exclude message for half.life calculations on flat lines

[Gero1999]

Do you think it might be worth adding a exclude message for this output case? Basically when all values are the same and we just obtain a flat line for the concentrations.... Currently when running pk.nca it produces NAs but leaves exclude without any reason for it. This is an example case

df <- Theoph %>%
  filter(Subject == 1) %>%
  mutate(conc = 1)
o_conc <- PKNCAconc(df, formula = "conc ~ Time | Subject")
o_data <- PKNCAdata(data.conc = o_conc, intervals = data.frame(lambda.z = TRUE, start = 0, end = Inf))
as.data.frame(pk.nca(o_data))

# A tibble: 11 × 6
   Subject start   end PPTESTCD            PPORRES exclude
   <ord>   <dbl> <dbl> <chr>                 <dbl> <chr>  
 1 1           0   Inf tmax                    0   NA     
 2 1           0   Inf tlast                  24.4 NA     
 3 1           0   Inf lambda.z               NA   NA     
 4 1           0   Inf r.squared              NA   NA     
 5 1           0   Inf adj.r.squared          NA   NA     
 6 1           0   Inf lambda.z.time.first    NA   NA     
 7 1           0   Inf lambda.z.time.last     NA   NA     
 8 1           0   Inf lambda.z.n.points      NA   NA     
 9 1           0   Inf clast.pred             NA   NA     
10 1           0   Inf half.life              NA   NA     
11 1           0   Inf span.ratio             NA   NA     

[billdenney]

Yes, there should be an exclude message here. The issue appears to come from here where all lambda.z are -Inf or 0:

pknca/R/half.life.R

Lines 232 to 243 in 3b72ee4

	mask_best <-
	half_lives_for_selection$lambda.z > 0 &
	if (min.hl.points == 2 & nrow(half_lives_for_selection) == 2) {
	rlang::warn(
	message = "2 points used for half-life calculation",
	class = "pknca_halflife_2points"
	)
	TRUE
	} else {
	half_lives_for_selection$adj.r.squared >
	(max(half_lives_for_selection$adj.r.squared, na.rm=TRUE) - adj.r.squared.factor)
	}

Can you please make a PR that does more detailed checking for this? When making that PR, can you please make it apply to both automatic and manual half-life calculations?

A simpler reprex is pk.calc.half.life(conc = rep(1, 5), time = 0:4, tmax = 0).

[Gero1999]

And 2 points used for half-life calculation should not be also a exclude message for r.squared, adj.r.squared rather than just a warning? It will make it more CDISC reportable and specific to the profile. Right now, this is what happens:

df <- Theoph %>%
  filter(Subject == 1) %>%
  filter(Time %in% c(2.02, 7.03, 12.12))
o_conc <- PKNCAconc(df, formula = "conc ~ Time | Subject")
o_data <- PKNCAdata(data.conc = o_conc, intervals = data.frame(lambda.z = TRUE, start = 0, end = Inf))
PKNCA.options(min.hl.points = 2)
as.data.frame(pk.nca(o_data))

No dose information provided, calculations requiring dose will return NA.
# A tibble: 12 × 7
   Subject start   end PPTESTCD            PPORRES PPANMETH exclude
   <ord>   <dbl> <dbl> <chr>                 <dbl> <chr>    <chr>  
 1 1           0   Inf tmax                 2.02   ""       NA     
 2 1           0   Inf tlast               12.1    ""       NA     
 3 1           0   Inf lambda.z             0.0450 ""       NA     
 4 1           0   Inf r.squared            1      ""       NA     
 5 1           0   Inf adj.r.squared       NA      ""       NA     
 6 1           0   Inf lambda.z.corrxy     -1      ""       NA     
 7 1           0   Inf lambda.z.time.first  7.03   ""       NA     
 8 1           0   Inf lambda.z.time.last  12.1    ""       NA     
 9 1           0   Inf lambda.z.n.points    2      ""       NA     
10 1           0   Inf clast.pred           5.94   ""       NA     
11 1           0   Inf half.life           15.4    ""       NA     
12 1           0   Inf span.ratio           0.331  ""       NA     
Warning messages:
1: n must be > 2 for adj.r.squared 
2: 2 points used for half-life calculation 

For me personally does not sound right to consider an r.squared of 2 points for any summary table, but what do you think?

[Gero1999]

although perhaps it makes more sense for this case to just create a exclude function to check for the 2 points case and exclude on parameters r.squared, adj.r.squared...

[billdenney]

For the two-points in a half life, the only way that a user would get that is if they intentionally requested two points (either by changing the minimum or by manually specifying points). (Aside, scientifically, I think that I've only ever done that on very rare occasions as sensitivity analysis.)

So, the user has specifically requested 2 points, my opinion is that we should allow it.

Specifically for r.squared and corrxy, I agree that they are not logical. I agree with your assessment that the best is probably to create an exclude function for them. The current default exclude mechanism doesn't allow for choosing within the parameters since they are all returned at once. I think that an even better solution would be to extend the way that theexclude attribute works so that it could also be a data.frame giving control over each parameter.

If you go down this path, please make it a separate PR.

[Gero1999]

Good point! I created an independent issue to think/work on this (#505)

On the other hand, the original issue of this thread is addressed in PR #504