Update exercises2.qmd with calc_bucher_naive, fix outstandR undefined…

NAMESPACE

@@ -30,6 +30,7 @@ S3method(print,summary.outstandR)
 S3method(summary,outstandR)
 export(calc_ALD_stats)
 export(calc_IPD_stats)
+export(calc_bucher_naive)
 export(calc_gcomp_bayes)
 export(calc_gcomp_ml)
 export(calculate_ate)

R/outstandR.R

@@ -160,12 +160,12 @@ outstandR <- function(ipd_trial, ald_trial, strategy,
   stats <- result_stats(ipd_stats, ald_stats, CI)
 
   naive_results <- calc_bucher_naive(
-    ipd_trial = ipd_trial_clean,
-    ald_trial = ald_trial_clean,
-    outcome_model = strategy$formula$outcome_model,
+    ipd_trial = ipd_trial,
+    ald_trial = ald_trial,
+    outcome_model = strategy$outcome_model,
     family = strategy$family,
     ref_trt = ref_trt,
-    scale = scale_internal
+    scale = scale
   )
 
   structure(

scripts/exercises2.qmd

@@ -289,7 +289,24 @@ prob_C_raw_BC <- BC_ALD_binY_contX |>
   pull(value)
 
 # 3. Naive Anchored Risk Difference
-naive_anchored_rd <- (prob_A_raw - prob_C_raw_AC) - (prob_B_raw - prob_C_raw_BC)
+# Explicit manual calculation (direct subtraction):
+naive_anchored_rd_explicit <- (prob_A_raw - prob_C_raw_AC) - (prob_B_raw - prob_C_raw_BC)
+
+# Or using outstandR's built-in calc_bucher_naive() function:
+# This calculates it on the log-odds scale and back-transforms, which is HTA best practice.
+naive_results_pkg <- calc_bucher_naive(
+  ipd_trial = AC_IPD_binY_contX,
+  ald_trial = BC_ALD_binY_contX,
+  outcome_model = lin_form_bin,
+  family = binomial(link = "logit"),
+  ref_trt = "C",
+  scale = "risk_difference"
+)
+naive_anchored_rd <- naive_results_pkg$Estimate
+
+# Note: You can also extract this directly from any of the outstandR output objects
+# since the naive Bucher comparison is run automatically under the hood:
+# naive_anchored_rd <- out_maic_rd$naive$Estimate
 
 # 4. Extract and bind results
 adjusted_results <- data.frame(
@@ -440,9 +457,6 @@ ggplot(data.frame(weights = out_maic$model$weights), aes(x = weights)) +
 
 If a large percentage of your weights are near zero, your MAIC estimates may be highly unstable, suggesting poor covariate overlap between your IPD and target population.
 
-<<<<<<< HEAD
-2.\ Next, let's explore the Bias-Variance trade-off by running a minimal, correctly specified MAIC balancing *only one* the effect modifiers.
-=======
 2.\ Fit a deliberately misspecified G-computation model to observe the consequences. We will purposefully drop the `trt:EM` interaction terms.
 
 ```{r}
@@ -466,7 +480,6 @@ cat("Correct ML G-comp Risk Difference (A vs B):", out_gcomp_ml_rd$results$contr
 ```
 
 3.\ Next, let's explore the Bias-Variance trade-off by running a minimal, correctly specified MAIC balancing *only one* effect modifier.
->>>>>>> f052889ab1209f78958e28b82c66c95af339a23c
 
 ```{r}
 #| echo: true
@@ -625,6 +638,16 @@ prob_A_naive <- odds_A_naive / (1 + odds_A_naive)
 
 naive_anchored_rd <- prob_A_naive - prob_B_raw
 
+# Alternatively, the above back-transformation can be done in one line using calc_bucher_naive():
+# naive_anchored_rd <- calc_bucher_naive(
+#   ipd_trial = AC_IPD_binY_contX,
+#   ald_trial = BC_ALD_binY_contX,
+#   outcome_model = lin_form_bin,
+#   family = binomial(link = "logit"),
+#   ref_trt = "C",
+#   scale = "risk_difference"
+# )$Estimate
+
 # 5. Naive Unanchored Risk Difference
 # A simple unadjusted direct comparison of A vs B
 naive_unanchored_rd <- prob_A_raw - prob_B_raw

tests/testthat/test-gcomp.R

@@ -35,7 +35,7 @@ test_that("different combinations of covariates in formula", {
                    ald_trial = BC_ALD,
                    strategy = strat_1234)
 
-  expect_length(res, 13)
+  expect_length(res, 14)
   # expect_equal(outstandR(AC_IPD, BC_ALD, strategy = strat_31))
   # expect_equal(outstandR(AC_IPD, BC_ALD, strategy = strat_13))
   # expect_equal(outstandR(AC_IPD, BC_ALD, strategy = strat_1))
@@ -67,7 +67,7 @@ test_that("different combinations of covariates in formula", {
                    ald_trial = BC_ALD,
                    strategy = strat_1234)
 
-  expect_length(res, 13)
+  expect_length(res, 14)
 
   # expect_equal(outstandR(AC_IPD, BC_ALD, strategy = strat_31))
   # expect_equal(outstandR(AC_IPD, BC_ALD, strategy = strat_13))

tests/testthat/test-maicstc.R

@@ -42,7 +42,7 @@ test_that("different combinations of covariates in formula MAIC", {
     formula = list(balance_model = as.formula("~ X1")))
 
   res <- outstandR(AC_IPD, BC_ALD, strategy = strat_1234)
-  expect_length(res, 13)
+  expect_length(res, 14)
 
   res <- outstandR(AC_IPD, BC_ALD, strategy = strat_123)
 

tests/testthat/test-mim.R

@@ -28,10 +28,10 @@ test_that("different combinations of covariates in formula", {
 
   out_1234 <- outstandR(AC_IPD, BC_ALD, strategy = strat_1234)
 
-  expect_length(out_1234, 13)
+  expect_length(out_1234, 14)
 
   expect_named(out_1234, expected = 
-                 c("results", "call", 
+                 c("results", "naive", "call", 
                    "outcome_model", "balance_model", "formula", 
                    "CI", "ref_trt", "ipd_comp", 
                    "ald_comp", "scale", "var_method", "family", "model"))