Efficacy and safety of psychostimulants for amphetamine and methamphetamine use disorders: a systematic review and meta-analysis

Background Amphetamine and methamphetamine use disorders are associated with severe health and social consequences. No pharmacological therapy has been approved for the treatment of these disorders. Psychostimulants can act as maintenance-like therapies for managing substance use among these patients. The aim of this study is to evaluate the literature examining the efficacy and safety of psychostimulant agents for increasing abstinence and treatment retention among patients with amphetamine and methamphetamine use disorders. Methods We searched MEDLINE, EMBASE, PsycInfo, Cochrane Central, and CINAHL from inception to August 2016. Selection of studies, data extraction, and risk of bias assessment were conducted independently by two reviewers. We conducted meta-analyses to provide a pooled summary estimate for included trials and report the review according to PRISMA guidelines. Results We identified and selected 17 studies with 1387 participants. Outcome reporting across trials was inconsistent, and the overall quality of evidence was very low due to high risk of bias and indirectness. A meta-analysis of five trials (642 participants) found no effect of psychostimulants for end-of-study abstinence (odds ratio = 0.97, 95% confidence interval 0.65 to 1.45). Additionally, the pooled estimate from 14 studies (1184 participants) showed no effect of psychostimulants for treatment retention (odds ratio = 1.20, 95% confidence interval = 0.91 to 1.58). The incidence of serious adverse events did not differ between intervention and placebo groups based on qualitative reports from trials. Conclusions Quantitative analyses showed no effect of psychostimulants for sustained abstinence or treatment retention. We also identified the need for more rigorous studies in this research area with clinician and patient important outcomes. Electronic supplementary material The online version of this article (doi:10.1186/s13643-016-0370-x) contains supplementary material, which is available to authorized users.


Background
Amphetamine-type stimulants, including methamphetamine, amphetamine, and dextroamphetamine, are highly addictive synthetic substances exhibiting increasing rates of abuse among individuals with wide-ranging sociodemographic profiles [1]. Methamphetamine use, production, and trafficking are significant problems in North America, with 61% of global methamphetamine seizures being reported within the continent in 2013 [2]. The use and abuse of amphetamine-type stimulants is a major public health concern given that initial exposure can potentially progress into dependence as rapidly as within a single week [1]. Amphetamine and methamphetamine use disorders (AMD), encompassing both abuse and dependence, are characterized by problematic patterns of the use of amphetamine-type stimulants, leading to development of tolerance and experiencing withdrawal symptoms [1]. Regular use of amphetamine-type stimulant substances can lead to adverse physical, psychological, and social consequences. Frequent methamphetamine use is associated with serious physical and psychological consequences such as increased risk of infectious diseases, cardiovascular pathology, cognitive impairment, severe dental decay, depression, psychotic disorders, and mortality [3,4]. Methamphetamine use is also associated with social consequences such as aggression, violence, and criminal activity [4,5]. Currently, there are no approved medications for the treatment of AMD [6,7]. Psychosocial therapies have shown positive outcomes, but results are inconsistent for treatment retention and many trials show no reduction in substance use at follow-up [8,9]. A previous research has called for future work to determine which medications are effective for relief of withdrawal symptoms and to be used in combination with psychosocial treatments [9]. Given the medical and social costs of AMD, it is necessary to identify an effective harm reduction therapy.
A number of randomized controlled trials (RCTs) have investigated the effectiveness of psychostimulants as harm reduction therapy for AMD. Psychostimulant agents are generally dopamine (DA) agonists that increase synaptic concentrations of DA with similar mechanisms of action as abused amphetamines and methamphetamines [10]. Abused amphetamines and methamphetamines increase synaptic concentrations of monoaminergic neurotransmitters, including DA [10]. The expectation is for psychostimulants to act as maintenance-type therapy for patients with AMD to relieve withdrawal symptoms and drug cravings, subsequently leading to reduced stimulant use. Previous reviews have been published [6,7,10,11], with only the most recent review by Pérez-Mañá et al. [7] conducting a meta-analysis. Pérez-Mañá and colleagues [7] found no effect of psychostimulants for AMD. However, a number of trials have been conducted with psychostimulants since the completion of this review. Our systematic review will broaden eligibility criteria by including adolescent and adult studies, and conducting subgroup analyses by age group and frequency of substance use, which may have an effect on treatment outcome [12,13]. Pérez-Mañá et al. [7] reported that the quality of evidence in this area was very low, indicating that the new trials published since then may have a substantial impact on the treatment outcome. Additionally, due to the small sample size of individual trials, an updated review is necessary to summarize the evidence.
The objective of the present review is to summarize the efficacy and safety of psychostimulant medications for the treatment of AMD in adolescents and adults. Specifically, we aim to summarize the current literature examining the efficacy of psychostimulant medications for (1) sustaining abstinence from illicit amphetamines and methamphetamines and (2) retention in treatment by conducting risk of bias and quality assessment of included trials, and computing quantitative pooled summary estimates. Secondary objectives of the review are to consider the potential modifiers of psychostimulant treatment efficacy and summarize the risk of serious adverse events (SAEs) related to psychostimulant therapy, as reported by the included studies.

Methods
The following methods were established a priori in an unpublished protocol that is available upon request from the authors. The review was submitted to PROS-PERO for registration but did not meet eligibility criteria at the time of submission. The review is reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [14] and an additional file shows the PRISMA checklist (see Additional file 1).

Data sources
The following databases were searched from inception until August 2016: MEDLINE, EMBASE, PsycINFO, Cochrane Central Library, and Cumulative Index to Nursing and Allied Health Literature (CINAHL). A broad search was employed in order to capture all relevant citations. An additional file outlines the search strategy, developed by the first author (M.B.) in collaboration with an expert health sciences librarian (N.B.) (Additional file 2). Furthermore, reference lists of past systematic reviews and included trials were manually searched for relevant studies. ClinicalTrials.gov was searched as a source for unpublished and ongoing trials.

Inclusion and exclusion criteria
The search was limited to studies with human participants. However, all databases were searched without restrictions on language or date of publication. We included RCTs comparing the use of any psychostimulant medication to an inactive control group (i.e., placebo) for the treatment of AMD. We excluded all study designs that were not RCTs, including non-randomized clinical trials and quasi-randomized studies. Participants included adolescents and adults ≥14 years of age and diagnosed with amphetamine or methamphetamine use disorder (including those with abuse or dependence), according to the Diagnostic and Statistical Manual of Mental Disorders (DSM) criteria. We included studies that used clinician-administered or self-reported validated instruments for diagnosis of AMD. We also included studies of participants diagnosed with DSM-5 criteria stimulant use disorder (including cocaine, amphetamine, and methamphetamine) and extracted relevant data about patients with only AMD. Trials of participants with cocaine use disorder alone were excluded. Studies examining the use of psychostimulants or central nervous system stimulants for the treatment of AMD were included in the review. Since "psychostimulant" agents represent a pharmacological effect rather than class of medications, a comprehensive list of these agents is not available. Instead, psychostimulants are dispersed over various pharmacological groups based on their indication. We utilized an approach described by Castells et al. [15] and Pérez-Mañá et al. [7] in previous reviews of stimulant use disorders to obtain a comprehensive list of psychostimulants. We identified drugs with psychostimulant modes of action from the Anatomical Therapeutic Chemical Classification [16] and the American Hospital Formulary Service Pharmacologic-Therapeutic Classification System [17] and included in the search strategy. The strategy included psychoanaleptics with mild stimulant effects (bupropion, modafinil) as well as classical stimulants (dextroamphetamine, methylphenidate, and dexmethylphenidate). Trials managing withdrawal rather than dependence were excluded. Trials were excluded if they assessed non-substance use outcomes, such as cognitive and neurological outcomes, after in-lab administration of amphetamines or methamphetamine.

Data collection and extraction
Two authors independently screened titles and abstracts of all retrieved citations, excluding studies that failed to meet eligibility criteria. Following title and abstract screening, two authors independently reviewed full-texts of relevant studies. Studies that satisfied all eligibility criteria were included for data extraction. We used Covidence systematic review software (Veritas Health Innovation, Melbourne, Australia) for screening and fulltext review. We resolved disagreements during the screening process by discussion to consensus. If they remained unresolved, a senior author was consulted to determine study eligibility. Two authors extracted data from included trials using a pilot tested data extraction form. An additional file includes the completed form with details extracted from individual studies (see Additional file 3), which was later exported into Review Manager Version 5.3 (The Cochrane Collaboration) for statistical analysis.

Risk of bias and quality assessment
Two raters independently assessed risk of bias of included trials using the Cochrane Collaboration's risk of bias tool [18]. Scores of high, low, or unclear risk of bias were assigned to trial factors including random sequence generation, allocation concealment, blinding of participants, personnel and outcome adjudicators, incomplete outcome assessment, selective reporting, and other sources of bias. We assessed publication bias visually by generating a funnel plot for primary outcomes. We used the risk of bias assessment to support conclusions regarding the overall quality of evidence in the review. We applied the Grading of Recommendations, Assessment, Development, and Evaluation (GRADE) framework for systematic reviews to the included trials to assign an overall outcome-specific rating for risk of bias, inconsistency, indirectness, imprecision, and publication bias [19]. We used GRADE Pro GDT software (http:// gdt.guidelinedevelopment.org/app/) to create evidence and summary of findings tables.

Data synthesis
We measured inter-rater agreement using the unweighted kappa statistic for full text review and risk of bias assessment and conducted meta-analyses using Review Manager Software Version 5.3 (The Cochrane Collaboration, London, UK).
The primary outcomes of this review were (1) abstinence from illicit amphetamines and methamphetamines and (2) retention in treatment. Abstinence was measured as the proportion of participants having substance-free urine tests in intervention and control groups. The longest substance-free period of time assessed across multiple studies was used to determine sustained abstinence. We measured treatment retention, or completion of treatment, as the proportion of individuals remaining in treatment at the end of the trial. The secondary outcome was the incidence of SAEs for different psychostimulant interventions in comparison with placebo. SAEs were defined as any medical or psychiatric event causing hospitalization or dropout from the study.
We summarized individual study results qualitatively and conducted meta-analyses to obtain pooled summary estimates (odds ratios), when interventions and outcomes were comparable between trials. We pooled studies broadly and conducted predetermined subgroup analyses to explore heterogeneity. We measured heterogeneity using the inconsistency index (I 2 ) for each meta-analysis, with I 2 >40% representing substantial heterogeneity. We applied the random-effects model with inverse variance methods for meta-analysis, as it accounts for within and between-study variability and considerable variability is expected between included trials [20]. We analyzed all data according to the intention-to-treat principle with the number of participants randomized being used as the denominator. If multiple trials reported overlapping of study participants, only the trial with the larger sample was included in the meta-analysis so as to not to violate the assumption of independence.

Subgroup analyses
We conducted the following a priori planned subgroup analyses: We planned an additional subgroup analysis to compare dosage of psychostimulant interventions but could not conduct due to similar doses between studies examining the same medications.

Sensitivity analysis
We conducted sensitivity analyses by removing studies with high risk of bias for each item on the Cochrane risk of bias tool in order to evaluate the robustness of the summary estimates for each outcome. We also conducted a sensitivity analysis by removing studies conducted within specific high-risk populations (participants with attentiondeficit hyperactivity disorder, criminal offenders, and men who have sex with men).

Description of studies
Following screening of 1251 unique citations for eligibility, 17 studies were included for data extraction and analysis in the review (Fig. 1). The unweighted kappa for full-text review for two independent reviewers was 0.84. We also calculated inter-rater agreement for the risk of bias assessment, yielding a kappa statistic of 0.83, indicating excellent levels of agreement between raters, as per Cochrane Handbook cutoffs.
Most studies enrolled participants with DSM-IV methamphetamine dependence, while two studies recruited participants with both amphetamine and methamphetamine dependence [28,32]. Furthermore, two other studies included participants with a diagnosis of amphetamine dependence alone and comorbid attention-deficit hyperactivity disorder (ADHD) [25,26]. There was only one trial conducted in adolescents [33], while the remaining trials were within adult samples, 18-65 years of age. The adolescent study was conducted among participants with ≤18 days of methamphetamine use in the month prior to baseline interview [33]. Two other studies were conducted among low-frequency adult methamphetamine users (≤29 days in the month prior to baseline interview) [12,13]. Men comprised >60% of the study sample in many studies, and two studies included only men [21,26]. Among those studies, Das et al. [21] specifically included men who have sex with men, a population at increased risk of methamphetamine use and abuse. A number of studies excluded individuals diagnosed with comorbid substance use disorders [13, 24-26, 28-30, 33, 35].
The majority of trials administered the psychostimulant intervention for 8-12 weeks, while few studies had longer treatment duration (20-24 weeks) [25,28,32]. Doses of medications were relatively consistent within psychostimulant type: however, the dose varied between different types of psychostimulants. All studies investigating modafinil administered doses of 200 mg, with the exception of study by Anderson et al. [35] having a second treatment group receiving 400 mg of modafinil. For the Anderson [35] trial, the number of events from the 200 and 400 mg treatment groups was combined in the meta-analysis and compared to the placebo group. Bupropion was consistently given in doses of 150 mg twice daily, and the maximum dose of methylphenidate ranged from 54 to 180 mg between studies. There were two trials of dextroamphetamine/dexamphetamine-both stabilizing participants at different doses. All trials were placebocontrolled, although most trials did not report the inactive substance in the placebo capsule. Two trials used lactose [23,27] and one used gelatin [32] as the inactive substance in the placebo group.

Risk of bias assessment
Figures 2 and 3 display a summary of the risk of bias assessment, conducted in duplicate by the review authors using the Cochrane risk of bias tool. A number of studies had unclear or high risk for at least one methodological criterion. Random sequence generation was adequately reported in less than half of the trials, and over 75% of the trials did not report allocation concealment adequately. While 16 of the 17 trials implemented blinding of participants and study personnel, all trials had unclear or high risk of bias for blinding of outcome assessment.  All studies used an inactive placebo pill of the same dosage as the intervention for the control group. The substance use outcomes are summarized as they are reported in the individual studies MA methamphetamine, SD standard deviation, NR not reported, GEE generalized estimating equations, OR odds ratio, CI confidence interval, MD mean difference Attrition was a major drawback in most trials, with >50% of participants dropping out before study completion. However, the majority of trials conducted an intention-to-treat analysis and had similar proportions of dropout in intervention and placebo groups. High risk of bias for incomplete outcome data was still an issue for 50% of the trials when reasons for dropouts were not stated, and it could not be determined whether reasons for dropout were similar for intervention and control groups. Risk of bias due to selective outcome reporting was primarily low, although study outcomes were reported inconsistently between trials.
All studies required written informed consent from participants and reported approval from an ethics review board. The majority of trials were not industry funded, but some study authors reported a conflict of interest. Other bias resulted from significant conflicts of interest (as judged by the rater), imbalanced randomization judged to have potential impact on the outcome, and underpowered based on presented sample size/power calculation. There were also a few studies with 17 or less participants in each treatment arm, which were likely biased towards type II errors due to small sample sizes [21,25,32,33].

Meta-analyses: efficacy of psychostimulants for abstinence from illicit amphetamine and methamphetamine
Five out of 17 included studies were included in the primary meta-analysis for abstinence from illicit amphetamines or methamphetamines [12,13,24,30,35]. All of the five studies included in the meta-analysis included participants with methamphetamine dependence (regular or low frequency of use) and measured "terminal abstinence," defined as methamphetamine-free urine screens during the final 2 weeks of the trial. The remaining studies were excluded from the meta-analysis because outcomes were not reported consistently, and the studies lacked sufficient information to determine the number of participants with sustained abstinence in intervention and control groups. Figure 4a shows the pooled summary estimate, based on five studies and a total sample size of 642 individuals. Of these participants, 353 received psychostimulant therapy and 289 received placebo. The pooled odds ratio for sustained abstinence was 0.97 (95% confidence interval 0.65 to 1.45, p value = 0.87), representing no significant effect of psychostimulants on abstinence from methamphetamines. This summary estimate was associated with low statistical heterogeneity, as indicated by an I 2 of 2% (chi-square = 4.08, degrees of freedom = 4, p

Subgroup analyses
Due to inclusion of five trials in the meta-analysis, we were unable to conduct subgroup analyses based on age (adolescent or adult), substance use disorder (amphetamine or methamphetamine use disorder), or treatment duration. The first subgroup analysis was conducted based on psychostimulant intervention. The modafinil group was associated with no heterogeneity (I 2 = 0%, p value = 0.78), while the bupropion group showed Fig. 4 a Forest plot for efficacy of psychostimulants on abstinence from illicit amphetamines or methamphetamines (measured by urinalysis). b Subgroup analysis by psychostimulant medication on abstinence from illicit amphetamines or methamphetamines. c Subgroup analysis by frequency of substance use on abstinence from illicit amphetamines or methamphetamines substantial heterogeneity (I 2 = 47%, p value = 0.15). However, both modafinil (OR = 0.86, 95% CI 0.46 to 1.61, p value = 0.65) and bupropion (OR = 1.12, 95% CI 0.54 to 2.33, p value = 0.76) showed no effect for sustained abstinence (Fig. 4b). The test for subgroup differences was not significant (I 2 = 0%, chi-square = 0.29, degrees of freedom = 1, p value = 0.59).
The second subgroup analysis was conducted for frequency of substance use and showed high heterogeneity among studies with low frequency (i.e., non-daily) substance users (I 2 = 74%, p value = 0.05). Figure 4c shows the forest plot. Studies with unspecified frequency of use were associated with no heterogeneity (I 2 = 0%, p value = 0.91). However, both low frequency (OR = 1.24, 95% CI 0.34 to 4.48, p value = 0.74) and unspecified frequency (OR = 0.92, 95% CI 0.54 to 1.54, p value = 0.74) substance use groups showed no effect for abstinence. The test for subgroup differences was again associated with no heterogeneity but remained non-significant (I 2 = 0%, chi-square = 0.18, degrees of freedom = 1, p value = 0.67).

Meta-analyses: efficacy of psychostimulants for retention in treatment
Fourteen studies were included in the meta-analysis for the outcome of retention in treatment, defined as the proportion of participants that completed the study (Fig. 5a). Two studies were excluded because information about dropouts was insufficient to determine the  [32,33], and a study by Elkashef et al. [22] was excluded due to an overlapping sample with Anderson et al. [12]. The pooled summary estimate was based on a collective 1184 participants, of whom 626 received psychostimulant interventions and 558 received placebo. The pooled OR (1.20) was non-significant (95% CI 0.91 to 1.58) and was associated with low heterogeneity (I 2 = 15%, chi-square = 15.34, degrees of freedom = 13, p value = 0.29). We explored sources of heterogeneity by prespecified subgroup analyses, but did not conduct a subgroup analysis of age because the adolescent study was excluded from the meta-analysis due to insufficient outcome data [33].

Subgroup analyses
The first subgroup analysis was based upon study inclusion of participants with amphetamine use disorder and methamphetamine use disorder (Fig. 5b). The test for subgroup differences was not statistically significant (I 2 = 0%, chi-square = 0.28, degrees of freedom = 1, p value = 0.59). The amphetamine use disorder subgroup was associated with substantial heterogeneity (I 2 = 57%) based on the cutoff of high heterogeneity at >40%, and the pooled OR of 1.54 was not statistically significant (95% CI 0.48 to 4.99, p value = 0.47). The methamphetamine use disorder subgroup was associated with no heterogeneity (I 2 = 0%), and a non-significant overall OR of 1.11 (95% CI 0.86 to 1.44, p value = 0.42). The test for subgroup differences was associated with no heterogeneity but was not significant (p value = 0.59).
We conducted a second subgroup analysis by specific type of psychostimulant medication, varying between two to six studies within the subgroups (Fig. 5c). A test for differences in effect between the aforementioned subgroups was not significant (chi-square = 2.31, degrees of freedom = 3, p value = 0.51), but associated with no heterogeneity (I 2 = 0%). No subgroups of psychostimulants showed a significant OR for treatment retention, and only the methylphenidate subgroup was associated with substantial heterogeneity (I 2 = 43%, p value = 0.13). There was a trend in both studies of dexamphetamine for favorable effects of psychostimulants; however, there was no significant pooled effect (OR = 2.50, 95% CI 0.80 to 7.87, p value = 0.12) and the 95% confidence interval was wide.
We conducted the third subgroup analysis (Fig. 5d) based upon frequency of substance use disorder. As previously defined, low frequency of substance use was characterized as substance use ≤29 days in the past month (i.e., non-daily substance use) and was compared to studies including participants with unspecified frequency of substance use. None of the subgroups showed a statistically significant pooled estimate for treatment retention; however, the low frequency subgroup only included two studies. The test for subgroup differences showed no heterogeneity and was not significant (chisquare = 0.05, degrees of freedom = 1, p value = 0.83).
The final subgroup analysis for this outcome was based on treatment duration, categorized as ≤12 weeks or >12 weeks (Fig. 5e). The test for subgroup differences was borderline significant and associated with substantial heterogeneity (I 2 = 74.5%, chi-square = 3.93, degrees of freedom = 1, p value = 0.05). The subgroup of studies with ≤12 weeks of treatment did not find a significant pooled OR. The pooled estimate of two studies >12 weeks was significant and associated with no heterogeneity. Results showed twice the increased odds of retention in treatment >12 weeks for the psychostimulant group with an OR = 2.45 (95% CI 1.14 to 5.25, p value = 0.02).

Psychostimulant safety: adverse events
Three of the 17 studies utilized a standardized instrument for investigating differences in adverse events between psychostimulant and placebo groups [23,28,34], while the remaining studies reported observed adverse events. Miles et al. [28] measured adverse events using the Udvalg Kliniske Undersogelser and found no differences between methylphenidate and placebo arms. Shearer et al. [34] reported no statistically significant univariate differences in mild, moderate, or serious adverse events between modafinil and placebo groups. Similarly, Galloway et al. [23] found no differences in reported adverse events between dextroamphetamine and placebo groups. Across all studies, mild adverse events differed by psychostimulant medication but these mild symptoms were not associated with dropout or discontinuation of medication in most studies.
While a number of studies reported SAEs including hospitalization due to suicidal ideation, seizures, pneumonia, and others, no study stated that the SAEs or dropouts were associated with active study medication. However, there was high inconsistency in reporting of adverse events, and it was often unclear whether SAEs resulted in complete discontinuation from the study. From the available information, we found there were 8 dropouts from modafinil groups [24,34,35], 11 from bupropion [12,13,22,30], and 7 from methylphenidate [27] (26 total) and a total of 11 from placebo groups [13,22,24,[26][27][28]30].

Sensitivity analyses
We conducted sensitivity analyses by removing studies within sub-populations one at a time. Das (2010) included a sample of men who have sex with men, Konstenius et al. [25] comprising of adults with comorbid ADHD, and Konstenius et al. [26] of men who are criminal offenders with ADHD were removed, but there were no differences in the results of the meta-analyses for the primary outcomes. Additionally, the pooled estimate did not change significantly for either primary outcome when we removed studies with high risk of bias for "selective outcome reporting" and "incomplete outcome data" one at a time. Figure 6a, b show funnel plots assessing publication bias for treatment retention. Figure 6a generates using the five studies from the meta-analysis assessing abstinence from amphetamine and methamphetamine use. The left side of the plot favors placebo, and the right side of the plot favors psychostimulants with regard to increased odds of abstinence from illicit stimulants. There is no evidence of asymmetry in the funnel plot based on five of the 17 included studies. However, it is expected that the plot will remain symmetrical after addition of the remaining studies since a number of studies with negative results have been published. Figure 6b assesses publication bias for treatment retention based on 14 included studies. There is likely no publication bias since

Discussion
This systematic review included 17 trials investigating the efficacy and safety of psychostimulants in patients diagnosed with AMD. Five out of 17 studies were included in the meta-analysis for the primary outcome of sustained abstinence. Studies included in the metaanalysis for abstinence all investigated psychostimulants for methamphetamine dependence, and no effect was found for this outcome. Subgroup analyses of abstinence also found no effect. For the outcome of treatment retention, 14 studies were included in the meta-analysis and the odds ratio was pooled. Overall, we found no significant effect of psychostimulants on treatment retention. However, the subgroup analyses for treatment retention demonstrated favorable effects for treatments with longer duration (>12 weeks). The literature has shown that patients enrolled in longer duration general substance use treatment programs tend to have more successful treatment outcomes [36]. These results are interesting because they indicate that participants in longer duration studies are at increased odds of retention until the end of treatment: however, this finding is based on a subgroup of two studies and cannot be generalized for all samples [26,28]. It is possible that longer duration treatment may also show increased effectiveness for substance use outcomes such as abstinence from amphetamines and methamphetamine, which is important to examine with further research. However, this finding of higher retention may also be due to factors such as standard of care, including improved therapeutic alliance or counseling community, in longer duration studies. As well, continued participation may have occurred due to expectation of early parole or fear of extended incarceration in the sample of criminal offenders recruited from prisons [26]. In the remaining subgroup analyses, we found no significant effects, although the direction of effects was consistent with our hypotheses. Other reviews have reported modafinil to be effective in reducing amphetamine and methamphetamine use [6,9]; however, these reviews did not quantitatively summarize the results. The present review found no effect of modafinil or bupropion on sustained abstinence from substances in the meta-analysis.
There was a high level of inconsistency in reporting adverse events and safety outcomes among included studies; therefore, the overall observed number of SAEs was determined for each psychostimulant intervention and placebo groups. There were no clear differences in reported adverse events based on raw numbers between intervention and placebo groups. However, adverse events may have been impacted by imperfect collection and reporting methods, since only three of 17 studies used standardized instruments to collect adverse event data [23,28,34]. Development of adverse event reporting standards for RCT's in this area and transparent reporting of the methods used to ascertain whether SAEs were associated with study medication is necessary. This is especially important due to the increasing number of trials conducted in recent years, which have shown no effect for reduction of substance use [12,13,28,33]. While it remains challenging to conduct clinical trials in the field of substance use disorders, improving the quality of future trials is important to detect a true treatment effect that is not significantly impacted by methodological limitations.
Previous reviews for pharmacological interventions have been conducted for general stimulant dependence [6,[9][10][11]37]; however, most reviews provide narrative summaries of results whereas only the most recent review is specific to AMD patients and conducts metaanalyses [7]. Since the publication of the recent review by Pérez-Mañá et al. [7], a reasonable number of new trials examining psychostimulants have been conducted [12,13,[26][27][28]31]. Though new trials in this area are being conducted, the methodology and reporting of outcomes remains largely unchanged, making it challenging to utilize trial results for quantitative analysis in systematic reviews. Substance use outcomes across the RCTs include the proportion of negative urine screens within a group, mean methamphetamine non-use days or weeks, longest period of time abstinent in days or weeks, mean number of negative urine screens, and number of participants with terminal or end-of-trial abstinence. Lack of uniformity in outcome reporting among trials investigating pharmacological therapies for AMD has resulted in several recent trials reporting group level outcomes such as overall proportion of negative urine screens in each trial arm [21,26,28,31,32,34]. Group level outcomes make it difficult to ascertain the effect of psychostimulant treatment on individual patients. In their systematic review, Pérez-Mañá et al. pooled the mean difference in number of negative urine screens per group as their primary outcome. Although this is an important outcome, it is statistically inappropriate to pool as a continuous outcome since the number of urine screens is a discrete value with some trials conducting as few as 10 urine tests [31]. For this reason, it is important for clinicians to identify a minimum clinically significant number of negative urine screens needed to achieve "success" in harm reduction treatment. This can allow future clinical trials to quantify the number of individuals achieving success in intervention and control groups as a dichotomous outcome and conduct a responder analysis [38].
However, this approach is also limited since there is currently no established threshold for negative urine screens needed to be "successful" in substance abuse treatment.
More than half of the trials were judged to have high risk of bias for at least one measure on the Cochrane risk of bias tool. The majority of studies did not mention allocation concealment, and incomplete outcome data (attrition) was a major source of bias. However, studies did conduct intention-to-treat analyses to compensate for the high attrition rate. Moreover, included trials had small samples (47% of trials had ≤30 participants per trial arm) and may have been underpowered to detect an effect at the individual study level. The GRADE evidence profile ( Table 2) and summary of findings (Table 3) display the overall quality of evidence as assessed using the GRADE framework and can be used to appraise the certainty of findings from this review. For the outcome of abstinence, risk of bias was very serious due to lack of allocation concealment and incomplete outcome data; however, treatment retention was only affected by lack of allocation concealment. There was a relatively low heterogeneity between studies for both primary outcomes;  The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different Low quality: Our confidence in the effect estimate is limited: The true effect may be substantially different from the estimate of the effect Very low quality: We have very little confidence in the effect estimate: The true effect is likely to be substantially different from the estimate of effect The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI) CI Confidence interval, OR Odds ratio a The majority of studies had high attrition bias (>50% dropout rate) and small sample sizes b 80% of studies did not mention allocation concealment, which may be a source of bias c Heterogeneity was not explained by subgroup analyses, as indicated by non-significant tests for subgroup differences d Studies investigating the efficacy of different psychostimulant drugs at varying doses were pooled e 95% confidence intervals are wide and there is a varying range of effect, with little overlap of confidence intervals from some studies f Populations varied across studies with certain studies including injection drug users, incarcerated individuals or participants with ADHD however, existing heterogeneity was not explained by subgroup analyses given that the tests for subgroup differences were not significant. Indirectness of the population and interventions affects the quality of the evidence. Many studies had strict eligibility criteria that excluded patients with comorbid substance use disorders [13, 24-26, 28-30, 33, 35], whereas patients in clinical practice frequently present with multiple substance use disorders [39]. Furthermore, some trials were conducted in specific populations such as individuals with ADHD, incarcerated individuals, or men who have sex with men [21,25,26]. Psychostimulant interventions may have differential treatment effects within such specific populations diagnosed with AMD. The ORs for abstinence and treatment retention varied across studies, with little to no overlap of 95% confidence intervals of individual study point estimates. There was no evidence of publication bias for either outcomes based on visual assessment of funnel plots. Overall, the quality of evidence in the review can be considered very low based on the GRADE assessment.

Strengths and limitations
This systematic review was conducted with an a priori design to assess the efficacy and safety of psychostimulants for AMD. We conducted a comprehensive search of psychostimulant agents and two reviewers independently conducted title and abstract screening, full text review, risk of bias assessment, and data extraction. This is the largest review conducted to date investigating psychostimulant agents for AMD including 17 RCTs; however, we were limited by the quality of evidence in individual trials. Reporting of outcomes across studies was very inconsistent and did not allow us to pool all of the available data. Furthermore, we only included published and available data from included trials rather than obtaining additional data for analyses. Published data were used in order to comment on the quality of outcome reporting in RCTs in this area. Clinician and patient important outcomes were chosen as primary and secondary outcomes for this review rather than group level outcomes reported by individual studies. Nevertheless, the small number of trials in the meta-analysis for abstinence limits the generalizability of the findings. As well, we created broad eligibility criteria to include adolescent studies, yet only one trial has been conducted that has found inversely significant results. For this reason, results of this review cannot be generalized to adolescent populations.
contributed to the conception and design of the study, study selection process, critical revision, and final review of the manuscript. LBB contributed to the study selection process, critical revision, and final review of the manuscript. NB contributed to the development of the search strategy and final review of the manuscript. NM contributed to the critical revision and final review of the manuscript. AW contributed to the critical revision and final review of the manuscript. LT contributed to the critical revision and final review of the manuscript. ZS contributed to the conception and design of the study, critical revision, and final review of the manuscript. All authors read and approved the final manuscript.