chnaged input formula so that uses : rather than * so that it doesnt implicitly include the effect modifiers also as prognostic variables

Author: n8thangreen

R/prep_data.R

@@ -10,17 +10,21 @@ prep_ipd <- function(form, data) {
 prep_ald <- function(form, data) {
 
   all_term_labels <- attr(terms(form), "term.labels")
+  trt_name <- get_treatment_name(form)
 
-  # Remove duplicates (since age and I(age^2) will both appear)
+  # pull out original of transformed variable
+  # remove duplicates (since X and I(X^2) will both appear)
   term.labels <- unique(gsub("I\\(([^)]+)\\^2\\)", "\\1", all_term_labels))
 
+  # remove treatment
+  # interaction separate by :
+  term.labels <- unlist(strsplit(term.labels, ":", fixed = TRUE))
+  term.labels <- setdiff(term.labels, trt_name)
+  
   mean_names <- paste0("mean.", term.labels)
   sd_names <- paste0("sd.", term.labels)  ##TODO: for maic do we need these?
   term_names <- c(mean_names, sd_names)
 
-  # remove treatment labels
-  term_names <- sort(term_names[!grepl(pattern = "trt", term_names)])
-  
   # replace outcome variable name
   response_var <- all.vars(form)[1]
   response_names <- gsub(pattern = "y", replacement = response_var,

tests/testthat/test-parse_formula.R

@@ -1,5 +1,5 @@
 
-# unit test functions in parse_formula.R
+# unit test functions in parse_formula.R and prep_data.R
 
 
 test_that("test get_eff_mod_names", {
@@ -39,3 +39,53 @@ test_that("get_covariate_names", {
   expect_equal(get_covariate_names(y ~ x*z + z), c("x","z"))
 })
 
+
+test_that("prep_ald differenf formula formats", {
+  
+  # mock data
+  data <- data.frame(
+    mean.X1 = 0.6065438,
+    sd.X1    = 0.3917644,
+    mean.X2  = 0.6273452,
+    sd.X2    = 0.4031580,
+    mean.X3  = 0.5579042,
+    sd.X3    = 0.4077155,
+    mean.X4  = 0.6122686,
+    sd.X4    = 0.4466750,
+    y.C.sum  = 36,
+    y.C.bar  = 0.5373134,
+    y.C.sd   = 0.5023689,
+    N.C      = 67,
+    y.B.sum  = 35,
+    y.B.bar  = 0.2631579,
+    y.B.sd   = 0.4420122,
+    N.B      = 133,
+    stringsAsFactors = FALSE
+  )
+  
+  form <- as.formula("y ~ X3 + X4 + trt*X1 + trt*X2")
+  
+  prep_ald(form, data)
+  
+  form <- as.formula("y ~ X3 + X4 + trt + X1 + X2 + trt*X1 + trt*X2")
+
+  prep_ald(form, data)
+  
+  form <- as.formula("y ~ X3 + X4 + trt*(X1 + X2)")
+  
+  prep_ald(form, data)
+  
+  form <- as.formula("y ~ X3 + X4 + trt:(X1 + X2)")
+  
+  prep_ald(form, data)
+  
+  form <- as.formula("y ~ X3 + X4 + trt:X1 + trt:X2")
+  
+  prep_ald(form, data)
+  
+  form <- as.formula("y ~ X3 + X4 + trt + trt:X1 + trt:X2")
+  
+  prep_ald(form, data)
+
+})
+

vignettes/Binary_data_example.Rmd

@@ -235,28 +235,42 @@ A `strategy` argument of `outstandR` takes functions called
 particular method required, e.g. `strategy_maic()` for MAIC. Each
 specific example is provided below.
 
-### MAIC
-
-Using the individual level data for *AC* firstly we perform
-non-parametric bootstrap of the `maic.boot` function with `R = 1000`
-replicates. This function fits treatment coefficient for the marginal
-effect for *A* vs *C*. The returned value is an object of class `boot`
-from the `{boot}` package. We then calculate the bootstrap mean and
-variance in the wrapper function `maic_boot_stats`.
+### Model formula
 
-Defining $X_1,X2$ as effect modifiers and $X_3,X_4$ as prognostic variables then the formula used in this model is
+Defining $X_1, X_2$ as effect modifiers, $X_3, X_4$ as prognostic variables and $Z$ the treatment indicator then the formula used in this model is
 
 $$
-y = X_3 + X_4 + \beta_t X_1 + \beta_t X_2
+y = X_3 + X_4 + Z + Z X_1 + Z X_2
 $$
 
 which corresponds to the following `R` `formula` object passed as an
 argument to the strategy function.
 
 ```{r}
-lin_form <- as.formula("y ~ X3 + X4 + trt*X1 + trt*X2")
+lin_form <- as.formula("y ~ X3 + X4 + trt + trt:X1 + trt:X2")
+```
+
+Note that the more succinct formula
+
+```{r, eval=FALSE}
+y ~ X3 + X4 + trt*(X1 + X2)
 ```
 
+Would also include $X_1, X_2$ as prognostic factors so in not equivalent, but could be modified as follows.
+
+```{r, eval=FALSE}
+y ~ X3 + X4 + trt*(X1 + X2) - X1 - X2
+```
+
+### MAIC
+
+Using the individual level data for *AC* firstly we perform
+non-parametric bootstrap of the `maic.boot` function with `R = 1000`
+replicates. This function fits treatment coefficient for the marginal
+effect for *A* vs *C*. The returned value is an object of class `boot`
+from the `{boot}` package. We then calculate the bootstrap mean and
+variance in the wrapper function `maic_boot_stats`.
+
 ```{r outstandR_maic}
 outstandR_maic <-
   outstandR(AC.IPD, BC.ALD,
@@ -359,7 +373,7 @@ the previous analysis but we now use the `strategy_stc()` strategy
 function instead and a formula with centered covariates.
 
 $$
-y = X_3 + X_4 + \beta_t(X_1 - \bar{X_1}) + \beta_t(X_2 - \bar{X_2})
+y = X_3 + X_4 + Z + Z (X_1 - \bar{X_1}) + Z (X_2 - \bar{X_2})
 $$
 
 However, `outstandR()` knows how to handle this so we can simply pass