The Supporting Information also reports a linear probability model of treatment discontinuation prior to 25 weeks and both logistic regression and linear probability models of treatment discontinuation prior to 16 weeks of treatment for the entire sample and prior to 32 weeks for those with prior TB treatment episodes.
The variables included in the Cox models for discontinuation of first treatment were sex, age group, TNF-antagonist, year of first treatment with a TNF-antagonist, and diagnosis (RA, ankylosing spondylitis, psoriatic arthritis, juvenile idiopathic arthritis, or others).
The variables included in the Cox models for discontinuation of the second treatment were the same plus the main reason of discontinuation of the first treatment.
The independent variables were grouped into four levels, in accordance with a theoretical hierarchical model for discontinuation of EBF.
The trial data showed that discontinuation rates were similar among all three interventions, justifying the assumption in the model of equal discontinuation rates for all active treatments.
The final effect was that the sum of model-specific estimates of discontinuation from all states in the model, including each type of adverse event, matches the annual CATIE phase 1 discontinuation rates for any cause.
In adjusted Cox proportional hazards models, the adjusted hazard of discontinuation was approximately 40% higher among SU initiators (adjusted hazard ratio [AHR] = 1.390, 95% CI = 1.363, 1.418) and TZD initiators (AHR = 1.402, 95% CI = 1.377, 1.427) compared with DPP-4i initiators during 1-year follow-up.
Because of this, investigators should examine the rates of discontinuation observed in Model 1 and Model 2 to determine their similarity.
A subanalysis used a Cox proportional hazards regression model to assess multivariate risk of discontinuation of specific first-line regimens in common use vs the EFV/FTC/TDF fixed-dose regimen.
For each 30-day interval, we fit a logistic model and estimated the probability of discontinuation or reinitiation given the most recent time-varying covariates in the 90 days prior to the 30-day interval in which discontinuation or reinitiation was thought to have occurred.
