Pooled logistic regression with MICE - unable to obtain confint about mice HOT 8 CLOSED

The summary.mira() function has changed in mice 3.0. The names of the columns now conform to the broom package. You can request the estimates directly in the linear or log odds scale, as follows

require(mice, warn.conflicts = FALSE)
packageVersion("mice")
set.seed(123)
nhanes$hyp <- as.factor(nhanes$hyp)

# Can run logistic regression on (non-imputed) dataset
model <- glm(hyp ~ bmi, family = binomial(link = 'logit'), nhanes)
model_or <- exp(cbind(OR = coef(model), confint(model)))

# Using the nhanes dataset as an example
imputed_data <- mice::mice(nhanes, m = 25, method = "pmm", 
                           maxit = 10, seed = 12345, print = FALSE)
imputed_model <- with(imputed_data, 
                      glm(hyp ~ bmi, family = binomial(link = 'logit')))

# linear scale
round(summary(pool(imputed_model), conf.int = TRUE), 3)

# log odds scale
round(summary(pool(imputed_model), conf.int = TRUE, exponentiate = TRUE), 3)

from mice.

That's hugely helpful - thank you for the very quick reply!

from mice.

Using mice v3.3.0, when running the above example code I get the following warning when pooling on the log odds scale using round(summary(pool(imputed_model), conf.int = TRUE, exponentiate = TRUE), 3):

Warning message:
In process_mipo(z, object, conf.int = conf.int, conf.level = conf.level,  : 
    Exponentiating coefficients, but model did not use a log or logit link function

This is despite computing the imputed_model object with family = binomial(link = 'logit')) as in the example code.

Is it safe to ignore this message, as we know the model was computed using a logit link function?

from mice.

I am experiencing a very similar issue as @jaminday - any update on this?

I am calling the function as follows:

imp_long %>%
 by(as.factor(.$.imp), glm, formula=k2, family=binomial(link="logit")) %>%
  pool() %>% 
   summary(conf.int=T, exponentiate=T) 

Warning message is the same as above.

from mice.

The warning comes from an inappropriate test of the mipo object, which does not preserve the model settings. The warning can be ignored. It is actually superfluous, and therefore removed in mice 3.4.2.

from mice.

The summary.mira() function has changed in mice 3.0. The names of the columns now conform to the broom package. You can request the estimates directly in the linear or log odds scale, as follows

require(mice, warn.conflicts = FALSE)
packageVersion("mice")
set.seed(123)
nhanes$hyp <- as.factor(nhanes$hyp)

# Can run logistic regression on (non-imputed) dataset
model <- glm(hyp ~ bmi, family = binomial(link = 'logit'), nhanes)
model_or <- exp(cbind(OR = coef(model), confint(model)))

# Using the nhanes dataset as an example
imputed_data <- mice::mice(nhanes, m = 25, method = "pmm", 
                           maxit = 10, seed = 12345, print = FALSE)
imputed_model <- with(imputed_data, 
                      glm(hyp ~ bmi, family = binomial(link = 'logit')))

# linear scale
round(summary(pool(imputed_model), conf.int = TRUE), 3)

# log odds scale
round(summary(pool(imputed_model), conf.int = TRUE, exponentiate = TRUE), 3)

Using mice v3.3.0, when running the above example code I get the following warning when pooling on the log odds scale using round(summary(pool(imputed_model), conf.int = TRUE, exponentiate = TRUE), 3):

Using mice v3.15.0, when running the above example code I get the following warning when pooling on the log odds scale using round(summary(pool(imputed_model), conf.int = TRUE, exponentiate = TRUE), 3):
Warning message: Error in Math.data.frame(list(term = 1:35, estimate = c(0.46500606968507, :
non-numeric-alike variable(s) in data frame: term.
I do not know why, This question has been bothering me for a long time, I hope you can reply.

from mice.

The column term from the summary output is categorical, so the warning indicates that the desired process in rounding makes no sense there.

from mice.

Very thanks! That's hugely helpful ! I have also two questions to ask:

1. I have been using MICE to generate imputed datasets, Is it possible to stratify this imputed datasets according to gender and then perform logistic regression. If so how can I stratify this imputed datasets?
2. I have been using MICE to generate imputed datasets, Is it possible to run BKMR package analysis (or WQS analysis)using this imputed datasets?
   Many thanks for any assistance.

from mice.