Add method argument to s_odds_ratio() and estimate_odds_ratio() (…

…#1320)

# Pull Request

Fixes #1314

DESCRIPTION

@@ -48,7 +48,7 @@ Imports:
     rlang (>= 1.1.0),
     scales (>= 1.2.0),
     stats,
-    survival (>= 3.2-13),
+    survival (>= 3.7-0),
     tibble (>= 2.0.0),
     tidyr (>= 0.8.3),
     utils

NEWS.md

@@ -12,6 +12,7 @@
 * Refactored `estimate_incidence_rate` to work as both an analyze function and a summarize function, controlled by the added `summarize` parameter. When `summarize = TRUE`, labels can be fine-tuned via the new `label_fmt` argument to the same function.
 * Added `fraction` statistic to the `analyze_var_count` method group.
 * Improved `summarize_glm_count()` documentation and all its associated functions to better describe the results and the functions' purpose.
+* Added `method` argument to `s_odds_ratio()` and `estimate_odds_ratio()` to control whether exact or approximate conditional likelihood calculations are used. 
 
 ### Bug Fixes
 * Added defaults for `d_count_cumulative` parameters as described in the documentation.

R/odds_ratio.R

@@ -13,6 +13,8 @@
 #' @inheritParams argument_convention
 #' @param .stats (`character`)\cr statistics to select for the table. Run `get_stats("estimate_odds_ratio")`
 #'   to see available statistics for this function.
+#' @param method (`string`)\cr whether to use the correct (`"exact"`) calculation in the conditional likelihood or one
+#'   of the approximations. See [survival::clogit()] for details.
 #'
 #' @note
 #' * This function uses logistic regression for unstratified analyses, and conditional logistic regression for
@@ -64,7 +66,8 @@ s_odds_ratio <- function(df,
                          .df_row,
                          variables = list(arm = NULL, strata = NULL),
                          conf_level = 0.95,
-                         groups_list = NULL) {
+                         groups_list = NULL,
+                         method = "exact") {
   y <- list(or_ci = "", n_tot = "")
 
   if (!.in_ref_col) {
@@ -83,6 +86,7 @@ s_odds_ratio <- function(df,
       y <- or_glm(data, conf_level = conf_level)
     } else {
       assert_df_with_variables(.df_row, c(list(rsp = .var), variables))
+      checkmate::assert_subset(method, c("exact", "approximate", "efron", "breslow"), empty.ok = FALSE)
 
       # The group variable prepared for clogit must be synchronised with combination groups definition.
       if (is.null(groups_list)) {
@@ -118,14 +122,21 @@ s_odds_ratio <- function(df,
         grp = grp,
         strata = interaction(.df_row[variables$strata])
       )
-      y_all <- or_clogit(data, conf_level = conf_level)
+      y_all <- or_clogit(data, conf_level = conf_level, method = method)
       checkmate::assert_string(trt_grp)
       checkmate::assert_subset(trt_grp, names(y_all$or_ci))
       y$or_ci <- y_all$or_ci[[trt_grp]]
       y$n_tot <- y_all$n_tot
     }
   }
 
+  if ("est" %in% names(y$or_ci) && is.na(y$or_ci[["est"]]) && method != "approximate") {
+    warning(
+      "Unable to compute the odds ratio estimate. Please try re-running the function with ",
+      'parameter `method` set to "approximate".'
+    )
+  }
+
   y$or_ci <- formatters::with_label(
     x = y$or_ci,
     label = paste0("Odds Ratio (", 100 * conf_level, "% CI)")
@@ -163,8 +174,6 @@ a_odds_ratio <- make_afun(
 #' @describeIn odds_ratio Layout-creating function which can take statistics function arguments
 #'   and additional format arguments. This function is a wrapper for [rtables::analyze()].
 #'
-#' @param ... arguments passed to `s_odds_ratio()`.
-#'
 #' @return
 #' * `estimate_odds_ratio()` returns a layout object suitable for passing to further layouting functions,
 #'   or to [rtables::build_table()]. Adding this function to an `rtable` layout will add formatted rows containing
@@ -193,15 +202,15 @@ estimate_odds_ratio <- function(lyt,
                                 groups_list = NULL,
                                 na_str = default_na_str(),
                                 nested = TRUE,
-                                ...,
+                                method = "exact",
                                 show_labels = "hidden",
                                 table_names = vars,
                                 var_labels = vars,
                                 .stats = "or_ci",
                                 .formats = NULL,
                                 .labels = NULL,
                                 .indent_mods = NULL) {
-  extra_args <- list(variables = variables, conf_level = conf_level, groups_list = groups_list, ...)
+  extra_args <- list(variables = variables, conf_level = conf_level, groups_list = groups_list, method = method)
 
   afun <- make_afun(
     a_odds_ratio,
@@ -230,6 +239,7 @@ estimate_odds_ratio <- function(lyt,
 #'
 #' Functions to calculate odds ratios in [estimate_odds_ratio()].
 #'
+#' @inheritParams odds_ratio
 #' @inheritParams argument_convention
 #' @param data (`data.frame`)\cr data frame containing at least the variables `rsp` and `grp`, and optionally
 #'   `strata` for [or_clogit()].
@@ -300,19 +310,20 @@ or_glm <- function(data, conf_level) {
 #' or_clogit(data, conf_level = 0.95)
 #'
 #' @export
-or_clogit <- function(data, conf_level) {
+or_clogit <- function(data, conf_level, method = "exact") {
   checkmate::assert_logical(data$rsp)
   assert_proportion_value(conf_level)
   assert_df_with_variables(data, list(rsp = "rsp", grp = "grp", strata = "strata"))
   checkmate::assert_multi_class(data$grp, classes = c("factor", "character"))
   checkmate::assert_multi_class(data$strata, classes = c("factor", "character"))
+  checkmate::assert_subset(method, c("exact", "approximate", "efron", "breslow"), empty.ok = FALSE)
 
   data$grp <- as_factor_keep_attributes(data$grp)
   data$strata <- as_factor_keep_attributes(data$strata)
 
   # Deviation from convention: `survival::strata` must be simply `strata`.
   formula <- stats::as.formula("rsp ~ grp + strata(strata)")
-  model_fit <- clogit_with_tryCatch(formula = formula, data = data)
+  model_fit <- clogit_with_tryCatch(formula = formula, data = data, method = method)
 
   # Create a list with one set of OR estimates and CI per coefficient, i.e.
   # comparison of one group vs. the reference group.

tests/testthat/_snaps/odds_ratio.md

@@ -97,3 +97,12 @@
       ———————————————————————————————————————————————————————————
       Odds Ratio (95% CI)                     1.24 (0.54 - 2.89) 
 
+# estimate_odds_ratio method argument works
+
+    Code
+      res
+    Output
+                                    A            B
+      ————————————————————————————————————————————
+      Odds Ratio (95% CI)   0.96 (0.85 - 1.08)    
+

tests/testthat/test-odds_ratio.R

@@ -30,11 +30,7 @@ testthat::test_that("or_clogit estimates right OR and CI", {
     stringsAsFactors = TRUE
   )
 
-  # https://github.com/therneau/survival/issues/240
-  withr::with_options(
-    opts_partial_match_old,
-    result <- or_clogit(data, conf_level = 0.95)
-  )
+  result <- or_clogit(data, conf_level = 0.95)
 
   # from SAS
   res <- testthat::expect_silent(result)
@@ -66,17 +62,13 @@ testthat::test_that("s_odds_ratio estimates right OR and CI (stratified analysis
     strata = factor(sample(c("C", "D"), 100, TRUE))
   )
 
-  # https://github.com/therneau/survival/issues/240
-  withr::with_options(
-    opts_partial_match_old,
-    result <- s_odds_ratio(
-      df = subset(dta, grp == "A"),
-      .var = "rsp",
-      .ref_group = subset(dta, grp == "B"),
-      .in_ref_col = FALSE,
-      .df_row = dta,
-      variables = list(arm = "grp", strata = "strata")
-    )
+  result <- s_odds_ratio(
+    df = subset(dta, grp == "A"),
+    .var = "rsp",
+    .ref_group = subset(dta, grp == "B"),
+    .in_ref_col = FALSE,
+    .df_row = dta,
+    variables = list(arm = "grp", strata = "strata")
   )
 
   res <- testthat::expect_silent(result)
@@ -94,19 +86,15 @@ testthat::test_that("s_odds_ratio returns error for incorrect groups", {
     "Arms A+B" = c("A", "B")
   )
 
-  # https://github.com/therneau/survival/issues/240
-  withr::with_options(
-    opts_partial_match_old,
-    result <- testthat::expect_error(s_odds_ratio(
-      df = subset(data, grp == "A"),
-      .var = "rsp",
-      .ref_group = subset(data, grp == "B"),
-      .in_ref_col = FALSE,
-      .df_row = data,
-      variables = list(arm = "grp", strata = "strata"),
-      groups_list = groups
-    ))
-  )
+  testthat::expect_error(result <- s_odds_ratio(
+    df = subset(data, grp == "A"),
+    .var = "rsp",
+    .ref_group = subset(data, grp == "B"),
+    .in_ref_col = FALSE,
+    .df_row = data,
+    variables = list(arm = "grp", strata = "strata"),
+    groups_list = groups
+  ))
 })
 
 testthat::test_that("estimate_odds_ratio estimates right OR and CI (unstratified analysis)", {
@@ -132,14 +120,10 @@ testthat::test_that("estimate_odds_ratio estimates right OR and CI (stratified a
     strata = factor(sample(c("C", "D"), 100, TRUE))
   )
 
-  # https://github.com/therneau/survival/issues/240
-  withr::with_options(
-    opts_partial_match_old,
-    result <- basic_table() %>%
-      split_cols_by(var = "grp", ref_group = "A", split_fun = ref_group_position("first")) %>%
-      estimate_odds_ratio(vars = "rsp", variables = list(arm = "grp", strata = "strata")) %>%
-      build_table(df = data)
-  )
+  result <- basic_table() %>%
+    split_cols_by(var = "grp", ref_group = "A", split_fun = ref_group_position("first")) %>%
+    estimate_odds_ratio(vars = "rsp", variables = list(arm = "grp", strata = "strata")) %>%
+    build_table(df = data)
 
   res <- testthat::expect_silent(result)
   testthat::expect_snapshot(res)
@@ -170,12 +154,67 @@ testthat::test_that("estimate_odds_ratio works with strata and combined groups",
       groups_list = groups
     )
 
-  # https://github.com/therneau/survival/issues/240
-  withr::with_options(
-    opts_partial_match_old,
-    result <- build_table(lyt = lyt, df = anl)
+  result <- build_table(lyt = lyt, df = anl)
+
+  res <- testthat::expect_silent(result)
+  testthat::expect_snapshot(res)
+})
+
+testthat::test_that("s_odds_ratio method argument works", {
+  set.seed(1)
+  nex <- 2000 # Number of example rows
+  dta <- data.frame(
+    "rsp" = sample(c(TRUE, FALSE), nex, TRUE),
+    "grp" = sample(c("A", "B"), nex, TRUE),
+    "f1" = sample(c("a1", "a2"), nex, TRUE),
+    "f2" = sample(c("x", "y", "z"), nex, TRUE),
+    strata = factor(sample(c("C", "D"), nex, TRUE)),
+    stringsAsFactors = TRUE
+  )
+
+  res <- s_odds_ratio(
+    df = subset(dta, grp == "A"),
+    .var = "rsp",
+    .ref_group = subset(dta, grp == "B"),
+    .in_ref_col = FALSE,
+    .df_row = dta,
+    variables = list(arm = "grp", strata = "strata"),
+    method = "approximate"
   )
 
+  testthat::expect_false(all(is.na(res$or_ci)))
+
+  # warning works
+  expect_warning(
+    s_odds_ratio(
+      df = subset(dta, grp == "A"),
+      .var = "rsp",
+      .ref_group = subset(dta, grp == "B"),
+      .in_ref_col = FALSE,
+      .df_row = dta,
+      variables = list(arm = "grp", strata = "strata")
+    )
+  )
+})
+
+testthat::test_that("estimate_odds_ratio method argument works", {
+  nex <- 2000 # Number of example rows
+  set.seed(12)
+  dta <- data.frame(
+    "rsp" = sample(c(TRUE, FALSE), nex, TRUE),
+    "grp" = sample(c("A", "B"), nex, TRUE),
+    "f1" = sample(c("a1", "a2"), nex, TRUE),
+    "f2" = sample(c("x", "y", "z"), nex, TRUE),
+    strata = factor(sample(c("C", "D"), nex, TRUE)),
+    stringsAsFactors = TRUE
+  )
+
+  lyt <- basic_table() %>%
+    split_cols_by(var = "grp", ref_group = "B") %>%
+    estimate_odds_ratio(vars = "rsp", variables = list(arm = "grp", strata = "strata"), method = "approximate")
+
+  result <- build_table(lyt, df = dta)
+
   res <- testthat::expect_silent(result)
   testthat::expect_snapshot(res)
 })