Efficacy and Tolerability of Different Interventions in Children and Adolescents with Attention Deficit Hyperactivity Disorder

Background Our study is an analysis of multiple publications involving assessing the comparable efficacy and tolerability of six interventions, which are lisdexamfetamine dimesylate (LDX), atomoxetine (ATX), methylphenidate (MPH), clonidine hydrochloride (CLON), guanfacine extended release (GXR), and bupropion, for young patients (6–18 years old) suffering from attention deficit hyperactivity disorder (ADHD). Methods A conventional meta-analysis (MA) was performed to give direct comparisons and a network meta-analysis (NMA) was used to show the combination of direct and indirect evidence. Ranking preference for all the interventions under a certain outcome was given by the surface of cumulative ranking curve area (SUCRA). Results Overall, 15,025 participants from 73 studies were involved in our analysis. In the pairwise MA, LDX was associated with less withdrawal than ATX for lack of efficacy. MPH showed less effectiveness than LDX according to ADHD Rating Scale score. Based on the analysis of our NMA, significant results of efficacy that LDX is a competitive drug were observed when evaluating LDX in comparison with other drugs except for CLON. ATX and GXR presented higher rates of abdominal pain morbidity versus inactive treatment. Conclusion The stimulants LDX and MPH are still highly recommended because they are highly effective and are tolerated well by patients. Among the non-stimulants, CLON can be taken into consideration for its appreciable effectiveness and tolerability. ATX and GXR can be seen as moderate choices.

and social life (1). The exact causes of ADHD are unclear, but it is thought that it is the result of an imbalance of catecholamine metabolism in the cerebral cortex, or inhibitory dopaminergic and decrease of noradrenergic activities, or a mixture of the two (1,4). Drugs therapies treating ADHD can be classified into dopaminergic and noradrenergic pathways (4). Several drugs have been employed for patients suffering from ADHD, including stimulants and non-stimulants. For example, atomoxetine (ATX) is widely used and is considered as a non-psychostimulant (5). It can reduce the symptoms of ADHD and has other clinical advantages such as drowsiness, comorbidities with tics, and anxiety (6). Bupropion (BUP) is a monocyclic phenylaminoketone structurally related to the phenylisopropylamines, with significant antidepressant effects (7). Clonidine hydrochloride (CLON) is α2-adrenergic agonist developed to reduce and delay the release and has recently been used in combination with psychostimulants to treat ADHD (8,9). Guanfacine is the most widely prescribed psychostimulants as the selective α2A-adrenoceptor agonist for the treatment of ADHD. The efficacy of guanfacine extended release (GXR) is considered to be among first-line treatments for ADHD (10). Lisdexamfetamine dimesylate (LDX) is a stimulant, normally used as a monotherapy or supplement for psychostimulants in ADHD treatment (10,11). In general, LDX is employed as prodrug due to its fast absorption and hydroxylation, which leads to a gradual and long lasting release (12)(13)(14). Methylphenidate (MPH) is a psych stimulant and is considered as the first-line therapy (6). It can increase the concentration of serotonin, dopamine, and norepinephrine to control the extent of inattention and impulsivity (5). MPH is also recognized for its significant antidepressant effects (7). All of the treatments described above have been effective in the treatment of ADHD, but there are aspects of their relative efficacy and safety that still remain unclear. Therefore, it is necessary to establish a system of evaluation in order to make comparisons among these various therapies.
A variety of studies have been conducted in the attempt to create a system of comparison for these diverse treatments,    however, a large proportion of these studies have only focused on the pairwise comparison with placebo-controlled treatment (7,(15)(16)(17)(18)(19), or were restricted to three or four medications (20). Besides, the majority of these current literatures for systematic analysis have only considered the efficacy of these therapies without incorporating an evaluation of their safety (21). Moreover, the characteristics of the samples chosen in some studies are limited to small subgroups and this has led to limitations in comparing the differences between treatments (4). More importantly, previous studies have used different indicators to measure the efficacy of treatment options. This has led to inconsistencies and contradictions that further increase the need to perform a network meta-analysis (NMA) to estimate the effectiveness and reliability of medications used to cure ADHD. Our study tries to combine studies involving seven interventions (non-stimulants: ATX, BUP, CLON, and GXR; stimulants: LDX and MPH). Seven outcomes have been considered, including ADHD Rating Scale (ADHD-RS), all cause withdrawal, withdrawal due to adverse event, withdrawal due to lack of efficacy, nausea, abdominal pain, and fatigue.

Publication search
A systematic search was conducted in PubMed, Embase, Cochrane library, and CNKI (up to March 29, 2017), aiming to retrieve randomized controlled trials (RCTs) related to drug therapy in children and juveniles with ADHD. We used the following key words: "randomized controlled trial, " "attention deficit hyperactivity disorder" (including synonyms), and "drug therapy. " The cited articles of references including RCTs, systematic reviews or meta-analyses were also searched manually as supplementary material. Ethical approval was not needed for this study.

inclusion criteria
Type of study: mostly RCTs of a minimum of 3-week duration will be included in this review.
Type of participants: children and adolescents aged between 6 and 18 who meet Diagnostic and Statistical Manual of Mental Disorders, 4th edition (DSM-IV) criteria for ADHD.
Type of interventions: studies involving direct comparison with one drug therapy against another or against placebo will be included. All target interventions are ATX, CLON, GXR, BUP, LDX, and MPH.
Type of outcome measures: the efficacy is evaluated by ADHD-RS as a continuous score. And the adverse effects (as a dichotomous outcome) for tolerability are all cause withdrawals, withdraw due to adverse event, withdrawal due to lack of efficacy, nausea, abdominal pain, or fatigue for tolerability. Weight loss is not included because there is limited outcome on it.

Data extraction
We extracted baseline data and evaluated the risk of bias, including selection bias, performance bias, detection bias and other bias, by the means of random sequence generation, allocation concealment, blinding, incomplete outcome data, and selective reporting. The assessment tool came from Cochrane Handbook (version 5.1.0) for RCTs. Data of interest were blinding, durations, diagnostic criteria, treatment, age of patients, number of patients, and assessment criteria for patients' conditions.

statistical analysis
Conventional meta-analysis (MA) was performed by STATA 12.0 software, which gave us direct comparisons among these drugs, in the forms of the mean deviation (MD) with the corresponding 95% confidence interval (CI) for the primary outcomes of ADHD-RS, and the pooled odds ratios (ORs) with 95% CI for the secondary outcomes of tolerability. Cochran's Q test (22) and the I 2 test (23) were utilized to assess the degree of heterogeneity among studies. A fixed-effects model (Mantel-Haenszel method) was utilized if significant heterogeneity did not exist (P > 0.05 or I 2 < 50%). Otherwise, a random-effects model was used. This NMA was conducted using STATA 12.0 software and WinBUGS software, which showed us the combination of direct and indirect evidence. A Markov chain Monte Carlo method was applied to build Bayesian networks. The data presented in our NMA were similar to that in a pairwise MA. We illustrated the comparison of efficacy and tolerability by computing the MD and OR with the corresponding 95% credible interval (CrI) separately. Ranking preference for all the interventions under a certain outcome was given by the surface of the SUCRA, the value of which was 1 for the best and 0 for the worst.
There are five domains are bias due to (1) the randomization process, (2) deviations from intended interventions, (3) missing outcome data, (4) measurement of the outcome, and (5) selection of the reported results (24). We used a comparison adjusted funnel plot to illustrate publication bias. Symmetry of the plots indicated no publication bias. P-value was a significant parameter to assess whether there was consistency in comparing direct and indirect evidence, and P < 0.05 showed a statistical inconsistency. The degree of consistency was illustrated by color in the heat plot. Figure 1 showed the process of literature retrieval and screening. A total of 2,024 studies were obtained after removing duplicates in the primary searches in databases and other sources. Among these studies, 1,567 trials were excluded by title and abstract, according to the inclusion criteria. Full-text reading finally enabled us to identify 73 qualitative trials as sources for data extraction (3, 5-16, 19, 25-83). Overall, 15,025 participants were involved in our analysis. Figures 2 and 3 showed the geometric distribution of RCTs for the included outcomes, which were related to different aspects in efficacy and tolerability. Placebo was taken as the control group in most RCTs. There were considerable quantities of patients in the research results for ATX, MPH, LDX, and GXR. However, the number of patients with BUP and CLON as interventions was limited. Baseline characteristics were listed in Table 1. Double blinding was adopted in most trials, with one single blinding and seven open-label trials.

result from Pairwise Ma
The results of the direct MA were shown in

results from nMa
Bayesian models allowed for more refined estimates. Comparisons without direct connection were compared indirectly through Bayesian NMA. Available data from NMA was recorded in Tables 3 and 4

ranking scheme Based on sUcra
The probability of being the best treatment was derived from SUCRA. The result was displayed in Figures 7 and 8. LDX and MPH could be considered as a group with the best comprehensive ranking score, including efficacy and tolerability. LDX had the highest probability of being the best efficacious drug therapy in decreasing ADHD symptoms (0.72) and remitting abdominal pain (0.82). MPH was located in the top three under all the outcomes except nausea. CLON ranked in the secondary group, but there was an absence of data related to nausea and abdominal pain. ATX and GXR had moderate rankings in efficacy, but were associated with the worst evaluation in the morbidity of adverse events. Though BUP was given a high ranking in reducing nausea, the results remained unclear due to the absence of essential information.

Publication Bias and consistency
The symmetry of the "comparison adjusted" funnel plots in Figure  S1 in Supplementary Material suggested no publication bias. The small-study effect was limited in our research. According to the heat plot in Figure S2 in Supplementary Material, there were no statistically inconsistent results between direct and indirect evidence according to the blue color in most area, except for some ambiguity between LDX and MPH.

DiscUssiOn
This comprehensive NMA was conducted to compare the efficacy and tolerability of all interventions used to treat young ADHD patients. Our NMA study gave a comprehensive evaluation that included direct and indirect evidence extracted from previous studies. This NMA's results indicated that the statistical differences between all interventions were compared successfully and their ranking orders under different criteria were determined, respectively. Therefore, it is possible for us to find the optimal treatment option by taking all the outcomes into account.
Stimulants, including LDX and MPH, have been comprehensively proven superior to non-stimulants in improving children and adolescents' symptoms of ADHD both in documents and our NMA study. LDX presented the best efficacy. Previous studies have demonstrated that the mechanism of LDX for treating ADHD was related to the sufficient gradual-release of d-amphetamine by hydrolysis from LDX (44). Due to this feature, LDX could provide continuous effects throughout the day. LDX is an effective medication, but its level of toxicity is generally attracted additional attention because it frequently results in adverse events. Nevertheless, it has been confirmed that its toxicity can be controlled via changing the daily dosage (44), which makes it very competitive and consistent with our NMA result.
Besides, our results suggested that MPH is a good candidate, especially with regards to fatigue and the rate of withdrawal, which makes MPHas the routine therapy clinically. LDX is used for patients with ADHD who have an inadequate response to MPH (12,14). Although MPH releases quickly in vivo as a stimulant, it can be controlled by modification, which is documented as osmotic release oral system (37,41,51,83), leading to moderate side effects (20). It is worth noting that the adverse responses that have been selected in our NMA are related to mild symptoms. There are still arguments concerning the stimulant therapies. For example, some researchers have mentioned that stimulants are not recommended for patients with various cardiovascular problems because there have been reports of sudden death at usual doses and serious cardiovascular adverse events (68).
Non-stimulant therapies can reduce safety concerns to some extent, so they are expected to be substitutes for stimulants. We give an appreciable ranking score to CLON, and its performance in ADHD-RS shows that it tends to be superior to the stimulant MPH. Although its characteristics of releasing in a twinkling induces unwanted effects, for instance, somnolence, restfulness, and sleepiness, and its half-life is relatively short, leading to more side effects, dose optimization and extended-release formulation could be adapted to address the problems (8). However, the probability of overestimation remains due to the relatively limited sample size and direct evidence is absent to provide robust support. ATX and GXR are located at moderate positions under symptom improvement and rate of withdrawal. There are not significant differences in ADHD-RS when compared with another therapy, which indicated comparable efficacy. This result is consistent with existed evidence. However, the unsatisfying SUCRA ranking scores of ATX and GXR in nausea, abdominal pain and fatigue do not mean they are poorly tolerated or unsafe. Statistics about BUP should be updated so that a more comprehensive estimate can be presented. The overall quality of literatures included can be considered relatively high in terms of publication time, the authority of the journals and the characteristics of the studies. Besides, this NMA study compared all available medications for ADHD and assessed their efficacy and tolerability under various outcomes. However, there are some limitations existing for our study. (1) It is difficult for us to exclude publication bias completely, especially for ADHD-RS and all-cause withdrawal. This may be a result of small sample size and lack of direct comparison between some therapies, such as GXR versus CLON, or CLON versus BUP. (2) Most of the studies compared active treatment to placebo and only a handful of comparative effectiveness studies. (3) Differences in samples, such as mean age, gender ratio, and growth background also affect the accuracy of our NMA. Thus, in order to take all the limitations mentioned above into consideration, better-designed RCT studies should be performed.
In summary, according to the results obtained from our NMA, the stimulants LDX and MPH are still highly recommended because they are highly efficacious and well tolerated by patients. Among the non-stimulants, CLON should be taken into consideration for its appreciable effectiveness and tolerability. ATX and GXR can be seen as moderate choices. We failed to evaluate BUP because of the lack of enhanced evidence. Even though this study provides some guidance for the effectiveness and safety of various medications, clinicians still need to use their professional judgment in assessing the benefits with the efficacy and reliability profile of the therapy, as well as patients' features and preferences in order to make final treatment decisions.

aUThOr cOnTriBUTiOns
Conception or design of the work: RL and ZM. Data acquisition, analysis, and interpretation: FY. Drafting the work: RL and ZM. Revising article critically for important intellectual content: SH. Final approval of the version to be published: all authors.

sUPPleMenTarY MaTerial
The Supplementary Material for this article can be found online at http://www.frontiersin.org/article/10.3389/fpsyt.2017.00229/ full#supplementary-material.  In combination, we show heat colors corresponding to the change in agreement between direct and indirect estimate. The blue color means the two data possess good consistency while red color indicates inconsistency.