Online databases will be searched for the relevant articles. The databases examined will include PubMed, Google Scholar, MEDLINE and EMBASE, from where articles will be systematically assessed to identify those relevant to our hypothesis. The search performed was as follows: (Hippocampus AND psychosis) OR (Hippocampal subfield AND psychosis) OR (Cornu ammonis AND psychosis) OR (Subiculum AND psychosis) OR (dentate AND psychosis) OR (Hippocampus AND Brief Limited Intermittent Psychotic Symptoms) OR (Hippocampal Subfield AND Brief Limited Intermittent Psychotic Symptoms) OR (Hippocampus AND Schizophrenia) OR (Hippocampus AND Bipolar Disorder) OR (Hippocampus AND Depression) OR (Hippocampus AND Schizoaffective Disorder) OR (Hippocampus AND Delusional Disorder) OR (Hippocampus AND Brief Psychotic Disorder) OR (Hippocampus AND Schizophreniform Disorder) OR (Hippocampus AND Medication Induced Psychotic Disorder) OR (Hippocampus AND Hallucinations) OR (Hippocampus AND Delusions) OR (Hippocampus AND Thought Disorder) OR (Hippocampus AND Catatonia) OR (Hippocampus AND Personality Disorder)
References from the output articles will also be checked, and articles which are pertinent to our study will also be incorporated into it. The search items will be rerun before publication to include newer studies which got added to the databases.
The studies to be included in this review will encompass MRI neuroimaging studies of patients with a definitive diagnosis of psychosis, and comparisons with healthy control participants. The studies which segmented the subfields of the hippocampus using either automatic or manual means will be included. Studies lacking a control group will be excluded. All the studies included will have been peer-reviewed. Although the search items were in English, we will include non-English studies, which we’ll get translated professionally, and contact the corresponding author with any confusions which may arise. Studies where illicit drug use is documented, or those with history of other significant psychiatric comorbidities will be excluded. Studies with participants that had medical psychosis will also be excluded since psychiatric psychosis is the focus of this review. Studies with an emphasis on ages above 65 or below 16 will also be excluded. Incomplete or ambiguous information will be clarified by contacting the corresponding authors of the respective studies. In instances where the results from a study has been reported in more than one article, the results from the article with greater sample size will be extracted.
Two researchers will independently conduct the search and selection procedures to gather suitable studies. We will include case-control, cohort, cross-sectional studies, randomised control trials and longitudinal studies. Inconsistencies regarding the studies gathered will be discussed and resolved with guidance from Dr Roddy.
One reviewer will independently perform all the data extraction on an excel spreadsheet, and another reviewer will then systemically check all the extracted information using the original articles to ensure accuracy. The information to be extracted includes the following:
Author and publication year
Cohort information (the specific psychotic illness(es) the study discusses)
Sample and descriptive demographic information (age, gender)
Age of onset, duration of illness, duration of untreated psychosis
Diagnostic method and quantification criteria for psychosis with results e.g. PANSS, BPRS, SAPS + SANS
Whether the participants were medicated or not. If so, which medication class was used?
Software used for hippocampal subfield volume determination
Type of MRI scanner used, and magnetic field strength
Volumetric information for each of the subfields in both hemispheres
Illicit drug use
Hippocampal Subfield Volumes
Quite often in neuroimaging studies, the definition and segmentation of subfields is conducted using differing methodologies and softwares: manual or automatic. Though the output from these should be very similarly, though we will be cautious of possible heterogeneity which may arise due to this by recording the software used in each instance, allowing for a possible explanation of heterogeneity if it arises during our analysis.
The subfield volume information we will extract will be that of the patients and controls within each of the studies. We will be extract volumes of the CA1, CA2, CA3, DG, CA4, Dentate Gyrus (DG), Subiculum, Presubiculum. In instances where composite volumes are presented in the paper, those will be recorded. CA4-DG and CA2-3 are pairs which can be difficult to tell differentiate, particularly using the automated softwares. Hence, we expect that we’ll frequently be recording composite volumes for those and will be running analysis on them as a composite.
Statistical analysis will be conducted using the metaphor and meta packages in R Studio 2020 (RStudio, PBC, Boston, MA; URL http://www.rstudio.com), which is an integrated development environment for R [39, 40]. Given the assumption of exchangeability in a random-effects model, it will be applied throughout our review to weight each study and control for potential heterogeneity [41, 42]. Potential heterogeneity we may encounter has been explored and includes the software used for subfield segmentation, duration of illness, type of psychotic disorder, age of onset, medication use, measurement on scales for psychosis and magnetic field strength of the MRI scanner used. Cohen’s d statistic or Hedges’ unbiased g will be computed, as appropriate, for an effect size of the difference between means of the patient and control groups. Potential type 1 inflation errors will be addressed using conservative correction measures such as false discovery rate  or Bonferroni .
Meta-regression will be employed for assessment of secondary and tertiary hypotheses that subfield volumes change with psychosis scores and with duration of illness. An analysis will be performed to examine the relationship between standardized mean differences (SMD) of the measurements on psychosis scales and Cohen’s d for volume changes in each of the subfields. Another analysis will examine the relationship between duration of illness and volume changes in each of the subfields. Regression will be performed using SPSS-26 (IBM SPSS Statistics 26 for Windows 10).
Assessment of between-study heterogeneity will be conducted using the Cochran’s Q, and the degree of heterogeneity will be quantified using the I2 statistic . This will give us the percentage of variability that is due to differences between studies compared to sampling variance. The interpretation of these I2 values will be 0.25 = low, 0.5 moderate, 0.75 = high. The significance threshold for establishing the studies are heterogenous will be 0.1. In instances where the Q-statistic is significant, Galbriath plots will be produced to supplement forest plots in determining the studies which have the largest influence on increasing the heterogeneity .
Quality of evidence will be evaluated using the Grading of Recommendations, Assessment, Development and Evaluations (GRADE) criteria with five domains of evidence being assessed (risk of bias, imprecision, inconsistency, indirectness, and publication bias) each according to four levels of quality (very low, low, moderate, and high) . Publication bias and small-study effects will be important features addressed as part of this review. These occur when mostly the significant findings get published . Small study bias refers to phenomena where studies with smaller samples, and less power, tend to report larger effect sizes . These will be verified by visual inspection of funnel plots and assessed using Eggers test . Studies which don’t appear adequately robust will be eliminated from the quantitative meta-analysis but may be documented narratively throughout the paper. Risk of bias (i.e. flaws in study design, conduct or analysis) will be assessed using Newcastle-Ottawa Scale, where studies will be graded according to three quality outcomes: group selection, group comparability, and outcome .
This study will obtain clinical, demographic, and methodological variants. A forest plot will be used to synthesise the total number of participants, studies, subfield volumes with mean differences, 95% confidence intervals, p values, and I2 statistics in graphical form . If a meta-analytical approach is not feasible based on heterogeneity and sample sizes, we plan to summarise the findings as a narrative systematic review.