Update pvalue_maxcombo() and move dplyr to Suggests

DESCRIPTION

@@ -1,7 +1,7 @@
 Package: simtrial
 Type: Package
 Title: Clinical Trial Simulation
-Version: 0.3.0.2
+Version: 0.3.0.3
 Authors@R: c(
  person("Keaven", "Anderson", email = "[email protected]", role = c("aut")),
  person("Yilong", "Zhang", email = "[email protected]", role = c("aut")),
@@ -39,7 +39,6 @@ Imports:
  Rcpp,
  data.table,
  doFuture,
- dplyr,
  foreach,
  future,
  magrittr,
@@ -53,6 +52,7 @@ Suggests:
  Matrix,
  bshazard,
  covr,
+ dplyr,
  ggplot2,
  gsDesign,
  knitr,

NAMESPACE

@@ -30,10 +30,6 @@ importFrom(data.table,setDF)
 importFrom(data.table,setDT)
 importFrom(data.table,setorderv)
 importFrom(doFuture,"%dofuture%")
-importFrom(dplyr,group_by)
-importFrom(dplyr,mutate)
-importFrom(dplyr,select)
-importFrom(dplyr,starts_with)
 importFrom(future,plan)
 importFrom(magrittr,"%>%")
 importFrom(methods,is)

R/pvalue_maxcombo.R

@@ -25,15 +25,12 @@
 #' that provided in the ADaM ADTTE format.
 #'
 #' @param z A dataset output from [wlr()]; see examples.
-#' @param dummy_var A dummy input that allows [dplyr::group_map()] to be used to
-#' compute p-values for multiple simulations.
 #' @param algorithm This is passed directly to the `algorithm` argument
 #' in [mvtnorm::pmvnorm()].
 #'
 #' @return A numeric p-value.
 #'
 #' @importFrom mvtnorm pmvnorm GenzBretz
-#' @importFrom dplyr select starts_with
 #'
 #' @export
 #'
@@ -68,24 +65,25 @@
 #' # MaxCombo power estimate for cutoff at max of targeted events, minimum follow-up
 #' p <- xx %>%
 #' group_by(sim) %>%
-#' group_map(pvalue_maxcombo) %>%
+#' group_map(~ pvalue_maxcombo(.x)) %>%
 #' unlist()
 #' mean(p < .025)
 pvalue_maxcombo <- function(
  z,
- dummy_var,
  algorithm = mvtnorm::GenzBretz(maxpts = 50000, abseps = 0.00001)) {
- ans <- (1 - mvtnorm::pmvnorm(
- lower = rep(
- z$z %>% min() %>% as.numeric(),
- nrow(z)
- ),
- corr = z %>%
- select(starts_with("v")) %>%
- data.matrix(),
+ zmin <- as.numeric(min(z$z))
+ lower_limits <- rep.int(zmin, times = nrow(z))
+
+ correlation_matrix <- z[, grep("^v", colnames(z))]
+ correlation_matrix <- data.matrix(correlation_matrix)
+
+ distribution <- mvtnorm::pmvnorm(
+ lower = lower_limits,
+ corr = correlation_matrix,
  algorithm = algorithm
- )[1]
- ) %>% as.numeric()
+ )
+
+ ans <- 1 - distribution[1]
 
- ans
+ as.numeric(ans)
 }

R/sim_fixed_n.R

@@ -104,7 +104,7 @@
 #' p <- xx %>%
 #' filter(cut != "Targeted events") %>%
 #' group_by(sim) %>%
-#' group_map(pvalue_maxcombo) %>%
+#' group_map(~ pvalue_maxcombo(.x)) %>%
 #' unlist()
 #'
 #' mean(p < .025)
@@ -113,7 +113,7 @@
 #' p <- xx %>%
 #' filter(cut == "Targeted events") %>%
 #' group_by(sim) %>%
-#' group_map(pvalue_maxcombo) %>%
+#' group_map(~ pvalue_maxcombo(.x)) %>%
 #' unlist()
 #'
 #' mean(p < .025)

R/sim_pw_surv.R

@@ -57,7 +57,6 @@
 #' i.e., 1 is a failure, 0 is a dropout.
 #'
 #' @importFrom data.table ":=" .N data.table setDF setDT setorderv
-#' @importFrom dplyr group_by mutate
 #' @importFrom tibble tibble
 #'
 #' @export

tests/testthat/generate-backwards-compatible-test-data.R

@@ -300,4 +300,35 @@ generate_test_data <- function(
  # Compute p-value of modestly weighted logrank of Magirr-Burman
  ex2 <- pnorm(z)
  saveRDS(ex2, file.path(outdir, "mb_weight_ex2.rds"))
+
+ # pvalue_maxcombo() ------------------------------------------------------------
+
+ # Example 1
+ set.seed(12345)
+ x <- sim_fixed_n(
+ n_sim = 1,
+ timing_type = 5,
+ rho_gamma = data.frame(
+ rho = c(0, 0, 1),
+ gamma = c(0, 1, 1)
+ )
+ )
+ ex1 <- pvalue_maxcombo(x)
+ saveRDS(ex1, file.path(outdir, "pvalue_maxcombo_ex1.rds"))
+
+ # Example 2
+ # Only use cuts for events, events + min follow-up
+ set.seed(12345)
+ xx <- sim_fixed_n(
+ n_sim = 100,
+ timing_type = 5,
+ rho_gamma = data.frame(
+ rho = c(0, 0, 1),
+ gamma = c(0, 1, 1)
+ )
+ )
+
+ # MaxCombo power estimate for cutoff at max of targeted events, minimum follow-up
+ ex2 <- as.numeric(by(xx, xx$sim, pvalue_maxcombo))
+ saveRDS(ex2, file.path(outdir, "pvalue_maxcombo_ex2.rds"))
 }

tests/testthat/test-backwards-compatibility.R

@@ -313,3 +313,35 @@ test_that("mb_weight()", {
  expect_equal(observed, expected)
 })
 
+test_that("pvalue_maxcombo()", {
+ # Example 1
+ set.seed(12345)
+ x <- sim_fixed_n(
+ n_sim = 1,
+ timing_type = 5,
+ rho_gamma = data.frame(
+ rho = c(0, 0, 1),
+ gamma = c(0, 1, 1)
+ )
+ )
+ observed <- pvalue_maxcombo(x)
+ expected <- readRDS("fixtures/backwards-compatibility/pvalue_maxcombo_ex1.rds")
+ expect_equal(observed, expected)
+
+ # Example 2
+ # Only use cuts for events, events + min follow-up
+ set.seed(12345)
+ xx <- sim_fixed_n(
+ n_sim = 100,
+ timing_type = 5,
+ rho_gamma = data.frame(
+ rho = c(0, 0, 1),
+ gamma = c(0, 1, 1)
+ )
+ )
+
+ # MaxCombo power estimate for cutoff at max of targeted events, minimum follow-up
+ observed <- as.numeric(by(xx, xx$sim, pvalue_maxcombo))
+ expected <- readRDS("fixtures/backwards-compatibility/pvalue_maxcombo_ex2.rds")
+ expect_equal(observed, expected)
+})

tests/testthat/test-independent_test_counting_process.R

@@ -13,7 +13,7 @@ surv_to_count <- function(time, status, trt, strats) {
  ) # ensure left continuous
  }
  km <- db %>%
- group_by(strats) %>%
+ dplyr::group_by(strats) %>%
  dplyr::do(tidy_survfit(Surv(time, status) ~ 1, data = .))
 
  # KM estimator by stratum and treatment Group Predicted at Specified Time
@@ -27,11 +27,11 @@ surv_to_count <- function(time, status, trt, strats) {
  # Number of Event
  .x2 <- data.frame(time = .survfit$time, n.event = .survfit$n.event) %>% subset(n.event > 0)
 
- merge(.x1, .x2, all = TRUE) %>% mutate(n.event = dplyr::if_else(is.na(n.event), 0, n.event))
+ merge(.x1, .x2, all = TRUE) %>% dplyr::mutate(n.event = dplyr::if_else(is.na(n.event), 0, n.event))
  }
 
  km_by_trt <- db %>%
- group_by(strats, trt) %>%
+ dplyr::group_by(strats, trt) %>%
  dplyr::do(pred_time = pred_survfit(km[km$strats == .$strats[1], ]$time,
  Surv(time, status) ~ 1,
  data = .
@@ -43,7 +43,7 @@ surv_to_count <- function(time, status, trt, strats) {
  # Log Rank Expectation Difference and Variance
  res <- merge(km, km_by_trt, all = TRUE) %>%
  dplyr::arrange(trt, strats, time) %>%
- mutate(
+ dplyr::mutate(
  OminusE = tn.event - tn.risk / n.risk * n.event,
  Var = (n.risk - tn.risk) * tn.risk * n.event * (n.risk - n.event) / n.risk^2 / (n.risk - 1)
  )

vignettes/pvalue_maxcomboVignette.Rmd

@@ -134,7 +134,7 @@ We now consider the example simulation from the `pvalue_maxcombo()` help file to
 # Only use cut events + min follow-up
 xx <- sim_fixed_n(n_sim = 100, timing_type = 5, rho_gamma = tibble(rho = c(0, 0, 1), gamma = c(0, 1, 1)))
 # MaxCombo power estimate for cutoff at max of targeted events, minimum follow-up
-p <- unlist(xx %>% group_by(sim) %>% group_map(pvalue_maxcombo))
+p <- unlist(xx %>% group_by(sim) %>% group_map(~ pvalue_maxcombo(.x)))
 mean(p < .001)
 ```
 
@@ -152,15 +152,15 @@ Now we compute a p-value separately for each cut type, first for targeted event
 
 ```{r,warning=FALSE,message=FALSE}
 # Subset to targeted events cutoff tests
-p <- unlist(xx %>% filter(cut == "Targeted events") %>% group_by(sim) %>% group_map(pvalue_maxcombo))
+p <- unlist(xx %>% filter(cut == "Targeted events") %>% group_by(sim) %>% group_map(~ pvalue_maxcombo(.x)))
 mean(p < .025)
 ```
 
 Now we use the later of targeted events and minimum follow-up cutoffs.
 
 ```{r,warning=FALSE,message=FALSE}
 # Subset to targeted events cutoff tests
-p <- unlist(xx %>% filter(cut != "Targeted events") %>% group_by(sim) %>% group_map(pvalue_maxcombo))
+p <- unlist(xx %>% filter(cut != "Targeted events") %>% group_by(sim) %>% group_map(~ pvalue_maxcombo(.x)))
 mean(p < .025)
 ```