Peripheral iron levels in children with attention-deficit hyperactivity disorder: a systematic review and meta-analysis

There is growing recognition that the risk of attention-deficit hyperactivity disorder (ADHD) in children may be influenced by micronutrient deficiencies, including iron. We conducted this meta-analysis to examine the association between ADHD and iron levels/iron deficiency (ID). We searched for the databases of the PubMed, ScienceDirect, Cochrane CENTRAL, and ClinicalTrials.gov up to August 9th, 2017. Primary outcomes were differences in peripheral iron levels in children with ADHD versus healthy controls (HCs) and the severity of ADHD symptoms in children with/without ID (Hedges’ g) and the pooled adjusted odds ratio (OR) of the association between ADHD and ID. Overall, seventeen articles met the inclusion criteria. Peripheral serum ferritin levels were significantly lower in ADHD children (children with ADHD = 1560, HCs = 4691, Hedges’ g = −0.246, p = 0.013), but no significant difference in serum iron or transferrin levels. In addition, the severity of ADHD was significantly higher in the children with ID than those without ID (with ID = 79, without ID = 76, Hedges’ g = 0.888, p = 0.002), and there was a significant association between ADHD and ID (OR = 1.636, p = 0.031). Our results suggest that ADHD is associated with lower serum ferritin levels and ID. Future longitudinal studies are required to confirm these associations and to elucidate potential mechanisms.


Results
Study selection. Figure 1 summarizes the details of the search results. In brief, a total of 46 studies entered the full-text review stage. Twenty-nine articles were excluded for various reasons including a lack of controls, non-clinical trials, not comparing iron between ADHD/controls, or review articles (see Supplementary Table 2). A list of excluded articles is presented in Supplementary Table 2. In total, twenty-two articles met the inclusion criteria. We were unable to conduct meta-analysis for four of the recruited studies, because we did not have enough studies (n < 3) for the outcome measurements such as hair iron level, food iron intake level, or plasma/ blood iron level in children with ADHD versus controls, or the prevalence of ADHD in children with and without ID [20][21][22][23] . Finally, we recruited seventeen studies in our meta-analysis 9,[24][25][26][27][28][29][30][31][32][33][34][35][36][37][38][39] , and the details are summarized in Table 1.
Characteristics and methodological quality of the included studies. Of the included studies, thirteen provided information about serum ferritin levels, six about serum iron levels, two about serum transferrin levels, three about the severity of ADHD symptoms in patients with ID compared to controls, and three provided data of adjusted odds ratios (ORs) with regards the association between ADHD and ID.
We were unable to conduct meta-analysis for the prevalence of ADHD in children with and without ID because we did not have enough studies (n < 3) 9 . Rather, we chose differences in the severity of ADHD symptoms in children with and without ID to indicate the presence of ADHD.

Study quality appraisal.
Regarding the methodological quality of the included studies, the average modified Newcastle-Ottawa Scale (NOS) score 40 for case-control trials in subgroup meta-analysis of comparisons of the peripheral iron levels in children with and without ADHD was 6.82 (high quality) with a standard deviation (SD) of 1.17 41 . The NOS score for cross-sectional trials in subgroup meta-analysis of comparisons of the peripheral iron levels in children with and without ADHD was 4.67 (high quality) (SD = 0.58) 41 . In addition, the average SCIENTIFIC RePoRTs | (2018) 8:788 | DOI: 10.1038/s41598-017-19096-x modified NOS scores for cohort trials and cross-sectional trials in subgroup meta-analysis of comparisons of the severity of ADHD symptoms in children with and without iron deficiency were 5.00 (low quality) (SD = 1.41) 40 and 5.75 (high quality) (SD = 2.50) 41 , respectively (Supplementary Table 3).
Meta-regression analysis. The ESs of differences in serum ferritin levels between the children with ADHD and controls had significantly positive association between mean age (slope = 0.187, k = 17, p = 0.025) but not significantly moderated by proportion of females (p = 0.061), percentage of combined subtypes of ADHD (p = 0.550), percentage of the inattention subtype of ADHD (p = 0.860), percentage of the hyperactivity/impulsivity subtype of ADHD (p = 0.884), cognition (in forms of mean IQ) (p = 0.574), sample size of the group (p = 0.740), and geographical latitude of where the study was conducted (p = 0.690) (Supplementary Table 4).
Subgroup meta-analysis. To identify potential sources of heterogeneity, we focused on the studies recruiting children who did not receive medications. Three studies with four datasets were included 34,38,42 , and the results of meta-analysis showed that there was no significant difference in serum ferritin level between the children with ADHD and the controls (k = 4, Hedges' g = −0.287, 95% CI = −0.812 to 0.237, p = 0.283).
Meta-regression analysis. The ESs of differences in serum iron levels between the children with ADHD and controls were not significantly moderated by mean age (p = 0.827), proportion of females (p = 0.958), sample size of the ADHD group (p = 0.388), and geographical latitude of where the study was conducted (p = 0.361) (Supplementary Table 4).
Meta-analysis of peripheral iron levels in the children with ADHD versus controls: peripheral transferrin. Meta-analysis demonstrated that there was no significant difference in serum transferrin level    36 . We could not perform meta-analysis on plasma transferrin or peripheral blood transferrin levels because there were fewer than three datasets.
Meta-regression analysis. Meta-regression and subgroup meta-analysis could not be performed as fewer than the minimal number of datasets were available. Meta-regression analysis. Meta-regression and subgroup meta-analysis could not be performed as fewer than the minimal number of datasets were available.   9 . Meta-regression and subgroup meta-analysis could not be performed as fewer than the minimal number of datasets were available.

Discussion
The main results of our review and meta-analysis suggest that serum ferritin levels, but not iron or transferrin levels, are significantly lower in children diagnosed with ADHD than in those without ADHD. However, when removal of one potential confounding study, the result of meta-analysis of serum iron changed into significantly lower serum iron level in the children with ADHD compared to the controls. Children with ID were also more likely to have ADHD and have more severe ADHD symptoms than those without ID. Our results are in general agreement with three previous meta-analyses [14][15][16] (Table 2). However, our findings not simply confirm the same results in previous reports but also added further information upon current scientific knowledge, such as a higher odds of ADHD and higher severity of ADHD symptoms in the patients with ID and significantly lower serum iron level in the children with ADHD compared to the controls in specific post-hoc meta-analysis, such as sensitivity testing. Specifically, we found that the ORs and symptom severity of ADHD were higher in the patients with ID. Together with a lower ferritin level, this result may suggest an association between a lower serum iron level and ADHD symptomatology. Moreover, although both our meta-analysis and most recent meta-analysis by Wang et al. showed no significant difference in serum iron levels between ADHD and control groups 16 , sensitivity test was not done in the study by Wang et al. After we re-investigated the potential sources of insignificant results through sensitivity test (Table 2), we found significantly lower serum iron method: Chemiluminescent method; CPRS: Conners' parents rating scales; CTRS: Conners' teacher rating scales; DSM-III: diagnostic and statistical manual of mental disorders, third edition; DSM-IV: diagnostic and statistical manual of mental disorders, fourth edition; Dx: diagnosis; Fer.: Ferrozine method; HC: health control; ICD-9: international statistical classification of diseases and related health problems 9th revision; ID: iron deficiency; IDA: iron deficiency anemia; MA: meta-analysis; Mx: medication; MP: methylphenidate; n/a: not available; OR: odds ratio; PDD: pervasive developmental disorder; RIA: Radio-immunoassay; RLS: restless leg syndrome; SDQ: Strengths and difficulties questionnaire; TBAQ-R: Toddler behavior assessment questionnairerevised; Tx: treatment. Based on the hypothesis of normal distribution of the peripheral iron levels and prevalence of ADHD, we merged the different outcomes from recruited studies into one single outcome, the Hedges' G.  levels in the children with ADHD compared to those without ADHD after removing the study by Chen et al. 27 . In the study by Chen and colleagues (2004), we identified several significantly different baseline variables between the ADHD and control groups, including higher food iron intake and higher food vitamin C intake in the ADHD group. Vitamin C is known to enhance iron absorption from food 43 . Knowing that human iron nutrients primarily come from food intake, the children with ADHD included in the study by Chen who had significantly higher iron and vitamin C intake would be expected to show elevated blood iron levels. Such information about food intake was not provided by the other studies included in our subgroup meta-analysis of serum iron levels 26,35,36,38 . Therefore, it was reasonable to exclude data from the study by Chen et al. 27 when interpreting the results of subgroup meta-analysis of serum iron levels. Following exclusion from this study, the results of subgroup meta-analysis of serum iron levels revealed a significantly lower serum iron level in the children with ADHD than in those without ADHD (Hedges' g = −0.186, 95% CI = −0.323 to −0.048, p = 0.008). This provides further evidence of a shortage of body iron stores in children diagnosed with ADHD. Taken together, our findings provide tentative evidence that deficient iron storage in children with ADHD may be involved in the pathophysiology of the condition. However, future longitudinal research is required to confirm/refute this tentative hypothesis.
The relationship between ID and ADHD may be explained by several possible pathophysiological mechanisms. First, low peripheral iron levels, indicating insufficient iron storage, may dysregulate dopaminergic neurons, which may play a prominent role in the pathoetiology of ADHD 5,44,45 . In brief, iron aids in dopamine synthesis by acting as a co-factor for tyrosine hydroxylase, which is a rate-limiting enzyme for the conversion of hydroxylation of tyrosine to L-DOPA, a precursor of dopamine 46 . Iron deficiency may therefore result in disruption of dopamine activity, as shown in several animal studies 12,47,48 . This dysregulation of dopaminergic neurons may further result in multiple frontal dysfunctions that mimic the symptoms of ADHD 49 . In addition, ADHD has been found to be more prevalent in patients with restless leg syndrome (RLS) (27.62% with a diagnosis of ADHD), and iron deficiency and dopamine system dysregulation has also been reported to play an important role in RLS 47,48,50 . Therefore, iron deficiency with resulting brain dopamine dysfunction may be a common pathway for the pathophysiology of both disorders. In addition to the dopamine theory, lower ferritin levels may provide indirect evidence of elevated oxidative stress 17 , and heightened oxidative stress has also been reported in patients with ADHD 51 . This increased oxidative stress burden may disturb neurodevelopmental trajectories and gene functions potentially predisposing to the onset of ADHD 51 . However, given the observational nature of our data, the precise mechanisms and directionality of any relationships we observed cannot be verified.
Our meta-analysis focused on peripheral iron levels rather than brain iron levels. The extent to which peripheral and brain iron levels are correlated remains unclear. Cortese and colleagues provided some evidence that brain iron levels in the bilateral thalamus were significantly lower in children with ADHD compared to healthy controls 28 . However, another MRI study by Adisetiyo et al. failed to find an association between brain iron levels and serum ferritin levels 42 . Therefore, further studies are required to explore the relationship between brain and serum iron levels.
In the main result of our meta-analysis, high heterogeneity had been detected through the results. To address the potential source of heterogeneity, we arranged subgroup meta-analysis and meta-regression to investigate it. In part of subgroup analysis, if we only included the studies that recruited children who did not take medications, the difference in ferritin level between the ADHD and control groups became non-significant. Although, no obvious interactions have been reported between most drugs and ferritin level, D' Amato suggested that methylphenidate may cause a poor appetite and possibly less iron intake in children with ADHD 52 . However, only four of the nineteen datasets recruited drug-free participants, and most of the other studies did not provide much information about the medications prescribed to their participants. Therefore, further studies are warranted including drug-free participants only to further investigate the relationship between methylphenidate and ferritin level in children with ADHD. In addition to the potentially confounding effect by medication, the different food intake pattern would also contribute impact on the iron storage. For example, Lane and the colleague (2014) provided evidences about the enhancing effect on iron absorption by vitamin C intake 43 . Studies recruited subjects with high vitamin C intake or iron supplementation would have confounded results compared to others 27 . Finally, in part of meta-regression, we tried to address the potential impact of some clinical variables on the peripheral iron levels. The differences in serum ferritin levels between the children with ADHD and the controls were only significantly moderated by mean age but not by other factors including proportion of females, subtype of ADHD, cognition (in forms of mean IQ), sample size of the ADHD group, and geographical latitude of where the study was conducted. We are uncertain about the significance of the finding that the difference in ferritin level in ADHD and control groups became larger with age because there were no previous reports or studies addressed this issue. However, previous studies showed that the reference range of ferritin level became wider and higher with age in pediatric population 53 . Therefore, further studies investigating ferritin level and ADHD symptoms may need to take this factor into account, before analyzing their study results. On the other hand, our meta-regression did not show significant associations between the prevalence of ADHD in children with ID and mean age, mean proportion of females, and geographic latitude of where the study was conducted.

Limitations
There are several limitations to the current study. First, the total number of included studies was relatively small, and therefore there was a risk of type I and type II errors. Second, our main targets focused on peripheral samples of iron status parameters rather than CNS parameters because few studies provided such information 28,42 . Third, we could not fully exclude the potential confounding effect of food iron intake on peripheral iron status because only a few studies provided such information 27 . In addition, we could not completely rule out the potential confounding effect of the ability of different types of assay to detect iron due to the inconsistent reporting of those data across the included studies. Fourth, we lacked comparisons of changes in iron status in the children with ADHD in long-term follow-up because there were few cohort studies 7,39 . Fifth, we could not perform subgroup meta-analysis of plasma ferritin, iron, or transferrin due to the limited number of datasets available in the studies. Finally, we could not perform meta-regression of peripheral iron levels and attention because of the limited data available.

Conclusions
The results of our meta-analysis suggest that children diagnosed with ADHD have lower serum ferritin levels compared to those without ADHD. We also observed that the children with ID were more likely to have ADHD and to suffer from more severe ADHD symptoms compared to those without ID. We therefore suggest that further studies are warranted to explore the benefits of iron supplementation in children with ADHD with ID, in particular those with more severe ADHD symptoms. However, given the cross-sectional nature of most of the available studies, further longitudinal and cohort studies are required to thoroughly evaluate the relationships between iron status and ADHD symptoms, and to elucidate the potential pathophysiological mechanisms. In addition, further studies may be needed to investigate the relationship between methylphenidate and ferritin level to exclude any potential effects of methylphenidate on oral iron intake.

Methods and Materials
The current study followed the Meta-analysis Of Observational Studies in Epidemiology (MOOSE) guidelines 54 (see Supplementary Table 1 and Supplementary Figure 1). The meta-analysis followed our previous defined but unpublished protocol and was approved by the Institutional Review Board of Tri-Service General Hospital (TSGHIRB: B-105-12).
Eligibility criteria. The inclusion criteria were: (a) observational studies, including a cohort or cross-sectional study design, comparing all peripheral iron status parameters including iron, ferritin, and transferrin levels in children with ADHD (confirmed by either a structured or non-structured diagnostic interview) and controls; and (b) clinical human studies. We excluded preclinical studies, review articles, meeting abstracts, or peer-reviewed original articles not conducted in humans.
Search strategy and study selection. Two independent authors conducted an electronic systematic literature search from inception to August 9 th , 2017 across PubMed, ScienceDirect, Cochrane CENTRAL, and ClinicalTrials.gov databases. We used the following keywords: "(iron OR ferritin OR ferrous) AND (Attention deficit hyperactivity disorder OR ADHD)". To expand our eligible list, we consulted the reference lists of the included articles and recent reviews 14,55,56 .
At the eligibility stage, two authors (YS Cheng and PT Tseng) screened the titles and abstracts of all results to assess potential eligibility. The authors then reviewed the full-text articles which were deemed potentially eligible. A final list of included studies was determined and any inconsistencies were resolved by consensus or via thorough discussion with a third reviewer (PY Lin).
Primary outcomes. We set the primary outcomes as the difference in peripheral iron levels (including iron, ferritin, or transferrin) between children with ADHD and controls, and the OR of ADHD or severity of ADHD symptoms between children with or without ID. If the data of interest were not available in the articles, we contacted the authors twice over a month to request the data. Data extraction. Two independent authors (YS Cheng and PT Tseng) extracted data using a predetermined list of variables of interest, which included: prevalence/incidence rates of ADHD, prevalence rates of iron deficiency, peripheral iron levels, amount of food iron intake, mean age, gender distribution in the form of the percentage of females, body mass index, percentage of ADHD subtypes, cognitive performance in the form of IQ, parental tobacco smoking/alcohol consumption, ethnicity (including African, Caucasian, Asian, and Hispanic), parental history of ADHD, geographical latitude of where the study was conducted, and the type of assay used to detect the iron level.
Assessment of study quality. We used the Newcastle-Ottawa Scale for cohort studies and case control studies. For cross-sectional studies, we used a modified version of the Newcastle-Ottawa Scale for observational studies to assess the quality of the included studies. This modified version of the Newcastle-Ottawa Scale score for observational studies ranges from zero to six, and a score greater than three was classified as a high-quality study. For case control studies, the Newcastle-Ottawa Scale score ranges from zero to ten, with a score greater than five being classified as a high-quality study 41 . For cohort studies, the Newcastle-Ottawa Scale score ranges from zero to nine, with a score of six or more being classified as a high-quality study 40 .
Statistical analysis. The current study was conducted in two parts. First, we analyzed the data considering iron in relation to the children with ADHD compared to the controls. Second, we analyzed the data about the severity of ADHD symptoms or OR of ADHD in the children with ID compared to the controls. To control for the potential confounding effects of clinical variables, we performed further meta-analysis based on the pooled adjusted OR from the recruited studies. In brief, we extracted the adjusted OR with regards to the association of ADHD in the children with ID from the recruited studies to calculate the pooled adjusted OR of the association between ADHD and ID.
Based on the presumed heterogeneity of background and population among the recruited studies, we conducted the meta-analyses with a random effects model rather than a fixed effects model 57 . In brief, random-effects modeling is more stringent than fixed-effects modeling and incorporates a between-study variance in the calculations 58 . For continuous outcomes (i.e. differences in iron levels between the children with ADHD and the controls), we calculated Hedges' g and 95% CI. For dichotomous outcomes (i.e. differences in the prevalence of ID between the children with ADHD and the controls) we calculated the OR and 95% CI.
Heterogeneity, publication bias, and sensitivity test. Heterogeneity was assessed using the Cochran Q test 59 . The I 2 statistic should be interpreted as the proportion of heterogeneity a study estimates that is due to heterogeneity 60 . For publication bias, we used direct inspection of funnel plots for fewer than 10 datasets 61 and Egger's regression test for 10 or more datasets 62 . We also used the Duval and Tweedie's trim-and-fill procedure to adjust the ESs when publication bias was evident 63 . We also used a sensitivity test with one study removed to investigate any potential outliers present in the recruited studies 64 .
Subgroup meta-analysis and meta-regression analysis. To discover any potential sources of heterogeneity, subgroup analyses were performed to explore potential interactions between clinical variables and peripheral iron status parameters in the children with ADHD compared to the controls. We only performed subgroup analysis whenever data from three independent datasets were available 65 . The main clinical target for subgrouping included sample sources, subjects who were or were not drug free, and the situation when blood was drawn. When data were available for a moderator in more than five studies, we performed unrestricted maximum likelihood random-effects meta-regression to explore any potential source of heterogeneity. The moderators of interest included mean age, female proportion, body mass index, percentage of ADHD subtypes, mean IQ, percentage of parental tobacco smoking/alcohol consumption, percentage of each ethnicity, percentage of parental history of ADHD, sample size of the disease groups, and geographical latitude of where the study was conducted.
The meta-analyses were conducted using Comprehensive Meta-Analysis software, version 3 (Biostat, Englewood, NJ). Data availability. The datasets generated and analyzed during the current study are available from the corresponding author on reasonable request.