Data Acquisition and Statistical Analyses

Chapter 6 An Investigation of the Modulation of MEG Resting-State

7.2 Background, Aims and Hypotheses

7.3.1 Data Acquisition and Statistical Analyses

The MRI and MEG procedures for data acquisition and analysis have been described in detail in previous chapters. The pipeline for amplitude-amplitude coupling and statistical analysis have also been described previously. Following on from the GMM approach used in all previous analyses, we limited analysis to look at only those “signal” connections present, in all studies, at a threshold of 50%. To be clear, a node-node connection was only analysed if it was identified in every separate GMM analysis as having a group-mean probability of being valid “signal”>50%. This left a more limited number of connections to take forward for subsequent analysis.

7.3.1.1 Meta-Analysis

Given that I have completed several similar studies, I went on to use a meta-analytic approach to pool all of the data and look for group differences. For this analysis, I excluded the Schizophrenia Study 1 eyes-closed data since this would then include the same cohort of individuals twice and all other studies were completed with eyes-open. Pooled mean difference and heterogeneity were analysed using the Review Manager programme, version 5.3 (Manager, 2014). The mean and standard deviation (of all connections that were valid in all studies) for each group for each study was used to calculate a standardised mean difference. This is the difference in mean outcome between groups divided by the standard deviation of outcome among participants. A weighted average of pooled data was then calculated by summing mean difference estimates multiplied by weights (where weight=1/(standard error2_{+ inter-trial variance) and dividing this by the sum of weights for}

all studies. The standard error of pooled effect size was then used to derive a confidence interval and p value to estimate the strength of evidence against the null hypothesis of no pooled effect. I used a random effects model (DerSimonian and Laird, 1986) whereby standard errors are adjusted to take account of heterogeneity. This is used when different studies are estimating different but related effects and takes into account both individual and between study variance (Deeks et al., 2008).

7.4 Results

Figure 7.1 below shows summarised alpha connectivity cohort-differences for all studies. Figure 7.1A shows a connectivity map of those 62 connections that were valid (mean(p)>50%) in all studies. Figure 7.1B shows t-test statistics for each of these valid connections in each study. Overall, there appears to be reduced posterior alpha connectivity

in these connections in most of the studies, with Study 2 (Nottingham MISP) appearing to have weaker connectivity reductions in patients than the other studies. In Study 7 (SPRING Recent Onset) connectivity appears much more mixed, with both reduced and increased connectivity in patients in these connections, but with weaker effect sizes. Study 5 (Ketamine) diverges significantly from the other studies with connectivity in these valid connections being increased with ketamine compared to placebo. Figure 7.1C shows boxplots of mean t-statistics across all of the valid connections for each study. These results echo those seen in Figure 7.1B, with connectivity in Study 5 (Ketamine) being increased and all other studies aside from Study 7 (SPRING Recent Onset) having reduced connectivity. In study 7, connectivity is slightly increased in cases but almost at zero (mean t-statistic) with no difference between cases and controls. Similar effect sizes are seen across the other studies showing reduced connectivity, except for Study 2 (Nottingham MISP) where the effects are much weaker. Figure 7.1D shows the ranked t-statistics for each connection for all studies. Again, results are similar, with connectivity in Study 5 (Ketamine) being increased with ketamine for most connections. In Study 7 (SPRING Recent Onset), the line crosses zero with some connections being increased and some being reduced in cases. Connectivity is mostly reduced in cases in Study 2 (Nottingham MISP) but with a weaker effect and again, the line crossing zero suggesting increased connectivity in some connections.

Figure 7.2 below shows summarised beta connectivity analysis for all studies. Figure 7.2A shows a connectivity map of only those connections that are valid in all studies. There were 68 connections that were valid at a threshold of 50%. Figure 7.2B shows t-tests of only these valid connections for each study. Reduced connectivity in these connections in cases is evident in most studies although the effects appear weaker in Study 2, Study 4 and Study 7. Mixed (both increased and decreased) connectivity is seen in Study 4, 5 and 7. This is also

Study 1: Study 1: Study 2: Study 3: Study 4: Study 5: Study 6: Study 7:

Open Closed Nottingham MISP Schizotypy Cardiff Schizotypy Nottingham Ketamine SPRING Established SPRING Recent

Mean t-statistic across all valid connections

B C St udy 1 : O pe n St udy 1 : C lo se d St udy 2 St udy 3 St udy 4 St udy 5 St udy 6 St udy 7

Sorted t-statistics for all valid connections

D Sorted Connections Study 1: Open Study 1: Closed Study 2 Study 3 Study 4 Study 5 Study 6 Study 7

Figure 7.1 Summarized Alpha Connectivity across all studies

A. Connectivity map showing only connections valid in all studies.

B. t-test of valid connections in each study. Blue=reduced connectivity in cases. Red=increased connectivity in cases.

C. Box plots of mean t-statistics across all valid connections for each study. D. Graph of sorted t-statistics for valid connections for each study.

evident when reviewing boxplots of mean t-statistics in Figure 7.2C. For studies 2-7, the mean is close to zero and boxes cross zero for studies 4-7 suggesting minimal differences between cases and controls. Again, when looking at t-statistics for each connection and study, lines are fairly close to zero aside from Study 1 (both conditions).