Use with simulated portfolio data to generate under-reporting stats for specified scenarios.
dataframe as returned by sim_test_data_portfolio
numeric, set maximum number of additional under-reporting sites, see details Default: 3
numeric vector, set under-reporting rates for scenarios Default: c(0.25, 0.5)
integer, denotes number of simulations, default = 1000
logical, calculates poisson.test pvalue
logical, calculates probability for getting a lower value
logical, use parallel processing see details, Default: FALSE
logical, show progress bar, Default: TRUE
named list of parameters passed to
site_aggr
, Default: list()
named list of parameters passed to
eval_sites
, Default: list()
dataframe with the following columns:
study identification
site identification
number of patients at site
number of patients at site with visit_med75
median(max(visit)) * 0.75
mean AE at visit_med75 site level
mean AE at visit_med75 study level
number of patients at site with visit_med75 at study excl site
additional sites with under-reporting patients
ratio of patients in study that are under-reporting
under-reporting rate
p-value as
returned by poisson.test
bootstrapped probability for having mean_ae_site_med75 or lower
adjusted p-values
adjusted bootstrapped probability for having mean_ae_site_med75 or lower
probability under-reporting as 1 - pval_adj, poisson.test (use as benchmark)
probability under-reporting as 1 - prob_low_adj, bootstrapped (use)
The function will apply under-reporting scenarios to each site. Reducing the number of AEs by a given under-reporting (ur_rate) for all patients at the site and add the corresponding under-reporting statistics. Since the under-reporting probability is also affected by the number of other sites that are under-reporting we additionally calculate under-reporting statistics in a scenario where additional under reporting sites are present. For this we use the median number of patients per site at the study to calculate the final number of patients for which we lower the AEs in a given under-reporting scenario. We use the furrr package to implement parallel processing as these simulations can take a long time to run. For this to work we need to specify the plan for how the code should run, e.g. plan(multisession, workers = 18)
sim_test_data_study
get_config
sim_test_data_portfolio
sim_ur_scenarios
get_portf_perf
# \donttest{
df_visit1 <- sim_test_data_study(n_pat = 100, n_sites = 10,
frac_site_with_ur = 0.4, ur_rate = 0.6)
df_visit1$study_id <- "A"
df_visit2 <- sim_test_data_study(n_pat = 100, n_sites = 10,
frac_site_with_ur = 0.2, ur_rate = 0.1)
df_visit2$study_id <- "B"
df_visit <- dplyr::bind_rows(df_visit1, df_visit2)
df_site_max <- df_visit %>%
dplyr::group_by(study_id, site_number, patnum) %>%
dplyr::summarise(max_visit = max(visit),
max_ae = max(n_ae),
.groups = "drop")
df_config <- get_config(df_site_max)
df_config
#> # A tibble: 20 × 6
#> study_id ae_per_visit_mean site_number max_visit_sd max_visit_mean n_pat
#> <chr> <dbl> <chr> <dbl> <dbl> <int>
#> 1 0001 0.366 0001 4.42 20 10
#> 2 0001 0.366 0002 4.03 21.4 10
#> 3 0001 0.366 0003 4.13 20.2 10
#> 4 0001 0.366 0004 2.58 18.3 10
#> 5 0001 0.366 0005 4.64 17.8 10
#> 6 0001 0.366 0006 2.37 17.6 10
#> 7 0001 0.366 0007 4.80 19.8 10
#> 8 0001 0.366 0008 2 20 10
#> 9 0001 0.366 0009 3.17 19.5 10
#> 10 0001 0.366 0010 6.57 19.9 10
#> 11 0002 0.489 0001 2.85 19.9 10
#> 12 0002 0.489 0002 3.31 18.1 10
#> 13 0002 0.489 0003 3.14 18.1 10
#> 14 0002 0.489 0004 4.74 20.7 10
#> 15 0002 0.489 0005 5.20 19.2 10
#> 16 0002 0.489 0006 3.30 21 10
#> 17 0002 0.489 0007 4.07 19.9 10
#> 18 0002 0.489 0008 4.53 18.5 10
#> 19 0002 0.489 0009 2.95 21.7 10
#> 20 0002 0.489 0010 3.36 20.2 10
df_portf <- sim_test_data_portfolio(df_config)
df_portf
#> # A tibble: 3,834 × 8
#> study_id ae_per_visit_mean site_number max_visit_sd max_visit_mean patnum
#> <chr> <dbl> <chr> <dbl> <dbl> <chr>
#> 1 0001 0.366 0001 4.42 20 0001
#> 2 0001 0.366 0001 4.42 20 0001
#> 3 0001 0.366 0001 4.42 20 0001
#> 4 0001 0.366 0001 4.42 20 0001
#> 5 0001 0.366 0001 4.42 20 0001
#> 6 0001 0.366 0001 4.42 20 0001
#> 7 0001 0.366 0001 4.42 20 0001
#> 8 0001 0.366 0001 4.42 20 0001
#> 9 0001 0.366 0001 4.42 20 0001
#> 10 0001 0.366 0001 4.42 20 0001
#> # ℹ 3,824 more rows
#> # ℹ 2 more variables: visit <int>, n_ae <int>
df_scen <- sim_ur_scenarios(df_portf,
extra_ur_sites = 2,
ur_rate = c(0.5, 1))
#> aggregating site level
#> prepping for simulation
#> generating scenarios
#> getting under-reporting stats
#> evaluating stats
df_scen
#> # A tibble: 140 × 14
#> study_id site_number n_pat n_pat_with_med75 visit_med75 mean_ae_site_med75
#> <chr> <chr> <int> <int> <dbl> <dbl>
#> 1 0001 0001 10 9 16 5.44
#> 2 0001 0001 10 9 16 2.72
#> 3 0001 0001 10 9 16 0
#> 4 0001 0001 10 9 16 2.72
#> 5 0001 0001 10 9 16 0
#> 6 0001 0001 10 9 16 2.72
#> 7 0001 0001 10 9 16 0
#> 8 0001 0002 10 9 16 4.89
#> 9 0001 0002 10 9 16 2.44
#> 10 0001 0002 10 9 16 0
#> # ℹ 130 more rows
#> # ℹ 8 more variables: mean_ae_study_med75 <dbl>, n_pat_with_med75_study <int>,
#> # extra_ur_sites <dbl>, frac_pat_with_ur <dbl>, ur_rate <dbl>,
#> # prob_low <dbl>, prob_low_adj <dbl>, prob_low_prob_ur <dbl>
df_perf <- get_portf_perf(df_scen)
df_perf
#> # A tibble: 27 × 5
#> fpr thresh extra_ur_sites ur_rate tpr
#> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1 0.001 0.858 0 0 0.05
#> 2 0.001 0.858 1 0 0.05
#> 3 0.001 0.858 2 0 0.05
#> 4 0.001 0.858 0 0.5 1
#> 5 0.001 0.858 1 0.5 1
#> 6 0.001 0.858 2 0.5 1
#> 7 0.001 0.858 0 1 1
#> 8 0.001 0.858 1 1 1
#> 9 0.001 0.858 2 1 1
#> 10 0.01 0.857 0 0 0.05
#> # ℹ 17 more rows
# }