Logistic regression was used to model variables associated with undetectable RNA PCR viral load.
In both models adjusted odds ratios (AORs), 95% confidence intervals (95%CI) and a test for the overall significance of the model variables with more than 2 levels (Type 3 analysis of effects test) were computed to assess pairwise comparisons.
In this model, variables with P-value <0.1 were entered.
Initial variables were selected by developing univariate models; variables with Wald p>0.2 were excluded from further analysis.
For the final multivariable model variables with p-value > 0.05 were excluded with backward elimination procedure [ 21].
This test was initially performed by including in the model variables with a univariate likelihood ratio p value < 0.20.
Covariates as described above were entered stepwise in the model; variables with a significance level higher than 0.1 were removed, as they did not contribute to the model.
Initially, associations between potential covariates were assessed by univariate log-binomial regression models; variables with p < 0.2 were included in the full model.
In further models, variables with a P < 0.10 in univariate linear regression analyses for associations with the CER were selected for multivariate analysis.
On the other hand, lasso and adaptive elastic net demonstrated better performance in terms of percentage of model variables with inclusion frequency over 0.9 (Fig. 2).
