anestistouloumis / multgee Goto Github PK

Dear Anestis Touloumis

First I want to thank you and gratulate you for this wonderful R package, multgee.

I am running an analysis on data collected with questionnaires, where each individual has answered multiple time. The response can both be formulated as a linear score ranging from 1 to around 30, but also as categories, 4 to 7 (depending on questionnaire). For the linear analysis I used a geeglm from package geepack. For the categorial analysis I first used ordgee package, but that gave inconsistent results. Sometimes it did not converge or gave other errors. I therefore turned to your package multgee and function ordLORgee.

ordLORgee always converges and returns results, but the estimates for all significant variables are not same direction as in other analyses, like geeglm. Where I expected negative estimates I now get positive.

I made sure that my ID is a number ranging from 1 … n and the data is ordered by the ID.
The response is a ordered factor. I also tested casting the response to a number.
Then there are covariates age and gender.
The variable of interest is the event, that only happens to some participants,

Here is example data (not real data):
ID2 gad_score_fct sex age_at_questionnaire event time

1 minimal anxiety male 66 0 0
1 minimal anxiety male 70 0 1
2 minimal anxiety male 61 0 0
2 minimal anxiety male 64 0 1
3 mild anxiety female 54 0 0
3 mild anxiety female 57 0 1

Can you see anything that might be causing this?

Best,
Gauti

Is it possible to use non-cluster weights (i.e. inverse probability weights) in ordLORgee? And if so, can robust standard errors still be produced?

Thanks,

Andrew

Hi Dr. Touloumis,

My question is that it seems take forever to run ordLORgee for conducting correlated ordinal regression.

The setting is look at the effect of one intervention (binary) on an ordinal categorical outcome (around 10 levels). The correlation of outcomes comes from that different observations correspond to different physicians (defining clusters). It is not a longitudinal setting, where we don't have or don't care observations at different time points for the same cluster. But when uniform/exchangeable correlation structure is assumed, the longitudinal setting is similar to our cluster setting.

My code is like
ordLORgee(Y~POI+X, data=data, id=id, LORstr = "uniform")
, where POI is the predictor of interest, X represents covariates to adjust for, id is the index of cluster and the data is already sorted by the id.

Specifically, the sample size is around 150-200 and the number of cluster is around 10-15.

It took really long and took up much memory. I don't know what's going on. Is that because of non-convergence of the algorithm with respect to my data?

Best,
Charlie

I have a dataset with 12 patients, each has one explanatory variable and one response variable, and each patient was measured 2000 times (I know it's huge!). When running ordLORgee the whole R session crashed, presumably because of a memory issue. I was wondering if there's any way to side step this memory issue when the repeat number is huge.
Thanks a lot!

Hi there,

I am using the following code:

nfse.fit = ordLORgee(formula = tumor_cell_evaluation ~ nfse, link = "logit", 
                      id = t_number, data = recgli, LORstr = "category.exch")
dim(recgli)
#383 73

I am looking to simply run this univariate analysis. This code works fine. When I subset my data down using the following code:

recgli_inCEL = recgli[which(recgli$in_CEL == 1),]
dim(recgli_inCEL)
# 158 73 

nfse.fit2 = ordLORgee(formula = tumor_cell_evaluation ~ nfse, link = "logit", 
                      id = t_number, data = recgli_inCEL, LORstr = "category.exch")

These are almost identical datasets, just one is subset from the other; yet when I run the first code, it works fine and the second code it throws this error: "Please insert initial values."

Why does this happen?

I tried to insert initial values using the following code:

coef_init = summary(nfse.fit)$coefficients[,1]
nfse.fit2 = ordLORgee(formula = tumor_cell_evaluation ~ nfse, link = "logit", 
                      id = t_number, data = recgli_inCEL, LORstr = "category.exch", 
                    bstart = coef_init)

And this did not work very well either. Please advise!

Thank you

Julia