The concept of hypofrontality has been central to functional imaging research in schizophrenia for n

1
Meta-Analyzing Abnormal Brain FunctionFrontal
Activation in Schizophrenia
K.E.Hill1, L.Mann1, C.M.E.Stephenson1,
P.J.McKenna1, I.Nimmo-Smith2, K.Laws3
1Department of Psychiatry, Fulbourn and
Addenbrookes Hospitals, Cambridge 2MRC Cognition
and Brain Sciences Unit, Cambridge 3Department of
Psychology, London Guildhall University emaildrke
hill_at_waitrose.com
INTRODUCTION
What is the 3D Kolmogorov-Smirnov statistic
(KS3)? The standard Kolmogorov-Smirnov test
seeks to test possible differences in
distributions of data. Data from 2 samples are
arranged in order and plotted on a cumulative
fraction chart. The point of maximal separation
of the 2 curves produced is the KS1, ie the
biggest discrepancy between the two cumulative
distribution plots. If significant, the null
hypothesis that there is no difference between
distributions of data sets can be rejected. This
method has been extended by Nimmo-Smith to cover
a 3 dimensional distribution. The algorithm for
KS3 can be described in spatial terms as
follows. A A base point is chosen in xyz space
and planes parallel to x0, y0 and z0 are drawn
through it dividing the space into 8 cells
(octants). B The percentages of each type of
point on the eight octants is calculated (eg peak
activation foci in controls and schizophrenics).
C The absolute value of the differences
between the two percentages is calculated for
each octant and the largest difference is
noted. D The base point is moved so that the
difference at C is maximised. This maximised
difference is KS3 and represents the biggest
discrepancy between the two samples that can be
achieved by such orthogonal cellular partitions.

• The concept of hypofrontality has been central
  to functional imaging research in schizophrenia
  for nearly 20 years. Less than half of around 30
  studies using a region-of-interest (ROI)
  methodology have found hypofrontality under
  resting conditions, but a more consistent finding
  has been reduced prefrontal activation by
  schizophrenic patients during performance of
  executive, vigilance and memory tasks1,2. Resting
  hypofrontality was supported in a recent
  meta-analysis of ROI studies, but this
  unexpectedly cast doubt on the reality of
  activation hypofrontality3.
• Functional imaging studies of schizophrenia now
  almost always use voxel-based analysis.
  Meta-analysis of such studies is not possible
  using conventional techniques based on means and
  standard deviations.
• Duncan and Owen4 devised a method for combining
  data across voxel-based studies and used this to
  meta-analyze task-related prefrontal activation
  in normal subjects. They combined peak activation
  foci from 20 studies utilising a variety of
  executive tasks, plotted these onto a rendered
  brain, and examined homogeneity of activated
  regions within and between tasks using a 3
  dimensional version of the Kolmogorov-Smirnov
  test (see Box). They found that three different
  executive tasks all recruited mid-dorsolateral,
  mid-ventrolateral and dorsal anterior cingulate
  regions, with little evidence of regional
  functional specificity for the tasks.

METHOD
We adapted Duncan and Owens methodology to pool
data from functional imaging studies comparing
schizophrenic patients and normal controls during
performance of tasks activating prefrontal
cortex. Standard search techniques were used to
identify PET and fMRI studies of schizophrenia.
To be included papers had to use diagnostic
criteria for schizophrenia and provide peak
activation co-ordinates for schizophrenics and
controls separately. We selected studies which
used tasks known to produce activation of the
prefrontal cortex in normal subjects. These
employed executive, memory, and working memory
paradigms. Following Duncan and Owen, we included
studies using different reference brains, and
using different significance thresholds. All
peak activation foci from the studies were
plotted onto a 3-D rendered brain. We then used
the 3-D Kolmogorov-Smirnov test to examine the
homogeneity of distributions for patients and
controls.
Figure Combined peak activation foci for
schizophrenic patients (top) and normal controls
(bottom)
RESULTS
The Figure shows the combined peak activation
foci for schizophrenic patients and normal
controls from seven studies (4 PET and 3 fMRI).
Comparing the two distributions using the 3
dimensional Kolmogorov-Smirnov revealed no
significant difference for PFC (KS3 statistic
0.20, P 0.966). There was also no difference in
the whole brain distributions (KS3 statistic
0.14, P 0.958). It is also possible to gain
some idea of the magnitude of task-related
activation in schizophrenia compared to normal,
albeit indirectly. The Table shows the numbers of
reported peak activations across the pooled
studies for the PFC and non-PFC areas. A binomial
test indicates that the number of activations in
the PFC is evenly distributed across the patients
and controls (P 0.302). There was no significant
difference in the distribution of peak
activations by Chi-squared (Chi squared 0.33,
P0.566).
CONCLUSIONS
REFERENCES
1. Chua SE, McKenna PJ. Schizophrenia - a brain
disease? Br J Psychiatry 1995 166 563-582. 2.
Weinberger DR, Berman KF. Prefrontal function in
schizophrenia confounds and controversies. Phil
Trans Roy Soc (B) 1996 351 1495-1503. 3. Hill
KE, Mann L, Laws K, McKenna PJ. Hypofrontality in
schizophrenia a meta-analysis of functional
imaging studies. Submitted 4. Duncan J, Owen AM.
Common regions of the human frontal lobe
recruited by diverse cognitive demands. TINS
2000 23475-482.
Although functional imaging studies of
schizophrenia typically find significant
differences in prefrontal activation between
patients and controls, when data are pooled
across studies there is little to suggest an
altered pattern of frontal regional functional
architecture in the disorder. Although limited
by indirect analysis, the pooled data also fail
to support the view that schizophrenic patients
show quantitatively less prefrontal cortex
activation than normal when performing tasks
which activate this brain region. These
findings are based on pooled data from a small
number of studies (N7), and so should be
interpreted with caution.