The proposed systematic review will be reported in accordance with Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA), and this protocol was reported in accordance with the PRISMA guidelines for systematic review protocols (PRISMA-P). This meta-analysis has been registered in PROSPERO.
Data collection
Two review authors will independently extract data. Discrepancies will be identified and resolved through discussion with a third reviewer. Missing data will be requested from study authors via email.
We will record the following data: (1) sample size (i.e., number of current AN patients, number of recovered AN patients, number of controls), (2) diagnostic classification system used (e.g., clinical interview, clinician evaluation), (3) participant demographics (e.g., mean age of participants in each group, race, ethnicity, income), (4) participant clinical characteristics (e.g., mean age of illness onset (if reported), type of treatment (if in treatment), duration of illness, mean score from eating disorder symptom measures given (e.g., Eating Disorder Examination), lowest reported weight, highest reported weight, and type of control group used, (5) medication information (i.e., number of patients described as drug-free and number of patients using antidepressants, mood stabilizers, anxiolytics, antihistamine, or antipsychotics, and (6) MRI acquisition (i.e., slice thickness and magnetic field strength).
Analytic plan
We aim to expand on prior work using a VBM approach by using a technique called SDM [
27]. SDM is a coordinate-based meta-analytic tool that is able to combine both T-maps and coordinates into a single analysis, thus maximizing inclusion of studies. This tool also allows for the use of meta-regression to control for moderators [
28]. SDM has been applied to meta-analytic work in a variety of psychiatric disorders, including one study in acute AN [
15]. We will use SDM in our work to complement our ROI meta-analyses. For studies that used VBM methodology, we will use SDM to test for convergence against the null hypothesis that reported findings follow a random spatial distribution across the brain. For studies that used an ROI approach, we will achieve further specificity by demonstrating and comparing convergence abnormality in discrete brain regions among individuals with AN, weight-recovered AN, and healthy controls. Below, we outline our specific analytic plan in relation to our research questions:
1) How do global and regional structural brain abnormalities differ between individuals with acute AN versus healthy controls (HC), weight-recovered AN versus HC, and acute AN vs. weight-recovered AN? We will conduct a voxel-based and regional meta-analysis to probe differences in grey matter between individuals with acute AN and HC, weight-recovered AN and HC, and acute versus weight-recovered AN. Statistical analyses will be performed using random-effects inverse-weighted variance models. Subgroup analyses will be conducted where there is sufficient data to investigate. Where statistical pooling is not possible, the findings will be presented in narrative form including tables and figures to aid in data presentation, where appropriate. We will conduct sensitivity analysis to test how robust our results are relative to variations in meta-analytic method, using effect sizes and Hedges’ g for each brain region in our analyses for ROI and VBM. To reduce the number of comparisons in the ROI analysis, we will focus on brain regions that are significantly different from those of control subjects in either the acute-AN or weight-recovered-AN meta-analysis.
Effect sizes will be expressed as either odds ratios (for dichotomous data), weighted, or standardized final post-intervention mean differences (for continuous data), and their 95% confidence intervals will be calculated for analysis. Heterogeneity will be assessed statistically using the standard chi-squared and
I2 tests. For continuous outcome measures, we will use Hedges’
g ([
29]; the Cohen’s effect size with correction for bias from small samples).
In the case of the VBM studies specifically, we will follow the procedure for SDM [
27]. This statistical software allows for researchers to include both studies that use peak coordinates and coordinates that use Statistical Parametric Mapping (SPM) t-maps in a single meta-analysis. This technique increases accuracy of effect size maps by using specific masks for structural MRI scans. Furthermore, SDM accounts for the effect size and sign of peak coordinates and weights calculations based on inter-study variance and heterogeneity.
2) How do clinical characteristics and physiological differences moderate structural brain abnormalities in individuals with AN and weight-recovered AN? Meta-regression will be employed to explore the potential effects of clinical characteristics, neuroradiological, and study quality. To reduce type I errors, we will select clinical variables based on key clinical questions as well as the availability of the variables in the studies. The following moderators will be considered: duration of illness, age of onset, body mass index (BMI), amount of weight gain from treatment (longitudinal studies and cross-sectional recovered AN studies), levels of depression, shape and weight concerns, body image, restriction severity, drive for thinness, frequency of binge and purge behaviors, percentage of females, and percentage of medicated participants. To explore the effects of neuroradiological techniques, we will consider the following variables: MRI slice thickness, image smoothing level, MRI field strength (Tesla), and inclusion of motion correction.
To assess whether the findings in our analysis will be related to different presentations of extreme restrictive eating: AN binge/purge subtype and AN restricting subtype AN, we will complete a supplementary set of meta-analyses in subjects based on diagnostic categories. These subgroup analyses will be exploratory in nature and will be completed to generate hypotheses to be tested in new primary studies. We will use diagnostic criteria from the Diagnostic and Statistical Manual of Mental Disorders, 5th Edition (DSM-5). Table
1 presents the diagnostic criteria for adolescents and adults for each subtype:
Table 1
Anorexia nervosa subtypes
Anorexia nervosa-restricting subtype | • No binge episodes in the past 3 months • No purge episodes (i.e., self-induced vomiting or the misuse of laxatives, diuretics, or enemas) in the past 3 months • Otherwise meet criteria for AN |
Anorexia nervosa-binge/purge subtype | • Recurrent binge episodes in the past 3 months AND/OR • Recurrent purge episodes in the past 3 months • Otherwise meet criteria for AN |
Assessing risk of bias
Our proposed method includes VBM and ROI analysis to reduce risk of bias. We will compare results from the VBM and ROI analysis to assess agreement, which has broader implications for neuroimaging meta-analysis of other disorders and has not been done in eating disorders. VBM and ROI results often differ for the following reasons: (1) VBM involves smoothing, which biases sensitivity to brain regions dependent on the size of the kernel, (2) ROI analyses only report on a subset of brain regions and are likely to be impacted by publication bias, (3) SDM (used for VBM studies) uses coordinate data; thus, the effect size is biased toward zero in brain regions where there are no significant clusters, and (4) VBM adjusts for total global brain volumes, whereas ROI studies use absolute volumes.
We anticipate that some studies will report measures from subgroups of patients and matched controls. For example, studies may compare short-term versus long-term recovery. We will incorporate these results in the meta-analysis as two different studies, which is consistent with other meta-analyses [
14,
15]. As this will increase risk for type I error, we will use a Bonferroni correction and indicate results that survive this correction. Finally, we anticipate that studies may use different methods to diagnose eating disorders in their samples. We will record the method used for diagnosis (e.g., DSM-5 versus prior editions of the DSM and/or clinical interview, such as the Eating Disorders Examination) and include these as a subgroup analysis to examine stability of results. This is particularly important for an eating disorder meta-analysis as the criteria for anorexia nervosa has significantly changed over the years to include a broader range of individuals.
The Grading of Recommendations, Assessment, Development and Evaluation (GRADE) approach for grading the certainty of evidence will be followed to assess estimates of relative risk and a ranking of the quality of the evidence based on the risk of bias, directness, heterogeneity, precision, and risk of bias of the review results.