2: Minimum prototype for pilot

DESCRIPTION

@@ -1,29 +1,39 @@
 Type: Package
 Package: tern.gee
-Title: TLGs using Generalized Estimating Equations (GEE)
-Version: 0.0.0.9001
-Date: 2022-08-22
-Authors@R:
-    person("insightsengineering", , , "insightsengineering@example.com", role = c("aut", "cre"))
-Description: Create TLGs using Generalized Estimating Equations (GEE).
+Title: Tables and Graphs for Generalized Estimating Equations (GEE) Model
+    Fits
+Version: 0.0.0.9002
+Date: 2022-10-25
+Authors@R: c(
+    person("Daniel", "Sabanés Bové", , "daniel.sabanes_bove@roche.com", role = c("aut", "cre")),
+    person("F. Hoffmann-La Roche AG", role = c("cph", "fnd"))
+  )
+Description: Create tables and graphs for GEE model fits.
 License: Apache License 2.0 | file LICENSE
 URL: https://github.com/insightsengineering/tern.gee/
 BugReports: https://github.com/insightsengineering/tern.gee/issues
 Depends:
-    R (>= 3.6)
+    R (>= 4.0),
+    tern (>= 0.7.9)
 Imports:
-    plumber,
-    shiny,
-    stringr
+    checkmate,
+    geeasy,
+    geepack,
+    nlme,
+    rtables,
+    stats,
+    formatters,
+    rtables
 Suggests:
-    future,
-    httr,
     knitr,
-    shinytest,
-    testthat (>= 2.0)
+    rmarkdown,
+    testthat (>= 3.1),
+    vdiffr
 VignetteBuilder:
     knitr
-biocViews:
+Remotes:
+    insightsengineering/tern@*release
+Config/testthat/edition: 3
 Encoding: UTF-8
 Language: en-US
 LazyData: true

NAMESPACE

@@ -1,6 +1,21 @@
 # Generated by roxygen2: do not edit by hand
 
-export(hello)
-export(plumber_api)
-export(shiny_app)
-importFrom(utils,packageName)
+S3method(QIC,tern_gee)
+S3method(VarCorr,tern_gee)
+S3method(as.rtable,tern_gee)
+S3method(lsmeans,tern_gee_logistic)
+export(a_lsmeans_logistic)
+export(as.rtable)
+export(fit_gee)
+export(lsmeans)
+export(s_lsmeans_logistic)
+export(summarize_gee_logistic)
+import(checkmate)
+import(rtables)
+importFrom(gee,gee)
+importFrom(geepack,QIC)
+importFrom(nlme,VarCorr)
+importFrom(rtables,add_colcounts)
+importFrom(stats,acf)
+importFrom(tern,as.rtable)
+importFrom(tern,f_conf_level)

NEWS.md

@@ -1,4 +1,7 @@
-# tern.gee 0.0.0.9001
-### New features
+# tern.gee 0.0.0.9002
 
-* New package
+### Miscellaneous
+
+* Initialize the package.
+* Update pkgdown site and README, add unit tests.
+* Add examples for `fit_gee`, `lsmeans`, `s_lsmeans_logistic`, `summarize_gee_logistic`.

R/data.R

@@ -0,0 +1,15 @@
+#' Example dataset for `tern.gee` package.
+#'
+#' Measurements of FEV1 (forced expired volume in one second) is a measure of
+#' how quickly the lungs can be emptied.
+#' Low levels of FEV1 may indicate chronic obstructive pulmonary disease (COPD).
+#' @format A `tibble` with 800 rows and 7 variables:
+#'
+#'   - `USUBJID`: price, in US dollars.
+#'   - `AVISIT`: visit number.
+#'   - `ARMCD`: treatment, `TRT` or `PBO`.
+#'   - `RACE`: 3-category race.
+#'   - `SEX`: sex.
+#'   - `FEV1_BL`: FEV1 at baseline (%).
+#'   - `FEV1`: FEV1 at study visits.
+"fev_data"

R/fit_gee.R

@@ -0,0 +1,157 @@
+vars_gee <- function(response = "AVAL",
+                     covariates = c(),
+                     id = "USUBJID",
+                     arm = "ARM",
+                     visit = "AVISIT") {
+  list(
+    response = response,
+    covariates = covariates,
+    id = id,
+    arm = arm,
+    visit = visit
+  )
+}
+
+#' @keywords internal
+build_formula <- function(vars) {
+  assert_list(vars)
+  arm_part <- if (is.null(vars$arm)) NULL else vars$arm
+  rhs_formula <- paste(
+    c(arm_part, vars$covariates),
+    collapse = " + "
+  )
+  stats::as.formula(paste(
+    vars$response,
+    "~",
+    rhs_formula
+  ))
+}
+
+#' @keywords internal
+build_family <- function(regression) {
+  assert_string(regression)
+
+  result_object <- switch(regression,
+    logistic = stats::binomial(link = "logit"),
+    stop(paste("regression type", regression, "not supported"))
+  )
+
+  result_class <- paste0("tern_gee_", regression)
+
+  list(
+    object = result_object,
+    class = result_class,
+    control = geeasy::geelm.control(scale.fix = TRUE)
+  )
+}
+
+#' @keywords internal
+build_cor_details <- function(cor_str, vars, data) {
+  assert_string(cor_str)
+  assert_list(vars)
+  assert_data_frame(data)
+
+  result_str <- switch(cor_str,
+    "unstructured" = "unstructured",
+    "toeplitz" = "m-dependent",
+    "compound symmetry" = "exchangeable",
+    "auto-regressive" = "ar1",
+    stop(paste("correlation structure", cor_str, "not available"))
+  )
+
+  result_mv <- switch(cor_str,
+    "unstructured" = 1,
+    "toeplitz" = nlevels(data[[vars$visit]]) - 1,
+    "compound symmetry" = 1,
+    "auto-regressive" = 1
+  )
+
+  list(
+    str = result_str,
+    mv = result_mv
+  )
+}
+
+order_data <- function(data, vars) {
+  assert_data_frame(data)
+  assert_list(vars)
+
+  if (is.character(data[[vars$visit]])) {
+    message(paste("visit variable", vars$visit, "will be coerced to factor for ordering"))
+    message("order is:")
+    data[[vars$visit]] <- factor(data[[vars$visit]])
+    cat(toString(levels(data[[vars$visit]])))
+  }
+  if (is.factor(data[[vars$id]]) || is.character(data[[vars$id]])) {
+    data[[vars$id]] <- as.integer(as.factor(data[[vars$id]]))
+  }
+  assert_numeric(data[[vars$id]])
+
+  right_order <- order(data[[vars$id]], data[[vars$visit]])
+  data[right_order, ]
+}
+
+#' Fit a GEE Model
+#'
+#' @param vars (`list`)\cr see [vars_gee()].
+#' @param data (`data.frame`)\cr input data.
+#' @param regression (`string`)\cr choice of regression model.
+#' @param cor_struct (`string`)\cr assumed correlation structure.
+#'
+#' @details The correlation structure can be:
+#' * `unstructured`: No constraints are placed on the correlations.
+#' * `toeplitz`: Assumes a banded correlation structure, i.e. the correlation
+#'    between two time points depends on the distance between the time indices.
+#' * `compound symmetry`: Constant correlation between all time points.
+#' * `auto-regressive`: Auto-regressive order 1 correlation matrix.
+#'
+#' @return Object of class `tern_gee` as well as specific to the kind of regression
+#'   which was used.
+#' @export
+#'
+#' @examples
+#' df <- fev_data
+#' df$AVAL <- as.integer(fev_data$FEV1 > 30)
+#'
+#' fit_gee(vars = vars_gee(arm = "ARMCD"), data = df)
+#'
+#' fit_gee(vars = vars_gee(arm = "ARMCD"), data = df, cor_struct = "compound symmetry")
+fit_gee <- function(vars = vars_gee(),
+                    data,
+                    regression = c("logistic"),
+                    cor_struct = c("unstructured", "toeplitz", "compound symmetry", "auto-regressive")) {
+  formula <- build_formula(vars)
+
+  regression <- match.arg(regression)
+  family <- build_family(regression)
+
+  data <- order_data(data, vars)
+  data[[".id"]] <- data[[vars$id]]
+  data[[".waves"]] <- as.integer(data[[vars$visit]])
+
+  cor_struct <- match.arg(cor_struct)
+  cor_details <- build_cor_details(cor_struct, vars, data)
+
+  fit <- geeasy::geelm(
+    formula = formula,
+    id = .id,
+    waves = .waves,
+    data = data,
+    family = family$object,
+    corstr = cor_details$str,
+    Mv = cor_details$mv,
+    control = family$control
+  )
+
+  fit$qic <- geepack::QIC(fit)
+  fit$visit_levels <- levels(data[[vars$visit]])
+  fit$vars <- vars
+  fit$data <- data
+  assert_factor(data[[vars$arm]])
+  fit$ref_level <- levels(data[[vars$arm]])[1L]
+
+  structure(
+    fit,
+    class = c(family$class, "tern_gee", class(fit))
+  )
+}

R/gee_methods.R

@@ -0,0 +1,45 @@
+#' Methods for GEE Models
+#'
+#' Additional methods which can simplify working with the GEE result object.
+#' @name gee_methods
+NULL
+
+#' @rdname gee_methods
+#' @importFrom nlme VarCorr
+#' @exportS3Method
+VarCorr.tern_gee <- function(x, sigma = 1, ...) {
+  dim_mat <- length(x$visit_levels)
+  tmp <- id_mat <- diag(dim_mat)
+  corest <- x$geese$alpha
+
+  # Start with lower-triangular matrix part.
+  lower_mat <- switch(x$corstr,
+    unstructured = , # Since this is the same as exchangeable, we can do this.
+    exchangeable = {
+      tmp[lower.tri(tmp)] <- corest
+      tmp
+    },
+    ar1 = {
+      row_col_diff <- row(tmp) - col(tmp)
+      tmp[lower.tri(tmp)] <- corest^(row_col_diff[lower.tri(row_col_diff)])
+      tmp
+    },
+    `m-dependent` = {
+      row_col_diff <- row(tmp) - col(tmp)
+      tmp[lower.tri(tmp)] <- corest[row_col_diff[lower.tri(row_col_diff)]]
+      tmp
+    }
+  )
+
+  # Construct the full symmetric matrix.
+  mat <- lower_mat + t(lower_mat) - id_mat
+  rownames(mat) <- colnames(mat) <- x$visit_levels
+  mat
+}
+
+#' @rdname gee_methods
+#' @importFrom geepack QIC
+#' @exportS3Method
+QIC.tern_gee <- function(object, ...) {
+  object$qic
+}