Page 142 - DUOKOPT BIBLIOBOOK

DUOKOPT - bibliography book - CONFIDENTIAL - document for exclusive use by personnel of Laboratoires Théa – DO NOT DISTRIBUTE
®

Acta Ophthalmologica 2010

Table 1. Summary of criteria for visual field progression. Kolmogorov–Smirnov test. If data
were normally distributed, a two-
Appearance of a new defect, defined as follows:
Cluster of three or more non-edge points, each of which decreases from baseline at p < 0.05 in tailed unpaired Student’s t-test was
three consecutive fields used to compare means between treat-
Cluster of two or more non-edge points, each of which decreases from baseline at p < 0.02 in ment groups for quantitative variables
three consecutive fields at baseline. Baseline comparisons used
Cluster of one or more non-edge points, each of which decreases from baseline at p < 0.01 in a generalized estimating equation
three consecutive fields (GEE) (Katz et al. 1994) with patient
Deepening of an existing defect, defined as follows: intraclass correlation.
Cluster of two or more points at p < 0.05 decreasing to p < 0.02 A two-way analysis of variance
Cluster of one or more points at p < 0.05 decreasing to p < 0.01
Cluster of two or more points at p < 0.02 decreasing to p < 0.01 (anova) was used to evaluate the effect
Expansion of an existing defect to contiguous previously normal area: of treatments and the time factor.
Cluster of three or more non-edge points, each of which decreases from baseline at p < 0.05 in When a normal distribution was not
three consecutive fields expected (time of follow-up), the non-
Cluster of two or more non-edge points, each of which decreases from baseline at p < 0.02 in parametric Kruskal–Wallis one-way
three consecutive fields anova was applied.
Cluster of one or more non-edge points, each of which decreases from baseline at p < 0.01 in Categorical variables were com-
three consecutive fields
pared using chi-square test and Fish-
er’s exact test, as required.
A conditional Cox hazard model
(screening, post-screening and baseline gression if a significant slope was with patient intraclass correlation was
visits), and all subsequent examina- detected in the same point in three used to estimate and test factors for
tions. Progression was considered to consecutive fields (similarly to the their association with progression in
have occurred if an existing scotoma event-based analysis). The criterion VF damage (event-based method).
deepened or expanded, or a new sco- for VF defect progression was a con- A conditional hazard model was used
toma appeared. Details on progression firmed progression in three or more for univariate and multivariate analy-
criteria have been previously reported test points (Vesti et al. 2003). sis. A backward strategy was adopted,
(Martinez & Sanchez 2005, 2008a). with a statistically significant cut-off
Progression criteria are summarized in Statistical analysis for variable screening of 0.05.
Table 1. If progression was detected, Visual field progression-free survival
two additional reliable tests (24-2 full- Sample size calculation was based on rates were plotted for treatment
threshold strategy) were performed the assumption that a 20% difference groups using Kaplan–Meier analysis
within 1 month to confirm VF impair- in VF progression between the study and were compared using a log-rank
ment. If all three tests confirmed VF groups was clinically relevant (two- test.
impairment, the eye was considered to tailed test). Potential predictors were catego-
have progressed and was dropped Approximately 64 patients were rized as having low or high values
from the study. required, given an a of 0.05 and a 1)b according to the median in the
of 0.70. A statistical power of 70% sample.
was chosen to decrease the risk of Factors associated with progression
Trend analyses of threshold sensitivity
a false negative result. A minimum in the univariate analysis at p £ 0.2
stata 9.0 (StataCorp LP, College Sta- recruitment of 80 patients was were included in the multivariate anal-
tion, TX, USA) was used to perform planned to allow for patient with- ysis.
point-wise liner regression. Point-wise drawals, and 80 and 81 patients were Kappa statistics were used to esti-
linear regression analysis (PLRA) was initially enrolled in the DT and BT mate agreement of the methods at
performed separately for each of the treatment groups, respectively. 5 years. Additionally, in order to make
52 test points of the 24-2 field area. Demographic and clinical variables a comparison between the event- and
The analysis was performed at each (age, sex, systolic, diastolic and mean trend-based methods, we classified eyes
visit, starting from the first follow-up arterial pressure, and use of antihyper- according to progression of VF defects
visit when the first four fields were tensive treatments) were based on the after 5 years of follow-up. This infor-
included in the calculation (screening, data collected from the 146 patients mation was used to plot Venn
post-screening, baseline and first fol- enrolled and retained in the study. diagrams (Chow & Ruskey 2004).
low-up visits), because regression anal- Ocular risk factors were based on Statistical analyses were performed
ysis is not likely to detect a trend analyses of 215 eyes in 146 patients. using stata 9.0 (StataCorp LP).
when fewer data are available (Hol- The sample for analysis included all
min & Krakau 1982; Fitzke et al. patients who received study medica- Results
1996). Test points with a significant tions and had at least a valid visit at
regression slope (p < 0.01) showing a month 12. Descriptive statistics (mean Of the 182 patients who were
sensitivity loss ‡ 1.0 dB⁄ year were [standard deviation]) and 95% confi- screened, 161 fulfilled the respective
identified for each follow-up visit dence intervals (95% CIs) were used demands of the inclusion and exclu-
(Fitzke et al. 1996; Vesti et al. 2003; for demographic, clinical and ocular sion criteria. Fifteen of these were lost
Nouri-Mahdavi et al. 2007). A point baseline characteristics. Data were to follow-up: six for personal reasons
was considered to have confirmed pro- tested for normal distribution using a (two in the DT and four in the BT

544
142