How to do decision curve analysis on multiple imputation datasets?

[YiJuChou]

Hi
I am currently working on research about validation of prediction model.
I have successfully use R to perform a decision curve analysis.
However, there is another dataset which includes some missing values.
Therefore, if I want to perform DCA on the dataset, I will have to do multiple imputation first.
I know this will generate few imputed datasets.
There are solutions to pool the results of other analyses such as multivariate logistic regression.
But when it comes to DCA, I couldn't find any solution to this.
Will be great if anybody has a hint.
Thanks a lot.

[shaunporwal]

Hi Yi-Ju,

Thank you for your question. You can do it in this order:

1. Calculate the DCA dataframe object for each of the imputed datasets.
2. Combine (average) the DCA object risk score columns from all of the multiple imputation DCA datasets. This should give you a dataframe with a net benefit estimate at each threshold.
3. Plot the results.

[YiJuChou]

First, thank you for your kindly reply.
But I am still a little beat confused about how to do the above three steps.
If there are explanations on technical aspect will be helpful.

[VickersA]

A simple alternative is just to create one large data set from the imputed data sets and run the decision curve on that. Because we aren't interested in 95%CI, you don't have to worry about Rubin's rules for combining data etc etc

[shaunporwal]

@YiJuChou

Given Dr. Vickers' answer, you can approach it this way code-wise:

Load necessary libraries

library(mice) # For multiple imputation
library(dcurves) # For decision curve analysis
library(dplyr) # For data manipulation
library(ggplot2) # For plotting

Simulating a dataset with missing values

set.seed(123) # For reproducibility
data <- tibble(
patientid = 1:100,
cancer = rbinom(100, 1, 0.3),
risk_group = sample(c("low", "intermediate", "high"), 100, replace = TRUE),
age = rnorm(100, mean = 65, sd = 10),
famhistory = sample(c(NA, 1, 0), 100, replace = TRUE),
marker = runif(100),
cancerpredmarker = runif(100, 0, 0.6)
)

Adjust 'm' as needed for the number of imputations

num_imputations <- 5

E.g., Impute missing values using 'mice' package, mice = 'Multivariate Imputation by Chained Equations'

mice_data <- mice(data, m = num_imputations, method = 'pmm', maxit = 5, print = FALSE)

Create one large dataset from all imputed datasets using the dynamic 'm'

all_imputed_data <- do.call(rbind, lapply(1:num_imputations, function(i) complete(mice_data, action = i)))

Now perform DCA on this combined dataset

dca(cancer ~ cancerpredmarker + famhistory,
data = all_imputed_data,
thresholds = seq(0, 0.35, by = 0.01)) %>%
plot(smooth = TRUE)