This was followed by a stepwise logistic regression with backward elimination to identify a multivariate predictor model.
In addition, we would like to point out that the 6-week pain intensity predictor alone had better ability to predict future pain score (R = 0.350 in the test set) compared to the full multivariate predictor model (R = 0.261 in the test set).
Final selection of multivariate predictors (model 4) was determined with stepwise analysis as a backward-stepping procedure based on a likelihood ratio test with a P value greater than 0.10 used for exclusion from the model.
The parameter estimates from the most efficient predictor multivariate models were used to construct subject-specific predictor scores.
Where both the Charlson score and one of its components were significant univariate predictors, additional multivariate models were created with only one of these two predictors, to ensure that colinearity did not mask underlying relationships.
Using these predictors, saturated multivariate models were constructed and subsequently reduced by backward elimination techniques.
This was done with the aim of deepening our understanding of the probable causalities and explanatory powers of predictors in multivariate models, but not with the aim of generating a predictive equation [69].
The effects of sociodemographic factors such as age or place of residence as well as Internet use were not maintained as statistically significant predictors in multivariate models.
New approaches in exposure assessment that consider physiologic parameters such as BMI and lactation were developed by Wolff et al. (2005) with the use of first-order pharmacokinetic and predictor-based multivariate models.
