All randomised and quasi‐randomised controlled trials directly comparing topical latanoprost and brimonidine in the treatment of OAG (primary or secondary), OHT or NTG as defined by the investigators were included. Studies needed to have measured efficacy, tolerability or both in humans. Comparisons between combinations of latanoprost and other anti‐glaucomatous agent(s) and brimonidine with the same anti‐glaucomatous agent(s) were accepted. There were no age or sex limitations. Trials with treatment duration <1 month for either intervention were excluded.
We comprehensively searched Medline via Ovid (1966—March week 2, 2006), Embase via Embase.com (1980—week 11, 2006), the Cochrane Central Register of Controlled Trials in the Cochrane Library (CENTRAL, Issue 1, 2006) and Scientific Citation Index Expanded (1945—March 2006; appendix A). The strategy included populations (OAG, OHT or NTG), interventions (latanoprost and brimonidine) and publication type (randomised or quasi‐randomised controlled trials). In addition, Current Controlled Trials, ClinicalTrials.gov, CenterWatch and the United Kingdom National Research Register were searched. There were no limitations on language, date or publication status.
References of included publications were reviewed until no further relevant studies were found. Authors were contacted to clarify duplications and trial methods, and to identify further relevant trials. When duplication was confirmed, only the most complete trial was included.
Two reviewers (ATF and SER) independently screened combined search results to determine trial eligibility and extract data on to a standardised form. Authors of trial, sample size, location, design, interventions, patient characteristics, baseline and endpoint values, trial quality (allocation concealment, blinding, measurement bias, completeness of follow‐up and intention‐to‐treat analysis) and adverse events were recorded. Disagreements were resolved by discussion or consensus involving a third reviewer (JCC) when required.
For efficacy, the mean IOP reduction (IOPR) from baseline to end point was determined. Daily mean values were analysed, given the tendency for IOP to fluctuate throughout the 24‐h cycle.17
For three trials,9,11,12
IOP at its peak effect after drug administration was used because daily mean values were not reported.
For tolerability, adverse events were analysed in the following subgroups: itch/discomfort, hyperemia, eyelid disorder, visual disturbance, conjunctival disorder, keratopathy, dry eye, hypertrichosis, increased iris pigmentation, fatigue and headache.
Quantitative data synthesis and analysis
Extracted data were pooled for summary estimates using Review Manager 4.2.7 (The Cochrane Collaboration). Continuous outcomes were expressed as weighted mean difference (WMD), with values >0 favouring latanoprost, and dichotomous outcomes as relative risk (RR), with values <1 favouring latanoprost. Both outcomes were reported with 95% confidence intervals.
For studies that only reported absolute values for IOP at baseline and end point, the IOPR and standard deviation (SD) of the IOPR (SDIOPR) were calculated as follows:
SDIOPR=√(SDbaseline2+SDend point2−2ρSDbaselineSDend point)
and was calculated from trials with known SDIOPR.
For studies that only reported standard errors (SEs), SD was calculated by the formula SD
SE×√n. Six trials contained at least a component of crossover design.11,12,17,18,19,20,21
For these trials, only the initial parallel phase was analysed. To minimise unit‐of‐analysis error, the sample size analysed in each arm was the true sample size divided by two.
For continuous data (IOP), the sample size was based on intention‐to‐treat analysis or last observation carried forward only if the authors of the trial clearly stated that this was the process used. Otherwise, available case analysis was used. For dichotomous data (adverse events), available case analysis was undertaken irrespective of how the original trialists explored the data to avoid imputing data.
Intertrial statistical heterogeneity was explored using the Cochran Q test, with calculated I2
indicating the percentage of the total variability in effect estimates among trials that is due to heterogeneity rather than chance.22
Results for efficacy and tolerability were calculated using a random effects model. This assumes that each study estimates different but related treatment effects, and is more conservative than a fixed effects model in the presence of heterogeneity.22
Subgroup analyses were determined a priori and included duration (<6 months,
6 months), study type (parallel v
crossover), peak versus trough IOP readings, glaucoma type (OAG/OHT v
NTG) and monotherapy versus adjunctive therapy. For trials in which end points were reported at more than one duration, only the longest duration was analysed.
Regression models exploring heterogeneity were performed with SAS statistical software V.8. A priori determined covariates included trial duration analysed, trial design (parallel v crossover), trial quality (allocation concealment) and treatment (monotherapy v adjunctive therapy). Publication bias was assessed by visually inspecting a funnel plot.