Attention-deficit/hyperactivity disorder and brain metabolites from proton magnetic resonance spectroscopy: a systematic review and meta-analysis protocol

Abstract Despite major advances in the study of the brain, investigations on neurochemistry in vivo still lack the solid ground of more established methods, such as structural and functional magnetic resonance imaging. Proton magnetic resonance spectroscopy (MRS) is a technique that might potentially fill in this gap. Nevertheless, studies using this approach feature great methodological heterogeneity, such as varying voxel of choice, differences on emphasized metabolites, and absence of a standardized unit. In this study, we present a methodology for creating a systematic review and meta-analysis for this kind of scientific evidence using the prototypical case of attention-deficit/hyperactivity disorder. Systematic review registration: International Prospective Register of Systematic Reviews (PROSPERO), CRD42018112418.


Introduction
Attention-deficit/hyperactivity disorder (ADHD) is a neuropsychiatric condition characterized by symptoms of inattention, hyperactivity, and impulsivity. 1 It is a prevalent disorder among children and adolescents that often persists into adulthood. 2,3 ADHD is also highly associated with comorbid psychiatric disorders 3 and negative outcomes, such as poor quality of life, 4 unemployment, 5 and increased mortality. 6 The economic burden associated with ADHD in the U.S. alone is estimated to be US$67-116B yearly due to lack of productivity. 5 Even so, key aspects of the neural basis of the disorder remain to be unmasked. 7,8 Since the 1990s, major advances in technology have made it possible to better understand brain diseases through the study of the brain in vivo. 9 Functional magnetic resonance imaging studies, for instance, have shown the role of frontoparietal and default mode network systems on ADHD. 10 Nevertheless, some areas of neuroscience, such as neurochemistry, have shown a more modest progress, still majorly relying on either post-mortem or more indirect approaches. 11 Given this framework, proton magnetic resonance spectroscopy (MRS) appears to be a link to partially fill in this gap and promote an upgraded landscape on the study of the biochemistry of living tissues. [12][13][14] MRS is a technique based on the concept of chemical shift, which describes how electronic shielding of an atomic nucleus embedded in a more complex chemical compound -i.e., a nucleus not free -changes its resonance frequency. 15 This allows us to establish different fingerprints to different molecules based on specific resonance frequencies of protons and estimate their amount on a certain pre-specified volume of interest. Following this principle, some molecules are classically found on MRS studies of neural tissue, such as n-acetylaspartate, creatine, choline, glutamate, myo-inositol, and lactate. 16 The first MRS studies of the brain of living organisms were performed in the 1980s, 17 Dissertations & Theses Global, using the keywords "ADHD" and/or "Attention-deficit hyperactivity disorder" for the condition, and "MRS," "MR spectroscopy," and/or simply "spectroscopy" for the technique. No restriction will be imposed on language or year of publication.
All entries will be recorded in a comprehensive list.
Duplicates will be removed and the remainder records will form the final list for eligibility evaluation. Once the studies are selected, we are going to look for further potential candidate articles in the reference lists of all the studies included (Table 1).

Study identification and selection
Two authors will independently analyze the whole list of studies to assess eligibility. Their evaluation will be matched for each entry and divergences (through percentage and Kappa statistics: κ ≡ (po -pe)/ (1-pe) = 1 -(1-po)/(1-pe), where po is the relative observed agreement among raters (accuracy), and pe is the hypothetical probability of chance agreement.
Whenever divergences are found, the articles will be finally assessed by a third author. In a first approach, studies not related to our investigation will be excluded on the basis of title and abstract. All remaining entries will be subjected to full-text reading, when other studies might be ruled out.
The following inclusion criteria will be considered: 1) all studies must contain at least one group with patients with ADHD and one group with healthy controls; 2) all studies must contain original proton MRS data on brain metabolites. When more than one diagnosis is considered, the studies will be included if there is a group comprised of patients with ADHD only. Cases will be defined as other psychiatric conditions will be required, neither for cases nor for controls; this information will nonetheless be used as a quality criterion to be evaluated in our bias assessment, as described hereinafter.

Study characteristics to be extracted
Extraction of data from each study will retrieve the following characteristics: year of publication, sample size of each group, male-to-female ratio in each group, aimed population (children/adolescents or adults), mean age of each group, regions of the brain studied, metabolites measured, and main results. The extraction will be performed by two authors, and divergences will be addressed by a third author. General principles of extraction followed the Cochrane Handbook for Systematic Reviews of Interventions. 27 Data directly related to the meta-analysis -i.e., mean metabolite measurement, standard deviation (or standard error), and sample size for each group (cases and controls) -will be recorded on independent worksheets, one for each brain region selected. The preferable sources are tables and written data on the article's full text. When the information is not detected on these formats, graph estimation using a digital ruler 28,29 will be performed. In case the previous methods are not available in the study or supplemental material, an email requesting the information will be dispatched for the correspondence address indicated in the authors' section of the article. If there is no response in two months, the data will be deemed as missing.
Two authors will collect the data independently, and disparities will be corrected by discussion consulting the original source.

Selection of brain regions for meta-analysis
Considering the heterogeneity of voxel choice among studies, it is important to delineate our approach on grouping data for the meta-analysis. While we anticipate that some degree of subjectivity will be unavoidable, we expect to proceed in a roughly systematic manner, as follows: on a first level, if enough studies are available, we will group data that target specifically the same regions (e.g., left dorsolateral prefrontal cortex); if not enough studies are available, we will combine more comprehensive data, but still on areas spatially related (e.g., left dorsolateral prefrontal cortex and left ventromedial prefrontal cortex); lastly, if the previous ScienceDirect tak ((ADHD OR "Attention-Deficit/Hyperactivity Disorder") AND ("Nuclear magnetic resonance spectroscopy" OR "NMR spectroscopy" OR MRS OR "MR spectroscopy")) SciELO (Nuclear magnetic resonance spectroscopy) OR (NMR spectroscopy) OR (MRS) OR (MR spectroscopy) AND (ADHD) OR (Attention-deficit hyperactivity disorder) Scopus TITLE-ABS-KEY ( ( adhd OR "Attention-deficit hyperactivity disorder" ) AND ( "Nuclear magnetic resonance spectroscopy" OR "NMR spectroscopy" OR mrs OR "MR spectroscopy" ) ) Web of Science TS = (ADHD OR Attention-deficit hyperactivity disorder) AND TS = (Nuclear magnetic resonance spectroscopy OR NMR spectroscopy OR MRS OR MR spectroscopy) methods are not feasible, we intend to incorporate data from similar structures, disregarding laterality.

Bias assessment
Risk of bias will be assessed using the Newcastle-

Meta-analysis
Studies will be grouped according to the aforementioned criteria and a meta-analysis will be performed for each metabolite of each region that meets a minimum of three values coming from at least three different studies. In articles with more than one ADHD Establish brain regions investigated in the study Create one worksheet for each brain region Is there more than one targeted case group? (e.g. "treatment naïve" and "on stimulants")  group available (e.g., "treatment naïve ADHD group"

META-ANALYSIS DATA EXTRACTION FOR EACH STUDY
and "on stimulants ADHD group") in a way that the data cannot be coupled, all ADHD groups will be matched against the control group, with the sample size of the healthy control group divided by the number of ADHD groups rounded down. As for the example given, if both a "treatment naïve group" and an "on stimulants group" are present, both will be fully used as case groups, but matched against a control group divided by two (the number of case groups that meets the eligibility criteria), considering that the control group would be otherwise counted twice. Additionally, when combining different data from the same study (e.g., disregarding lateralization in a study with data from both sides of the brain, hence using two different sets of data from the same study, one for each side), both cases and controls will be divided by two and rounded down. These steps aim to avoid overestimation of sample sizes and both may be applied to the same work simultaneously if needed.

Standardized mean differences obtained through
Hedge's G method with random effects will be employed to determine pooled effect sizes. Significance will be established by a Z-test. Inverse variance will be used to determine individual study weights. Studies will also be assessed on their heterogeneity using the χ 2 and I 2 tests, considering p ≤ 0.01 as statistically significant.
Low, moderate, and high heterogeneity values will be assumed from I 2 values of 25, 50, and 75%, respectively. 31 In light of the fact that the possible use of effect sizes from the same study would artificially decrease heterogeneity, we will employ a three-level analysis adopting "study" as an extra random variable.
The R package meta will be used to assess standardized mean differences and heterogeneity values as well as to generate forest plots. 32

Sensitivity of analysis
We will perform sensitivity analyses in order to evaluate methodological disparities that might account for effect size differences. First, we intend to use the jackknife method to assess how dissonant studies might be affecting the outcome, excluding one result at a time from each meta-analysis; second, we will execute the analyses again after excluding studies with high risk of bias or rated as "unclear risk of bias" Are there three or more studies targeting specifically the same brain area on the same side of the brain? (e.g. 4 studies on left DLPFC)

CLUSTERING BRAIN AREAS FOR DATA ANALYSIS
Are there three or more studies targeting areas roughly on the same brain region at the same side of the brain? (e.g. left DLPFC and left VMPFC)

No
Are there three or more studies targeting areas roughly on the same brain region disregarding laterality?

No
There is not enough data.
Perform the analysis using this cluster.
Perform the analysis using this cluster.
Perform the analysis using this cluster. through a score related to our bias assessment from the NOS, as previously described, excluding studies rated as lower than one standard deviation from the mean (bootstrapping will be used to acquire normal distributions and set the thresholds in case the data is non-parametric); and third, we will divide the groups by age, gender, and field strength of the machinery, when enough data is available. Finally, we foresee an additional analysis including only published works to be performed as well.

Discussion
The MRS seems to be a useful technique for in vivo brain investigations, especially when considering the limited availability of methods for studying neurochemistry in living neural tissue. The main limitation of this approach is that its use is only possible when high concentrations of metabolites are available (in the millimolar range), once the magnetic resonance method is poorly sensitive. Also, on a practical level, there is a significant spectral overlap among many compounds, making its discrimination a particular challenge. 15  is not a primary concern for study purposes -; prespecifying brain regions to be studied as inclusion criteria 37 ; choosing a preferable brain side of source when information from both hemispheres is available 38 ; and stratifying the data in brain lobes. 39 Meta-analyses also diverge on how to address metabolite measures.
While most works screen for all the main proton MRS molecules, 36,37,[39][40][41] some choose to focus on specific frequency ranges. 38 We chose to preliminarily include all studies that met our eligibility criteria and subsequently delineate a hierarchical order of preference for merging the outcomes. With this approach, we hope to compile the data in a way that is both comprehensive and as specific as possible. Another limitation of our protocol refers to our broad inclusion criteria, which will include different diagnostic methods and mostly rely on the judgment of the authors of each study, probably blending different diagnostic rationales and different presentations of the same disorder. Nevertheless, at present, this seems to be the only way to pool a reasonable amount of data for our purposes. Also, our primary goal is to investigate MRS metabolites in the general diagnostic standard currently available. As we are still working on a symptom-based approach for diagnosis in psychiatry, biological parameters still have to derive from nonbiological clustering, and only subsequent analyses will be able to provide data for different methods in the future.

Conclusion
In summary, we believe that our protocol of systematic review and meta-analysis might prove helpful in broadening the use of MRS data. MRS is a potentially fruitful approach for neuroscientific endeavor, and the body of knowledge already generated from it can be used to draw a unified understanding of metabolite concentrations for specific disorders.
The heterogeneity in brain areas studied and the need for broad inclusion criteria due to the relatively low number of studies available would probably be the main limitations of our approach. Even so, we hope that our methodology will support us to gather multiple estimates of how much of a substance is present in living neural tissue, opening a vast window of investigation for which the full employment remains to be explored. Therefore, following this protocol will potentially allow us to use the MRS data produced so far to enlighten our current understanding of either ADHD or any other brain disorder. Last, in succeeding in our purpose, we would conceivably provide a background capable of encouraging more systematic approaches to MRS studies.