Table 12.

Comparisons among the approaches

VariableCox Regression with Baseline ExposureCox Regression with Time-Varying ExposureMSM
Effects estimated“Long-term” effects“Short-term” effectsEffects of the whole exposure history
Correctly estimate effects of time-updated exposure when time-dependent confounding is presentN/ANoYes
AssumptionsNoninformative censoring, proportional hazardNoninformative censoring, proportional hazardNoninformative censoring, proportional hazard, exchangeability, positivity, and correct specification of weight models
ProgrammingEasyEasyTedious
  • MSM, marginal structural model; N/A, not applicable.