Skip to main content

Advertisement

Table 4 Issues in model developmenta

From: Developing risk prediction models for type 2 diabetes: a systematic review of methodology and reporting

Variables Data
Sample size, median (IQR)  
   Development cohortb 2,562 (1,426 to 4,965)
   Validation cohortsc 1,895 (1,253 to 4,398)
Treatment of continuous risk predictors, n (%)  
   All kept continuous 13 (30%)
   All categorised/dichotomised 21 (49%)
   Some categorised, some not 6 (14%)
   Unclear 3 (7%)
Treatment of missing data, n (%)  
   Not mentioned 16 (41%)
   Complete case 21 (54%)
   Multiple imputation 1 (3%)
   Other (for example, surrogate splitter for regression trees) 1 (3%)
Model-building strategy, n (%)  
   Stepwise, forward selection, backward elimination 20 (51%)
   All significant in univariate analysis 2 (5%)
   Other 12 (31%)
   Unclear 5 (13%)
Overfitting mentioned or discussed, n (%) 5 (13%)
  1. aIQR, interquartile range; bsample size not reported in four studies; csample size not reported in two studies and unclear in one study.