Add SyntheticData generation script to data-raw

Author: Matt Secrest

data-raw/SyntheticData.R

@@ -0,0 +1,108 @@
+## code to prepare `SyntheticData` dataset
+## Based on real clinical trial summary statistics (SMA study)
+## Covariates are generated via copula to mimic realistic correlation structure
+
+library(copula)
+library(rdborrow)
+
+set.seed(202403)
+
+# ===== Copula for correlated covariates =====
+normal <- copula::normalCopula(param = c(0.8), dim = 4, dispstr = "ar1")
+
+# ===== Generate internal (RCT) covariates =====
+# x1 = SMA_Type (0 = Type II, 1 = Type III)
+# x2 = SMN2_Copy_Number (0 = 3 copies, 1 = 4 copies)
+# x3 = Scoliosis (0 = No, 1 = Yes)
+# x4 = Age_Enrollment
+X_int <- simulate_X_copula(
+   n = 200,
+   p = 4,
+   cp = normal,
+   margins = c("binom", "binom", "binom", "exp"),
+   paramMargins = list(
+      list(size = 1, prob = 0.7),
+      list(size = 1, prob = 0.9),
+      list(size = 1, prob = 0.3),
+      list(rate = 1 / 10)
+   )
+)
+
+X_int$x4 <- round(X_int$x4) + 1
+# x5 = Baseline outcome (Y0), modeled as linear function of covariates
+X_int$x5 <- 30 + 10 * X_int$x1 + 7 * X_int$x2 +
+   (-6) * X_int$x3 + (-0.5) * X_int$x4 +
+   rnorm(200, mean = 0, sd = 10)
+
+# ===== Generate external control covariates =====
+X_ext <- simulate_X_copula(
+   n = 100,
+   p = 4,
+   cp = normal,
+   margins = c("binom", "binom", "binom", "exp"),
+   paramMargins = list(
+      list(size = 1, prob = 0.7),
+      list(size = 1, prob = 0.9),
+      list(size = 1, prob = 0.3),
+      list(rate = 1 / 10)
+   )
+)
+
+X_ext$x4 <- round(X_ext$x4) + 1
+# External baseline outcome has different intercept/coefficients
+X_ext$x5 <- 50 + 10 * X_ext$x1 + 2 * X_ext$x2 +
+   (-1) * X_ext$x3 + (-0.3) * X_ext$x4 +
+   rnorm(100, mean = 0, sd = 10)
+
+# ===== Outcome model specifications =====
+varnames <- c("1", paste0("x", 1:5))
+
+# Coefficients: intercept, SMA_Type, SMN2_Copy, Scoliosis, Age, Y0
+# Treatment effect (A) and noise SD noted in comments
+model_form_x_t1 <- setNames(
+   c(10.0, 0.05, -1.5, -1.0, -0.2, -0.1), varnames
+) # effect = 1.5, sigma = 4.0
+
+model_form_x_t2 <- setNames(
+   c(6.0, 0.5, -0.5, -1.0, -0.3, -0.06), varnames
+) # effect = 1.8, sigma = 4.0
+
+model_form_x_t3 <- setNames(
+   c(5.0, 1.9, 1.4, -1.3, -0.4, -0.15), varnames
+) # effect = 1.6, sigma = 4.0
+
+model_form_x_t4 <- setNames(
+   c(1.2, 1.0, 2.0, -0.5, -0.4, -0.10), varnames
+) # effect = 2.5, sigma = 5.0
+
+outcome_model_specs <- list(
+   list(
+      effect = 0, model_form_x = model_form_x_t1,
+      noise_mean = 0, noise_sd = 4
+   ),
+   list(
+      effect = 1.0, model_form_x = model_form_x_t2,
+      noise_mean = 0, noise_sd = 4
+   ),
+   list(
+      effect = 2.0, model_form_x = model_form_x_t3,
+      noise_mean = 0, noise_sd = 4
+   ),
+   list(
+      effect = 5.0, model_form_x = model_form_x_t4,
+      noise_mean = 0, noise_sd = 4
+   )
+)
+
+# ===== Simulate trial data =====
+SyntheticData <- simulate_trial(
+   X_int,
+   X_ext,
+   num_treated = 100,
+   OLE_flag = TRUE,
+   T_cross = 2,
+   outcome_model_specs
+)
+
+# ===== Save to data/ =====
+usethis::use_data(SyntheticData, overwrite = TRUE)