Estimating curvilinear time-varying treatment effects: Combining g-estimation of structural nested mean models with time-varying effect models for longitudinal causal inference. | Synapse