Skip to contents

Statistical Methods

Source: vignettes/methods_stat.Rmd
methods_stat.Rmd

Statistical method types

The {chef} framework categorizes statistical functions into three main categories depending on how they operate over stratification levels and treatment levels. Operations can either be done once per stratification level or treatment level, or they can operate across levels.

Figure 1 below shows a typical table from a safety analysis and color codes each result based on the function type that creates it:

Figure 1

Figure 1


Table 1 below provides a description of each of the three function types, color coded to the outputs they produce in table 1 above:

Table 1
Name Description Example function Example application
stat_by_strata_and_trt One output per combination of strata level and treatment level for each strata type (e.g. SEX, AGEGR, etc): n_subj() - Counts number of subjects in each stratification level and treatment level How many female subjects are in TreatmentA (answer = 120)
stat_by_strata_across_trt One output per strata level, across treatment levels for each strata type (e.g. SEX, AGEGR, etc): RR() - Calculates the relative risk for the outcome/event within a stratification level What is the RR of having the outcome for females in Treatment A as compared to females in Treatment B (answer = 15)
stat_across_strata_across_trt One output per strata type (e.g. SEX, AGEGR, etc), across treatment levels p_val_interaction() - Test of interaction effect for stratification level Is the risk of having the outcome different across the strata of SEX

The function type also determines when the function is applied. In our example from figure 1, a stat_by_strata_by_trt function will be called once for each combination of stratification level and treatment level, as well as once for TOTALS for each treatment level:

  1. SEX == “MALE” & TRT == “TreatmentA”
  2. SEX == “FEMALE” & TRT == “TreatmentA”
  3. SEX == “MALE” & TRT == “TreatmentB”
  4. SEX == “FEMALE” & TRT == “TreatmentB”
  5. AGE == “<65” & TRT == “TreatmentA”
  6. AGE == “>=65” & TRT == “TreatmentA”
  7. AGE == “<65” & TRT == “TreatmentB”
  8. AGE == “>=65” & TRT == “TreatmentB”
  9. TOTAL_ == “total” & TRT == “TreatmentA”
  10. TOTAL_ == “total” & TRT == “TreatmentB”

Whereas a stat_across_strata_across_trt will only be called once per stratification “group”, i.e.:

  1. strata_var == “SEX”
  2. strata_var == “AGE”

Project-specific statistical functions

If you need a statistical function that does not exist in {chefStats}, we writing it and adding it the chefStats package, that way it can be used by others - please see the documentation in the chefStats package for more details.

But sometimes this might not make sense. In that case you can always reference your own custom function in the endpoint specification stored in the R/ directory of your project folder.

For example you want a new function my_custom_stats_function of type stat_by_strata_by_trt . You first define the function:

my_custom_function <- function(
    # function arguments here
  ){
  # Function does something here
}

Then once it’s save to R/, you can reference it in the endpoint specification just like chefStats functions:

chef::mk_endpoint_str(
  ..., # other arguments to mk_endpoint_str
  stat_by_strata_by_trt = my_custom_function)

Important: if the custom function relies on any external packages, you need to add those packages to the packages.R file in R/.