TABLE 4.

Noncausal associations and estimated causal effects of each additional month until switching therapy on CD4 T-cell count 8 months after loss of suppression among subjects with virologic failure, sampled from a cohort of HIV^*-infected persons in San Francisco, California, 2000–2004^†

Model and term	Estimate (cells/μl) per month of waiting to switch therapy	95% confidence interval‡
Regression-based estimates (residual confounding)
Model: E(Y(8)|C) = β0 + β1C
β1	4.8	−8.1, 17.9
Model: E(Y(8)|C,CD4(0) = β0 + β1C + β2CD4(0) + β3C × CD4(0)
β1	−9.5	−19.3, 0.2
β3	0.05	0.02, 0.08
Marginal structural model-based estimates
Model: E(Yc(8)) = β0 + β1c
β1	−4.8	−20.3, 10.6
Model: E(Yc(8)|CD4(0) = β0 + β1c + β2CD4(0) + β3c × CD4(0)
β1	−11.3	−21.3, −1.3
β3	0.05	0.02, 0.08

^*HIV, human immunodeficiency virus.

^†“C” and “c” denote, respectively, the observed and counterfactual numbers of months after baseline (virologic failure) at which the participant switched therapy from the original nonsuppressive regimen. “CD4(0)” denotes the observed CD4 T-cell count at baseline (time of virologic failure).

^‡Based on robust standard error estimates using generalized estimating equations.