When either creatinine or eGFR change was not linear with respect to a continuous predictor, the predictor was transformed into a binary variable, the cut-off value being the value minimizing the Bayesian Information Criterion of the model.
Where similar clinical features were described by more than one significant predictor, the predictor with the higher level of significance in univariate analyses was entered into multivariate analyses, in order to avoid co-aggregation of predictors; e.g. baseline CIND rather than the baseline CAMCOG score as both these tests assess cognition.
We calculated performance metrics for each of the 120 predictors (40 predictors × three endpoints) and compared the estimated accuracy in the training set with the observed accuracy in the independent validation set.
We also compared the probability of selecting the correct model which we estimated as the proportion of cases where the true predictor or predictors (in the case of two true predictors) was selected and no other variables were selected.
If the predictor or predictors in question are modifiable, this may help to make treatment more effective (Adamson et al., 2009).
The algorithms are based on a predictor corrector implementation where the predictors reside in add-in FORTRAN routines and the existing nonlinear equation solvers in ASPEN PLUS equipment models serve as the correctors.
It is important to recognise that the use of multiple predictors assumes that the predictors are independent of each other.
Those predictors were also different from the predictors of being a positive role model and meritorious.
This shows that the comparison between different predictors is complicated, because the predictors were developed and optimized on different datasets.
Different studies used differing prediction targets, candidate predictors, and populations, affecting the predictors included in the final predictive models.
