These 3 factors remained significantly and independently more associated with ST-45 infection than with other sequence types when combined in a multivariate model (tested by using likelihood ratio tests) (Table 2).
Although the presence of microalbuminuria was different among HbA1c quintiles in univariate analysis, this difference was not maintained in various multivariate models tested.
In our study, whereas it was consistently associated with DFS and DDFS, it was not retained as an independent prognosticator in all multivariate models tested.
The final multivariate model showed tests of standing (IRR 2.28, 95% CI 1.54 – 3.36) and leaning balance (IRR 1.78, 95% CI 1.20 – 2.62) and a measure of depressive symptomatology (IRR 2.31, 95% CI 1.59 – 3.36) to be significant and independent predictors of falling.
Model fit for each multivariate model was tested using the Hosmer-Lemeshow test for goodness of fit [ 22] developed for testing logistic modelling.
As outlined by Hosmer and Lemeshow (9 ), the 5 steps of purposeful selection are 1) test for univariate significance (p<0.25); 2) build the multivariate model (p<0.05); 3) test for confounding; 4) assess continuous variables for linearity; and, 5) test for effect modification and include interaction terms that are both significant (p<0.05) and biologically plausible.
For each individual variable, the effect of the variable on treatment outcome in the multivariate model was tested by the Wald chi-square test.
Hosmer Lemeshow goodness of fit of the multivariate model was tested on each imputed dataset, with p values ranging from 0.22 to 0.96 CI confidence interval, HSCT hematopoietic stem cell transplantation, ICU intensive care unit, RBC red blood cell, OR odds ratio, SOFA sequential organ failure assessment.
An interaction variable between the difference in wait time and the dummy variable "period" was included in the multivariate model to test the hypothesis that the publication of wait time information led to more patients selecting the site with shorter wait time.
Stratified analysis and multivariate model interaction testing did not yield any significant effect modifier.
