Hello!
I'm currently playing which my data and am unsure about the statistical test to use.
I have MRI data from 25 patients. I have 4 (predetermined) lesion types and a control area from which I obtained MRI values. Not all patients have all lesion types, some patients have several of the same lesion types. I would like to know if MRI values differ significantly between the lesion types (and control). I have possibly two covariates - scanner type (I used two) and area size.
Data looks something like this;
scanner subject lesiontype area value
1 1 1 4.5 850
1 1 3 3.7 456
1 1 3 3.3 387
1 2 2 5.4 658
etc
I don't think an ANCOVA would work as the data isn't independent (subject 1 has both lesion type 1 and 3). Repeated measures doesn't work because of missing data (subject 1 doesn't have lesion type 2). Regression analysis seems to work but I may violate independence here too.
Next I grouped my data together to get only one average value for lesion type;
scanner subject lesiontype area value
1 1 1 4.5 850
1 1 3 3.3 387
1 2 2 5.4 658
etc
I still have missing data (lesion type 2 for subject 1), but was now able to do a Generalised Estimating equations or GEE (for longitudinal data, nested data or correlated data). Here I used subject as subject variable and lesiontype as within subject variable (which I can recode to run several models and compare the groups).
Unfortunately I don't know enough about this statistical test to know whether this is the way forward? I can only find examples for longitudinal data. Will there be issues here? Which statistical test would be best to analyse this data?
Many, many thanks!
I'm currently playing which my data and am unsure about the statistical test to use.
I have MRI data from 25 patients. I have 4 (predetermined) lesion types and a control area from which I obtained MRI values. Not all patients have all lesion types, some patients have several of the same lesion types. I would like to know if MRI values differ significantly between the lesion types (and control). I have possibly two covariates - scanner type (I used two) and area size.
Data looks something like this;
scanner subject lesiontype area value
1 1 1 4.5 850
1 1 3 3.7 456
1 1 3 3.3 387
1 2 2 5.4 658
etc
I don't think an ANCOVA would work as the data isn't independent (subject 1 has both lesion type 1 and 3). Repeated measures doesn't work because of missing data (subject 1 doesn't have lesion type 2). Regression analysis seems to work but I may violate independence here too.
Next I grouped my data together to get only one average value for lesion type;
scanner subject lesiontype area value
1 1 1 4.5 850
1 1 3 3.3 387
1 2 2 5.4 658
etc
I still have missing data (lesion type 2 for subject 1), but was now able to do a Generalised Estimating equations or GEE (for longitudinal data, nested data or correlated data). Here I used subject as subject variable and lesiontype as within subject variable (which I can recode to run several models and compare the groups).
Unfortunately I don't know enough about this statistical test to know whether this is the way forward? I can only find examples for longitudinal data. Will there be issues here? Which statistical test would be best to analyse this data?
Many, many thanks!