All available covariates have been considered as potential risk factors and have been included in the multiple models using the forward stepwise approach, i.e. the variables were sequentially added to an "empty" (intercept only) model one at a time giving priority to those that had shown the strongest evidence of association at the univariate stage (smallest P value).
Univariable analysis was performed to examine the association of each risk factor with stillbirth or very low Apgar scores and factors which reached statistical significance at P < 0.1 were entered into the multivariable model using the forward stepwise approach.
"Not excellent" was the referent group Model I investigates the individual PS items controlling for self-rated health, treatment history, stage at diagnosis and age Model II investigates the overall PS item controlling for self-rated health, treatment history, stage at diagnosis and age The results of both models were confirmed using the forward stepwise approach.
Using the forward selection approach, we selected five independent marginal SNPs (i.e. rs3733588, rs874432, rs4697695, rs9291683 and rs11734783) capturing most marginal effects of the region, where the first three are within SLC2A9 and the latter two are intergenic between WDR1 and ZNF518B.
Using the forward selection approach described earlier, six independent imputed SNPs were selected: rs938558, rs4428284, rs4697695 (also typed), rs10489074, rs4481233 (also typed) and rs16895984 (also typed), where rs10489074 and rs16895984 are intergenic between WDR1 and ZNF518B and the rest are within SLC2A9.
Using the forward stepwise LDA approach, we generated and tested the models derived from all possible gene pairs independently for each of the 10 training sets.
Genome-wide QTL scans were performed using the forward selection interval mapping approach (Carlborg et al., 2005).
The proposed motion planning approach using the forward kinematics and the perturbation method can be applied to solve the path planning problem with geometrical singularity and task priority requirements, such as obstacle avoidance and singularity avoidance.
A full model was constructed with these covariates using the forward addition and backward deletion approach.
Although there are very efficient coalescent-based tools to simulate data, we preferred to use the forward simulation approach because it exactly models the solution to the forward diffusion Equation 3 and because it can be adapted to incorporate the effects of natural selection more easily than coalescent-based models.
