We used backward-selection logistic regression to systematically evaluate and remove variables that did not significantly contribute to the overall model (p ≥0.10).
A backward selection multiple logistic regression analysis was performed by first including all parameters in a multivariate model and subsequently leaving out the parameters with the largest p values until no parameter with a p value greater than 0.25 was included.
Potential variables for inclusion in the propensity score (age at baseline, BMI, systolic blood pressure, low-density lipoprotein (LDL -cholesterol and years of education) were expLDL -cholesterolc regression with andackward selection procedure (p<0.25 as selection criterion).
Next, we performed multivariable logistic regression with a backward selection procedure, i.e., stepwise deletion of the variables that contributed least to the model that predicts use of CAM practitioners until all remaining variables contributed significantly at p < 0.05 level.
Variables associated with non-compliance with P<0.20 were entered in a multivariate-adjusted logistic regression model with a backward selection procedure and a significance level of P=0.05.
The multivariate logistic regression analysis used a backward selection process to eliminate nonsignificant variables from the model (criteria for elimination from the model set at p > 0.05).
Covariates found to be associated with attributable mortality on univariate analysis at a level of significance P < 0.1 were eligible for inclusion in a multivariate logistic regression model using a backward selection procedure.
Covariates found to be associated with bacteremia on univariate analysis at a level of significance P < 0.2 were eligible for inclusion in a multivariate logistic regression model using a backward selection procedure [ 21].
Variables associated with 'referral to a specialist' with P < 0.20 in univariate analysis were entered in a multivariate adjusted logistic regression model with a backward selection procedure and a P = 0.05 significance level to identify factors influencing 'referral to a specialist.' Odds ratios (OR) and 95% confidence intervals (CI) were also estimated.
As we hypothesized that using different cut offs or a different definition for erosive joints (according to the SENS method) may affect the predictive ability of the prediction rule, data on the number of erosive joints were added to the logistic regression model with a backward selection procedure that was used to derive the prediction rule [ 7].
A logistic regression model was built using a backward selection algorithm and SNPs nominally associated with nephropathy in our population.
