Cavum septum pellucidum and first-episode psychosis: A meta-analysis

Objectives To investigate the prevalence and changes of cavum septum pellucidum (CSP) in first-episode psychosis (FEP) patients. Methods Medline, Embase, and the Cochrane Central Register of Controlled Trials (CENTRAL) were searched to identify eligible studies comparing FEP patients and healthy controls from inception to Feb 29, 2016. Results Ten cross-sectional studies and three longitudinal studies reported in ten articles met our criteria. Our meta-analysis found no significant differences in the prevalence of either “any CSP” (OR = 1.41; 95% CI 0.90–2.20; p = 0.13; I2 = 52.7%) or “large CSP” (OR = 1.10; 95% CI 0.77–1.58; p = 0.59; I2 = 24.1%) between FEP patients and healthy controls. However, the heterogeneity analysis of the prevalence of “any CSP” suggested bias in outcome reporting. Conclusions The results based on current evidence suggest it is unclear whether “any CSP” is a risk factor for FEP due to the heterogeneity of the studies. There is insufficient evidence to support that “large CSP” is a possible risk factor for FEP.

The CSP is the space that remains when the leaflets of the septum pellucidum (SP) do not fuse [3]. Serving as a relay station in the limbic system, the SP is thought to connect the hypothalamic autonomic system to the hippocampus, amygdala, and habenula and regulate brainstem reticular formation [3,24,25]. The SP closes within one month of birth in 15% of subjects and within 6 months in 85% of subjects [26]. The normal fusion of the SP is also associated with an enlargement of the amygdala, hippocampus, and corpus callosum [27]. Recent comparison studies have reported that CSP length in psychosis patients shows negative correlations with the relative volume of the bilateral amygdala, hippocampus, and left posterior parahippocampal gyrus and an association with a shorter adhesio interthalamica (AI) [19,22,28,29]. Others, however, have argued that there is no association between CSP length and the morphology of the anterior cingulate cortex, hippocampus and fornix or the absence of the AI [7,19,20]. Comparisons between first-episode psychosis (FEP) patients and individuals with chronic schizophrenia have suggested that those volumetric reductions may be due to degenerative processes after illness onset [30,31], and chronic schizophrenia patients show an increased prevalence of clinically significant brain abnormalities [32]. Additionally, a meta-analysis of longitudinal magnetic resonance imaging (MRI) studies on patients with schizophrenia and psychotic disorders showed increased rates of lateral ventricle dilation after years of illness [33]. A recent longitudinal study also reported that CSP length increased at a higher rate in FEP patients, which may explain the higher prevalence of CSP in chronic cases [34], whereas increased CSP length in patients may be caused by the effects of antipsychotics or the duration of illness [35].
Thus, whether the CSP may serve as a risk factor for psychosis or is only a reflection of neuroanatomical changes in individuals with chronic psychotic disorders remains ambiguous. Therefore, we conducted a meta-analysis to assess the association between the CSP and FEP.

Methods
We conducted this study according to the standards of the Meta-analysis of Observational Studies in Epidemiology (MOOSE) guidelines [36]. We searched records from Medline, Embase, and the Cochrane Central Register of Controlled Trials (CENTRAL) from inception to Feb 29, 2016 using the following terms: "septum pellucidum", "septi pellucidi", "psychosis", "psychotic", and "schizophrenia" (see search strategies in the appendix). The inclusion criteria were as follows: 1) use of MRI to assess the CSP; 2) a population diagnosed with FEP; 3) a comparative group of healthy subjects; and 4) publication in English. The exclusion criteria were as follows: studies on infants or children. Teams of two trained and paired reviewers screened eligible titles, abstracts, and full texts independently, evaluated the risk of bias, and collated data from each study meeting our criteria. Disagreements between reviewers were resolved through discussion or judged by a third reviewer. Stata version 12.0 was used to analyse the outcome data. Dichotomous data were pooled using the odds ratios (ORs), and the continuous data were pooled using the mean differences (MDs) and associated 95% confidence intervals (CIs). The heterogeneity of the statistical models was examined via the χ 2 test and the I 2 statistic. The random effects model was used when I 2 >50%, and the fixed effects model was used when I 2 <50% [37]. A funnel plot and Egger's test were used to examine publication bias. To explore the source of heterogeneity, subgroup analyses for the different CSP measurement and assessment (qualitative or quantitative) methods were conducted. A meta-regression model was estimated using the residual maximum likelihood (REML) method, with the different assessment methods, CSP prevalence in the healthy controls and key sample characteristics as predictors. A sensitivity analysis was conducted using the leave-one-out method.

Evidence from cross-sectional studies
The 10 cross-sectional studies enrolled 749 FEP patients (mean age range, 21.5-30.3 years; mean range of FEP duration, 0.1-1.0 years; Table 1) and 727 healthy controls. To assess CSPs, four studies (thickness of MRI slices >3 mm) used qualitative assessments, including three studies that used grading systems based on visual inspection of coronal MRI slices [14,16,27] and one that used the "with/without" method [45]. The other six studies used quantitative assessment, by multiplying the number of coronal MRI slices (thickness 1.5 mm) on which the CSP could be visualized [8,20,29,34] or measuring the anteroposterior length of the CSP in millimetres [19,28] (Table 2). Three of the 10 studies reported a significantly higher prevalence of "any CSP" in FEP patients [14,16,28], and one study reported a significantly higher prevalence of "large CSP" in FEP patients [8]. The other six studies found no significant difference in the prevalence of either "any CSP" or "large CSP" between FEP patients and healthy controls [19,20,27,29,34,45] (Table 2).
Prevalence of "any CSP" in FEP Patients. The funnel plot and publication bias test for the prevalence of "any CSP" showed no significant results (Fig 2, Egger's test p = 0.20). The group of tests that included the qualitative assessment showed a significantly higher prevalence of "any CSP" in the FEP patients than in the healthy controls (OR = 1.94; 95% CI 1.14-3.30; p = 0.02; I 2 = 0%), whereas the group of tests that included the quantitative assessment showed no significant results (OR = 1.19; 95% CI 0.63-2.26; p = 0.59; I 2 = 68.4%). The overall risk of "any CSP" in the FEP patients was not significantly different from that in the healthy controls (OR = 1.41; 95% CI 0.90-2.20; p = 0.13; I 2 = 52.7%) (Fig 3).
The univariable meta-regression analyses showed no statistically significant association between the OR of "any CSP" and the four predictors (assessment method: p = 0.42; publication year: p = 0.80; prevalence of CSP in the healthy controls: p = 0.08; mean age of the FEP patients: p = 0.15), whereas the other three predictors (sex ratio of the FEP patients: p = 0.04; sex ratio of the healthy controls: p = 0.01; mean age of the healthy controls: p = 0.02) showed statistically significant associations with the OR of "any CSP". Some of the studies reported an extreme gender imbalance (male/female>2) in the FEP patients [8,19,20,28,45] (Fig 4 (1)) and in the healthy controls [8,14,16,19,20,[27][28][29]34,45] (Fig 4 (2)). One study [28] reported a much older mean age of the healthy controls than the other studies (Fig 4(3)) and of the FEP patients in the same study (Table 1). The sensitivity analysis conducted using the leave-one-out method suggested instability in outcome reporting. The prevalence of "any CSP" was higher in the FEP patients (OR = 1.82; 95% CI 1.36-2.43; p<0.001; I 2 = 29.3%) after one study [8] was excluded.
Prevalence of "large CSP" in the FEP Patients. When comparing the prevalence of "large CSP" between the FEP patients and healthy controls, the funnel plot and publication bias test showed no significant results (Fig 5, Egger's test p = 0.91). In the subgroup analyses of the risk of "large CSP", neither the qualitative assessment group (OR = 3.15; 95% CI 0.51-19.48; p = 0.22; I 2 = 0%) nor the quantitative assessment group (OR = 1.05; 95% CI 0.72-1.51; p = 0.81; I 2 = 45.1%) showed significant differences, and there was no significant difference in the overall risk (OR = 1.10; 95% CI 0.77-1.58; p = 0.59; I 2 = 24.1%) (Fig 6). The univariable meta-regression analyses showed no statistically significant association between the OR of "large CSP" and the seven predictors (assessment method: p = 0.32; publication year: p = 0.37; prevalence of CSP in the healthy controls: p = 0.46; sex ratio in the FEP patients: p = 0.80; sex ratio in the healthy controls: p = 0.68; mean age of the FEP patients: p = 0.75; and mean age of the healthy controls: p = 0.50). The sensitivity analysis conducted using the leave-one-out method suggested no significant differences in outcome between studies.

Evidence from longitudinal studies
Three studies included eligible baseline and follow-up reports [19,29,34], two studies maintained the same sample subjects [19,29], and one study was able to assess some of the subjects after follow-up [34]. Therefore, to study changes in CSP length after follow-up (mean duration range 1.1-2.6 years), a total of 120 patients were assessed (Table 3). Two studies reported that the CSP length of subjects was stable in both FEP patients and healthy controls [19,29]. With a larger sample, another study [34] reported that CSP length increased in both FEP patients and healthy controls but increased more significantly in FEP patients (Table 4).
Increases in CSP length in FEP Patients. Data on increases in CSP length after follow-up were transformed into the mean difference ± SD of the differences in each group (Table 4). A comparison of increases in CSP length between FEP patients and healthy controls showed no significant differences (mean difference = 0.31, 95% CI -0.32-0.93, p = 0.34, I 2 = 0%) (Fig 10).

Main findings
The results from recent cross-sectional studies [8,14,16,19,20,[27][28][29]34,45] showed that the CSP prevalence and length were not statistically different between FEP patients and healthy controls. However, the a meta-regression analysis comparing the prevalence of "any CSP" between FEP patients and healthy controls according to the sex ratio of both groups and the mean age of the healthy controls suggested study heterogeneity, and the sensitivity analysis suggested that the OR was not stable. Our study found no evidence in support of a significant difference in the prevalence of "large CSP" between FEP patients and healthy controls. Meanwhile, the prevalence of CSP was not higher in the FESZ group. After a few years of follow-up, the evidence from recent longitudinal studies [19,29,34] showed no significant difference in the magnitude of the increase in CSP length between FEP patients and healthy controls.

Measurement and assessment of the CSP
One significant limitation in our analysis was the mixture of MRI measurements, which may result a variations in CSP prevalence because of potentially missing smaller CSPs [20,21]. In the cross-sectional studies, the prevalence of "any CSP" in FEP patients varied from 3.3% [45] to 94.3% [34] and from 0.0% [45] to 89.7% [20] in healthy controls. However, earlier studies [14,16,27,45] used thicker (!3.0 mm) MRI slices and reported a lower prevalence of "any CSP" than later studies (MRI thickness approximately 1.0 mm).
Different assessments of CSP were combined with the MRI methods. Earlier studies used qualitative assessments. For example, Degreef [16] used visual inspection with a grading system from 0 to 3 (representing absent, questionable, small, moderate and large), and other Cavum septum pellucidum and first-episode psychosis studies [14,27] used similar methods, except for that of Borgwardt [45], which detected a normal CSP variant in only one FEP patient using the with/without classification in a qualitative approach. However, both the thickness of MRI slices and the measurements seemed unsatisfactory for identifying CSP prevalence. Moreover, visual inspections were actually based on the width of the CSP rather than the length measurement in the studies using quantitative assessments.
As described by Nopoulos [10], the length of the CSP can be calculated based on the number of thin-slice MRI slices, and later studies preferred this quantitative method. However, this method remains questionable. Davidson [19] claimed that a normal CSP may be associated with the SP due to its lengthwise stretching, and the assessment of only CSP length may not be sensitive enough to detect existing changes [29]. For example, Choi [49] conducted a grading measurement of length, width and overall size, and de Souza Crippa [7] measured the volume of the CSP using voxels.
To explore the source of heterogeneity of the findings, we performed subgroup analyses of the outcomes reported according to the different CSP measurement and assessment (qualitative vs. quantitative) methods. Our analysis of "any CSP" prevalence (Fig 3) showed that although the overall comparison between FEP patients and healthy controls showed no significant difference, the subgroup using qualitative assessments did show a significant difference in FEP patients. However, this difference has minimal practical value because of the small sample size. Furthermore, the study heterogeneity estimated by the sensitivity analysis may also be   Cavum septum pellucidum and first-episode psychosis explained by the measurement and assessment methods. After the exclusion of one study [8] (MRI thickness = 0.9375 mm) that detected a higher prevalence of normal variance of CSP, the prevalence of "any CSP" was significantly higher in FEP patients than in healthy controls.  Cavum septum pellucidum and first-episode psychosis

Other limitations
The instability of diagnoses could be another limitation in this analysis. A meta-analysis estimating the diagnostic stability of FEP reported that the prospective diagnostic stability in schizophrenia was high, with no significant ICD/DSM difference, while the stabilities of other first-episode psychotic diagnoses were low [50]. From the ten included studies, 292 FEP patients were diagnosed with FESZ [8,16,19,20,29,34], and the other subjects were diagnosed with a mixture of first-episode schizophrenia spectrum disorders, affective psychosis, or psychoses not otherwise specified [8,14,20,27,28,34,45]; seven studies reported a confirmation of diagnoses after follow-up [8,16,19,20,27,29,34], and two studies reported a few cases subsequently confirmed as schizophrenia [28,45]. Although some studies suggested that psychosis may share genome linkage [51,52] and abnormalities in neurodevelopment [8,9,53], other studies suggested that structural abnormalities differ according to diagnosis and stage [54][55][56].
In considering the potential variations based on the diagnosis, we analysed the prevalence of CSP in FESZ patients as a separate group (Fig 8; Fig 9). There were also some limitations caused by sample selection. First, the comparison of the gender effect on CSP was not clear. Three studies [8,27,29] reported no gender difference in CSP prevalence between FEP patients and controls, and two studies [20,29] reported no gender difference in CSP length. On the other hand, one study [34] reported that males had a higher prevalence of a large CSP when all subjects were pooled. However, gender distributions of the included studies were not equally balanced between males and females. For example, a total of 749 FEP patients comprised 477 (63.68%) males and 272 (36.32%) females (Table 1), and the meta-regression by sex ratio (male/female) for the prevalence of "any CSP" suggested heterogeneity in both the FEP patients and healthy controls. Second, the recruitment of FEP patients and healthy controls may have caused a selection bias. Most FEP patients were recruited from hospitals [8,14,16,19,20,[27][28][29]45], except for one study [34] that reported recruitment from a population. Five studies recruited controls from the community [8,14,19,27,34], and other studies [16,20,28,29,45] recruited controls from a mixture of groups, including hospital staff, university students and the community (Table 1).
Although we included longitudinal studies [19,29,34] to explore the increase in the CSP length, the sample sizes were small (FEP patient sample sizes range, 20-75 subjects), and the mean durations of follow-up were short (mean follow-up range, 1.1-2.6 years). The result of our analysis showed no difference in the increase in the CSP length between FEP patients and healthy controls, but one study [34] with a larger sample reported a more significant increase in the CSP length in FEP patients and suggested a main effect of time. Similar to other longterm studies on brain morphometric changes in patients with psychoses [48,57], future studies with longer follow-up times are needed to evaluate the interaction between CSP growth and illness duration.

Conclusions
Although the combined analysis of the prevalence of "any CSP" showed no statistical significance, the evaluation of study heterogeneity suggested that it is unclear whether "any CSP" is a risk factor for FEP. The current evidence suggests no significant difference in the prevalence of "large CSP" between FEP individuals and healthy controls. Therefore, this evidence is insufficient to support "large CSP" as a risk factor for FEP. The CSP length seems stable during relatively short follow-up times. However, measuring only the CSP length may not be sufficiently sensitive to detect changes in the CSP. Additional longitudinal studies are needed to explore the relationship between morphometric changes in the CSP and the progression of psychosis.