Types of studies

We will include only RCTs in this review.

Types of participants

We will include RCTs that enrolled adult participants (18 years or older) with primary open angle glaucoma (POAG) or ocular hypertension (OHT). We will not include RCTs that evaluated only participants with a secondary form of glaucoma, such as pigmentary or pseudoexfoliative glaucoma.

Monotherapy is the typical first‐line treatment, but participants at an advanced stage of disease when first diagnosed may require more aggressive therapy (e.g., combination therapy) as the first treatment. Therefore, we will include RCTs in which participants were receiving some form of drug treatment at the time of study enrollment, as well as RCTS in which participants had never been treated. We will not restrict any studies with regard to gender, ethnicity or comorbidity. We expected that POAG and OHT definition criteria vary considerably across different studies, with various combinations of diagnostic criteria for intraocular pressure (IOP), visual field (VF) and optic nerve head (ONH) damages. We will accept any POAG and OHT definitions as used in the RCTs. We will accept any stage of the disease, with no restriction in any direction (very early stages as well as more advanced stages).

Types of interventions

We will consider RCTs that compared the following FC therapies and monotherapies in any combination for the intent of lowering IOP. We will include multi‐arm RCTs. All included interventions are considered to be "jointly randomizable", satisfying the assumption for network meta‐analysis (see more details under "Evaluating the assumption for network meta‐analysis").

FC therapies:

• Pilocarpine plus epinephrine or betaxolol

• Timolol plus a carbonic anhydrase inhibitor (dorzolamide or brinzolamide)

• Timolol plus a prostaglandin analog (bimatoprost, travoprost, or latanoprost)

• Timolol plus brimonidine

• Brinzolamide plus brimonidine

• Timolol plus dorzolamide and brimonidine

Monotherapies:

• Parasympathomimetics (e.g., pilocarpine)

• β‐blockers (e.g., betaxolol, carteolol, levobunolol, timolol)

• Carbonic anhydrase inhibitor (e.g., brinzolamide)

• α‐2 adrenergic agonist (e.g., brimonidine)

• Prostaglandin analogs (e.g., bimatoprost, travoprost, latanoprost)

Types of outcome measures

Electronic searches

We will search the Cochrane Register for Controlled Trials (CENTRAL) (which contains the Cochrane Eye and Vision Group Trials Register) (latest issue), Ovid MEDLINE, Ovid MEDLINE In‐Process and Other Non‐Indexed Citations, Ovid MEDLINE Daily, Ovid OLDMEDLINE (January 1946 to present), EMBASE (January 1980 to present), PubMed (1948 to present), the metaRegister of Controlled Trials (mRCT) (www.controlled‐trials.com), ClinicalTrials.gov (www.clinicaltrials.gov) and the World Health Organization (WHO) International Clinical Trials Registry Platform (ICTRP) (www.who.int/ictrp/search/en). We will not use any date or language restrictions in the electronic search for trials.

See: Appendices for details of search strategies for CENTRAL (Appendix 1), MEDLINE (Appendix 2), EMBASE (Appendix 3), PubMed (Appendix 4), mRCT (Appendix 5), ClinicalTrials.gov (Appendix 6) and the ICTRP (Appendix 7).

Searching other resources

We will search the references of included studies to identify additional studies that could be included. We will search the FDA and European Medicine Agency websites for RCTs submitted for regulatory approval.

Selection of studies

Two review authors independently will assess the titles and abstracts of all studies identified by the electronic and manual searches. We will label each record at this stage as "definitely relevant," "possibly relevant" or "definitely not relevant." We will exclude records labeled as "definitely not relevant" by both review authors. We will retrieve full‐text reports of records labeled as "definitely relevant" or "possibly relevant" and two review authors independently will assess whether they meet the inclusion criteria. We will resolve any disagreement when present at any stage through discussion. When necessary, we will consult a third review author or contact the study investigators for more information to determine eligibility.

Data extraction and management

Two review authors independently will extract data related to study design and methods, participant characteristics, interventions, and outcomes using an electronic form developed and maintained in the Systematic Review Data Repository (SRDR, srdr.ahrq.gov). We will compare data extracted by the two review authors and resolve discrepancies by discussion or, when necessary, by consulting a third review author. One review author will enter data into RevMan 5.3 (Revman 2014) and a second review author will verify the data entered.

Assessment of risk of bias in included studies

Two review authors will evaluate the following domains for risk of bias: sequence generation and allocation sequence concealment (selection bias), masking of participants (performance bias) and outcome assessors (detection bias), incomplete outcome data (attrition bias), selective outcome reporting (reporting bias), and other potential sources of bias. We will evaluate each study and assign for each 'Risk of bias' domain a judgment of "low risk of bias," "high risk of bias" or "unclear risk of bias" following criteria outlined in Chapter 8 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). We will resolve discrepancies by discussion or, when necessary, by consulting a third review author or by contacting the study investigators for more information. If we do not receive a response from study investigators within six weeks, we will use the information available in published reports to judge the risk of bias.

Measures of treatment effect

We will calculate mean differences with 95% confidence intervals (CIs) for continuous outcomes, including the primary outcome and secondary outcomes #1 to 4. For dichotomous outcomes (ONH/RNFL damage progression evidence and adverse events), we will use summary risk ratios with 95% CIs. We will generate probabilities of each treatment taking each possible rank and the cumulative ranking probabilities, known as "sucra" plots (Chaimani 2013; Salanti 2011). Sucra plots show the relative probability of each intervention being among the best treatment options.

Unit of analysis issues

We expect that a majority of trials will have randomized either individual participants or one eye per participant. In this case, the unit of analysis will be individual participant. When two eyes of the same participant have been randomized independently, we will use appropriate estimates from the trial report that accounted for inter‐eye correlation of measurements and outcomes between the two eyes of each participant. Because we plan to conduct network meta‐analyses, we will account for the correlation among arms within multi‐arm trials (see details under Data synthesis).

Dealing with missing data

We will contact study investigators in an attempt to obtain missing data. We will use available information when we do not receive an answer within six weeks. We will not impute missing data.

Assessment of heterogeneity

We will evaluate clinical, methodological and statistical heterogeneity. We will assess participant characteristics, interventions and outcomes as well as study design and risk of bias, to assess clinical and methodological heterogeneity. We will use the I² and tau² values, and Q statistics test to quantify the statistical heterogeneity. We will consider I² > 75% as the cut‐off point to identify the presence of considerable heterogeneity.

Evaluating the assumption for network meta‐analysis

The consistency assumption of network meta‐analysis implies that participants included in the network could hypothetically be randomized to any of the interventions represented in the network (Salanti 2012). Interventions defined in our review are comparable in that any of them could theoretically be chosen as a treatment modality for POAG/OHT participants who were receiving some form of drug treatment or presented with advanced stage of disease at the time of study enrollment. We will use our qualitative assessment (i.e., evaluation of participant characteristics, interventions, outcomes, and design characteristics) to inform the assessment of consistency assumption. We will compare the distributions of characteristics that may modify the treatment effect.

We will further evaluate statistical disagreement of direct and indirect evidence, known as inconsistency, by a loop‐specific approach and by fitting design‐by‐treatment interaction models (Dias 2013; Jackson 2014; White 2011; White 2012). We will fit an inconsistency network meta‐analysis model when statistical inconsistency is present and cannot be explained.

We will conduct network meta‐analysis using a suite of "network" commands in STATA 13® (Stata 2013).

Assessment of reporting biases

When a sufficient number of studies is included (10 or more), we will assess the evidence of reporting bias by examining asymmetry in the funnel plot of effect estimates. We will check whether the trials submitted to the FDA have been published; if so, we will compare outcome data between the FDA documents and the study publications.

Data synthesis

When no substantial clinical and methodological heterogeneity is present (as defined above in Assessment of heterogeneity), we first will conduct pair‐wise meta‐analyses for all direct comparisons using a random effects model. We will assume comparison‐specific statistical heterogeneity (i.e., each pair‐wise meta‐analysis will have a separate estimate of statistical heterogeneity) as well as a common heterogeneity across all comparisons (i.e., all pair‐wise meta‐analyses will have a common estimate of statistical heterogeneity). We then will fit network meta‐analysis models using the multivariate meta‐regression approach proposed by White et al (Jackson 2014; White 2011; White 2012). This model also handles multi‐arm trials. We will report direct estimates of treatment effects and estimates from the network meta‐analysis.

Sensitivity analysis

We will conduct a sensitivity analysis to evaluate the effect of excluding studies judged as having a high risk of bias on allocation concealment. We also will conduct a sensitivity analysis excluding studies sponsored by drug companies.