assign_quartile() function

[callistosp]

Problem: existing functions in R which assign quartiles will assign different quartiles to the same numeric value to force an equal number of values into each quartile. Instead, we would like a function that will calculate the quartile values and assigns values to each quartile while ensuring that the same value will not appear in multiple quartiles, resulting in an interpretable numeric range for each quartile.

Proposed solution: this function was adapted from @graceannobrien's code on a recent project

assign_quartile <- function(.df, .cov){
  qt <- .df %>% 
    distinct(USUBJID,{{.cov}}) %>% 
    mutate(
      COV_DRANK = dense_rank({{.cov}}),
      COV_DRANK_MAX = max(COV_DRANK, na.rm = T),
      COV_DRANK_FINAL = COV_DRANK/COV_DRANK_MAX,
      COVQ = case_when(
        COV_DRANK_FINAL <= 0.25 ~ 1,
        COV_DRANK_FINAL <= 0.5 & COV_DRANK_FINAL > 0.25 ~ 2,
        COV_DRANK_FINAL <= 0.75 & COV_DRANK_FINAL > 0.5 ~ 3,
        COV_DRANK_FINAL > 0.75 ~ 4
      )
    ) %>% select(USUBJID, COVQ)
  
  names(qt)[2] <- paste0(str_to_upper(deparse(substitute(.cov))),"Q")
  
  return(full_join(.df, qt, by="USUBJID"))
}

[graceannobrien]

@michaelmcd18 I made dummy dataset to test if worked for duplicates as lowest or highest number like we discussed, and it puts them into same groups how we want.

temp <- tibble( nam= c(0.1, 0.1, 0.6, 0.55, 0.58, 0.99, 9, 9, 9, 1.25, 1.58, 1.58, 1.58, 1.58))
temp %>%  
      mutate(dnam = dense_rank(nam),     
                    dnam_max = max(dnam, na.rm = T),
                    nam_final = dnam/dnam_max)
... add quartile code etc. 

# A tibble: 14 × 4
     nam  dnam dnam_max nam_final
   <dbl> <int>    <int>     <dbl>
 1  0.1      1        8     0.125
 2  0.1      1        8     0.125
 3  0.55     2        8     0.25 
 4  0.58     3        8     0.375
 5  0.6      4        8     0.5  
 6  0.99     5        8     0.625
 7  1.25     6        8     0.75 
 8  1.58     7        8     0.875
 9  1.58     7        8     0.875
10  1.58     7        8     0.875
11  1.58     7        8     0.875
12  9        8        8     1    
13  9        8        8     1    
14  9        8        8     1 

[callistosp]

Was just looking at large simulation results, and we found an issue with this function. If there are a large amount of duplicated values in a data set, the function doesn't equally allocate to quartiles. If you switch dense_rank for min_rank, then it accounts for duplicates in the calculations and the results are as expected.

[callistosp]

relevant example from the documentation:

• min_rank() gives every tie the same (smallest) value so that c(10, 20, 20, 30) gets ranks c(1, 2, 2, 4). It's the way that ranks are usually computed in sports and is equivalent to rank(ties.method = "min").
• dense_rank() works like min_rank(), but doesn't leave any gaps, so that c(10, 20, 20, 30) gets ranks c(1, 2, 2, 3).

[callistosp]

revisiting this function, I think it would be helpful to have the option to also append a column with the values of the quartile. Here is how I am currently creating such a column:

exp_quartiles <- exposures %>% 
  assign_quartile(CMAXSS) %>% 
  group_by(CMAXSSQ) %>% 
  mutate(
    CMAXSSQV = paste(pmtables::sig(min(CMAXSS)), 
                     "-", pmtables::sig(max(CMAXSS)), "nM")
  ) %>% 
  ungroup()