Dear Kathryn,
At 13:45 14/06/99 -0400, [log in to unmask] wrote:
| I am trying to compare FDG-PET scans from 44 individual epilepsy
| patients to a group of 8 normals using spm. I've used the "Compare
| Groups: Single Scan Per Subject" function to compare each individual
| patient scan to the group of 8 normals, with so-so results.
The problem here is one of power. With only eight normal subjects to
represent the normal population, and especially given the multiple
comparisons corrections implicated in neuroimaging, you have very little
power to detect all but the largest discrepancies between a single patient
and the normal population. These "prediction interval" type of analyses,
comparing a single patient to the normal population, have n-1 degrees of
freedom, where n is the number of normal controls. The test is essentially
a
two-sample t-test with only one subject in one of the groups.
You have clearly put a considerable amount of effort into collating the 44
patients data, but the analyses are let down by the small normal control
sample. I recommend you get some more normal controls, so you have at least
12. For sample sizes up to 12 adding additional subjects gives considerable
gain in terms of detectable normalised change. After 20 subjects there is a
diminishing return in terms of power.
Our anecdotal evidence so far (mainly from functional fMRI & PET data)
suggests that reasonable population effects require about 12 subjects,
whilst population differences need at least nine per group. (All the above
being random effects analyses.) In these situations it is usually
inter-subject variability that dominates, such that the various imaging
modalities are comparable in terms of subject numbers required.
----------------
| Would it be more powerful to compare a single patient scan to a single
| mean image of the normals, using either the "Compare Groups" function
| or the "Single Subject: Replication of Conditions" function? I'm
| unsure as to whether this would make a difference in the statistical
| results, but when I attempt the latter option it produces a blank spm
| report.
Clearly not more powerful! You can't statistically test for differences
between two images, because there is no information available on the
variability of the images.
The "prediction interval" analysis you first carried out actually compares
the single patient scan to the mean of the normals, using the variability o
the normals as "yardstick" against which to assess the significance of any
differences.
By the way, the "Compare groups" and "Single Subject: Replication of
Conditions" models of SPM96/7-PET are basically the same model - the names
of the effects differ, and the latter model offers the option to include
covariates and to grand mean scale by group. In SPM99, the term "groups"
has
been replaced with "populations" where the fixed effects model effects
valid
population inference (i.e. with one scan per subject, as here).
----------------
Hope this helps,
-andrew
+- Dr Andrew Holmes ------------------ mailto:[log in to unmask] -+
| ___ __ ___ Robertson Centre for Biostatistics |
| ( ,) / _)( ,) Boyd Orr Building, |
| ) \( (_ ) ,\ University Avenue, |
| (_)\_)\__)(___/ Glasgow. G12 8QQ Scotland, UK. |
+----------------------------------------- http://www.rcb.gla.ac.uk/ -+
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|