how to deal with the error in using jstable::TableSubgroupMultiGLM( ) for subgroup analysis of multiple logistic regression

[agnesisyss]

hi, kim, I am agnes.
Thanks for your useful package.
I am a tiro of R， I use this package jstable::TableSubgroupMultiGLM( ) for subgroup analysis of multiple logistic regression
below is my code:
data.design <- svydesign(id = ~sdmvpsu,
weights = nhs_wt,
data = A)
TableSubgroupMultiGLM(YX, var_subgroups = c("sex","eth","maritalstatus"), #both Y and X are categorical variables

•                   data = data.design, 
  

•                   family = "gaussian", 
  

•                   line = TRUE)
  

then, there report the error :
Error in glm.fit(x = c(1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, :
NA/NaN/Inf in 'y'
In addition: Warning messages:
1: In Ops.factor(y, mu) : ‘-’ not meaningful for factors
2: In Ops.factor(eta, offset) : ‘-’ not meaningful for factors
3: In Ops.factor(y, mu) : ‘-’ not meaningful for factors

would your please tell me how to deal with this error above, thanks a lot.

[jinseob2kim]

Can you convert Y variable to integer or numeric?(0, 1)

Please show me Y variable or your dataset

[agnesisyss]

hi Kim：
thanks a lot for your reply, I have write an e-mail([email protected]) to you, but i am not sure wether you can receive.
the attachment is my data for analysis.
Y is dependent variable，and X ，age, sex,race, education and poverty are independent variables.

                                                                                                           thanks a lot
                                                                                                            Sisi

recode.csv

[jinseob2kim]

hi Kim： thanks a lot for your reply, I have write an e-mail([email protected]) to you, but i am not sure wether you can receive. the attachment is my data for analysis. Y is dependent variable，and X ，age, sex,race, education and poverty are independent variables.

                                                                                                           thanks a lot
                                                                                                            Sisi

recode.csv

Sorry. Now, my function can't support X with 3 category: only 2 category OK. The below code is example with "Y ~ sex"

library(survival);library(jstable);library(survey);library(data.table);library(magrittr)

a <- fread("recode.csv")
for (v in c("sex", "race","education")){
  a[[v]] <- factor(a[[v]])
}

data.design <- svydesign(id = ~sdmvpsu, weights = ~nhs_wt, data = a)
TableSubgroupMultiGLM(Y~sex, var_subgroups = c("race","education"), data = data.design, family = "binomial") 
                      
                

[ltj-github]

hi Kim： thanks a lot for your reply, I have write an e-mail([email protected]) to you, but i am not sure wether you can receive. the attachment is my data for analysis. Y is dependent variable，and X ，age, sex,race, education and poverty are independent variables.嗨，金[email protected]附件是我的分析资料。Y为因变量，X、年龄、性别、种族、教育程度和贫困程度为自变量。

                                                                                                           thanks a lot
                                                                                                            Sisi

recode.csv

Sorry. Now, my function can't support X with 3 category: only 2 category OK. The below code is example with "Y ~ sex"抱歉现在，我的函数不能支持3类X：只有2类OK。下面的代码是“Y ~ sex”的示例

library(survival);library(jstable);library(survey);library(data.table);library(magrittr)

a <- fread("recode.csv")
for (v in c("sex", "race","education")){
  a[[v]] <- factor(a[[v]])
}

data.design <- svydesign(id = ~sdmvpsu, weights = ~nhs_wt, data = a)
TableSubgroupMultiGLM(Y~sex, var_subgroups = c("race","education"), data = data.design, family = "binomial") 
                      
                

Hi ，I'm using this code and I'm getting ：
data.design <- svydesign(id = ~sdmvpsu, weights = ~nhs_wt, data = a)

TableSubgroupMultiGLM(Y~sex, var_subgroups = c("race","education"), data = data.design, family = "binomial")
Error in purrr::map():
ℹ In index: 1.
Caused by error in solve.default():
! The system is computationally singular: the inverse condition number=3.26124e-19
Run rlang::last_trace() to see where the error occurred.

[jinseob2kim]

I think this is a converge issue. Can you try each subgroup analysis without my package?

Ex) svyglm(Y~sex, design = subset(a, race==1), family = "quasibinomial")

[ltj-github]

When I use "svyglm(Y~sex, design = subset(a, race==1), family = "quasibinomial")", I encounter the error "Error in UseMethod("svyglm", design) : "svyglm" does not apply to "c('tbl_df', 'tbl', 'data.frame')"Method of target object". However, when I change the data to weighted data "data.design", I get the result. But using "TableSubgroupMultiGLM" still does not work.

[jinseob2kim]

Can you get result of all subset variable/value combinations? If you share your data, I run

[ltj-github]

At present, my classification data is named in English, and the gtsummary package is usually used. I will convert the variables to the form of 0,1. Maybe you will use it more conveniently.

[jinseob2kim]

Can you try other subgroup analysis?

svyglm(Y~sex, design = subset(a, race==0), family = "quasibinomial")
svyglm(Y~sex, design = subset(a, sex==0), family = "quasibinomial")
svyglm(Y~sex, design = subset(a, sex==1), family = "quasibinomial")
svyglm(Y~sex, design = subset(a, education==0), family = "quasibinomial")
svyglm(Y~sex, design = subset(a, education==1), family = "quasibinomial")

[ltj-github]

data.csv
Here's my data.

[ltj-github]

Can you try other subgroup analysis?你能尝试其他的亚组分析吗？

svyglm(Y~sex, design = subset(a, race==0), family = "quasibinomial")
svyglm(Y~sex, design = subset(a, sex==0), family = "quasibinomial")
svyglm(Y~sex, design = subset(a, sex==1), family = "quasibinomial")
svyglm(Y~sex, design = subset(a, education==0), family = "quasibinomial")
svyglm(Y~sex, design = subset(a, education==1), family = "quasibinomial")

This is the result of trying, only the last one will get the result

a <- fread("recode.csv")
for (v in c("sex", "race","education")){

• a[[v]] <- factor(a[[v]])
• }

a <- svydesign(id = sdmvpsu, weights = nhs_wt, data = a)
svyglm(Ysex, design = subset(a, race==0), family = "quasibinomial")
Error in contrasts<-(*tmp*, value = contr.funs[1 + isOF[nn]]) :
contrasts can be applied only to factors with 2 or more levels
svyglm(Ysex, design = subset(a, sex==0), family = "quasibinomial")
Error in contrasts<-(*tmp*, value = contr.funs[1 + isOF[nn]]) :
contrasts can be applied only to factors with 2 or more levels
svyglm(Ysex, design = subset(a, sex==1), family = "quasibinomial")
Error in contrasts<-(*tmp*, value = contr.funs[1 + isOF[nn]]) :
contrasts can be applied only to factors with 2 or more levels
svyglm(Ysex, design = subset(a, education==0), family = "quasibinomial")
Error in contrasts<-(*tmp*, value = contr.funs[1 + isOF[nn]]) :
contrasts can be applied only to factors with 2 or more levels
svyglm(Y~sex, design = subset(a, education==1), family = "quasibinomial")
1 - level Cluster Sampling design (with replacement)
With (3) clusters.
subset(a, education == 1)

Call: svyglm(formula = Y ~ sex, design = subset(a, education == 1),
family = "quasibinomial")

Coefficients:
(Intercept) sex2
-1.9718 -0.5729

Degrees of Freedom: 1823 Total (i.e. Null); 1 Residual
Null Deviance: 1143
Residual Deviance: 1131 AIC: NA

[jinseob2kim]

I run the code below with your data

library(survey);library(data.table);library(magrittr)

a <- fread("data (8).csv")
for (v in c("Sex", "Race","education.attainment")){
  a[[v]] <- factor(a[[v]])
}

svyglm(Y~Sex, design = subset(data.design, Race == 2), family = quasibinomial()) %>% summary

Then, P value can't be calculated. So interaction P can't be calculated too.

Call:
svyglm(formula = Y ~ Sex, design = subset(data.design, Race == 
    2), family = quasibinomial())

Survey design:
subset(data.design, Race == 2)

Coefficients:
            Estimate Std. Error t value Pr(>|t|)
(Intercept) -2.04358    0.13378 -15.275      NaN
Sex2         0.59439    0.08481   7.008      NaN

Zero or negative residual df; p-values not defined

(Dispersion parameter for quasibinomial family taken to be 1.002519)

Number of Fisher Scoring iterations: 4

[ltj-github]

Thank you for your response. I saw a related article that seems to be about issues with the survey package. The author of the article contacted Professor Thomas, the author of the survey package, and he said that the anova.svyglm function needs to be rewritten in order to work properly. The professor mentioned that the next version may improve this.