Bug in `AUC.competing.risks` with `cutpoints`

[loffleraSMH]

Thanks for maintaining this awesome package!

I noticed a small bug in AUC.competing.risks: When running the Score function on a competing risk model, I get the following error when specifying cutpoints:
Error in cutpoint.helper.fun(FPR, TPR, risk, ipcwCases, ipcwControls1, : object 'SE.NPV' not found.

I believe the issue can be resolved by changing line 162 in AUC.competing.risks to NPV <- SE.NPV <- NA.
Similarly, line 141 could be changed to PPV <- SE.PPV <- NA.

The following code replicates the issue:

library(riskRegression)
library(prodlim)
library(survival)

set.seed(123)

# Number of patients
n_patients <- 100

# Generate random patient data
my_data <- data.frame(
  id = 1:n_patients,
  age = sample(18:90, n_patients, replace = TRUE),
  sex = sample(c("Male", "Female"), n_patients, replace = TRUE),
  time = round(runif(n_patients, 30, 1000)),
  status = sample(c(0, 1, 2), n_patients, replace = TRUE, prob = c(0.5, 0.3, 0.2))  # Status: censored, disease, death
)

head(my_data)


model <- riskRegression::FGR(Hist(time, status) ~ age + sex,
                             data = my_data,
                             cause = 1
)

score_result <- Score(
  list(cr_model = model),
  data = my_data,
  formula = Hist(time, status) ~ 1,
  cause = 1,
  times = 365,
  metrics = c("auc", "brier"),
  summary = c("risks", "ipa"),
  plots = c("calibration", "ROC"),
  conf.int = TRUE,
  cutpoints = c(0.05, 0.1)
)

On a related note: Why are PPV and NPV only returned when Score is run with cutpoints input? The reason I ran into this bug was because I'm specifically interested in PPV, which is currently not returned when no cutpoints are provided. Would it be possible to enable PPV (and NPV) without cutpoints input by adding something like this after line 59 of AUC.competing.risks?

aucDT[, PPV := cumsum(ipcwCases) / (cumsum(ipcwCases) + cumsum(ipcwControls1) + cumsum(ipcwControls2))]

Thanks,
Anne

[tagteam]

Hej Anne thanks for reporting this bug. I have fixed the issue according to your proposal and some other improvements on github. will push to CRAN ASAP ;) best Thomas Anne Loffler ***@***.***> writes:

…

Thanks for maintaining this awesome package! I noticed a small bug in `AUC.competing.risks`: When running the `Score` function on a competing risk model, I get the following error when specifying cutpoints: `Error in cutpoint.helper.fun(FPR, TPR, risk, ipcwCases, ipcwControls1, : object 'SE.NPV' not found`. I believe the issue can be resolved by changing line 162 in `AUC.competing.risks` to `NPV <- SE.NPV <- NA`. Similarly, line 141 could be changed to `PPV <- SE.PPV <- NA`. The following code replicates the issue: ``` library(riskRegression) library(prodlim) library(survival) set.seed(123) # Number of patients n_patients <- 100 # Generate random patient data my_data <- data.frame( id = 1:n_patients, age = sample(18:90, n_patients, replace = TRUE), sex = sample(c("Male", "Female"), n_patients, replace = TRUE), time = round(runif(n_patients, 30, 1000)), status = sample(c(0, 1, 2), n_patients, replace = TRUE, prob = c(0.5, 0.3, 0.2)) # Status: censored, disease, death ) head(my_data) model <- riskRegression::FGR(Hist(time, status) ~ age + sex, data = my_data, cause = 1 ) score_result <- Score( list(cr_model = model), data = my_data, formula = Hist(time, status) ~ 1, cause = 1, times = 365, metrics = c("auc", "brier"), summary = c("risks", "ipa"), plots = c("calibration", "ROC"), conf.int = TRUE, cutpoints = c(0.05, 0.1) ) ``` On a related note: Why are PPV and NPV only returned when `Score` is run with `cutpoints` input? The reason I ran into this bug was because I'm specifically interested in PPV, which is currently not returned when no cutpoints are provided. Would it be possible to enable PPV (and NPV) without cutpoints input by adding something like this after line 59 of `AUC.competing.risks`? ``` aucDT[, PPV := cumsum(ipcwCases) / (cumsum(ipcwCases) + cumsum(ipcwControls1) + cumsum(ipcwControls2))] ``` Thanks, Anne

-- 7LL-1 Make peace with your past so it doesn’t spoil your present.

[loffleraSMH]

Thanks so much for the quick fix! :-)