clean.conc.blq: consider LLOQ

[Gero1999]

Currently the App assumes that all LLOQ data will be coded with 0s. This makes that all "keep" actions in clean.conc.blq are indeed the same as setting =0.

I wonder if it will make sense for this function to allow the user the option to set its own LLOQ limit. We can either follow a global strategy (set one LLOQ for all compounds in 1 PKNCAdata) or similar to the parameter units (allow user to define a LLOQ for each analyte). Here are some ideas on how this could be implemented:

1. As an optional list element for the conc.blq argument (i.e, list(first="keep", middle = 0, last = "drop", lloq = 0.05).

clean.conc.blq <- function(conc, time,
                           ...,
                           options=list(),
                           conc.blq=NULL,
                           conc.na=NULL,
                           check=TRUE) {
                           
### Rest of the code

lloq <- if (!is.null(conc.blq$lloq)) conc.blq$lloq else 0
# If all measurements are BLQ
    if (all(ret$conc == lloq)){
      # Apply "first" BLQ rule to everything for tfirst/tlast
      tfirst <- max(ret$time)
      tlast <- tfirst + 1
      
### Rest of the code      

2. Included in the global PKNCA.options, perhaps even as a data.frame for each analyte. I guess then can be obtained in the options argument in clean.conc.blq

[billdenney]

I'm not sure that I understand the difference you're asking for here.

Are you suggesting that data which has not had the LLOQ yet applied to it to have PKNCA to do that data management? For example if you had a concentration value of 0.02 and set lloq to 0.05 like your example, PKNCA would then apply that limit for you (so, PKNCA would set the concentration to 0)?

[Gero1999]

It would actually only affect the mask variables of the function, without transforming the original data. So it is just to determine which concentration values the blq rule applies. This in my opinion can have multiple advantages. such as

• In the cases when you want to keep the original concentration value with the action "keep" you are not just forced to keep the premade 0s, but rather the original concentrations.

• It opens up new possibilities for other useful BLQ operations that are not keep or drop (i.e, lloq or lloq/2)

A dummy example of what I was exactly thinking would be:

# Sample data
data <- data.frame(
  analyte = "a1",
  time = c(0, 1, 2, 3, 4),
  conc = c(0.01, 0.5, 0.01, 1.5, 0.01)
)

# Example of list BLQ for PKNCA.options
conc.blq <- list(first = 0, middle = "keep", last = "lloq", lloq = 0.05)

#Expected result
data <- data.frame(
  analyte = "a1",
  time = c(0, 1, 2, 3, 4),
  conc = c(0, 0.5, 0.01, 1.5, 0.05)
)

[billdenney]

Applying the BLQ processing (0.01 < 0.05 from your example) should be done as data management before data input to PKNCA. BLQ handling becomes complex quickly; it is not always the same per analyte and per study. (As an example, I have had an analyte with four different LLOQ values within one study because of dilution and an assay update during the study.) Trying to handle that within PKNCA is a data management, not NCA, task.

And, keeping 0.01 which is BLQ in the data is against regulatory standards that I'm aware of. Values that are BLQ should be considered as BLQ without information of what the value is. (So, keeping the 0.01 at 2 hours would not be a typical analysis nor what we would support within PKNCA directly-- that is another reason it's a data management task before PKNCA rather than within.)

With that in mind, the behavior you're wanting is already present:

library(PKNCA)
#> 
#> Attaching package: 'PKNCA'
#> The following object is masked from 'package:stats':
#> 
#>     filter
  
data <- data.frame(
  analyte = "a1",
  time = c(0, 1, 2, 3, 4),
  conc = c(0, 0.5, 0, 1.5, 0)
)

result <- data.frame(
  analyte = "a1",
  time = c(0, 1, 2, 3, 4),
  conc = c(0, 0.5, 0.01, 1.5, 0.05)
)

clean.conc.blq(
  data$conc, data$time,
  conc.blq=list(first = 0, middle = "keep", last=0.05), conc.na=0.5
)
#>   conc time
#> 1 0.00    0
#> 2 0.50    1
#> 3 0.00    2
#> 4 1.50    3
#> 5 0.05    4

^{Created on 2025-02-14 with reprex v2.1.1}

[Gero1999]

You are actually right , it is probably over complicating it. I just found that for aspects related with certain pk plots or even in some preclinical projects they still like to use 1/2 LLOQ imputation. But as you pointed out I can simply do the value assignation myself. So this proposal might be a wrong call, I will close this discussion!