AUCINT differences with mixed BLQ and above LOQ time points

[billdenney]

@billdenney I am also noticing this issue with AUCINT compared to Winonlin

Using this dummy dataset
example-ADNCA.csv

library(PKNCA)
library(dplyr)

# Load the data
data <- read.csv("example-ADNCA.csv", na.strings = c("", "NA"))

data <- data %>%
  filter(USUBJID == "S1-01",
         PARAM == "DrugA",
         PCSPEC == "SERUM")

# Load the PK concentration data
df_conc <- data

# Dosing dataset
df_dose <- df_conc %>%
  group_by(ATPTREF) %>%
  slice(1) %>%
  mutate(AFRLT = AFRLT - ARRLT)

# Create PKNCA objects
pknca_conc <- PKNCA::PKNCAconc(
  df_conc,
  formula = AVAL ~ AFRLT | STUDYID + PCSPEC + DOSETRT + USUBJID / PARAM,
  time.nominal = "NFRLT",
  concu = "AVALU",
  timeu = "RRLTU",
)

pknca_dose <- PKNCA::PKNCAdose(
  data = df_dose,
  formula = DOSEA ~ AFRLT | STUDYID + DOSETRT + USUBJID,
  route = "intravascular",
  time.nominal = "NFRLT",
  duration = "ADOSEDUR",
  doseu = "DOSEU"
)

intervals <-
  data.frame(
    start = 0, end = 24,
    ATPTREF = "DOSE1",
    auclast = TRUE,
    aucint.last = TRUE,
    clast.obs = TRUE,
    tlast = TRUE
  )


pknca_data_object <- PKNCA::PKNCAdata(
  data.conc = pknca_conc,
  data.dose = pknca_dose,
  intervals = intervals,
  options = list(
    auc.method = "lin up/log down",
    progress = FALSE,
    keep_interval_cols = "ATPTREF"
  ),
  impute = "start_conc0"
)
## Calculate the NCA parameters
results_obj <- as.data.frame(pk.nca(pknca_data_object))
print(results_obj)

If I run this code, AUCLast is 12.3 but AUCint is 20.1
In winlonlin, AUC 0 to 24 is 12.98.

Not sure what the reason for such a large difference is? Or which one is correct...
As far as I can tell I have followed the same method for both.

Originally posted by @js3110 in #431 (comment)

[billdenney]

From the discussion thread:

It appears that WNL is reporting AUCall for the AUC0-24 since there are BLQ values starting at 12 hours in your dataset.

For the AUCint in PKNCA, PKNCA is interpolating between your two above limit of quantification measures at 7.59 and 24.654 hr. I think that this is a bug.

For "which one is correct", my opinion is that the answer is probably somewhere in between WNL and PKNCA's current behavior.

PKNCA's behavior should change in the following ways:

1. Be more dose-aware, and do not extrapolate over a dose with AUCint, unless the dose is within the interval (your dose occurs at 24.1 hr; so extrapolation should not use any data after that time, unless your interval were >24.1 hr)
2. If half-life is estimable, extrapolate using the half-life (this is currently done in other cases, but not the case here)
3. If half-life is not estimable, use AUCall

We can then use PPANMETH to clarify what method was used.