when mean_ae_site_med75 == mean_ae_study_med75 overreporting must be 0 (

#57)

R/simaerep.R

@@ -411,7 +411,14 @@ eval_sites <- function(df_sim_sites,
 
     if (! under_only) {
       df_out <- df_out %>%
-        mutate(prob_high = 1 - .data$prob_low) %>%
+        mutate(
+          prob_high = 1 - .data$prob_low,
+          prob_high = ifelse(
+            .data$mean_ae_site_med75 == .data$mean_ae_study_med75,
+            1,
+            .data$prob_high
+          )
+        ) %>%
         arrange(.data$study_id, .data$prob_high) %>%
         mutate(
           prob_high_adj = p.adjust(.data$prob_high, method = method),

tests/testthat/test_over_reporting.R

@@ -43,3 +43,23 @@ test_that("simaerep() - under_only = FALSE return over-reporting statistics", {
   ))
 
 })
+
+test_that(paste("simaerep() - under_only = FALSE - over-reporting must be zero when",
+                "mean_ae_site_med75 == mean_ae_study_med75"), {
+
+  df_visit <- sim_test_data_study(
+    ae_per_visit_mean = 0
+  ) %>%
+  mutate(study_id = "A")
+
+  aerep <- simaerep(df_visit, under_only = FALSE)
+
+  zeros <- aerep$df_eval %>%
+    filter(mean_ae_study_med75 == mean_ae_site_med75) %>%
+    summarise(across(c(prob_low_prob_ur, prob_high_prob_or), sum)) %>%
+    summarise(across(c(prob_low_prob_ur, prob_high_prob_or), ~ . == 0)) %>%
+    unlist()
+
+  expect_true(all(zeros))
+
+})

vignettes/over.Rmd

@@ -76,7 +76,7 @@ setdiff(colnames(aerep_ovr$df_eval), colnames(aerep_def$df_eval))
 
 # Analyze
 
-## Probability getting a lower or equal AE count
+## Probability getting a lower AE count
 
 
 ```{r}
@@ -107,11 +107,10 @@ ggExtra::ggMarginal(p, groupColour = TRUE, type = "density")
 
 We can see that we have a gap for the default setting in the generated probabilities. The values filling the gap can be interpreted as the probability of having a `higher` site average than originally observed.
 
-We need to be aware that values in the simulation `equal` to the site mean will count towards the under-reporting probability. This can potentially lead to artifacts with low AE counts.
 
 ## Over-Reporting
 
-We can add the over-reporting probability as (1- under-reporting probability)
+We can add the over-reporting probability as (1- under-reporting probability), for cases when mean_ae_site_med75 is equal to mean_ae_study_med75 over-reporting probability will always be zero.
 
 ```{r}