 |
 |
Many thanks, Anderson! Eyesha Date: Wed, 4 May 2016 09:06:49 +0100 From: [log in to unmask] Subject: Re: [FSL] Statistical analyses- which mean values to use To: [log in to unmask] Hi Eyesha, Please, see below: On 3 May 2016 at 21:44, Eyesha Younas <[log in to unmask]> wrote: Hi Anderson, Re: You may want to study the signs of the effect of NP in each group to see if a positive interaction means one of the following: - NP and DTI have positive relationship in patients, but negative in controls. - NP and DTI have strongly positive relationship in patients, but weakly positive or zero in controls. - NP and DTI have weakly negative or zero relationship in patients, but strongly negative in controls. This follow up analysis can be done with fsl_glm, as only the test statistics are needed (i.e., not the actual p-values). A few things: 1- I get the 4D images for t-statistics as output of fsl_glm. How do I ascertain the above points with these? 2- For some reason I was unable to use my hand-written matrix with intercept for this part. I switched to the GLM set up gui and set up something as follows (consider 3 controls and 3 patients): controls patients NP NP EV1 EV2 EV3 EV4 1 0 40 0 1 0 58 0 1 0 17 0 0 1 0 27 0 1 0 11 0 1 0 43 These values are not de-meaned and I am using the demean option with fsl_glm. Let's not use -demean in this design, as it causes rank-deficiency in this case (it may be fixed internally but we should have the model right, as opposed to depending on the application to fix it). My contrast is: 0 0 1 -10 0 -1 1 These contrasts are fine with this design to test the interaction, which you do in randomise. Here in fsl_glm we are interested on the signs of each of the EVs separtely, so we want these contrasts: 0 0 1 0 0 0 0 1 Then in the areas where the interaction was found significant with randomise, you check whether these contrasts have positive or negative signs, indicating for each group whether the association of NP and the DTI measure is positive or negative. All the best, Anderson Does this seem alright? Many thanks for rescuing me. Eyesha Date: Sat, 30 Apr 2016 09:35:13 +0100 From: [log in to unmask] Subject: Re: [FSL] Statistical analyses- which mean values to use To: [log in to unmask] Hi Eyesha, Please, see below: On 29 April 2016 at 22:39, Eyesha Younas <[log in to unmask]> wrote: Hi Anderson, A few other questions also came to my mind. 1. Orthogonality means that there is no interaction between group and NP? It means that there is no correlation (once the intercept has been taken into account) between group and NP. It means that there is a perfect balance of NP values among patients and controls. If patients have in general worse NP scores than controls, then orthogonality isn't present. 2. In the setup that you suggested, we are looking at positive interaction [0 0 0 1 0 ...] and negative interaction [0 0 0 -1 0 ...]. Is it possible that I get significance for both, in different areas though? Yes. 3. If I do find significance for NP interaction, then what's the next step? You may want to study the signs of the effect of NP in each group to see if a positive interaction means one of the following: - NP and DTI have positive relationship in patients, but negative in controls. - NP and DTI have strongly positive relationship in patients, but weakly positive or zero in controls. - NP and DTI have weakly negative or zero relationship in patients, but strongly negative in controls. This follow up analysis can be done with fsl_glm, as only the test statistics are needed (i.e., not the actual p-values). 4. For: "If not significant, you can investigate the effect of group and NP with additional contrasts then.". Then I treat NP like any other co-variate (of no interest) like age and sex. I guess I will just need to look at the contrasts: [0 1 0 0 ...] and [0 -1 0 0 ...] for group ( cont > pat and pat > cont respectively- considering EV2 as the group). )[0 0 1 0 ...] and [ 0 0 -1 0 ...] for NP (Considering EV3 as NP). Exactly. 5. How do I handle time since injury? It is something which is 0 for all controls. I am assuming that I am demeaning the age, gender and NP test performance scores. Can I demean this time variable too? The overall intercept and the intercept per group are coded though EV1 and EV2 jointly, so mean centering isn't needed, but doing so won't hurt. It may be more clear, though, if this variable were mean-centered within the patient group only (thus, remaining as zero for controls). Also what contrast do I construct for this? Same as if there were values also for the controls, i.e., same as usual. Can I do an interaction term for this as well? Yes, multiply this EV (that has zeroes for controls) by the other EV that you'd like to test the interaction with, say, age for instance. There is no need to test an interaction of time since injury with group. 6. For age and sex, do I simply quote that these were included as covariates while performing t-test or do I still need to perform a significance test for these and state that there were no significant difference for these? Depends on what the reviewers ask. I wouldn't test needlessly, but would include. Will the contrast [0 0 0 0 1 ...] indicate that controls mean age was significantly higher than patients'? (Considering EV5 as the age). This contrast will test an effect of age across both groups. To have the effect of age for a single group, the age EV would have to be split in two, and each tested separately, but then again probably an interaction group by age would be more interesting. Hope this helps. All the best, Anderson This is my first time performing DTI analyses and I greatly appreciate your help. Eyesha Date: Fri, 29 Apr 2016 10:35:21 +0100 From: [log in to unmask] Subject: Re: [FSL] Statistical analyses- which mean values to use To: [log in to unmask] Hi Eyesha, Please, see below: On 29 April 2016 at 03:23, Eyesha <[log in to unmask]> wrote: Dear experts, I have a question about statistical analyses of DTI data. I am comparing DTI parameters between two groups (controls and patients) and eventually I also want to look at the correlation between performance on neuropsychological tests and DTI parameters. This is my plan of action (after performing the basic DTI processing) after much thought: A- Run the randomise tool to obtain t-statistics and p-values for voxel-wise comparison between the two groups, corrected for multiple comparisons, B- Threshold the output at the desired p-value. I am choosing 0.05 and further dividing it by 4 to account for multiple analyses (4 DTI parameters: FA, MD, RD and L1). C- Use the thresholded image to find tracks with significantly different values. D- Get the mean value in each of the significant track (here I use only that part of the tract where significant differences were found, i.e. only the part of the tract that was included in the thresholded image and not the entire track). E- Use the mean values found in D to investigate the correlation with NP test performance in the patient group. F- Use regression to account for variables like age, sex and disease stage. If the whole sample (patients and controls) were used for this follow up test, this would be a circular analysis, unless NP were perfectly orthogonal to group. Using only one of the groups, this isn't really an issue. However, it's still problematic in other grounds: suppose there is a correlation in that particular area found in A above in the patient group. How can one know if that is a noteworthy result if one hasn't looked into the control group? What if there is similar correlation between the DTI measures and NP in the control group in that particular region? Then there is nothing interesting to be reported as far as NP goes, and although the correlation may be significant in patients, it's still a false positive. Conversely, if there is a different correlation between NP and DTI measures according to group, then the difference between groups in terms of DTI varies according to NP, and the initial analysis of group differences is likewise misleading. To deal with this, perhaps be best thing is to drop these multiple models/stages and run a single one, that has: EV1: interceptEV2: group (+1/-1)EV3: NPEV4: NP*group (interaction)EV5, EV6, etc: age, sex, disease stage, etc. Then test for the interaction with contrasts [0 0 0 1 0 ...] and [0 0 0 -1 0 ...]. If significant, then it's something interesting. If not significant, you can investigate the effect of group and NP with additional contrasts then. The point where I got stuck was if I should use only the tracts with significant difference for further analyses or all the tracts. And if I should just use the mean value in the entire tract from the skeletonized parametric image, or un-skeletonized mean value in the entire tract or the skeletonized value from the significantly different region of the tract? I see people doing all of these things but I am not convinced for all. Here I would use the whole brain (voxelwise). Hope this helps. All the best, Anderson Thank you for your expert opinion and help. Eyesha