This review provides an overview of structural MRI findings in schizophrenia over the past decade, beginning in 1988, when major improvements in technology occurred, and ending in May 1998. It builds on, and updates our earlier review (Shenton et al 1997
). Because of the large number of studies cited, more than 170, we use a table to present, for each ROI, the number of studies with positive and negative findings. The text itself summarizes the findings and their implications for each ROI, but space constraints do not allow us to describe each individual study. To qualify for inclusion in the review, studies must have: 1) appeared in peer–reviewed journals; 2) had 10 or more subjects; 3) have included a control group; 4) had quantitative measurements; and 5) have included schizophrenic patients as a major focus. This review counts the same group of subjects only once for each ROI in those cases where the same subjects appear in more than one paper, such as, for example, where schizophrenics in one paper are contrasted with controls and with another pathological group in a separate publication.
Some discussion of the scope of this review may be useful. First, while we believe our conclusion about quality of studies improving after 1987 is generally true, this neglects the pioneering study of Andreasen et al (1986)
, the first quantitative controlled study of adequate sample size, which may be regarded as ushering in the “next generation” of MRI studies. Other quality studies may also have been omitted, but a rule setting the starting point in a technologically rapidly changing field is needed, although we acknowledge any date so selected is somewhat arbitrary.
Second, we need to comment on the tabular presentation of the studies in . One broad approach for reviews is to survey the literature, and then to provide an informed opinion as to the summary trends and findings, with specific citations to drive home the points. Most reviews, in fact, adopt this approach; however, we note that using such an approach leads to much of the review’s summarizing process remaining opaque to the reader, who, without reading the literature, cannot formulate a judgment of the accuracy of the review’s conclusions based on the data presented. We thus decided to present a summary of the studies in tabular form, to assist the reader in forming an independent conclusion. We further decided to summarize the results of studies in terms of whether each study did or did not report an abnormality for a particular ROI.
Summary of MRI Studies Reporting Positive and Negative Findings in Schizophrenia, 1987–May 1998
A possible alternative to the comprehensive survey of studies in our review is meta–analysis, which essentially involves weighting each individual study by a function of its N
and effect size, and then using this information to produce an estimate of the combined effect size (see Glass et al 1981
; Hunter and Schmidt 1990
; Petitti 1994
; Rosenthal 1987
). While we are quite sympathetic to the desire to use all of the information in studies, we note that the combination of all of the studies surveyed here in a region–by–region meta–analysis would be premature, potentially misleading, and, at worst, invalid, since: 1) MR scanner technology in the past decade has been changing rapidly, and studies are therefore not quantitatively comparable; 2) the extent of detailed (anatomically based ROI) information used in measurement of images varies; 3) there is a wide difference in moderator variables of subject gender, chronicity (age of onset), medication, parental SES, etc.; and finally, 4) meta–analysis, especially of MRI studies, is beset with the difficulty of estimating the number of studies with negative findings that did not get published [Rosenthal’s “file drawer” problem (1987)].
We concluded that, for a review of this broad a scope, covering more than a decade and all published studies, meta–analysis was not the most appropriate approach, as it would yield a false sense of numerical exactness. We suggest that one cannot simultaneously do a comprehensive review of a decade of studies, the goal of this review, and also perform a valid meta–analysis, unless the variables 1 to 3 above have remained constant, a constancy for which there is little evidence.
We, nonetheless, are sympathetic to readers who would like some more quantitative information and for this reason we have provided a table of subject N
for each study in a summary table. It will be clear that the statistical power of negative studies with small subject N
is less, and that their results consequently are less convincing than those negative studies with a larger N
. We also, following a suggestion of Rosenthal (1987)
, computed the probability of the observed positive and negative statistical findings, using a two–tailed probability, and the alpha level of p
<.05 of the studies. The reader will note that this procedure, like the more standard meta–analyses, assumes comparability of the studies with respect to measurements and subjects, although it does not assume normality of the distributions. We use this probability statistic with caution, since, in our opinion, it is hazardous to assume that the studies are comparable in methodology and subject characteristics; however, it may be of interest to the reader that, if one does assume comparability, the binomial theorem computation (using p
<.05 for a positive study) shows that all ROI surveyed in show a two–tailed p
<.05 for the number of positive studies, except for the fourth ventricle and cerebellum, and all ROI are less than or equal to .002 except for the occipital lobe (.004). Again, we urge caution on use of this probability estimate because comparability is not strict and an (unknown) percentage of studies with negative results may not get published.
This review begins with a description of what structural MRI can tell us about schizophrenia, followed by a description of what we consider to be important design features of an MRI study, and then by a review of the MRI findings to date in schizophrenia.