Analyze a stepped wedge dataset

Usage

analyze(
  dat,
  method = "mixed",
  estimand_type = "TATE",
  estimand_time = c(1, attr(dat, "n_seq")),
  exp_time = "IT",
  cal_time = "categorical",
  family = stats::gaussian,
  re = c("clust", "time"),
  corstr = "exchangeable",
  offset = NULL,
  n_knots_exp = 4,
  n_knots_cal = 4
)

Arguments

dat: A dataframe containing the stepped wedge trial data.
method: A character string; either "mixed", for a mixed-effects model, or "GEE", for generalized estimating equations.
estimand_type: One of c("TATE", "PTE"); "TATE" represents the time-averaged treatment effect and "PTE" represents the point treatment effect.
estimand_time: An integer vector of length 1 or 2. When estimand_type="TATE", `estimand_time` must be a numeric vector of length 2, representing the start and end times of the exposure time period to average over. When estimand_type="PTE", `estimand_time` must be a numeric vector of length 1, representing the time period of interest. See examples.
exp_time: One of c("IT", "ETI", "NCS", "TEH"); model for exposure time. "IT" encodes an immediate treatment model with a single treatment effect parameter. "ETI" is an exposure time indicator model, including one indicator variable for each exposure time point. "NCS" uses a natural cubic spline model for the exposure time trend. "TEH" includes a random slope term in the model, allowing the treatment effect to vary by timepoint.
cal_time: One of c("categorical", "NCS", "linear", "none"); model for calendar time. "categorical" uses indicator variables for discrete time points, as in the Hussey and Hughes model. "NCS" uses a natural cubic spline, useful for datasets with continuous time. "linear" uses a single slope parameter. "none" assumes that there is no underlying calendar time trend.
family: A family object; see documentation for `glm`.
re: A character vector of random effects to include; only relevant if method="mixed" is used. Possible random effects include "clust" (random intercept for cluster), "time" (random intercept for cluster-time interaction), "ind" (random intercept for individuals; appropriate when a cohort design is used), "tx" (random treatment effect)
corstr: One of c("independence", "exchangeable", "ar1"); only relevant if method="GEE" is used. Defines the GEE working correlation structure; see the documentation for `geepack::geeglm`.
offset: A linear predictor offset term; see docs for `lme4::lmer`.
n_knots_exp: An integer; only relevant when exp_time="NCS". Specifies the number of knots to use for exposure time, including boundary knots. The spline basis includes an intercept, and the degree of the basis is equal to the number of knots.
n_knots_cal: An integer; only relevant when exp_time="NCS". Specifies the number of knots to use for calendar time, including boundary knots. The spline basis includes an intercept, and the degree of the basis is equal to the number of knots.

Value

A list with the model object, model type as a string, estimand type as a string, numeric treatment effect estimate, numeric treatment effect standard error, treatment effect 95 dataframe with treatment effect at each exposure timepoint.

Examples

# Load data
test_data <- load_data(time ="period", cluster_id = "cluster", individual_id = NULL,
treatment = "trt", outcome = "outcome_cont", data = sw_data_example)

# Analysis example 1: TATE estimand for exposure times 1 through 4
results_tate <- analyze(dat = test_data, method = "mixed", estimand_type = "TATE",
estimand_time = c(1, 4), exp_time = "ETI")

results_tate

# Analysis example 2: PTE estimand for exposure time 3
results_pte <- analyze(dat = test_data, method = "mixed", estimand_type = "PTE",
estimand_time = 3, exp_time = "ETI")

results_pte