓

✓

   

✓

✓

Xiang et al. (2000) [23]

Unclear

✓

    

✓

 
  1. *A specified value for t for the C-index is not always required — if the model assumes proportional hazards then the C-index should remain the same regardless of time point
  2. **Lowsky et al. (2012) [25] used the Integrated Brier Score with added inverse probability of censoring weights. This is referred to as the IPEC in the paper
  3. ***Omurlu et al. (2009) [24] were unclear in what the estimands were for their simulation study