Package 'brms.mmrm'

Description

The mixed model for repeated measures (MMRM) is a popular model for longitudinal clinical trial data with continuous endpoints, and brms a is powerful and versatile package for fitting Bayesian regression models. The brms.mmrm R package leverages brms to run MMRMs, and it supports a simplified interfaced to reduce difficulty and align with the best practices of the life sciences.

References

• Bürkner, P.-C. (2017), "brms: An R package for Bayesian multilevel models using Stan," Journal of Statistical Software, 80, 1–28. https://doi.org/10.18637/jss.v080.i01.

• Holzhauer, B., and Weber, S. (2024), "Bayesian mixed effects model for repeated measures," in Applied Modeling in Drug Development, Novartis AG. https://opensource.nibr.com/bamdd/src/02h_mmrm.html.

• Mallinckrodt, C. H., Lane, P. W., Schnell, D., and others (2008), "Recommendations for the primary analysis of continuous endpoints in longitudinal clinical trials," Therapeutic Innovation and Regulatory Science, 42, 303–319. https://doi.org/10.1177/009286150804200402.

• Mallinckrodt, C. H., and Lipkovich, I. (2017), Analyzing longitudinal clinical trial data: A practical guide, CRC Press, Taylor & Francis Group.

Description

Create a cell-means-like informative prior archetype with a special fixed effect to represent the average across time.

Usage

brm_archetype_average_cells(
  data,
  intercept = FALSE,
  baseline = !is.null(attr(data, "brm_baseline")),
  baseline_subgroup = !is.null(attr(data, "brm_baseline")) && !is.null(attr(data,
    "brm_subgroup")),
  baseline_subgroup_time = !is.null(attr(data, "brm_baseline")) && !is.null(attr(data,
    "brm_subgroup")),
  baseline_time = !is.null(attr(data, "brm_baseline")),
  covariates = TRUE,
  prefix_interest = "x_",
  prefix_nuisance = "nuisance_"
)

Arguments

Details

This archetype has a special fixed effect for each treatment group to represent the mean response averaged across all the time points.

To illustrate, suppose the dataset has two treatment groups A and B, time points 1, 2, and 3, and no other covariates.

Let mu_gt be the marginal mean of the response at group g time t given data and hyperparameters. The model has fixed effect parameters beta_1, beta_2, ..., beta_6 which express the marginal means mu_gt as follows:

  `mu_A1 = 3 * beta_1 - beta_2 - beta_3`
  `mu_A2 = beta_2`
  `mu_A3 = beta_3`

  `mu_B1 = 3 * beta_4 - beta_5 - beta_6`
  `mu_B2 = beta_5`
  `mu_B3 = beta_6`

For group A, beta_1 is the average response in group A averaged across time points. You can confirm this yourself by expressing the average across time (mu_A1 + mu_A2 + mu_A3) / 3 in terms of the ⁠beta_*⁠ parameters and confirming that the expression simplifies down to just beta_1. beta_2 represents the mean response in group A at time 2, and beta_3 represents the mean response in group A at time 3. beta_4, beta_5, and beta_6 are analogous for group B.

Value

A special classed tibble with data tailored to the cell-means-like time-averaged archetype. The dataset is augmented with extra columns with the "archetype_" prefix, as well as special attributes to tell downstream functions like brm_formula() what to do with the object.

Prior labeling for brm_archetype_average_cells()

Within each treatment group, the initial time point represents the average, and each successive time point represents the response within that actual time. To illustrate, consider the example in the Details section. In the labeling scheme for brm_archetype_average_cells(), you can label the prior on beta_1 using brm_prior_label(code = "normal(1.2, 5)", group = "A", time = "1"). Similarly, you cal label the prior on beta_5 with brm_prior_label(code = "normal(1.3, 7)", group = "B", time = "2"). To confirm that you set the prior correctly, compare the brms prior with the output of summary(your_archetype). See the examples for details.

Nuisance variables

In the presence of covariate adjustment, functions like brm_archetype_successive_cells() convert nuisance factors into binary dummy variables, then center all those dummy variables and any continuous nuisance variables at their means in the data. This ensures that the main model coefficients of interest are not implicitly conditional on a subset of the data. In other words, preprocessing nuisance variables this way preserves the interpretations of the fixed effects of interest, and it ensures informative priors can be specified correctly.

Prior labeling

Informative prior archetypes use a labeling scheme to assign priors to fixed effects. How it works:

1. First, assign the prior of each parameter a collection
  of labels from the data. This can be done manually or with
  successive calls to [brm_prior_label()].
2. Supply the labeling scheme to [brm_prior_archetype()].
  [brm_prior_archetype()] uses attributes of the archetype
  to map labeled priors to their rightful parameters in the model.

For informative prior archetypes, this process is much more convenient and robust than manually calling brms::set_prior(). However, it requires an understanding of how the labels of the priors map to parameters in the model. This mapping varies from archetype to archetype, and it is documented in the help pages of archetype-specific functions such as brm_archetype_successive_cells().

See Also

Other informative prior archetypes: brm_archetype_average_effects(), brm_archetype_cells(), brm_archetype_effects(), brm_archetype_successive_cells(), brm_archetype_successive_effects()

Examples

set.seed(0L)
data <- brm_simulate_outline(
  n_group = 2,
  n_patient = 100,
  n_time = 4,
  rate_dropout = 0,
  rate_lapse = 0
) |>
  dplyr::mutate(response = rnorm(n = dplyr::n())) |>
  brm_data_change() |>
  brm_simulate_continuous(names = c("biomarker1", "biomarker2")) |>
  brm_simulate_categorical(
    names = c("status1", "status2"),
    levels = c("present", "absent")
  )
dplyr::select(
  data,
  group,
  time,
  patient,
  starts_with("biomarker"),
  starts_with("status")
)
archetype <- brm_archetype_average_cells(data)
archetype
summary(archetype)
formula <- brm_formula(archetype)
formula
prior <- brm_prior_label(
  code = "normal(1, 2.2)",
  group = "group_1",
  time = "time_2"
) |>
  brm_prior_label("normal(1, 3.3)", group = "group_1", time = "time_3") |>
  brm_prior_label("normal(1, 4.4)", group = "group_1", time = "time_4") |>
  brm_prior_label("normal(2, 2.2)", group = "group_2", time = "time_2") |>
  brm_prior_label("normal(2, 3.3)", group = "group_2", time = "time_3") |>
  brm_prior_label("normal(2, 4.4)", group = "group_2", time = "time_4") |>
  brm_prior_archetype(archetype)
prior
class(prior)
if (identical(Sys.getenv("BRM_EXAMPLES", unset = ""), "true")) {
tmp <- utils::capture.output(
  suppressMessages(
    suppressWarnings(
      model <- brm_model(
        data = archetype,
        formula = formula,
        prior = prior,
        chains = 1,
        iter = 100,
        refresh = 0
      )
    )
  )
)
suppressWarnings(print(model))
brms::prior_summary(model)
draws <- brm_marginal_draws(
  data = archetype,
  formula = formula,
  model = model
)
summaries_model <- brm_marginal_summaries(draws)
summaries_data <- brm_marginal_data(data)
brm_plot_compare(model = summaries_model, data = summaries_data)
}

Description

Create a treatment effect informative prior archetype with a special fixed effect to represent the average across time.

Usage

brm_archetype_average_effects(
  data,
  intercept = FALSE,
  baseline = !is.null(attr(data, "brm_baseline")),
  baseline_subgroup = !is.null(attr(data, "brm_baseline")) && !is.null(attr(data,
    "brm_subgroup")),
  baseline_subgroup_time = !is.null(attr(data, "brm_baseline")) && !is.null(attr(data,
    "brm_subgroup")),
  baseline_time = !is.null(attr(data, "brm_baseline")),
  covariates = TRUE,
  prefix_interest = "x_",
  prefix_nuisance = "nuisance_"
)

Arguments

Details

This archetype has a special fixed effect for each treatment group to represent the mean response averaged across all the time points, and treatment effects are explicitly parameterized.

To illustrate, suppose the dataset has two treatment groups A (placebo/reference group) and B (active/non-reference group), time points 1, 2, and 3, and no other covariates. Let mu_gt be the marginal mean of the response at group g time t given data and hyperparameters. The model has fixed effect parameters beta_1, beta_2, ..., beta_6 which express the marginal means mu_gt as follows:

`mu_A1 = 3 * beta_1 - beta_2 - beta_3`
`mu_A2 = beta_2`
`mu_A3 = beta_3`

`mu_B1 = 3 * beta_1 - beta_2 - beta_3 + 3 * beta_4 - beta_5 - beta_6`
`mu_B2 = beta_2 + beta_5`
`mu_B3 = beta_3 + beta_6`

For group A, beta_1 is the average response in group A averaged across time points. You can confirm this yourself by expressing the average across time (mu_A1 + mu_A2 + mu_A3) / 3 in terms of the ⁠beta_*⁠ parameters and confirming that the expression simplifies down to just beta_1. beta_2 represents the mean response in group A at time 2, and beta_3 represents the mean response in group A at time 3. beta_4 is the treatment effect of group B relative to group A, averaged across time points. beta_5 is the treatment effect of B vs A at time 2, and beta_6 is analogous for time 3.

Value

A special classed tibble with data tailored to the treatment effect time-averaged archetype. The dataset is augmented with extra columns with the "archetype_" prefix, as well as special attributes to tell downstream functions like brm_formula() what to do with the object.

Prior labeling for brm_archetype_average_effects()

Within each treatment group, the initial time point represents the average, and each successive time point represents the response within that actual time. To illustrate, consider the example in the Details section. In the labeling scheme for brm_archetype_average_effects(), you can label the prior on beta_1 using brm_prior_label(code = "normal(1.2, 5)", group = "A", time = "1"). Similarly, you cal label the prior on beta_5 with brm_prior_label(code = "normal(1.3, 7)", group = "B", time = "2"). To confirm that you set the prior correctly, compare the brms prior with the output of summary(your_archetype). See the examples for details.

Nuisance variables

In the presence of covariate adjustment, functions like brm_archetype_successive_cells() convert nuisance factors into binary dummy variables, then center all those dummy variables and any continuous nuisance variables at their means in the data. This ensures that the main model coefficients of interest are not implicitly conditional on a subset of the data. In other words, preprocessing nuisance variables this way preserves the interpretations of the fixed effects of interest, and it ensures informative priors can be specified correctly.

Prior labeling

Informative prior archetypes use a labeling scheme to assign priors to fixed effects. How it works:

1. First, assign the prior of each parameter a collection
  of labels from the data. This can be done manually or with
  successive calls to [brm_prior_label()].
2. Supply the labeling scheme to [brm_prior_archetype()].
  [brm_prior_archetype()] uses attributes of the archetype
  to map labeled priors to their rightful parameters in the model.

For informative prior archetypes, this process is much more convenient and robust than manually calling brms::set_prior(). However, it requires an understanding of how the labels of the priors map to parameters in the model. This mapping varies from archetype to archetype, and it is documented in the help pages of archetype-specific functions such as brm_archetype_successive_cells().

See Also

Other informative prior archetypes: brm_archetype_average_cells(), brm_archetype_cells(), brm_archetype_effects(), brm_archetype_successive_cells(), brm_archetype_successive_effects()

Examples

set.seed(0L)
data <- brm_simulate_outline(
  n_group = 2,
  n_patient = 100,
  n_time = 4,
  rate_dropout = 0,
  rate_lapse = 0
) |>
  dplyr::mutate(response = rnorm(n = dplyr::n())) |>
  brm_data_change() |>
  brm_simulate_continuous(names = c("biomarker1", "biomarker2")) |>
  brm_simulate_categorical(
    names = c("status1", "status2"),
    levels = c("present", "absent")
  )
dplyr::select(
  data,
  group,
  time,
  patient,
  starts_with("biomarker"),
  starts_with("status")
)
archetype <- brm_archetype_average_effects(data)
archetype
summary(archetype)
formula <- brm_formula(archetype)
formula
prior <- brm_prior_label(
  code = "normal(1, 2.2)",
  group = "group_1",
  time = "time_2"
) |>
  brm_prior_label("normal(1, 3.3)", group = "group_1", time = "time_3") |>
  brm_prior_label("normal(1, 4.4)", group = "group_1", time = "time_4") |>
  brm_prior_label("normal(2, 2.2)", group = "group_2", time = "time_2") |>
  brm_prior_label("normal(2, 3.3)", group = "group_2", time = "time_3") |>
  brm_prior_label("normal(2, 4.4)", group = "group_2", time = "time_4") |>
  brm_prior_archetype(archetype)
prior
class(prior)
if (identical(Sys.getenv("BRM_EXAMPLES", unset = ""), "true")) {
tmp <- utils::capture.output(
  suppressMessages(
    suppressWarnings(
      model <- brm_model(
        data = archetype,
        formula = formula,
        prior = prior,
        chains = 1,
        iter = 100,
        refresh = 0
      )
    )
  )
)
suppressWarnings(print(model))
brms::prior_summary(model)
draws <- brm_marginal_draws(
  data = archetype,
  formula = formula,
  model = model
)
summaries_model <- brm_marginal_summaries(draws)
summaries_data <- brm_marginal_data(data)
brm_plot_compare(model = summaries_model, data = summaries_data)
}

Description

Create an informative prior archetype for cell means.

Usage

brm_archetype_cells(
  data,
  intercept = FALSE,
  baseline = !is.null(attr(data, "brm_baseline")),
  baseline_subgroup = !is.null(attr(data, "brm_baseline")) && !is.null(attr(data,
    "brm_subgroup")),
  baseline_subgroup_time = !is.null(attr(data, "brm_baseline")) && !is.null(attr(data,
    "brm_subgroup")),
  baseline_time = !is.null(attr(data, "brm_baseline")),
  covariates = TRUE,
  clda = FALSE,
  prefix_interest = "x_",
  prefix_nuisance = "nuisance_"
)

Arguments

Details

In this archetype, each fixed effect is a cell mean: the group mean for a given value of treatment group and discrete time (and subgroup level, if applicable).

Value

A special classed tibble with data tailored to the successive differences archetype. The dataset is augmented with extra columns with the "archetype_" prefix, as well as special attributes to tell downstream functions like brm_formula() what to do with the object.

Prior labeling for brm_archetype_cells()

Within each treatment group, each model parameter is a cell mean, and the labeling scheme in brm_prior_label() and brm_prior_archetype() translate easily. For example, brm_prior_label(code = "normal(1.2, 5)", group = "B", time = "VISIT2") declares a normal(1.2, 5) prior on the cell mean of treatment group B at discrete time point VISIT2. To confirm that you set the prior correctly, compare the brms prior with the output of summary(your_archetype). See the examples for details.

Nuisance variables

In the presence of covariate adjustment, functions like brm_archetype_successive_cells() convert nuisance factors into binary dummy variables, then center all those dummy variables and any continuous nuisance variables at their means in the data. This ensures that the main model coefficients of interest are not implicitly conditional on a subset of the data. In other words, preprocessing nuisance variables this way preserves the interpretations of the fixed effects of interest, and it ensures informative priors can be specified correctly.

Prior labeling

Informative prior archetypes use a labeling scheme to assign priors to fixed effects. How it works:

1. First, assign the prior of each parameter a collection
  of labels from the data. This can be done manually or with
  successive calls to [brm_prior_label()].
2. Supply the labeling scheme to [brm_prior_archetype()].
  [brm_prior_archetype()] uses attributes of the archetype
  to map labeled priors to their rightful parameters in the model.

For informative prior archetypes, this process is much more convenient and robust than manually calling brms::set_prior(). However, it requires an understanding of how the labels of the priors map to parameters in the model. This mapping varies from archetype to archetype, and it is documented in the help pages of archetype-specific functions such as brm_archetype_successive_cells().

See Also

Other informative prior archetypes: brm_archetype_average_cells(), brm_archetype_average_effects(), brm_archetype_effects(), brm_archetype_successive_cells(), brm_archetype_successive_effects()

Examples

set.seed(0L)
data <- brm_simulate_outline(
  n_group = 2,
  n_patient = 100,
  n_time = 4,
  rate_dropout = 0,
  rate_lapse = 0
) |>
  dplyr::mutate(response = rnorm(n = dplyr::n())) |>
  brm_data_change() |>
  brm_simulate_continuous(names = c("biomarker1", "biomarker2")) |>
  brm_simulate_categorical(
    names = c("status1", "status2"),
    levels = c("present", "absent")
  )
dplyr::select(
  data,
  group,
  time,
  patient,
  starts_with("biomarker"),
  starts_with("status")
)
archetype <- brm_archetype_cells(data)
archetype
summary(archetype)
formula <- brm_formula(archetype)
formula
prior <- brm_prior_label(
  code = "normal(1, 2.2)",
  group = "group_1",
  time = "time_2"
) |>
  brm_prior_label("normal(1, 3.3)", group = "group_1", time = "time_3") |>
  brm_prior_label("normal(1, 4.4)", group = "group_1", time = "time_4") |>
  brm_prior_label("normal(2, 2.2)", group = "group_2", time = "time_2") |>
  brm_prior_label("normal(2, 3.3)", group = "group_2", time = "time_3") |>
  brm_prior_label("normal(2, 4.4)", group = "group_2", time = "time_4") |>
  brm_prior_archetype(archetype)
prior
class(prior)
if (identical(Sys.getenv("BRM_EXAMPLES", unset = ""), "true")) {
tmp <- utils::capture.output(
  suppressMessages(
    suppressWarnings(
      model <- brm_model(
        data = archetype,
        formula = formula,
        prior = prior,
        chains = 1,
        iter = 100,
        refresh = 0
      )
    )
  )
)
suppressWarnings(print(model))
brms::prior_summary(model)
draws <- brm_marginal_draws(
  data = archetype,
  formula = formula,
  model = model
)
summaries_model <- brm_marginal_summaries(draws)
summaries_data <- brm_marginal_data(data)
brm_plot_compare(model = summaries_model, data = summaries_data)
}

Description

Create an informative prior archetype for a simple treatment effect parameterization.

Usage

brm_archetype_effects(
  data,
  intercept = FALSE,
  baseline = !is.null(attr(data, "brm_baseline")),
  baseline_subgroup = !is.null(attr(data, "brm_baseline")) && !is.null(attr(data,
    "brm_subgroup")),
  baseline_subgroup_time = !is.null(attr(data, "brm_baseline")) && !is.null(attr(data,
    "brm_subgroup")),
  baseline_time = !is.null(attr(data, "brm_baseline")),
  covariates = TRUE,
  clda = FALSE,
  prefix_interest = "x_",
  prefix_nuisance = "nuisance_"
)

Arguments

Details

In this archetype, each fixed effect is either a placebo response or a treatment effect.

To illustrate, suppose the dataset has two treatment groups A and B, time points 1, 2, and 3, and no other covariates. Assume group A is the reference group (e.g. placebo). Let mu_gt be the marginal mean of the response at group g time t given data and hyperparameters. The model has fixed effect parameters beta_1, beta_2, ..., beta_6 which express the marginal means mu_gt as follows:

  `mu_A1 = beta_1`
  `mu_A2 = beta_2`
  `mu_A3 = beta_3`

  `mu_B1 = beta_1 + beta_4`
  `mu_B2 = beta_2 + beta_5`
  `mu_B3 = beta_3 + beta_6`

Above, beta_2 is the group mean of treatment group A at time 2, and beta_5 is the treatment effect of B relative to A at time 2.

Value

A special classed tibble with data tailored to the successive differences archetype. The dataset is augmented with extra columns with the "archetype_" prefix, as well as special attributes to tell downstream functions like brm_formula() what to do with the object.

Prior labeling for brm_archetype_effects()

In the reference group (e.g. placebo) each fixed effect is a cell mean at a time point. In each non-reference group, each fixed effect is the treatment effect relative to the reference (at a time point). The labeling scheme in brm_prior_label() and brm_prior_archetype() translate straightforwardly. For example, brm_prior_label(code = "normal(1.2, 5)", group = "A", time = "2") declares a normal(1.2, 5) on beta_2 (cell mean of the reference group at time 2). Similarly, brm_prior_label(code = "normal(1.3, 6)", group = "B", time = "2") declares a normal(1.3, 6) prior on the treatment effect of group B relative to group A at discrete time point 2. To confirm that you set the prior correctly, compare the brms prior with the output of summary(your_archetype). See the examples for details.

Nuisance variables

In the presence of covariate adjustment, functions like brm_archetype_successive_cells() convert nuisance factors into binary dummy variables, then center all those dummy variables and any continuous nuisance variables at their means in the data. This ensures that the main model coefficients of interest are not implicitly conditional on a subset of the data. In other words, preprocessing nuisance variables this way preserves the interpretations of the fixed effects of interest, and it ensures informative priors can be specified correctly.

Prior labeling

Informative prior archetypes use a labeling scheme to assign priors to fixed effects. How it works:

1. First, assign the prior of each parameter a collection
  of labels from the data. This can be done manually or with
  successive calls to [brm_prior_label()].
2. Supply the labeling scheme to [brm_prior_archetype()].
  [brm_prior_archetype()] uses attributes of the archetype
  to map labeled priors to their rightful parameters in the model.

For informative prior archetypes, this process is much more convenient and robust than manually calling brms::set_prior(). However, it requires an understanding of how the labels of the priors map to parameters in the model. This mapping varies from archetype to archetype, and it is documented in the help pages of archetype-specific functions such as brm_archetype_successive_cells().

See Also

Other informative prior archetypes: brm_archetype_average_cells(), brm_archetype_average_effects(), brm_archetype_cells(), brm_archetype_successive_cells(), brm_archetype_successive_effects()

Examples

set.seed(0L)
data <- brm_simulate_outline(
  n_group = 2,
  n_patient = 100,
  n_time = 4,
  rate_dropout = 0,
  rate_lapse = 0
) |>
  dplyr::mutate(response = rnorm(n = dplyr::n())) |>
  brm_data_change() |>
  brm_simulate_continuous(names = c("biomarker1", "biomarker2")) |>
  brm_simulate_categorical(
    names = c("status1", "status2"),
    levels = c("present", "absent")
  )
dplyr::select(
  data,
  group,
  time,
  patient,
  starts_with("biomarker"),
  starts_with("status")
)
archetype <- brm_archetype_effects(data)
archetype
summary(archetype)
formula <- brm_formula(archetype)
formula
prior <- brm_prior_label(
  code = "normal(1, 2.2)",
  group = "group_1",
  time = "time_2"
) |>
  brm_prior_label("normal(1, 3.3)", group = "group_1", time = "time_3") |>
  brm_prior_label("normal(1, 4.4)", group = "group_1", time = "time_4") |>
  brm_prior_label("normal(2, 2.2)", group = "group_2", time = "time_2") |>
  brm_prior_label("normal(2, 3.3)", group = "group_2", time = "time_3") |>
  brm_prior_label("normal(2, 4.4)", group = "group_2", time = "time_4") |>
  brm_prior_archetype(archetype)
prior
class(prior)
if (identical(Sys.getenv("BRM_EXAMPLES", unset = ""), "true")) {
tmp <- utils::capture.output(
  suppressMessages(
    suppressWarnings(
      model <- brm_model(
        data = archetype,
        formula = formula,
        prior = prior,
        chains = 1,
        iter = 100,
        refresh = 0
      )
    )
  )
)
suppressWarnings(print(model))
brms::prior_summary(model)
draws <- brm_marginal_draws(
  data = archetype,
  formula = formula,
  model = model
)
summaries_model <- brm_marginal_summaries(draws)
summaries_data <- brm_marginal_data(data)
brm_plot_compare(model = summaries_model, data = summaries_data)
}

Description

Create an informative prior archetype where the fixed effects are successive differences between adjacent time points.

Usage

brm_archetype_successive_cells(
  data,
  intercept = FALSE,
  baseline = !is.null(attr(data, "brm_baseline")),
  baseline_subgroup = !is.null(attr(data, "brm_baseline")) && !is.null(attr(data,
    "brm_subgroup")),
  baseline_subgroup_time = !is.null(attr(data, "brm_baseline")) && !is.null(attr(data,
    "brm_subgroup")),
  baseline_time = !is.null(attr(data, "brm_baseline")),
  covariates = TRUE,
  clda = FALSE,
  prefix_interest = "x_",
  prefix_nuisance = "nuisance_"
)

Arguments

Details

In this archetype, each fixed effect is either an intercept on the first time point or the difference between two adjacent time points, and each treatment group has its own set of fixed effects independent of the other treatment groups.

To illustrate, suppose the dataset has two treatment groups A and B, time points 1, 2, and 3, and no other covariates. Let mu_gt be the marginal mean of the response at group g time t given data and hyperparameters. The model has fixed effect parameters beta_1, beta_2, ..., beta_6 which express the marginal means mu_gt as follows:

  `mu_A1 = beta_1`
  `mu_A2 = beta_1 + beta_2`
  `mu_A3 = beta_1 + beta_2 + beta_3`

  `mu_B1 = beta_4`
  `mu_B2 = beta_4 + beta_5`
  `mu_B3 = beta_4 + beta_5 + beta_6`

For group A, beta_1 is the time 1 intercept, beta_2 represents time 2 minus time 1, and beta_3 represents time 3 minus time 2. beta_4, beta_5, and beta_6 behave analogously for group B.

Value

A special classed tibble with data tailored to the successive differences archetype. The dataset is augmented with extra columns with the "archetype_" prefix, as well as special attributes to tell downstream functions like brm_formula() what to do with the object.

Nuisance variables

In the presence of covariate adjustment, functions like brm_archetype_successive_cells() convert nuisance factors into binary dummy variables, then center all those dummy variables and any continuous nuisance variables at their means in the data. This ensures that the main model coefficients of interest are not implicitly conditional on a subset of the data. In other words, preprocessing nuisance variables this way preserves the interpretations of the fixed effects of interest, and it ensures informative priors can be specified correctly.

Prior labeling for brm_archetype_successive_cells()

Within each treatment group, each intercept is labeled by the earliest time point, and each successive difference term gets the successive time point as the label. To illustrate, consider the example in the Details section. In the labeling scheme for brm_archetype_successive_cells(), you can label the prior on beta_1 using brm_prior_label(code = "normal(1.2, 5)", group = "A", time = "1"). Similarly, you cal label the prior on beta_5 with brm_prior_label(code = "normal(1.3, 7)", group = "B", time = "2"). To confirm that you set the prior correctly, compare the brms prior with the output of summary(your_archetype). See the examples for details.

Prior labeling

Informative prior archetypes use a labeling scheme to assign priors to fixed effects. How it works:

1. First, assign the prior of each parameter a collection
  of labels from the data. This can be done manually or with
  successive calls to [brm_prior_label()].
2. Supply the labeling scheme to [brm_prior_archetype()].
  [brm_prior_archetype()] uses attributes of the archetype
  to map labeled priors to their rightful parameters in the model.

For informative prior archetypes, this process is much more convenient and robust than manually calling brms::set_prior(). However, it requires an understanding of how the labels of the priors map to parameters in the model. This mapping varies from archetype to archetype, and it is documented in the help pages of archetype-specific functions such as brm_archetype_successive_cells().

See Also

Other informative prior archetypes: brm_archetype_average_cells(), brm_archetype_average_effects(), brm_archetype_cells(), brm_archetype_effects(), brm_archetype_successive_effects()

Examples

set.seed(0L)
data <- brm_simulate_outline(
  n_group = 2,
  n_patient = 100,
  n_time = 4,
  rate_dropout = 0,
  rate_lapse = 0
) |>
  dplyr::mutate(response = rnorm(n = dplyr::n())) |>
  brm_data_change() |>
  brm_simulate_continuous(names = c("biomarker1", "biomarker2")) |>
  brm_simulate_categorical(
    names = c("status1", "status2"),
    levels = c("present", "absent")
  )
dplyr::select(
  data,
  group,
  time,
  patient,
  starts_with("biomarker"),
  starts_with("status")
)
archetype <- brm_archetype_successive_cells(data)
archetype
summary(archetype)
formula <- brm_formula(archetype)
formula
prior <- brm_prior_label(
  code = "normal(1, 2.2)",
  group = "group_1",
  time = "time_2"
) |>
  brm_prior_label("normal(1, 3.3)", group = "group_1", time = "time_3") |>
  brm_prior_label("normal(1, 4.4)", group = "group_1", time = "time_4") |>
  brm_prior_label("normal(2, 2.2)", group = "group_2", time = "time_2") |>
  brm_prior_label("normal(2, 3.3)", group = "group_2", time = "time_3") |>
  brm_prior_label("normal(2, 4.4)", group = "group_2", time = "time_4") |>
  brm_prior_archetype(archetype)
prior
class(prior)
if (identical(Sys.getenv("BRM_EXAMPLES", unset = ""), "true")) {
tmp <- utils::capture.output(
  suppressMessages(
    suppressWarnings(
      model <- brm_model(
        data = archetype,
        formula = formula,
        prior = prior,
        chains = 1,
        iter = 100,
        refresh = 0
      )
    )
  )
)
suppressWarnings(print(model))
brms::prior_summary(model)
draws <- brm_marginal_draws(
  data = archetype,
  formula = formula,
  model = model
)
summaries_model <- brm_marginal_summaries(draws)
summaries_data <- brm_marginal_data(data)
brm_plot_compare(model = summaries_model, data = summaries_data)
}

Description

Create an informative prior archetype where the fixed effects are successive differences between adjacent time points and terms in non-reference groups are treatment effects.

Usage

brm_archetype_successive_effects(
  data,
  intercept = FALSE,
  baseline = !is.null(attr(data, "brm_baseline")),
  baseline_subgroup = !is.null(attr(data, "brm_baseline")) && !is.null(attr(data,
    "brm_subgroup")),
  baseline_subgroup_time = !is.null(attr(data, "brm_baseline")) && !is.null(attr(data,
    "brm_subgroup")),
  baseline_time = !is.null(attr(data, "brm_baseline")),
  covariates = TRUE,
  clda = FALSE,
  prefix_interest = "x_",
  prefix_nuisance = "nuisance_"
)

Arguments

Details

Within the reference treatment group (e.g. placebo), each fixed effect is either an intercept on the first time point or the difference between two adjacent time points. In each non-reference treatment group, each model parameter is defined as an effect relative to the reference group.

To illustrate, suppose the dataset has two treatment groups A and B, time points 1, 2, and 3, and no other covariates. Say group A is the reference group (e.g. placebo). Let mu_gt be the marginal mean of the response at group g time t given data and hyperparameters. The model has fixed effect parameters beta_1, beta_2, ..., beta_6 which express the marginal means mu_gt as follows:

  `mu_A1 = beta_1`
  `mu_A2 = beta_1 + beta_2`
  `mu_A3 = beta_1 + beta_2 + beta_3`

  `mu_B1 = beta_1 + beta_4`
  `mu_B2 = beta_1 + beta_2 + beta_4 + beta_5`
  `mu_B3 = beta_1 + beta_2 + beta_3 + beta_4 + beta_5 + beta_6`

For group A, beta_1 is the time 1 intercept, beta_2 represents time 2 minus time 1, and beta_3 represents time 3 minus time 2. beta_4 is the treatment effect of group B relative to group A at time 1. beta_5 is the treatment effect of the difference between times 2 and 1, relative to treatment group A. Similarly, beta_6 is the treatment effect of the difference between times 3 and 2, relative to treatment group A.

Value

A special classed tibble with data tailored to the successive differences archetype. The dataset is augmented with extra columns with the "archetype_" prefix, as well as special attributes to tell downstream functions like brm_formula() what to do with the object.

Prior labeling for brm_archetype_successive_effects()

Within each treatment group, each intercept is labeled by the earliest time point, and each successive difference term gets the successive time point as the label. To illustrate, consider the example in the Details section. In the labeling scheme for brm_archetype_successive_effects(), you can label the prior on beta_1 using brm_prior_label(code = "normal(1.2, 5)", group = "A", time = "1"). Similarly, you cal label the prior on beta_5 with brm_prior_label(code = "normal(1.3, 7)", group = "B", time = "2"). To confirm that you set the prior correctly, compare the brms prior with the output of summary(your_archetype). See the examples for details.

Nuisance variables

In the presence of covariate adjustment, functions like brm_archetype_successive_cells() convert nuisance factors into binary dummy variables, then center all those dummy variables and any continuous nuisance variables at their means in the data. This ensures that the main model coefficients of interest are not implicitly conditional on a subset of the data. In other words, preprocessing nuisance variables this way preserves the interpretations of the fixed effects of interest, and it ensures informative priors can be specified correctly.

Prior labeling

Informative prior archetypes use a labeling scheme to assign priors to fixed effects. How it works:

1. First, assign the prior of each parameter a collection
  of labels from the data. This can be done manually or with
  successive calls to [brm_prior_label()].
2. Supply the labeling scheme to [brm_prior_archetype()].
  [brm_prior_archetype()] uses attributes of the archetype
  to map labeled priors to their rightful parameters in the model.

For informative prior archetypes, this process is much more convenient and robust than manually calling brms::set_prior(). However, it requires an understanding of how the labels of the priors map to parameters in the model. This mapping varies from archetype to archetype, and it is documented in the help pages of archetype-specific functions such as brm_archetype_successive_cells().

See Also

Other informative prior archetypes: brm_archetype_average_cells(), brm_archetype_average_effects(), brm_archetype_cells(), brm_archetype_effects(), brm_archetype_successive_cells()

Examples

set.seed(0L)
data <- brm_simulate_outline(
  n_group = 2,
  n_patient = 100,
  n_time = 4,
  rate_dropout = 0,
  rate_lapse = 0
) |>
  dplyr::mutate(response = rnorm(n = dplyr::n())) |>
  brm_data_change() |>
  brm_simulate_continuous(names = c("biomarker1", "biomarker2")) |>
  brm_simulate_categorical(
    names = c("status1", "status2"),
    levels = c("present", "absent")
  )
dplyr::select(
  data,
  group,
  time,
  patient,
  starts_with("biomarker"),
  starts_with("status")
)
archetype <- brm_archetype_successive_effects(data)
archetype
summary(archetype)
formula <- brm_formula(archetype)
formula
prior <- brm_prior_label(
  code = "normal(1, 2.2)",
  group = "group_1",
  time = "time_2"
) |>
  brm_prior_label("normal(1, 3.3)", group = "group_1", time = "time_3") |>
  brm_prior_label("normal(1, 4.4)", group = "group_1", time = "time_4") |>
  brm_prior_label("normal(2, 2.2)", group = "group_2", time = "time_2") |>
  brm_prior_label("normal(2, 3.3)", group = "group_2", time = "time_3") |>
  brm_prior_label("normal(2, 4.4)", group = "group_2", time = "time_4") |>
  brm_prior_archetype(archetype)
prior
class(prior)
if (identical(Sys.getenv("BRM_EXAMPLES", unset = ""), "true")) {
tmp <- utils::capture.output(
  suppressMessages(
    suppressWarnings(
      model <- brm_model(
        data = archetype,
        formula = formula,
        prior = prior,
        chains = 1,
        iter = 100,
        refresh = 0
      )
    )
  )
)
suppressWarnings(print(model))
brms::prior_summary(model)
draws <- brm_marginal_draws(
  data = archetype,
  formula = formula,
  model = model
)
summaries_model <- brm_marginal_summaries(draws)
summaries_data <- brm_marginal_data(data)
brm_plot_compare(model = summaries_model, data = summaries_data)
}

Description

Create a dataset to analyze with an MMRM.

Usage

brm_data(
  data,
  outcome,
  baseline = NULL,
  group,
  subgroup = NULL,
  time,
  patient,
  covariates = character(0L),
  missing = NULL,
  reference_group,
  reference_subgroup = NULL,
  reference_time = NULL,
  role = NULL,
  level_baseline = NULL,
  level_control = NULL
)

Arguments

Value

A classed tibble with attributes which denote features of the data such as the treatment group and discrete time variables.

Preprocessing

The preprocessing steps in brm_data() are as follows:

• Perform basic assertions to make sure the data and other arguments are properly formatted.

• Convert the group and time columns to character vectors.

• Sanitize the levels of the group and time columns using make.names(unique = FALSE, allow_ = TRUE) to ensure agreement between the data and the output of brms.

• For each implicitly missing outcome observation, add explicit row with the outcome variable equal to NA_real_. Missing values in the predictors are implicitly filled using zoo::na.locf() on within each patient, which is not valid for time-varying covariates. If any covariates are time-varying, please manually perform this step before calling brm_data().

• Arrange the rows of the data by group, then patient, then discrete time.

• Select only the columns of the data relevant to an MMRM analysis.

Separation string

Post-processing in brm_marginal_draws() names each of the group-by-time marginal means with the delimiting character string from Sys.getenv("BRM_SEP", unset = "|"). Neither the column names nor element names of the group and time variables can contain this string. To set a custom string yourself, use Sys.setenv(BRM_SEP = "YOUR_CUSTOM_STRING").

See Also

Other data: brm_data_change(), brm_data_chronologize()

Examples

set.seed(0)
data <- brm_simulate_simple()$data
colnames(data) <- paste0("col_", colnames(data))
data
brm_data(
  data = data,
  outcome = "col_response",
  group = "col_group",
  time = "col_time",
  patient = "col_patient",
  reference_group = "group_1",
  reference_time = "time_1"
)

Description

Convert a dataset from raw response to change from baseline.

Usage

brm_data_change(data, name_change = "change", name_baseline = "baseline")

Arguments

Value

A classed tibble with change from baseline as the outcome variable and the internal attributes modified accordingly. A special baseline column is also created, and the original raw response column is removed. The new baseline column is comprised of the elements of the response variable corresponding to the reference_time argument of brm_data().

If there is a column to denote missing values for simulation purposes, e.g. the "missing" column generated by brm_simulate_outline(), then missing baseline values are propagated accordingly such that change from baseline will be missing if either the post-baseline response is missing or the baseline response is missing.

See Also

Other data: brm_data(), brm_data_chronologize()

Examples

set.seed(0)
data <- brm_data(
  data = dplyr::rename(brm_simulate_simple()$data, y_values = response),
  outcome = "y_values",
  group = "group",
  time = "time",
  patient = "patient",
  reference_group = "group_1",
  reference_time = "time_1"
)
data
attr(data, "brm_outcome")
attr(data, "brm_baseline")
attr(data, "brm_reference_time")
changed <- brm_data_change(data = data, name_change = "delta")
changed
attr(changed, "brm_outcome")
attr(changed, "brm_baseline")
attr(data, "brm_reference_time")

Description

Convert the discrete time variable into an ordered factor.

Usage

brm_data_chronologize(
  data,
  order = NULL,
  levels = NULL,
  time = attr(data, "brm_time")
)

Arguments

Details

Most MMRMs should use an ordered factor for the time column in the data. This way, individual time points are treated as distinct factor levels for the purposes of fixed effect parameterizations (see the "Contrasts" section), and the explicit ordering ensures that informative prior archetypes and ARMA-like correlation structures are expressed correctly. Without the ordering, problems can arise when character vectors are sorted: e.g. if AVISIT has levels ⁠"VISIT1", "VISIT2", ..., "VISIT10"⁠, then brms will mistake the the order of scheduled study visits to be ⁠"VISIT1", "VISIT10", "VISIT2", ...⁠, which is not chronological.

You can easily turn the time variable into an ordered factor using brm_data_chronologize(). Either supply an explicit character vector of chronologically-ordered factor levels in the levels argument, or supply the name of a time-ordered variable in the order argument.

brm_data_chronologize() can be called either before or just after brm_data(), but in the former case, the discrete time variable needs to be specified explicitly in time argument. And in the latter, brm_data_chronologize() must be called before any of the informative prior archetype functions such as brm_archetype_successive_cells().

Value

A data frame with the time column as an ordered factor.

Contrasts

Ordinarily, ordered factors automatically use polynomial contrasts from contr.poly(). This is undesirable for MMRMs, so if the time variable is an ordered factor, then brm_data() manually sets contrasts(data[[time]]) to a set of treatment contrasts using contr.treatment(). If you prefer different contrasts, please manually set contrasts(data[[time]]) to something else after calling brm_data().

See Also

Other data: brm_data(), brm_data_change()

Examples

data <- brm_simulate_outline(n_time = 12, n_patient = 4)
data$AVISIT <- gsub("_0", "_", data$time)
data$AVISITN <- as.integer(gsub("time_", "", data$time))
data[, c("AVISIT", "AVISITN")]
sort(unique(data$AVISIT)) # wrong order
data1 <- brm_data_chronologize(data, time = "AVISIT", order = "AVISITN")
sort(unique(data1$AVISIT)) # correct order
levels <- paste0("time_", seq_len(12))
data2 <- brm_data_chronologize(data, time = "AVISIT", levels = levels)
sort(unique(data2$AVISIT)) # correct order

Description

Build a model formula for an MMRM, either for a generic brm_data() dataset or an informative prior archetype.

Usage

brm_formula(
  data,
  model_missing_outcomes = FALSE,
  check_rank = TRUE,
  sigma = brms.mmrm::brm_formula_sigma(data = data, check_rank = check_rank),
  correlation = "unstructured",
  autoregressive_order = 1L,
  moving_average_order = 1L,
  residual_covariance_arma_estimation = FALSE,
  ...
)

## Default S3 method:
brm_formula(
  data,
  model_missing_outcomes = FALSE,
  check_rank = TRUE,
  sigma = brms.mmrm::brm_formula_sigma(data = data, check_rank = check_rank),
  correlation = "unstructured",
  autoregressive_order = 1L,
  moving_average_order = 1L,
  residual_covariance_arma_estimation = FALSE,
  intercept = TRUE,
  baseline = !is.null(attr(data, "brm_baseline")),
  baseline_subgroup = !is.null(attr(data, "brm_baseline")) && !is.null(attr(data,
    "brm_subgroup")),
  baseline_subgroup_time = !is.null(attr(data, "brm_baseline")) && !is.null(attr(data,
    "brm_subgroup")),
  baseline_time = !is.null(attr(data, "brm_baseline")),
  covariates = TRUE,
  group = TRUE,
  group_subgroup = !is.null(attr(data, "brm_subgroup")),
  group_subgroup_time = !is.null(attr(data, "brm_subgroup")),
  group_time = TRUE,
  subgroup = !is.null(attr(data, "brm_subgroup")),
  subgroup_time = !is.null(attr(data, "brm_subgroup")),
  time = TRUE,
  center = TRUE,
  ...,
  effect_baseline = NULL,
  effect_group = NULL,
  effect_time = NULL,
  interaction_baseline = NULL,
  interaction_group = NULL
)

## S3 method for class 'brms_mmrm_archetype'
brm_formula(
  data,
  model_missing_outcomes = FALSE,
  check_rank = TRUE,
  sigma = brms.mmrm::brm_formula_sigma(data = data, check_rank = check_rank),
  correlation = "unstructured",
  autoregressive_order = 1L,
  moving_average_order = 1L,
  residual_covariance_arma_estimation = FALSE,
  ...,
  warn_ignored = TRUE
)

Arguments

Value

An object of class "brmsformula" returned from brms::brmsformula(). It contains the fixed effect mapping, correlation structure, and residual variance structure.

brm_data() formulas

For a brm_data() dataset, brm_formula() builds an R formula for an MMRM based on the details in the data and your choice of mapping. Customize your mapping by toggling on or off the various TRUE/FALSE arguments of brm_formula(), such as intercept, baseline, and group_time. All plausible additive effects, two-way interactions, and three-way interactions can be specified. The following interactions are not supported:

• Any interactions with the concomitant covariates you specified in the covariates argument of brm_data().

• Any interactions which include baseline response and treatment group together. Rationale: in a randomized controlled experiment, baseline and treatment group assignment should be uncorrelated.

Formulas for informative prior archetypes

Functions like brm_archetype_successive_cells() tailor datasets to informative prior archetypes. For these specialized tailored datasets, brm_formula() works differently. It still applies the variance and correlation structure of your choosing, and it still lets you choose whether to adjust for nuisance covariates, but it no longer lets you toggle on/off individual terms in the model, such as intercept, baseline, or group. Instead, to ensure the correct interpretation of the parameters, brm_formula() uses the ⁠x_*⁠ and ⁠nuisance_*⁠ columns generated by brm_archetype_successive_cells( prefix_interest = "x_", prefix_nuisance = "nuisance_").

Parameterization

For a formula on a brm_data() dataset, the formula is not the only factor that determines the fixed effect mapping. The ordering of the categorical variables in the data, as well as the contrast option in R, affect the construction of the model matrix. To see the model matrix that will ultimately be used in brm_model(), run brms::make_standata() and examine the X element of the returned list. See the examples below for a demonstration.

See Also

Other models: brm_formula_sigma(), brm_model()

Examples

set.seed(0)
data <- brm_data(
  data = brm_simulate_simple()$data,
  outcome = "response",
  group = "group",
  time = "time",
  patient = "patient",
  reference_group = "group_1",
  reference_time = "time_1"
)
brm_formula(data)
brm_formula(data = data, intercept = FALSE, baseline = FALSE)
formula <- brm_formula(
  data = data,
  intercept = FALSE,
  baseline = FALSE,
  group = FALSE
)
formula
# Standard deviations of residuals are distributional parameters that can
# regress on variables in the data.
homogeneous <- brm_formula_sigma(data, time = FALSE)
by_group <- brm_formula_sigma(data, group = TRUE, intercept = TRUE)
homogeneous
by_group
brm_formula(data, sigma = homogeneous)
brm_formula(data, sigma = by_group)
# Optional: set the contrast option, which determines the model matrix.
options(contrasts = c(unordered = "contr.SAS", ordered = "contr.poly"))
# See the fixed effect mapping you get from the data:
head(brms::make_standata(formula = formula, data = data)$X)
# Specify a different contrast method to use an alternative
# mapping when fitting the model with brm_model():
options(
  contrasts = c(unordered = "contr.treatment", ordered = "contr.poly")
)
# different model matrix than before:
head(brms::make_standata(formula = formula, data = data)$X)
# Formula on an informative prior archetype:
data <- brm_simulate_outline(
  n_group = 2,
  n_patient = 100,
  n_time = 4,
  rate_dropout = 0,
  rate_lapse = 0
) |>
  dplyr::mutate(response = rnorm(n = dplyr::n())) |>
  brm_data_change() |>
  brm_simulate_continuous(names = c("biomarker1", "biomarker2")) |>
  brm_simulate_categorical(
    names = "biomarker3",
    levels = c("present", "absent")
  )
archetype <- brm_archetype_successive_cells(data)
formula <- brm_formula(data = archetype)
formula

Description

Parameterize standard deviations using a formula for the sigma argument of brm_formula().

Usage

brm_formula_sigma(
  data,
  check_rank = TRUE,
  intercept = FALSE,
  baseline = FALSE,
  baseline_subgroup = FALSE,
  baseline_subgroup_time = FALSE,
  baseline_time = FALSE,
  covariates = FALSE,
  group = FALSE,
  group_subgroup = FALSE,
  group_subgroup_time = FALSE,
  group_time = FALSE,
  subgroup = FALSE,
  subgroup_time = FALSE,
  time = TRUE
)

Arguments

Details

In brms, the standard deviations of the residuals are modeled through a parameter vector called sigma. brms.mmrm always treats sigma as a distributional parameter (https://paulbuerkner.com/brms/articles/brms_distreg.html). brm_formula_sigma() lets you control the parameterization of sigma. The output of brm_formula_sigma() serves as input to the sigma argument of brm_formula().

The default sigma formula is sigma ~ 0 + time, where time is the discrete time variable in the data. This is the usual heterogeneous variance structure which declares one standard deviation parameter for each time point in the data. Alternatively, you could write brm_formula_sigma(data, intercept = TRUE, time = FALSE). This will produce sigma ~ 1, which yields a single scalar variance (a structure termed "homogeneous variance").

With arguments like baseline and covariates, you can specify extremely complicated variance structures. However, if baseline or covariates are used, then the output of brm_marginal_draws() omit effect size due to the statistical challenges of calculating marginal means of draws of sigma for this uncommon scenario.

Value

A base R formula with S3 class "brms_mmrm_formula_sigma". This formula controls the parameterization of sigma, the linear-scale brms distributional parameters which represent standard deviations.

See Also

Other models: brm_formula(), brm_model()

Examples

set.seed(0)
data <- brm_data(
  data = brm_simulate_simple()$data,
  outcome = "response",
  group = "group",
  time = "time",
  patient = "patient",
  reference_group = "group_1",
  reference_time = "time_1"
)
homogeneous <- brm_formula_sigma(data, time = FALSE, intercept = TRUE)
by_group <- brm_formula_sigma(data, group = TRUE, intercept = TRUE)
homogeneous
by_group
brm_formula(data, sigma = homogeneous)
brm_formula(data, sigma = by_group)

Description

Marginal summaries of the data.

Usage

brm_marginal_data(
  data,
  level = 0.95,
  use_subgroup = !is.null(attr(data, "brm_subgroup"))
)

Arguments

Value

A tibble with one row per summary statistic and the following columns:

• group: treatment group.

• subgroup: subgroup level. Only included if the subgroup argument of brm_marginal_data() is TRUE.

• time: discrete time point.

• statistic: type of summary statistic.

• value: numeric value of the estimate.

The statistic column has the following possible values:

• mean: observed mean response after removing missing values.

• median: observed median response after removing missing values.

• sd: observed standard deviation of the response after removing missing values.

• lower: lower bound of a normal equal-tailed confidence interval with confidence level determined by the level argument.

• upper: upper bound of a normal equal-tailed confidence interval with confidence level determined by the level argument.

• n_observe: number of non-missing values in the response.

• n_total: number of total records in the data for the given group/time combination, including both observed and missing values.

See Also

Other marginals: brm_marginal_draws(), brm_marginal_draws_average(), brm_marginal_grid(), brm_marginal_probabilities(), brm_marginal_summaries()

Examples

set.seed(0L)
data <- brm_data(
  data = brm_simulate_simple()$data,
  outcome = "response",
  group = "group",
  time = "time",
  patient = "patient",
  reference_group = "group_1",
  reference_time = "time_1"
)
brm_marginal_data(data = data)

Description

Get marginal posterior draws from a fitted MMRM.

Usage

brm_marginal_draws(
  model,
  data = model$brms.mmrm_data,
  formula = model$brms.mmrm_formula,
  transform = brms.mmrm::brm_transform_marginal(data = data, formula = formula,
    average_within_subgroup = average_within_subgroup),
  effect_size = attr(formula, "brm_allow_effect_size"),
  average_within_subgroup = NULL,
  use_subgroup = NULL,
  control = NULL,
  baseline = NULL
)

Arguments

Value

A named list of tibbles of MCMC draws of the marginal posterior distribution of each treatment group and time point. These marginals are also subgroup-specific if brm_formula() included fixed effects that use the subgroup variable originally declared in brm_data(). In each tibble, there is 1 row per posterior sample and one column for each type of marginal distribution (i.e. each combination of treatment group and discrete time point. The specific tibbles in the returned list are described below:

• response: on the scale of the response variable.

• difference_time: change from baseline: the response at a particular time minus the response at baseline (reference_time). Only returned if the reference_time argument of brm_data() was not NULL (i.e. if a baseline value for the time variable was identified).

• difference_group: treatment effect: These samples depend on the values of reference_group and reference_time which were originally declared in brm_data(). reference_group is the control group, and reference_time is baseline. If baseline was originally given (via reference_time in brm_data()), then difference_time is the change-from-baseline value of each active group minus that of the control group. Otherwise, if baseline is omitted (i.e. reference_time = NULL (default) in brm_data()), then difference_time is the raw response at each active group minus that of the control group.

• difference_subgroup: subgroup differences: the difference_group at each subgroup level minus the difference_group at the subgroup reference level (reference_subgroup). Only reported if a subgroup analysis was specified through the appropriate arguments to brm_data() and brm_formula().

• effect: effect size, defined as the treatment difference divided by the residual standard deviation. Omitted if the effect_size argument is FALSE or if the brm_formula_sigma() includes baseline or covariates.

• sigma: posterior draws of linear-scale marginal standard deviations of residuals. Omitted if the effect_size argument is FALSE or if the brm_formula_sigma() includes baseline or covariates.

Baseline

The returned values from brm_marginal_draws() depend on whether a baseline time point was declared through the reference_time argument of brm_data(). If reference_time was not NULL, then brm_marginal_draws() will calculate change from baseline, and it will calculate treatment differences as differences between change-from-baseline values. If reference_time was not NULL, then brm_marginal_draws() will not calculate change from baseline, and it will calculate treatment differences as differences between response values.

Separation string

Post-processing in brm_marginal_draws() names each of the group-by-time marginal means with the delimiting character string from Sys.getenv("BRM_SEP", unset = "|"). Neither the column names nor element names of the group and time variables can contain this string. To set a custom string yourself, use Sys.setenv(BRM_SEP = "YOUR_CUSTOM_STRING").

See Also

Other marginals: brm_marginal_data(), brm_marginal_draws_average(), brm_marginal_grid(), brm_marginal_probabilities(), brm_marginal_summaries()

Examples

if (identical(Sys.getenv("BRM_EXAMPLES", unset = ""), "true")) {
set.seed(0L)
data <- brm_data(
  data = brm_simulate_simple()$data,
  outcome = "response",
  group = "group",
  time = "time",
  patient = "patient",
  reference_group = "group_1",
  reference_time = "time_1"
)
formula <- brm_formula(
  data = data,
  baseline = FALSE,
  baseline_time = FALSE
)
tmp <- utils::capture.output(
  suppressMessages(
    suppressWarnings(
      model <- brm_model(
        data = data,
        formula = formula,
        chains = 1,
        iter = 100,
        refresh = 0
      )
    )
  )
)
brm_marginal_draws(data = data, formula = formula, model = model)
}

Description

Simple un-weighted arithmetic mean of marginal MCMC draws across time points.

Usage

brm_marginal_draws_average(draws, data, times = NULL, label = "average")

Arguments

Value

A named list of tibbles of MCMC draws of the marginal posterior distribution of each treatment group and time point (or group-by-subgroup-by-time, if applicable). See brm_marginal_draws() for the full details of the return value. The only difference is that brm_marginal_draws_average() returns a single pseudo-time-point to represent the average across multiple real time points.

Separation string

Post-processing in brm_marginal_draws() names each of the group-by-time marginal means with the delimiting character string from Sys.getenv("BRM_SEP", unset = "|"). Neither the column names nor element names of the group and time variables can contain this string. To set a custom string yourself, use Sys.setenv(BRM_SEP = "YOUR_CUSTOM_STRING").

See Also

Other marginals: brm_marginal_data(), brm_marginal_draws(), brm_marginal_grid(), brm_marginal_probabilities(), brm_marginal_summaries()

Examples

if (identical(Sys.getenv("BRM_EXAMPLES", unset = ""), "true")) {
set.seed(0L)
data <- brm_data(
  data = brm_simulate_simple()$data,
  outcome = "response",
  group = "group",
  time = "time",
  patient = "patient",
  reference_group = "group_1",
  reference_time = "time_1"
)
formula <- brm_formula(
  data = data,
  baseline = FALSE,
  baseline_time = FALSE
)
tmp <- utils::capture.output(
  suppressMessages(
    suppressWarnings(
      model <- brm_model(
        data = data,
        formula = formula,
        chains = 1,
        iter = 100,
        refresh = 0
      )
    )
  )
)
draws <- brm_marginal_draws(data = data, formula = formula, model = model)
brm_marginal_draws_average(draws = draws, data = data)
brm_marginal_draws_average(
  draws = draws,
  data = data,
  times = c("time_1", "time_2"),
  label = "mean"
)
}

Description

Describe the column names of the data frames output by brm_marginal_draws().

Usage

brm_marginal_grid(data, formula)

Arguments

Details

Useful for creating custom posterior summaries from the draws.

Value

A data frame with a name column with the names of columns of data frames in brm_marginal_draws(), along with metadata to describe which groups, subgroups, and time points those columns correspond to.

See Also

Other marginals: brm_marginal_data(), brm_marginal_draws(), brm_marginal_draws_average(), brm_marginal_probabilities(), brm_marginal_summaries()

Examples

data <- brm_simulate_outline()
brm_marginal_grid(data, brm_formula(data))
data <- brm_simulate_outline(n_subgroup = 2L)
brm_marginal_grid(data, brm_formula(data))

Description

Marginal probabilities on the treatment effect for an MMRM.

Usage

brm_marginal_probabilities(draws, direction = "greater", threshold = 0)

Arguments

Value

A tibble of probabilities of the form ⁠Prob(treatment effect > threshold | data)⁠ and/or ⁠Prob(treatment effect < threshold | data)⁠. It has one row per probability and the following columns: * group: treatment group. * subgroup: subgroup level, if applicable. * time: discrete time point, * direction: direction of the comparison in the marginal probability: "greater" for >, "less" for < * threshold: treatment effect threshold in the probability statement. * value: numeric value of the estimate of the probability.

See Also

Other marginals: brm_marginal_data(), brm_marginal_draws(), brm_marginal_draws_average(), brm_marginal_grid(), brm_marginal_summaries()

Examples

if (identical(Sys.getenv("BRM_EXAMPLES", unset = ""), "true")) {
set.seed(0L)
data <- brm_data(
  data = brm_simulate_simple()$data,
  outcome = "response",
  group = "group",
  time = "time",
  patient = "patient",
  reference_group = "group_1",
  reference_time = "time_1"
)
formula <- brm_formula(
  data = data,
  baseline = FALSE,
  baseline_time = FALSE
)
tmp <- utils::capture.output(
  suppressMessages(
    suppressWarnings(
      model <- brm_model(
        data = data,
        formula = formula,
        chains = 1,
        iter = 100,
        refresh = 0
      )
    )
  )
)
draws <- brm_marginal_draws(data = data, formula = formula, model = model)
brm_marginal_probabilities(draws, direction = "greater", threshold = 0)
}

Description

Summary statistics of the marginal posterior of an MMRM.

Usage

brm_marginal_summaries(draws, level = 0.95)

Arguments

Value

A tibble with one row per summary statistic and the following columns:

• marginal: type of marginal distribution. If outcome was "response" in brm_marginal_draws(), then possible values include "response" for the response on the raw scale, "change" for change from baseline, and "difference" for treatment difference in terms of change from baseline. If outcome was "change", then possible values include "response" for the response one the change from baseline scale and "difference" for treatment difference.

• statistic: type of summary statistic. "lower" and "upper" are bounds of an equal-tailed quantile-based credible interval.

• group: treatment group.

• subgroup: subgroup level, if applicable.

• time: discrete time point.

• value: numeric value of the estimate.

• mcse: Monte Carlo standard error of the estimate. The statistic column has the following possible values:

• mean: posterior mean.

• median: posterior median.

• sd: posterior standard deviation of the mean.

• lower: lower bound of an equal-tailed credible interval of the mean, with credible level determined by the level argument.

• upper: upper bound of an equal-tailed credible interval with credible level determined by the level argument.

See Also

Other marginals: brm_marginal_data(), brm_marginal_draws(), brm_marginal_draws_average(), brm_marginal_grid(), brm_marginal_probabilities()

Examples

if (identical(Sys.getenv("BRM_EXAMPLES", unset = ""), "true")) {
set.seed(0L)
data <- brm_data(
  data = brm_simulate_simple()$data,
  outcome = "response",
  group = "group",
  time = "time",
  patient = "patient",
  reference_group = "group_1",
  reference_time = "time_1"
)
formula <- brm_formula(
  data = data,
  baseline = FALSE,
  baseline_time = FALSE
)
tmp <- utils::capture.output(
  suppressMessages(
    suppressWarnings(
      model <- brm_model(
        data = data,
        formula = formula,
        chains = 1,
        iter = 100,
        refresh = 0
      )
    )
  )
)
draws <- brm_marginal_draws(data = data, formula = formula, model = model)
suppressWarnings(brm_marginal_summaries(draws))
}

Description

Fit an MMRM model using brms.

Usage

brm_model(
  data,
  formula,
  ...,
  prior = NULL,
  family = brms::brmsfamily(family = "gaussian"),
  imputed = NULL
)

Arguments

Value

A fitted model object from brms, with new list elements brms.mmrm_data and brms.mmrm_formula to capture the data and formula supplied to brm_model(). See the explanation of the data argument for how the data is handled and how it relates to the data returned in the brms.mmrm_data attribute.

Parameterization

For a formula on a brm_data() dataset, the formula is not the only factor that determines the fixed effect mapping. The ordering of the categorical variables in the data, as well as the contrast option in R, affect the construction of the model matrix. To see the model matrix that will ultimately be used in brm_model(), run brms::make_standata() and examine the X element of the returned list. See the examples below for a demonstration.

See Also

Other models: brm_formula(), brm_formula_sigma()

Examples

if (identical(Sys.getenv("BRM_EXAMPLES", unset = ""), "true")) {
set.seed(0L)
data <- brm_data(
  data = brm_simulate_simple()$data,
  outcome = "response",
  group = "group",
  time = "time",
  patient = "patient",
  reference_group = "group_1",
  reference_time = "time_1"
)
formula <- brm_formula(
  data = data,
  baseline = FALSE,
  baseline_time = FALSE
)
# Optional: set the contrast option, which determines the model matrix.
options(contrasts = c(unordered = "contr.SAS", ordered = "contr.poly"))
# See the fixed effect mapping you get from the data:
head(brms::make_standata(formula = formula, data = data)$X)
# Specify a different contrast method to use an alternative
# mapping when fitting the model with brm_model():
options(
  contrasts = c(unordered = "contr.treatment", ordered = "contr.poly")
)
# different model matrix than before:
head(brms::make_standata(formula = formula, data = data)$X)
tmp <- utils::capture.output(
  suppressMessages(
    suppressWarnings(
      model <- brm_model(
        data = data,
        formula = formula,
        chains = 1,
        iter = 100,
        refresh = 0
      )
    )
  )
)
# The output is a brms model fit object with added list
# elements "brms.mmrm_data" and "brms.mmrm_formula" to track the dataset
# and formula used to fit the model.
model$brms.mmrm_data
model$brms.mmrm_formula
# Otherwise, the fitted model object acts exactly like a brms fitted model.
suppressWarnings(print(model))
brms::prior_summary(model)
}

Description

Visually compare the marginals of multiple models and/or datasets.

Usage

brm_plot_compare(
  ...,
  marginal = "response",
  compare = "source",
  axis = "time",
  facet = c("group", "subgroup")
)

Arguments

Details

By default, brm_plot_compare() compares multiple models and/or datasets side-by-side. The compare argument selects the primary comparison of interest, and arguments axis and facet control the arrangement of various other components of the plot. The subgroup variable is automatically included if and only if all the supplied marginal summaries have a subgroup column.

Value

A ggplot object.

See Also

Other visualization: brm_plot_draws()

Examples

if (identical(Sys.getenv("BRM_EXAMPLES", unset = ""), "true")) {
set.seed(0L)
data <- brm_data(
  data = brm_simulate_simple()$data,
  outcome = "response",
  group = "group",
  time = "time",
  patient = "patient",
  reference_group = "group_1",
  reference_time = "time_1"
)
formula <- brm_formula(
  data = data,
  baseline = FALSE,
  baseline_time = FALSE
)
tmp <- utils::capture.output(
  suppressMessages(
    suppressWarnings(
      model <- brm_model(
        data = data,
        formula = formula,
        chains = 1,
        iter = 100,
        refresh = 0
      )
    )
  )
)
draws <- brm_marginal_draws(data = data, formula = formula, model = model)
suppressWarnings(summaries_draws <- brm_marginal_summaries(draws))
summaries_data <- brm_marginal_data(data)
brm_plot_compare(
  model1 = summaries_draws,
  model2 = summaries_draws,
  data = summaries_data
)
brm_plot_compare(
  model1 = summaries_draws,
  model2 = summaries_draws,
  marginal = "difference"
)
}

Description

Visualize posterior draws of marginals.

Usage

brm_plot_draws(draws, axis = "time", facet = c("group", "subgroup"))

Arguments

Value

A ggplot object.

See Also

Other visualization: brm_plot_compare()

Examples

if (identical(Sys.getenv("BRM_EXAMPLES", unset = ""), "true")) {
set.seed(0L)
data <- brm_data(
  data = brm_simulate_simple()$data,
  outcome = "response",
  group = "group",
  time = "time",
  patient = "patient",
  reference_group = "group_1",
  reference_time = "time_1"
)
formula <- brm_formula(
  data = data,
  baseline = FALSE,
  baseline_time = FALSE
)
tmp <- utils::capture.output(
  suppressMessages(
    suppressWarnings(
      model <- brm_model(
        data = data,
        formula = formula,
        chains = 1,
        iter = 100,
        refresh = 0
      )
    )
  )
)
draws <- brm_marginal_draws(data = data, formula = formula, model = model)
brm_plot_draws(draws = draws$difference_time)
}

Description

Create a brms prior for fixed effects in an archetype.

Usage

brm_prior_archetype(label, archetype)

Arguments

Value

A brms prior object that you can supply to the prior argument of brm_model().

Prior labeling

Informative prior archetypes use a labeling scheme to assign priors to fixed effects. How it works:

1. First, assign the prior of each parameter a collection
  of labels from the data. This can be done manually or with
  successive calls to [brm_prior_label()].
2. Supply the labeling scheme to [brm_prior_archetype()].
  [brm_prior_archetype()] uses attributes of the archetype
  to map labeled priors to their rightful parameters in the model.

For informative prior archetypes, this process is much more convenient and robust than manually calling brms::set_prior(). However, it requires an understanding of how the labels of the priors map to parameters in the model. This mapping varies from archetype to archetype, and it is documented in the help pages of archetype-specific functions such as brm_archetype_successive_cells().

See Also

Other priors: brm_prior_label(), brm_prior_simple(), brm_prior_template()

Examples

set.seed(0L)
data <- brm_simulate_outline(
  n_group = 2,
  n_patient = 100,
  n_time = 3,
  rate_dropout = 0,
  rate_lapse = 0
) |>
  dplyr::mutate(response = rnorm(n = dplyr::n())) |>
  brm_simulate_continuous(names = c("biomarker1", "biomarker2")) |>
  brm_simulate_categorical(
    names = c("status1", "status2"),
    levels = c("present", "absent")
  )
archetype <- brm_archetype_successive_cells(data)
dplyr::distinct(data, group, time)
prior <- NULL |>
  brm_prior_label("normal(1, 1)", group = "group_1", time = "time_1") |>
  brm_prior_label("normal(1, 2)", group = "group_1", time = "time_2") |>
  brm_prior_label("normal(1, 3)", group = "group_1", time = "time_3") |>
  brm_prior_label("normal(2, 1)", group = "group_2", time = "time_1") |>
  brm_prior_label("normal(2, 2)", group = "group_2", time = "time_2") |>
  brm_prior_label("normal(2, 3)", group = "group_2", time = "time_3") |>
  brm_prior_archetype(archetype = archetype)
prior
class(prior)

Description

Label an informative prior for a parameter using a collection of levels in the data.

Usage

brm_prior_label(label = NULL, code, group, subgroup = NULL, time)

Arguments

Value

A tibble with one row per model parameter and columns for the Stan code, treatment group, subgroup, and discrete time point of each parameter. You can supply this tibble to the label argument of brm_prior_archetype().

Prior labeling

Informative prior archetypes use a labeling scheme to assign priors to fixed effects. How it works:

1. First, assign the prior of each parameter a collection
  of labels from the data. This can be done manually or with
  successive calls to [brm_prior_label()].
2. Supply the labeling scheme to [brm_prior_archetype()].
  [brm_prior_archetype()] uses attributes of the archetype
  to map labeled priors to their rightful parameters in the model.

For informative prior archetypes, this process is much more convenient and robust than manually calling brms::set_prior(). However, it requires an understanding of how the labels of the priors map to parameters in the model. This mapping varies from archetype to archetype, and it is documented in the help pages of archetype-specific functions such as brm_archetype_successive_cells().

See Also

Other priors: brm_prior_archetype(), brm_prior_simple(), brm_prior_template()

Examples

set.seed(0L)
data <- brm_simulate_outline(
  n_group = 2,
  n_patient = 100,
  n_time = 3,
  rate_dropout = 0,
  rate_lapse = 0
) |>
  dplyr::mutate(response = rnorm(n = dplyr::n())) |>
  brm_simulate_continuous(names = c("biomarker1", "biomarker2")) |>
  brm_simulate_categorical(
    names = c("status1", "status2"),
    levels = c("present", "absent")
  )
archetype <- brm_archetype_successive_cells(data)
dplyr::distinct(data, group, time)
label <- NULL |>
  brm_prior_label("normal(1, 1)", group = "group_1", time = "time_1") |>
  brm_prior_label("normal(1, 2)", group = "group_1", time = "time_2") |>
  brm_prior_label("normal(1, 3)", group = "group_1", time = "time_3") |>
  brm_prior_label("normal(2, 1)", group = "group_2", time = "time_1") |>
  brm_prior_label("normal(2, 2)", group = "group_2", time = "time_2") |>
  brm_prior_label("normal(2, 3)", group = "group_2", time = "time_3")
label

Description

Generate a simple prior for a brms MMRM.

Usage

brm_prior_simple(
  data,
  formula,
  intercept = "student_t(3, 0, 2.5)",
  coefficients = "student_t(3, 0, 2.5)",
  sigma = "student_t(3, 0, 2.5)",
  unstructured = "lkj(1)",
  autoregressive = "",
  moving_average = "",
  compound_symmetry = "",
  correlation = NULL
)

Arguments

Details

In brm_prior_simple(), you can separately choose priors for the intercept, model coefficients, log-scale standard deviations, and pairwise correlations between time points within patients. However, each class of parameters is set as a whole. In other words, brm_prior_simple() cannot assign different priors to different fixed effect parameters.

Value

A classed data frame with the brms prior.

See Also

Other priors: brm_prior_archetype(), brm_prior_label(), brm_prior_template()

Examples

set.seed(0L)
data <- brm_simulate_outline()
data <- brm_simulate_continuous(data, names = c("age", "biomarker"))
formula <- brm_formula(
  data = data,
  baseline = FALSE,
  baseline_time = FALSE,
  check_rank = FALSE
)
brm_prior_simple(
  data = data,
  formula = formula,
  intercept = "student_t(3, 0, 2.5)",
  coefficients = "normal(0, 10)",
  sigma = "student_t(2, 0, 4)",
  unstructured = "lkj(2.5)"
)

Description

Template for the label argument of brm_prior_archetype().

Usage

brm_prior_template(archetype)

Arguments

Details

The label argument of brm_prior_archetype() is a tibble which maps Stan code for univariate priors to fixed effect parameters in the model. Usually this tibble is built gradually using multiple calls to brm_prior_label(), but occasionally it is more convenient to begin with a full template and manually write Stan code in the code column. brm_prior_template() creates this template.

Value

A tibble with one row per fixed effect parameter and columns to map Stan code to each parameter. After manually writing Stan code in the code column of the template, you can supply the result to the label argument of brm_prior_archetype() to build a brms prior for your model.

Prior labeling

Informative prior archetypes use a labeling scheme to assign priors to fixed effects. How it works:

1. First, assign the prior of each parameter a collection
  of labels from the data. This can be done manually or with
  successive calls to [brm_prior_label()].
2. Supply the labeling scheme to [brm_prior_archetype()].
  [brm_prior_archetype()] uses attributes of the archetype
  to map labeled priors to their rightful parameters in the model.

For informative prior archetypes, this process is much more convenient and robust than manually calling brms::set_prior(). However, it requires an understanding of how the labels of the priors map to parameters in the model. This mapping varies from archetype to archetype, and it is documented in the help pages of archetype-specific functions such as brm_archetype_successive_cells().

See Also

Other priors: brm_prior_archetype(), brm_prior_label(), brm_prior_simple()

Examples

set.seed(0L)
data <- brm_simulate_outline(
  n_group = 2,
  n_patient = 100,
  n_time = 3,
  rate_dropout = 0,
  rate_lapse = 0
) |>
  dplyr::mutate(response = rnorm(n = dplyr::n())) |>
  brm_simulate_continuous(names = c("biomarker1", "biomarker2")) |>
  brm_simulate_categorical(
    names = c("status1", "status2"),
    levels = c("present", "absent")
  )
archetype <- brm_archetype_successive_cells(data)
label <- brm_prior_template(archetype)
label$code <- c(
  "normal(1, 1)",
  "normal(1, 2)",
  "normal(1, 3)",
  "normal(2, 1)",
  "normal(2, 2)",
  "normal(2, 3)"
)
brm_prior_archetype(label = label, archetype = archetype)

Description

Change the center of a nuisance variable of an informative prior archetype.

Usage

brm_recenter_nuisance(data, nuisance, center)

Arguments

Details

By "centering vector y at scalar x", we mean taking the difference z = y - x. If x is the mean, then mean(z) is 0. Informative prior archetypes center nuisance variables at their means so the parameters can be interpreted correctly for setting informative priors. This is appropriate most of the time, but sometimes it is better to center a column at a pre-specified scientifically meaningful fixed number. If you want a nuisance column to be centered at a fixed value other than its mean, use brm_recenter_nuisance() to shift the center. This function can handle any nuisance variable

Value

An informative prior archetype data frame with one of the variables re-centered.

Examples

set.seed(0L)
data <- brm_simulate_outline(
  n_group = 2,
  n_patient = 100,
  n_time = 4,
  rate_dropout = 0,
  rate_lapse = 0
) |>
  dplyr::mutate(response = rnorm(n = dplyr::n())) |>
  brm_data_change() |>
  brm_simulate_continuous(names = c("biomarker1", "biomarker2")) |>
  brm_simulate_categorical(
    names = c("status1", "status2"),
    levels = c("present", "absent")
  )
archetype <- brm_archetype_cells(data)
mean(archetype$nuisance_biomarker1) # after original centering
center <- mean(data$biomarker1)
center # original center, before the centering from brm_archetype_cells()
attr(archetype$nuisance_biomarker1, "brm_center") # original center
max(abs((data$biomarker1 - center) - archetype$nuisance_biomarker1))
# Re-center nuisance_biomarker1 at 0.75.
archetype <- brm_recenter_nuisance(
  data = archetype,
  nuisance = "nuisance_biomarker1",
  center = 0.75
)
attr(archetype$nuisance_biomarker1, "brm_center") # new center
mean(archetype$nuisance_biomarker1) # no longer equal to the center
# nuisance_biomarker1 is now as though we centered it at 0.75.
max(abs((data$biomarker1 - 0.75) - archetype$nuisance_biomarker1))

Description

Simulate and append non-time-varying categorical covariates to an existing brm_data() dataset.

Usage

brm_simulate_categorical(data, names, levels, probabilities = NULL)

Arguments

Details

Each covariate is a new column of the dataset with one independent random categorical draw for each patient, using a fixed set of levels (via base::sample() with replace = TRUE). All covariates simulated this way are independent of everything else in the data, including other covariates (to the extent that the random number generators in R work as intended).

Value

A classed tibble, like from brm_data() or brm_simulate_outline(), but with new categorical covariate columns and with the names of the new covariates appended to the brm_covariates attribute. Each new categorical covariate column is a character vector, not the factor type in base R.

See Also

Other simulation: brm_simulate_continuous(), brm_simulate_outline(), brm_simulate_prior(), brm_simulate_simple()

Examples

data <- brm_simulate_outline()
brm_simulate_categorical(
  data = data,
  names = c("site", "region"),
  levels = c("area1", "area2")
)
brm_simulate_categorical(
  data = data,
  names = c("site", "region"),
  levels = c("area1", "area2"),
  probabilities = c(0.1, 0.9)
)

Description

Simulate and append non-time-varying continuous covariates to an existing brm_data() dataset.

Usage

brm_simulate_continuous(data, names, mean = 0, sd = 1)

Arguments

Details

Each covariate is a new column of the dataset with one independent random univariate normal draw for each patient. All covariates simulated this way are independent of everything else in the data, including other covariates (to the extent that the random number generators in R work as intended).

Value

A classed tibble, like from brm_data() or brm_simulate_outline(), but with new numeric covariate columns and with the names of the new covariates appended to the brm_covariates attribute.

See Also

Other simulation: brm_simulate_categorical(), brm_simulate_outline(), brm_simulate_prior(), brm_simulate_simple()

Examples

data <- brm_simulate_outline()
brm_simulate_continuous(
  data = data,
  names = c("age", "biomarker")
)
brm_simulate_continuous(
  data = data,
  names = c("biomarker1", "biomarker2"),
  mean = 1000,
  sd = 100
)

Description

Begin creating a simulated dataset.

Usage

brm_simulate_outline(
  n_group = 2L,
  n_subgroup = NULL,
  n_patient = 100L,
  n_time = 4L,
  rate_dropout = 0.1,
  rate_lapse = 0.05
)

Arguments

Value

A classed data frame from brm_data(). The data frame has one row per patient per time point and the following columns:

• group: integer index of the treatment group.

• patient: integer index of the patient.

• time: integer index of the discrete time point.

See Also

Other simulation: brm_simulate_categorical(), brm_simulate_continuous(), brm_simulate_prior(), brm_simulate_simple()

Examples

brm_simulate_outline()

Description

Simulate the outcome variable from the prior predictive distribution of an MMRM using brms.

Usage

brm_simulate_prior(
  data,
  formula,
  prior = brms.mmrm::brm_prior_simple(data = data, formula = formula),
  ...
)

Arguments

Details

brm_simulate_prior() calls brms::brm() with sample_prior = "only", which sets the default intercept prior using the outcome variable and requires at least some elements of the outcome variable to be non-missing in advance. So to provide feasible and consistent output, brm_simulate_prior() temporarily sets the outcome variable to all zeros before invoking brms::brm().

Value

A list with the following elements:

• data: a classed tibble with the outcome variable simulated as a draw from the prior predictive distribution (the final row of outcome in the output). If you simulated a missingness pattern with brm_simulate_outline(), then that missingness pattern is applied so that the appropriate values of the outcome variable are set to NA.

• model: the brms model fit object.

• model_matrix: the model matrix of the fixed effects, obtained from brms::make_standata().

• outcome: a numeric matrix with one column per row of data and one row per saved prior predictive draw.

• parameters: a tibble of saved parameter draws from the prior predictive distribution.

See Also

Other simulation: brm_simulate_categorical(), brm_simulate_continuous(), brm_simulate_outline(), brm_simulate_simple()

Examples

if (identical(Sys.getenv("BRM_EXAMPLES", unset = ""), "true")) {
set.seed(0L)
data <- brm_simulate_outline()
data <- brm_simulate_continuous(data, names = c("age", "biomarker"))
data$response <- rnorm(nrow(data))
formula <- brm_formula(
  data = data,
  baseline = FALSE,
  baseline_time = FALSE
)
tmp <- utils::capture.output(
  suppressMessages(
    suppressWarnings(
      out <- brm_simulate_prior(
        data = data,
        formula = formula
      )
    )
  )
)
out$data
}

Description

Simple function to simulate a dataset from a simple specialized MMRM.

Usage

brm_simulate_simple(
  n_group = 2L,
  n_patient = 100L,
  n_time = 4L,
  hyper_beta = 1,
  hyper_tau = 0.1,
  hyper_lambda = 1
)

Arguments

Details

Refer to the methods vignette for a full model specification. The brm_simulate_simple() function simulates a dataset from a simple pre-defined MMRM. It assumes a cell means structure for fixed effects, which means there is one fixed effect scalar parameter (element of vector beta) for each unique combination of levels of treatment group and discrete time point. The elements of beta have independent univariate normal priors with mean 0 and standard deviation hyper_beta. The residual log standard deviation parameters (elements of vector tau) have normal priors with mean 0 and standard deviation hyper_tau. The residual correlation matrix parameter lambda has an LKJ correlation prior with shape parameter hyper_lambda.

Value

A list of three objects:

• data: A tidy dataset with one row per patient per discrete time point and columns for the outcome and ID variables.

• model_matrix: A matrix with one row per row of data and columns that represent levels of the covariates.

• parameters: A named list of parameter draws sampled from the prior:

  • beta: numeric vector of fixed effects.

  • tau: numeric vector of residual log standard parameters for each time point.

  • sigma: numeric vector of residual standard parameters for each time point. sigma is equal to exp(tau).

  • lambda: correlation matrix of the residuals among the time points within each patient.

  • covariance: covariance matrix of the residuals among the time points within each patient. covariance is equal to diag(sigma) %*% lambda %*% diag(sigma).

See Also

Other simulation: brm_simulate_categorical(), brm_simulate_continuous(), brm_simulate_outline(), brm_simulate_prior()

Examples

set.seed(0L)
simulation <- brm_simulate_simple()
simulation$data

Description

Transformation from model parameters to marginal means.

Usage

brm_transform_marginal(
  data,
  formula,
  average_within_subgroup = NULL,
  prefix = "b_"
)

Arguments