Skip to main content

Advertisement

Table 5 Reproducibility and validation of models

From: Reporting performance of prognostic models in cancer: a review

Topic % (n = 47) articles
Model validation included 34 (16)
Validation dataset*  
   Same data (bootstrap) 13 (6)
   Same population, new data** 23 (11)
   External (that is, new population setting) 11 (5)
   Larger series including original sample 0 (0)
Validation of models  
   Final model with same coefficients and variables 26 (12)
   Unclear reporting 9 (4)
Modifications suggested to model in light of validation? 0 (0)
  1. * Five studies included validation with more than one dataset.
  2. ** Methods used by studies were as follows: five studies used a random split, two used temporal split, five used cross validation, one used the jacknife method. Two studies used two and three methods respectively.