Hypothesis testing: Adjusting for covariates in discrete time models

Hi everyone,

I’m using a discrete time model (Gompertz model/logistic regression using cloglog-Link function which equals the Cox interval-censored model) to estimate the effect of a treatment on the uptake of aftercare. I have an RCT with 5 measurement points (baseline, 3 weeks, 6 weeks, 3 months, 6 months after randomization). My intention is to test the hypothesis, that the type of treatment has an effect on the uptake of aftercare (intervention vs control/TAU).

In Cox PH models, I know that adjusting for as many covariates as possible (without overfitting) is the best possible way to test a hypothesis and to get the most accurate values for hypothesis testing. In how far is this the case in the discrete time model using Gompertz regression? Should I also adjust for as many covariates as possible (theoretical background assured)?

Assuming the covariates are measured at time zero or before and that pathways are also respected otherwise, the benefits of adjusting for more covariates that are potentially associated with output are quite general. That is, up to the point of overfitting. Think of the simplest situation which is the binary logistic regression model. Accounting for outcome heterogeneity get’s the exposure’s odds ratio farther from 1.0 and more correct.