This post hoc pooled analysis integrated and reanalyzed the source data from three randomized clinical trials (two phase III studies and a phase IV study) and two international real-world evidence observational studies that compared preservative-containing latanoprost (PCL) and preservative-free latanoprost (Monoprost®; PFL) (Table 1) [9, 11, 13, 14].
The source data from each of the five studies were acquired and reanalyzed in a new database. Each study was approved by the ethics review board in each country and adhered to the tenets of the Declaration of Helsinki and all relevant local regulations. As this is a review of data from previous studies, online registration as a clinical trial was not required.
The primary objective was tolerance by the evaluation of conjunctival hyperemia with each topical treatment. Most of the studies included here graded conjunctival hyperemia with the McMonnies/Chapman-Davies (MC-D) scale and one study used the Efron scale . To standardize the grading of conjunctival hyperemia, the Efron scale was converted to the MC-D scale as previously described and scored as absent (MC-D grade 0 = grade 0 Efron), mild (MC-D grade 1 or 2 = grade 1 Efron), moderate (MC-D grade 3 = grade 2 Efron), and severe (MC-D grade 4 or 5 = grade 3 or 4 Efron) .
The secondary criteria for the evaluation of tolerance included a change in signs and symptoms of OSD with a reduction indicating an improvement in tolerability and increased OSD indicating lower tolerability. Tolerability was based on a composite OSD score integrating ocular signs and symptoms between instillation of drops, patient and investigator satisfaction, and ocular and systemic adverse events. For standardization of ocular signs and symptoms, a five-parameter OSD composite score (0–100) was calculated from weighted parameters: the five variables assessed for this composite score were weighted as follows: eyelid redness (15), eyelid swelling (15), corneal staining (25), conjunctival staining (25), and tear film breakup time (TBUT) (20) (Table 2).
The weighting assigned to each of the ocular signs for the calculation of this score was defined by the expert authors. Other secondary objectives compared the IOP-lowering efficacy between medications and ocular signs and symptoms. Ocular signs and symptoms included eye dryness/tearing/foreign body sensation, irritation/stinging/burning, itching/sticky eye sensation, itching, red eye/lid redness, blurred vision, pain or discomfort, lid swelling, fluorescein staining (corneal and conjunctival), and TBUT.
Patient and investigator satisfaction with treatment were graded as “very unsatisfied”, “unsatisfied”, “satisfied”, and “very satisfied”.
This pooled study was based on individual patient data (IPD) from the five studies (two phase III, one phase IV, and two real-world evidence studies). These studies were selected owing to common inclusion criteria: all patients included were diagnosed with glaucoma and treated with PL or PFL and there was complete information on conjunctival hyperemia and IOP. All relevant data from the individual datasets were combined into one dataset.
Data Collection and Endpoints
Data were collected on patient demographics including gender and age (Table 1), IOP for efficacy; conjunctival hyperemia and ocular signs for tolerability, patient satisfaction, and adverse events. For each individual study, all the data were collected and verified (Table 2).
Patients who did not meet inclusion and non-inclusion criteria were excluded from the analysis. Inclusion of IPD in the final database was verified by the investigator and the statistician. Improvement of conjunctival hyperemia was defined as a decrease in the grade of conjunctival hyperemia between baseline (day 0) and day 84 after initiating treatment.
Analysis and Statistics
Descriptive statistics were provided depending on the nature of the considered criterion: number of observed (and missing, if any) values, mean, standard deviation, median, first and third quartiles, as well as minimum and maximum, for quantitative data; number of observed (and missing, if any) values, number and percentage of patients for qualitative data. All data were presented as overall values according to the studies.
A pooled analysis of IPD was performed to analyze primary and secondary endpoints. Randomized clinical trials and observational studies were combined to produce this pooled analysis. Shrier et al. have demonstrated that combining data from randomized clinical trials and non-randomized clinical trials produces more relevant estimators, reducing bias and mitigating the limitations of each type of study .
Evaluation of the one-step and two-step approach yielded very similar results, indicating that the analysis was robust with either approach; hence, we elected to go with the two-step analysis. A two-step IPD pooled analysis approach was chosen as this approach allows the use of forest plots to estimate odds ratio (OR) and 95% confidence intervals (CIs) and to illustrate heterogeneity and individual and pooled results. The I2 statistic was used to determine heterogeneity, which measures inconsistency (percentage of total variation across studies due to heterogeneity) of effects. As a result of the different study designs (e.g., crossover, parallel group, randomized, naive patients, and switching between different medications), heterogeneity between studies was examined by checking the results of the I2 statistic both for primary and secondary endpoint analyses. If heterogeneity was evident, it was incorporated using random effects modelling.
Two-Step IPD Pooled Analysis
The first step of the two-step IPD pooled analysis approach consists of analyzing the studies separately and individually. For each study, a logistic regression model was performed to provide estimates, variances, and covariances of the mean difference between treatments (including OR and 95% CIs). Logistic regression was used with the presence or absence of improvement (for both endpoints) as the binary dependent variable and treatment as a categorical independent variable.
The second step of the two-step approach combines the study estimates. The variances and covariances obtained in the first step are used as the variance and covariance of the residual errors. All studies were weighted on the inverse of the variance (within- and between-studies variances) to obtain the results of the pooled analysis. For the IPD pooled analysis on the primary and secondary endpoints, statistical analysis was performed using SAS® 9.4 software (SAS Institute Inc., Cary, NC, USA).