Fit survival models to estimate exposure-specific hazards for G-computation approach

fit_model_0() fits a Cox model to estimate risk for unexposed individuals on the original time scale. Includes all individuals, censoring exposed individuals at their time of exposure. By default, model is adjusted for by <covariates>, included as simple linear terms.

fit_model_1() fits a Cox model to estimate risk for exposed individuals on the time scale of time since exposure. Includes exposed individuals who remain at risk tau days after exposure. Individuals are additionally censored at censor_time days after exposure to avoid extrapolation beyond the time period of interest. By default, model is adjusted for <covariates>, included as simple linear terms, and exposure time is included as a natural cubic spline with 4 degrees of freedom.

Usage

fit_model_0(
  data,
  outcome_time,
  outcome_status,
  exposure,
  exposure_time,
  covariates,
  formula_0 = NULL
)

fit_model_1(
  data,
  outcome_time,
  outcome_status,
  exposure,
  exposure_time,
  covariates,
  tau,
  censor_time = NULL,
  formula_1 = NULL
)

Arguments

data: A data frame with one row per individual containing the columns named in outcome_time, outcome_status, exposure, exposure_time, and any variables listed in covariates.
outcome_time: Name of the time-to-event/censoring variable. Time should be measured from a given time origin (e.g. study start, enrollment, or age) for all individuals.
outcome_status: Name of the event indicator. The underlying column should be numeric (1 = event, 0 = censored).
exposure: Name of the exposure indicator. The underlying column should be numeric (1 = exposed during follow-up, 0 = never exposed during follow-up).
exposure_time: Name of the time to exposure, measured from the chosen time origin; use NA if not exposed. Time must be measured in the same units (e.g. days) as that used for outcome_time.
covariates: Character vector of covariates to adjust for when fitting the hazard models. These covariates should include all known confounders of exposure and censoring measured at the chosen time origin.
formula_0: Optional right hand side of the formula for model 0. By default, uses covariates.
tau: Non-negative numeric value specifying the time after exposure that should be excluded from the risk evaluation period. This argument is primarily intended for vaccination exposures, where it is common to exclude the time after vaccination when immunity is still building. Time must be measured in the same units as that used for outcome_time and exposure_time and should reflect the biological understanding of when vaccine-induced immunity develops (usually 1-2 weeks). For non-vaccine exposures, tau can be set to 0 (no delay period).
censor_time: Time after exposure at which exposed individuals are censored during model fitting to prevent extrapolation. By default, no censoring is applied.
formula_1: Optional right hand side of the formula for model 1. By default, uses covariates plus natural spline of vaccination day (4 df). Default NULL

Value

A fitted coxph object with additional component $data containing the analysis dataset used for fitting:

For fit_model_0(): includes the survival tuple (Y, event) and covariates adjusted for in model, where Y is the time from time origin to first of endpoint, censoring or exposure time (for exposed individuals).
For fit_model_1(): includes the survival tuple (T1, event), <exposure_time>, and covariates adjusted for in model, where T1 is the time from exposure to endpoint or censoring, with additional censoring by censor_time. Only includes exposed individuals at risk tau days after exposure.