Types of studies

We will include randomised controlled trials (RCTs) for intended beneficial outcomes. We will include cross‐over designs if the order of receiving treatments was randomised.

To improve our ability to detect data on adverse effects, we will include RCTs and observational studies (such as cohort studies, case control studies, case series and case reports), which we will use specifically to obtain data on harmful outcomes.

Our rationale for selecting studies according to type of outcome is that RCTs are designed to measure efficacy outcomes (Califf 2012), and are inadequate to assess adverse effects due to limitations in sample size, outcome assessment and follow‐up of participants, among other reasons (Fortin 2006; Saini 2014; Zorzela 2014). On the other hand, observational, non‐randomised studies, are important study designs to capture data on harmful outcomes (Loke 2011; O’Connor 2011).

Types of participants

Adults aged 18 years or older with a diagnosis of attention deficit hyperactivity disorder (ADHD) according to the Diagnostic and Statistical Manual of Mental Disorders (DSM)Third Edition (DSM‐III; APA 1980), Third Edition Revised (DSM‐III‐R; APA 1987), Fourth Edition (DSM‐IV; APA 1994) or Fifth Edition (DSM‐5; APA 2013), or with a diagnosis of hyperkinetic disorders according to the International Statistical Classification of Diseases and Related Health Problems Ninth Revision (ICD‐9), and Tenth Revision (ICD‐10) (WHO 1992).

Types of interventions

Immediate‐release methylphenidate administered at any dosage as part of any treatment regimen, compared with placebo or other pharmacological interventions.

Types of outcome measures

Electronic searches

We developed two search strategies: a core search strategy to identify RCTs with the intention of extracting data on beneficial and harmful outcomes, and a second search strategy to identify non‐randomised trials with the intention of extracting data related to harmful effects.

Our core search strategy includes both text word and indexed terms for the population of interest (patients with ADHD) and the intervention (methylphenidate). It is limited to RCTs using the Cochrane Highly Sensitive Search Strategy (Lefebvre 2011). We designed this strategy for MEDLINE (Appendix 1) and will adapt it to search the databases listed below.

1. Cochrane Central Register of Controlled Trials (CENTRAL; current issue) in the Cochrane Library, which includes the Cochrane Developmental, Psychosocial and Learning Problems Specialised Register

2. Embase Ovid (1980 onwards)

3. MEDLINE Ovid (1946 onwards)

4. MEDLINE In‐Process & Other Non‐Indexed Citations Ovid (current issue)

5. MEDLINE Epub Ahead of Print Ovid (current issue)

6. CINAHL EBSChost (Cumulative Index to Nursing and Allied Health Literature; 1980 onwards)

7. PsycINFO Ovid (1806 onwards)

8. Science Citation Index Web of Science (SCI; 1970 onwards)

9. Social Sciences Citation Index Web of Science (SSCI; 1970 onwards)

10. Conference Proceedings Citation Index — Science Web of Science (CPCI‐S; 1990 to current)

11. Conference Proceedings Citation Index — Social Science & Humanities Web of Science (CPCI‐SS&H; 1990 to current)

12. Cochrane Database of Systematic Reviews (CDSR; current issue) part of the Cochrane Library

13. Database of Abstracts of Reviews of Effects (DARE; Final Issue: 2015 Issue 2) part of the Cochrane Library

14. ClinicalTrials.gov (clinicaltrials.gov)

15. Open Trials (opentrials.net)

16. Drug Industry Documents (www.industrydocumentslibrary.ucsf.edu/drug)

17. World Health Organization International Clinical Trials Registry Platform (WHO ICTRP; apps.who.int/trialsearch)

Our search strategy to find data related to adverse effects is also designed for MEDLINE (Appendix 2) and is sensitive for all study designs (including cohort studies, case‐control studies, case series and case reports. It includes an adapted version of the filter for adverse effects published by Golder 2006. Together with the databases listed above, we will also include the following two databases in the adverse effects search, which we selected on the basis of previous research on searching for adverse drug effects (Golder 2012).

1. TOXLINE (US National Library of Medicine; 1940 onwards; toxnet.nlm.nih.gov/newtoxnet/toxline.htm).

2. British Library Direct (ondemand.bl.uk/onDemand/home).

We will search each database from inception onwards with no restrictions on date or language. The literature search will be performed by one author (SG) and another author will peer review the searches, in accordance with recommendations in the literature (McGowan 2015).

Searching other resources

We will contact specialists in the subject area and check citations of included studies to identify additional studies not captured by the electronic searches. We will also search for internal reports and conference proceedings on the websites of the European Medicines Agency (EMA; www.ema.europa.eu/ema), and the US Food and Drug Administration (FDA; www.fda.gov).

Selection of studies

We will use the reference management software, EndNote (EndNote 2017), to merge titles returned from the searches and remove any duplicates. Two review authors will then independently screen the titles and abstracts to remove irrelevant records. Next, we will retrieve the full texts of potentially relevant reports, which we will screen for eligibility using a standardised study selection form; this form will highlight our complete inclusion and exclusion criteria. At this stage, we will link together multiple reports of the same study. We will contact authors of studies whenever there is insufficient information available to decide whether or not a study is eligible for inclusion. We will resolve disagreements in the selection process by consensus or by consulting a third review author. We will record our decisions in a PRISMA diagram (Moher 2009).

Data extraction and management

We will develop and use a standardised data extraction form. Two review authors will independently extract data from each included study and we will resolve disagreement by discussion or by consulting a third review author. We will obtain additional information from the authors of the selected studies when necessary. Our data extraction form will be tailored to record the following data.

1. Trial design

2. Trial setting

3. Characteristics of the participants

4. Characteristics of the treatment

5. Comparator interventions and any cointerventions

6. Methods used to measure the outcomes and follow‐up duration

7. Outcomes and outcomes measures (any measures related to primary or secondary outcomes, as described under Types of outcome measures)

8. Financial conflict of interests

We will also extract data on the following essential elements related to harmful outcomes (Zorzela 2016).

1. How and when the adverse effect was ascertained

2. Methods to assess possible causality

Assessment of risk of bias in included studies

We will assess the risk of bias in RCTs across the following seven domains, using Cochrane’s tool for assessing risk of bias (Higgins 2017).

1. Sequence generation (selection bias)

2. Allocation sequence concealment (selection bias)

3. Blinding of participants and personnel (performance bias)

4. Blinding of outcome assessment (detection bias)

5. Incomplete outcome data (attrition bias)

6. Selective outcome reporting (reporting bias)

7. Other potential bias*

Two review authors will independently perform this task, resolving any disagreements by discussion or by consulting a third review author.

We plan to assess the risk of bias resulting from some domains for different groups or outcomes separately, in accordance with the instructions outlined by Cochrane (Higgins 2017). Specifically, we will:

1. assess blinding of participants and personnel (performance bias) separately for: a) participants and b) personnel;

2. assess blinding of outcome assessment (detection bias) separately for: a) beneficial outcomes and b) harmful outcomes; and

3. assess incomplete outcome data (attrition bias) separately for: a) primary beneficial outcomes and b) harmful outcomes.

We will rate the risk of bias in each domain as high, low or unclear, and we will provide a statement to support each of our judgements in accordance with the tool.

If necessary, we will revise the 'Risk of bias' domains to be assessed in accordance with new guidelines that may follow the release of the new Cochrane 'Risk of bias' tool (Sterne 2016), including issues regarding risk of bias of cross‐over trials. We may also adjust our 'Risk of bias' assessment regarding vested interests according to any new instrument adopted by Cochrane addressing this issue, and which may accompany the new Cochrane 'Risk of bias' tool, soon to be released.

We will assess the risk of bias of non‐randomised studies retrieved from the additional search for harmful effects in a two‐stage process. First, we will apply a minimised tool, based on the Risk Of Bias In Non‐randomized Studies — of Interventions (ROBINS‐I) (Higgins 2016; Sterne 2016), to evaluate design‐specific domains related to risk of bias. Then, we will consider additional domains to assess the risk of bias regarding adverse effect outcomes.

The rationale for applying a minimised tool is justified by evidence that the risk of bias for adverse effects may be distinct from biases applicable to outcomes of benefit (CRD 2009; Loke 2011; Viswanathan 2008a). For instance, confounding may impose a threat to the internal validity of measurements related to beneficial effects but not to harmful effects since "confounding by indication mainly influences treatment decisions with respect to outcomes about which the clinicians are primarily concerned” (Reeves 2011).

The minimised 'Risk of bias' tool will include the following 'Risk of bias' domains, with their respective signalling questions (Higgins 2016; Sterne 2016).

1. Selection of participants

   1. Was selection of participants into the study (or into the analysis) based on participant characteristics observed after the start of intervention?

   2. Do start of follow‐up and start of intervention coincide for most participants?

   3. Were adjustment techniques used that are likely to correct for the presence of selection biases?

2. Classification of interventions

   1. Were intervention groups clearly defined?

   2. Was the information used to define intervention groups recorded at the start of the intervention?

   3. Could classification of intervention status have been affected by knowledge of the outcome or risk of the outcome?

3. Departures from intended interventions

   1. Were there deviations from the intended intervention beyond what would be expected in usual practice?

   2. Were these deviations from intended intervention unbalanced between groups and likely to have affected the outcome?

   3. Were important cointerventions balanced across intervention groups?

   4. Was the intervention implemented successfully for most participants?

   5. Did study participants adhere to the assigned intervention regimen?

   6. Was an appropriate analysis used to estimate the effect of starting and adhering to the intervention?

The risk of bias regarding adverse effect outcomes will be based on current guidelines (Chou 2010; CRD 2009; Higgins 2016; Loke 2011; NHIBI 2014; Viswanathan 2008b; Sterne 2016), and will mainly build on the McMaster Quality Assessment Scale of Harms (McHarm) for primary studies (Viswanathan 2008b). The McHarm tool comprises the following signalling questions.

1. Were the harms pre‐defined using standardised or precise definitions?

2. Were serious events precisely defined?

3. Were severe events precisely defined?

4. Were the number of deaths in each study group specified or were the reason(s) for not specifying them given?

5. Was the mode of harms collection specified as active?

6. Was the mode of harms collection specified as passive?

7. Did the study specify who collected the harms?

8. Did the study specify the training or background of who ascertained the harms?

9. Did the study specify the timing and frequency of collection of the harms?

10. Did the author(s) use a standard scale(s) or checklist(s) for harms collection?

11. Did the authors specify if the harms reported encompass all the events collected or a selected sample?

12. Was the number of participants that withdrew or lost to follow‐up specified for each study group?

13. Was the total number of participants affected by harms specified for each study arm?

14. Did the author(s) specify the number for each type of harmful event for each study group?

15. Did the author(s) specify the type of analysis undertaken for harms data?

* Some review author teams have included financial conflict of interest as a 'Risk of bias' domain (Jorgensen 2016). However, this element does not reflect an independent methodological domain and its inclusion is considered inappropriate according to Cochrane standards (Higgins 2017). Complying with Cochrane guidelines, we plan to extract data about studies' sources of funding for analysis and interpretation matters, but we will not include it in the tool itself. Specific 'Risk of bias' domains that are recognised as being influenced by financial conflicts of interest are included in the tool; for example, incomplete outcome data and selective outcome reporting. The domain 'selective outcome reporting' is also considered when evaluating the quality of the evidence, which we are also planning to perform (see 'Summary of findings' table under Data synthesis). Other aspects that may be taken into consideration when interpreting data on financial conflict of interests are inappropriate comparator and selection of a sample not representative or healthier than the target population. This is not a restrictive list and other aspects that may be perceived as influencing the results due to vested interests will be fully and carefully evaluated.

Measures of treatment effect

Unit of analysis issues

Dealing with missing data

We will base the analysis on intention‐to‐treat (ITT) data from the individual clinical trials accounting for dropout data. We will access the studies’ registries, when available, and contact authors to obtain complete information regarding dropouts, withdrawals and other missing data not fully reported in the reports of included studies. In the event that study authors do not — or can not — provide missing data, we will take this into consideration in the 'Risk of bias' analysis.

Assessment of heterogeneity

We will avoid excessive methodological heterogeneity by combining data, whenever appropriate, among studies with similar designs.

We will assess statistical heterogeneity between studies using the I² statistic for quantification of variability. We will also use the Chi² test by a setting statistical significance at a level of 0.10 to improve power. We will pursue a Sensitivity analysis whenever a result demonstrates significant heterogeneity (P value less than 0.10 or an I² statistic greater than 50%) (Deeks 2017).

We will investigate clinical heterogeneity by subgroup analysis (see Subgroup analysis and investigation of heterogeneity).

Assessment of reporting biases

We will use funnel plots to investigate the presence of publication bias (the selective publication of trials with positive findings), and other small‐study effects, among the studies included in the review (Sterne 2011). We plan to use Egger's test to assess for funnel plot asymmetry (Egger 1997), providing 10 or more trials are included in a meta‐analysis; a funnel plot with fewer studies would not have the power to distinguish chance from real asymmetry. We will consult a statistician in situations where we are unable to interpret the asymmetries objectively and when we may consider alternative statistical tests (Sterne 2011).

Data synthesis

Whenever appropriate, we will combine the beneficial and harmful effects through a meta‐analysis using the generic inverse variance technique. The inverse variance method is a “common and simple version of the meta‐analysis” and is the method implemented in the software where Cochrane Reviews are developed (Deeks 2017). We will perform a meta‐analysis using a random‐effects model in all cases where a meta‐analysis is possible. Where we detect considerable heterogeneity (I² statistic greater than 75%), particularly in the presence of high inconsistency in the direction of effect, we will not calculate the average effect of the intervention through a meta‐analysis (Deeks 2017).

Subgroup analysis and investigation of heterogeneity

We plan to explore potential sources of heterogeneity if available data from the studies allow us to stratify patient subgroups by the following characteristics.

1. Age of participants (trials with participants aged 19 to 35 years, 36 to 54 years or aged 55 years or more)

2. Sex (female versus male)

3. Dosage of methylphenidate (low dose (30 mg or less) versus high dose (more than 30 mg))

4. Multimorbidities (participants with multimorbidity versus participants without multimorbidity)

5. Type of clinical scales used in diagnosis

6. Duration of treatment (short‐term trials (six months or less) versus long‐term trials (more than six months))

7. Subtype of ADHD (predominantly inattentive type, or hyperactive or impulsive type, or combined type)

We will calculate a pooled effect size for each subgroup.

Sensitivity analysis

We will repeat the analyses using a fixed‐effect model to evaluate the impact of the choice of the model and the different study features in results of the meta‐analysis.

If there is an adequate number of studies (two or more), we will perform sensitivity analyses to assess the robustness of our results to decisions made in the development of the review. Specifically, we will reanalyse the data excluding:

1. studies that are judged to be at high risk of selection bias, performance bias, detection bias or reporting bias;

2. studies in which more than 20% of participants were lost to follow‐up;

3. studies with different designs (e.g. parallel versus cross‐over trial); and

4. unpublished versus published studies.