Predictive Factors for Glaucomatous Progression in the Advanced Intervention Study

Kouros Nouri-Mahdavi, MD,1 Douglas Hoffman, BA,1 Anne L. Coleman, MD, PhD,1 Gang Liu, MS,2 Gang Li, PhD,2 Douglas Gaasterland, MD,3 Joseph Caprioli, MD1

Purpose: To investigate the risk factors associated with visual field (VF) progression in the Advanced Glaucoma Intervention Study (AGIS) with pointwise linear regression (PLR) analysis of serial VFs. Design: Prospective, multicenter, randomized clinical trial. Participants: Five hundred nine of 401 patients from the AGIS with a baseline VF score of Յ16, Ն7VF examinations, and Ն3 years of follow-up were selected. Main Outcome Measure: Visual field progression. Methods: This is a cohort study of patients enrolled in a prospective randomized clinical trial (AGIS). Worsening of a test location on PLR analysis was defined as a change of threshold sensitivity of Ն1.00 decibels a year, with PՅ0.01. Visual field progression was defined as worsening of at least 2 test locations within a Glaucoma Hemifield Test cluster with PLR analysis. Multivariate logistic regression was used to determine risk factors associated with VF worsening. Intraocular pressure (IOP) fluctuation was defined as standard deviation of the IOP at all visits after the initial surgery. Results: The mean (Ϯ standard deviation) follow-up time and baseline AGIS score were 7.4 (Ϯ1.7) years and 7.7 (Ϯ4.4), respectively. Visual field progression was detected with PLR analysis in 151 eyes (30%). Older age at the initial intervention (P ϭ 0.0012; odds ratio [OR], 1.30; 95% confidence interval [CI], 1.11–1.50), larger IOP fluctuation (P ϭ 0.0013; OR, 1.31; 95% CI, 1.12–1.54), increasing number of glaucoma interventions (P ϭ 0.01; OR, 1.74; 95% CI, 1.14–2.64), and longer follow-up (P ϭ 0.02; OR, 1.19; 95% CI, 1.03–1.38) were associated with increased odds of VF progression. When regression analyses were repeated in eyes with and without a history of cataract extraction, IOP fluctuation was the only variable to be consistently associated with VF progression. Conclusion: Both increasing age and greater IOP fluctuation increase the odds of VF progression by 30% (for each 5-year increment in age and 1-mmHg increase in IOP fluctuation). The higher risk conferred by IOP fluctuation was consistently observed in eyes with and without a history of cataract extraction. 2004;111:1627–1635 © 2004 by the American Academy of Ophthalmology.

Over the last 2 decades a number of studies have addressed the issue of risk factors associated with or predictive for glaucoma progression.1–21 A better understanding of clini- Originally received: January 7, 2004. Accepted: February 9, 2004. Manuscript no. 240023. cal risk factors for worsening of glaucoma may help us 1 Glaucoma Division, Jules Stein Institute, University of California develop new strategies to improve glaucoma care. A major Los Angeles, Los Angeles, California. obstacle has been the lack of a uniformly accepted and 2 Department of Biostatistics, UCLA School of Public Health, Los Ange- efficient approach to detect glaucoma progression. Evalua- les, California. tion of visual field (VF) series remains the clinical method 3 University Ophthalmology Consultants of Washington, Washington, DC. most frequently used to assess the course of glaucoma and Presented at: American Academy of Ophthalmology Annual Meeting, the efficacy of its treatment. November, 2003; Anaheim, California. The Advanced Glaucoma Intervention Study (AGIS) Supported by an unrestricted grant from Research to Prevent Blindness, used a single method for longitudinal evaluation of VFs. New York, New York, and the National Institutes of Health, Bethesda, Due to the lack of a gold standard and only fair concordance Maryland (grant no.: R01 EY12738). among various analytic methods, this could cause important The authors do not have any commercial or proprietary interest in any of associations to be missed or spurious ones to be detected. the products or companies cited in the article. Likewise, they have no We compared the results of an independent approach, point- financial interest in and have not received payment as a consultant, re- viewer, or evaluator from any of the companies mentioned. wise linear regression (PLR) analysis, with the AGIS 22 Correspondence to Joseph Caprioli, MD, Glaucoma Division, Jules Stein method in a subgroup of patients from that study. The 2 Eye Institute, 100 Stein Plaza, Los Angeles, CA 90095. E-mail: methods were concordant in two thirds of study eyes. [email protected]. The primary objective of this report is to identify the risk

© 2004 by the American Academy of Ophthalmology ISSN 0161-6420/04/$–see front matter 1627 Published by Elsevier Inc. doi:10.1016/j.ophtha.2004.02.017 Ophthalmology Volume 111, Number 9, September 2004 factors associated with progression of VF damage in the omitting the last threshold in a series and (2) after deleting the AGIS with PLR analysis to evaluate sequential VFs for threshold before last for the same series. This approach has been clinically and statistically significant change. shown, in simulation experiments, to be more specific than using all the data points for a single regression analysis, and it maintains a sensitivity comparable to other stringent algorithms used for the same purpose, such as two of two5 or three of four.4,6,7 Regression Materials and Methods slopes were considered statistically and clinically significant if Ն1.00 decibels (dB)/year or ՅϪ1.0 dB/year in the presence of The AGIS design and methods are described in detail elsewhere PՅ0.01. and are summarized here.2,23 Thirty-five- to 80-year-old phakic For evaluation of VF series, we used the most rigorous and patients with open-angle glaucoma no longer controlled by max- clinically relevant set of criteria explored in the aforementioned imally tolerated medical treatment were recruited. Eligible eyes investigation, the 2-point Glaucoma Hemifield Test change crite- had a best-corrected visual acuity (VA) score of at least 56 letters rion. According to this, a VF series is considered to be changing if (Early Treatment Diabetic Retinopathy Study) and met specified 2 test locations belonging to the same Glaucoma Hemifield Test criteria for combinations of consistently elevated intraocular pres- cluster demonstrate change in the same direction. This set of sure (IOP), glaucomatous VF defect, and optic disc rim deteriora- criteria was found to be the most conservative among different tion.2 Between 1988 and 1992, investigators at 12 participating PLR approaches. It yielded the smallest number of progressing VF AGIS clinical centers enrolled 789 eyes of 591 patients. Eyes were series, minimized the number of improving VF series, and dem- randomly assigned to 1 of 2 surgical intervention sequences: argon onstrated the highest agreement with AGIS criteria. laser (ALT)––trabeculectomy Visual field progression according to AGIS criteria was defined (ATT) or trabeculectomy–ALT–trabeculectomy (TAT). Data in as the first occurrence in an eye, at 3 consecutive 6-month visits, this report are based on a database closure of March 31, 2001. The of worsening in the VF defect score of Ն4 from the baseline value. institutional review boards at each of the participating centers Changes in the AGIS VF defect score were measured from prein- approved the AGIS protocol, and all patients gave informed con- tervention reference values. sent. Visual field outcomes from PLR were classified as progressing Visual field tests were conducted with a Humphrey Visual or nonprogressing. Improving and stable eyes were categorized as Field Analyzer I (Carl Zeiss Ophthalmic Systems, Inc., Dublin, nonprogressing. Associations between VF progression and various CA) set for the central 24-2 threshold test, size III white stimulus, preoperative and postoperative potential risk factors were evalu- and full threshold strategy, with the foveal threshold test turned on. ated with multivariate logistic regression.25 The 24-2 program of the Humphrey Visual Field Analyzer records It has been shown that both eyes of the same patient are at least data from 55 locations in the VF, all of which, except the locations partially correlated with respect to progression of the VF.8 We above and below the blind spot and the foveal threshold, are used used a generalized linear mixed model to account for the intereye for calculating the AGIS VF score.3 Scoring is based on the correlation. The generalized linear mixed model is a commonly number, pattern, and depth of depression of threshold sensitivities used random-effects model that permits the data to exhibit corre- as found in the Humphrey total deviation plot. Points are awarded lation and nonconstant variability and allows the response to come for the presence of a nasal defect (a cluster of Ն3 depressed from several distributions such as binomial, Poisson, and ␥. locations in the nasal field), nasal step (Ն1 depressed locations in Preoperative and postoperative factors that were associated the nasal field, in the absence of depression in any of the 3 with VF progression in univariate analyses (␹2 test, unpaired t test, locations on the opposite side of the horizontal midline), and or Wilcoxon rank sum test, depending on the type of data) at a P hemifield defect (cluster of Ն3 depressed sites in a hemifield). The value of Յ0.20 were included in the final model. In addition, we translation of an array of VF thresholds into a single number included all the variables that might potentially predict or con- simplifies the comparison of test results and the determination of found detection of VF progression from a clinical point of view progression or stability. Visual field defect scores ranged from 0 (Table 1). Mean IOP was calculated by averaging all the available (no defect) to 20 (advanced glaucoma). Visual field measurements IOPs starting at 3 and 6 months after the initial intervention and were made at baseline, 3 months after initial intervention, and at every 6 months thereafter. Standard deviation of the IOP at all each 6-month follow-up examination. Baseline or reference mea- visits after the initial surgery was used as a surrogate for IOP surements were performed after the eligibility measurements but fluctuation. The cutoff point for classification by logistic regres- before the first surgical intervention. sion was set at 0.50. Any variable with a P value of Յ0.05 was From the original pool of the recruited patients (789 eyes of considered statistically significant. 591 patients), 509 eyes of 401 patients meeting the following criteria were selected for this study: ● a reference VF score of Յ16 Results ● at least 3 years of follow-up ● a minimum of 7 VF examinations with a reliability score of A total of 151 eyes (29.9%) progressed according to the PLR Յ2. criteria, whereas 138 eyes (27.1%) showed progression based on AGIS criteria (␬ ϭ 0.30, percentage agreement ϭ 64%). The Statistical Methods characteristics of the study sample are presented in Table 2 based on occurrence of progression according to PLR criteria. Four eyes SPSS statistical software24 was used to perform PLR analysis. Our were considered indeterminate based on our PLR criteria and were methodology is described in detail elsewhere.22 We used the excluded from further analysis. Table 3 describes the results of the two-omitting regression algorithm recently described by Gardiner multivariate logistic regression. Four variables were associated and Crabb for definition of change versus stability at each point.4 with a higher probability of VF progression in decreasing magni- In summary, in this technique a test location is considered to be tude: older age at the time of first intervention, greater IOP progressing or improving during the follow-up period only if the fluctuation, increasing number of glaucoma interventions, and regression slope is statistically and clinically significant (as defined longer follow-up. Additionally, 2 other variables, male gender and below) in both of the following regression analyses: (1) after lower baseline IOP, demonstrated a possible association with VF

1628 Nouri-Mahdavi et al Predictive Factors for Visual Field Progression in the AGIS

Table 1. Final Independent Variables Explored in Logistic be associated with progression of VFs. Figure 1 shows the change Regression Models of AGIS score over time according to IOP fluctuation. Eyes with an IOP fluctuation of Ͻ3 mmHg remained stable over the course Numeric of follow-up (PϾ0.05), whereas eyes with an IOP fluctuation of Age Ն3 mmHg demonstrated significant progression (P ϭ 0.0006, Refractive error regression slope, 0.026/month). Baseline visual acuity score Baseline AGIS score Baseline IOP and no. of medications Average IOP during follow-up Discussion Average number of medications during follow-up IOP fluctuation over time (SD of IOP) The baseline characteristics of our subgroup of AGIS pa- Length of follow-up tients are similar to those of the original cohort of patients No. of glaucoma interventions described in previous AGIS reports.2 In this investigation, Categorical Gender we used rigorous and clinically relevant PLR criteria to Race define the main outcome measure, VF worsening. We de- Educational level tected VF progression in approximately 30% of the eyes, Presence or absence of diabetes which is consistent with the number of eyes showing pro- Presence or absence of systemic hypertension gression in the same cohort according to the AGIS criteria Intervention sequence Vertical cup/disc ratio (27%). We found that greater intervisit IOP fluctuation and older age at the time of first intervention were the most consistent predictors for VF progression. Each increased the AGIS ϭ Advanced Glaucoma Intervention Study; IOP ϭ intraocular odds for VF progression by approximately 30% for each pressure; SD ϭ standard deviation. 1-mmHg increase in IOP fluctuation and 5-year increment in age, respectively. The AGIS is a multicenter randomized clinical trial de- signed to evaluate best management strategies after maxi- progression (0.05ϽPϽ0.10). Excluding the variables related to glaucoma severity at baseline (AGIS VF score and cup/disc ratio mal effective medical treatment has failed to control IOP. A at baseline) from the logistic regression analysis to determine risk major outcome evaluated in the AGIS was the stability or factors resulted in the same 4 variables mentioned above being progression of VFs as related to IOP control after the initial 9 associated with increased odds of field progression. In addition, intervention. Eyes with IOPs of Ͻ18 mmHg at all visits male gender (P ϭ 0.0476, odds ratio [OR] ϭ 1.596) was also during the first 6 years of follow-up were least likely to associated with a statistically significant increase in the odds of VF show worsening. A lower mean IOP during the 18 months progression. after the initial intervention also predicted a better func- We repeated the same analyses on eyes belonging to the ATT tional outcome. The covariates for which the regression and TAT intervention sequences separately (Table 4). For the ATT models were corrected were age at randomization, gender, intervention sequence, higher number of glaucoma interventions, race, presence or absence of diabetes, intervention se- older age at first intervention, greater IOP fluctuation, longer follow-up, and male gender were significantly associated with VF quence, baseline IOP, and baseline VF score. progression, in that order. For the TAT sequence, 3 covariates To date, all AGIS reports have used the AGIS scoring showed a possible association with VF progression system for evaluating VF-related outcomes. The AGIS cri- (0.05ϽPϽ0.10): older age at first intervention (P ϭ 0.08), pres- teria for detection of VF progression are more conservative ence of diabetes (P ϭ 0.092), and greater IOP fluctuation (P ϭ than alternative methods such as Glaucoma Change Proba- 0.097). bility Maps (used in the Early Manifest Glaucoma Trial), Development and removal of cataracts are important confound- the Collaborative Initial Glaucoma Treatment Study crite- ing variables to consider. To address this issue, we repeated the ria,10,11 or PLR.12 Given the lack of a widely accepted logistic regression analysis in the 2 groups of eyes with or without external standard for evaluation of longitudinal field series, cataract surgery over the course of the study (Table 5). One it seems advisable to assess such data with various analyt- hundred ninety-seven eyes (38.7%) had cataract surgery after the first intervention. In the group of eyes that did not undergo cataract ical approaches. We have demonstrated, in a previous in- vestigation, that PLR agrees with the AGIS scoring system surgery (312 eyes), older age, greater IOP fluctuation, lower base- 22 line IOP, and longer follow-up were associated with VF worsen- in about two thirds of eyes. Although PLR has been ing. In the group of eyes that had had cataract surgery, male gender criticized because of the long follow-up needed before and greater IOP fluctuation were associated with VF progression. progression can be demonstrated,12–14 this is not a major As a comparison, we performed the same analyses using the drawback in this study, because all the patients had at least AGIS criteria for definition of VF outcomes. The following pre- 7 VFs and the mean follow-up was 7.7 years. Additionally, dictive factors were found on multivariate logistic regression: PLR has been proved to be highly specific in simulation lower baseline AGIS score (PϽ0.0001), greater IOP fluctuation 12 Ͻ Ͼ ϭ experiments. (P 0.0001), vertical cup/disc ratio 0.60 (P 0.001), greater In a complementary study, the AGIS investigators deter- number of glaucoma surgeries (P ϭ 0.008), longer follow-up (P ϭ ϭ mined the predictive variables for a sustained reduction of 0.013), and older age at first intervention (P 0.028). After 15 exclusion of baseline factors related to glaucoma severity (cup/disc the VF or VA after an initial intervention. About 30% of ratio and AGIS score at baseline), greater IOP fluctuation (P ϭ eyes had a sustained decrease of VF (SDVF), described as 0.0017) and increasing number of glaucoma interventions (P ϭ the first occurrence in an eye at 3 consecutive 6-month visits 0.0098) were the only statistically significant risk factors found to of either an increase in the VF defect score of Ն4 from the

1629 Ophthalmology Volume 111, Number 9, September 2004

Table 2. Characteristics of the Study Sample in Stable and Progressing Eyes with the 2-Point Glaucoma Hemifield Test Change Used as the Criterion for Progression of Visual Field Series

Progressing Nonprogressing N% N %P Value Total 151 29.9 354 70.1 Eye Right 77 51 162 45.8 0.281 Left 74 49 192 54.2 Gender Male 80 53.0 154 43.5 0.051 Female 71 47.0 200 56.5 Race African American 84 55.6 197 55.6 0.953 Caucasion 65 43.0 151 42.7 Hispanic 2 1.3 6 1.7 Age at first intervention (yrs) Mean 75.1 71.6 0.00008* SD 8.4 9.9 Range 47–97 43–89 Spherical equivalent (diopters) Mean 0.2 Ϫ0.3 0.149 SD 2.6 3 Range Ϫ12.0–7.0 Ϫ20–6.0 Visual acuity score Mean 80.0 79.7 0.78 SD 8.5 8.8 Range 56–97 56–100 Vertical cup/disc ratio Ͻ0.7 30 19.9 84 23.7 0.355 Ն0.7 121 80.1 270 76.3 Marital status Not married 63 41.8 165 46.6 0.312 Married 88 58.2 189 53.4 Education Lower than grade 12 62 41.1 143 40.4 0.889 Grade 12 or higher 89 58.9 211 59.6 Hypertension Yes 91 60.3 171 48.3 0.014† No 60 39.7 183 51.7 Diabetes Yes 32 21.2 76 21.5 0.945 No 119 78.8 278 78.5 History of vascular diseases No 123 81.5 282 79.7 0.643 Yes 28 18.5 72 20.3 Use of systemic ␤-blockers No 138 91.4 320 90.4 0.725 Yes 13 8.6 34 9.6 Intervention sequence ATT 74 49.0 182 51.4 0.62 TAT 77 51.0 172 48.6 Cataract surgery Yes 67 44.4 129 36.4 0.09 No 84 55.6 225 63.6 Length of follow-up (yrs) Mean 7.7 7.2 0.001‡ SD 1.6 1.8 Range 3.5–10.2 3.0–10.7 No. of visual field exams Mean 16.3 15.0 0.001‡ SD 3.3 4 Range 7–21 7–22 Average IOP over time (mmHg) Mean 15.4 14.5 0.008* SD 3.0 3.2 Range 6.0–25.9 4.4–24.6 IOP fluctuation (mmHg) Mean 4.0 3.4 0.0001‡ SD 2.0 1.3 Range 1.5–14.5 1.0–9.9 (Continued)

1630 Nouri-Mahdavi et al Predictive Factors for Visual Field Progression in the AGIS

Table 2. (Continued.)

Progressing Nonprogressing N% N %P Value Average no. of medications over time Mean 1.6 1.4 0.015‡ SD 0.8 1.0 Range 0–3.9 0–4 No. of glaucoma surgeries Mean 1.5 1.3 0.00003‡ SD 0.7 0.5 Range 1–31–3 Baseline AGIS score Mean 7.6 7.7 0.79 SD 4.4 4.5 Range 0–16 0–16 Baseline IOP (mmHg) Mean 23.2 23.5 0.38 SD 6.0 5.8 Range 6–50 9–47 No. of medications (baseline) Mean 2.8 2.7 0.208 SD 0.9 0.9 Range 1–40–4

AGIS ϭ Advanced Glaucoma Intervention Study; ATT ϭ argon laser trabeculoplasty–trabeculectomy–trabeculectomy; IOP ϭ intraocular pressure; SD ϭ standard deviation; TAT ϭ trabeculectomy–argon laser trabeculoplasty–trabeculectomy. The data do not include 4 eyes that had indeterminate results on pointwise linear regression analysis. *Unpaired t test. †Chi-square test. ‡Wilcoxon rank sum test. baseline value or a VF defect score of 19 or 20. Several significant risk factors for SDVF in the ATT sequence, and associations were found to be important. Eyes with less of diabetes was an additional significant risk factor for SDVF a baseline VF defect had an increased risk of subsequent in the TAT sequence. SDVF. Male gender and worse baseline VA were additional In the present study, older age at the time of first inter-

Table 3. Results of Logistic Regression Using All the Variables Mentioned in Table 1

95% CI for Odds Ratio P Odds Variable Value Ratio Lower Upper Age (per 5 yrs) 0.0012 1.289 1.109 1.498 Gender (reference, female) 0.0554 1.576 0.989 2.512 Race (reference, Caucasion) 0.7928 1.073 0.633 1.818 Education (reference, less than grade 12) 0.6904 1.108 0.669 1.835 Presence of hypertension (yes) 0.1097 1.491 0.914 2.432 Presence of diabetes (yes) 0.2888 0.724 0.398 1.316 Refractive error (reference, less than Ϫ4.0 D) Ͼ1.0 D 0.5821 0.733 0.233 2.310 Ϫ1.0–1.0 D 0.5377 0.796 0.375 1.688 Ϫ4.0 to less than Ϫ1.0 D 0.3013 0.752 0.432 1.311 Visual acuity score (no. of letters) 0.7593 1.005 0.975 1.035 Vertical cup/disc ratio (reference, Ͻ0.7) 0.2387 0.700 0.380 1.289 Baseline AGIS visual field score 0.4488 0.980 0.930 1.033 Intervention sequence (reference, ATT) 0.1183 1.468 0.904 2.385 Baseline IOP (per 1 mmHg) 0.07 0.962 0.921 1.003 No. of medications at baseline 0.9234 0.986 0.739 1.316 Mean IOP during follow-up (per 1 mmHg) 0.1333 1.076 0.977 1.185 IOP fluctuation during follow-up (per 1 mmHg) 0.0013 1.310 1.115 1.538 Mean no. of medications during follow-up 0.2998 1.191 0.853 1.662 No. of glaucoma surgeries 0.0103 1.736 1.143 2.636 Cataract surgery (yes) 0.4872 1.180 0.731 1.903 Length of follow-up (yrs) 0.0223 1.189 1.026 1.379

AGIS ϭ Advanced Glaucoma Intervention Study; ATT ϭ argon laser trabeculoplasty–trabeculectomy–trabeculectomy; CI ϭ confidence interval; D ϭ diopters; IOP ϭ intraocular pressure.

1631 Ophthalmology Volume 111, Number 9, September 2004

Table 4. Results of Logistic Regression Divided by Intervention Sequence

95% CI for Odds Ratio P Odds Intervention Sequence Value Ratio Lower Upper ATT Age (per 5 yrs) 0.0219 1.407 1.064 1.860 Gender (reference, female) 0.0436 2.231 1.024 4.861 Race (reference, Caucasion) 0.6968 0.845 0.360 1.983 Education (reference, less than grade 12) 0.4153 1.433 0.601 3.414 Presence of hypertension (yes) 0.2087 1.713 0.738 3.975 Presence of diabetes (yes) 0.7694 1.151 0.447 2.967 Refractive error (reference, less than Ϫ4.0 D) Ͼ1.0 D 0.8181 0.889 0.070 11.239 Ϫ1.0–1.0 D 0.7521 1.247 0.203 7.649 Ϫ4.0 to less than Ϫ1.0 D 0.9041 1.118 0.317 3.937 Visual acuity score (no. of letters) 0.8178 1.006 0.950 1.066 Vertical cup/disc ratio (reference, Ͻ0.7) 0.2859 0.518 0.103 2.607 Baseline AGIS visual field score 0.2651 0.949 0.858 1.049 Baseline IOP (per 1 mmHg) 0.2419 0.956 0.883 1.036 No. of medications at baseline 0.1929 0.708 0.406 1.234 Mean IOP during follow-up (per 1 mmHg) 0.8654 1.014 0.842 1.222 IOP fluctuation during follow-up (per 1 mmHg) 0.0366 1.411 1.027 1.939 Mean no. of medications during follow-up 0.6872 1.125 0.592 2.137 No. of glaucoma surgeries 0.0077 3.416 1.512 7.715 Cataract surgery (yes) 0.3436 1.452 0.624 3.378 Length of follow-up (yrs) 0.0374 1.369 1.023 1.832 TAT Age (per 5 yrs) 0.0824 1.237 0.969 1.578 Gender (reference, female) 0.7906 1.097 0.553 2.175 Race (reference, white) 0.1251 1.850 0.842 4.067 Education (reference, less than grade 12) 0.9627 0.982 0.460 2.095 Presence of hypertension (yes) 0.2202 1.569 0.762 3.228 Presence of diabetes (yes) 0.0918 0.422 0.154 1.152 Refractive error (reference, less than Ϫ4.0 D) Ͼ1.0 D 0.2386 0.593 0.057 6.133 Ϫ1.0–1.0 D 0.7307 0.826 0.196 3.478 Ϫ4.0 to less than Ϫ1.0 D 0.5679 0.573 0.187 1.751 Visual acuity score (no. of letters) 0.9038 1.003 0.953 1.055 Vertical cup/disc ratio (reference, Ͻ0.7) 0.7494 1.158 0.395 3.396 Baseline AGIS visual field score 0.6640 1.018 0.932 1.113 Baseline IOP (per 1 mmHg) 0.2578 0.962 0.896 1.033 No. of medications at baseline 0.5535 1.132 0.750 1.707 Mean IOP during follow-up (per 1 mmHg) 0.2261 1.097 0.937 1.285 IOP fluctuation during follow-up (per 1 mmHg) 0.0968 1.250 0.955 1.635 Mean no. of medications during follow-up 0.1069 1.543 0.898 2.652 No. of glaucoma surgeries 0.7819 0.908 0.436 1.893 Cataract surgery (yes) 0.6884 0.850 0.346 2.089 Length of follow-up (yrs) 0.2588 1.137 0.899 1.438

AGIS ϭ Advanced Glaucoma Intervention Study; ATT ϭ argon laser trabeculoplasty–trabeculectomy–trabeculectomy; CI ϭ confidence interval; D ϭ diopters; IOP ϭ intraocular pressure; TAT ϭ trabeculectomy–argon laser trabeculoplasty–trabeculectomy. vention was 1 of the 2 risk factors demonstrating the stron- mostly capsular glaucoma, Bergea et al found that IOP gest association with progression of VFs. Several other variation (range and peak) and mean IOP had a direct studies have reported a correlation of age with progression relationship with VF decay over 2 years of follow-up.32 of glaucomatous damage,15–21 whereas others, including a Asrani et al found a strong relationship between large diur- report from the Collaborative Normal-Tension Glaucoma nal fluctuations of IOP, measured with home tonometry, and Study Group, have not confirmed such conclusions.26–28 glaucoma progression.33 The mean office IOP had no pre- An important finding of this study was that IOP fluctu- dictive value, and mean home IOP showed a less significant ation is 1 of 2 variables consistently associated with VF association with glaucoma progression. Due to the design of progression. Intraocular pressure fluctuation was not ex- the AGIS, we could only evaluate the effect of intervisit IOP plored in previous AGIS reports. Our findings corroborate fluctuation; therefore, our results cannot be compared di- other reports regarding the significance of IOP fluctuation as rectly with theirs. Stewart et al also reported findings similar a predictor for VF worsening.1,18,29–33 Werner et al pointed to our results in a retrospective multicenter study.18 A to IOP fluctuation as a risk factor for glaucoma progression higher number of second and third glaucoma interventions after trabeculectomy.1 In a group of patients affected with was observed in eyes with progressive glaucoma in this

1632 Nouri-Mahdavi et al Predictive Factors for Visual Field Progression in the AGIS

Table 5. Results of Logistic Regression Based on Performance of Cataract Surgery

95% CI for Odds Ratio P Odds Value Ratio Lower Upper No cataract surgery Age (per 5 yrs) 0.0086 1.304 1.075 1.581 Gender (reference, female) 0.9774 1.009 0.558 1.822 Race (reference, Caucasion) 0.8749 1.058 0.525 2.130 Education (reference, less than grade 12) 0.5033 1.256 0.643 2.452 Presence of hypertension (yes) 0.2900 1.407 0.746 2.655 Presence of diabetes (yes) 0.5358 0.789 0.371 1.676 Refractive error (reference, less than Ϫ4.0 D) 1.000 Ͼ1.0 D 0.4000 0.476 0.053 4.252 Ϫ1.0–1.0 D 0.8909 1.074 0.276 4.188 Ϫ4.0 to less than Ϫ1.0 D 0.7688 0.893 0.329 2.425 Visual acuity score (no. of letters) 0.3178 0.979 0.939 1.021 Vertical cup/disc ratio (reference, Ͻ0.7) 0.4741 0.758 0.327 1.756 Baseline AGIS visual field score 0.7662 0.990 0.924 1.061 Baseline IOP (per 1 mmHg) 0.0350 0.935 0.878 0.995 No. of medications at baseline 0.9807 0.995 0.679 1.459 Intervention sequence (reference, ATT) 0.0174 2.293 1.170 4.492 Mean IOP during follow-up (per 1 mmHg) 0.1681 1.091 0.962 1.239 IOP fluctuation during follow-up (per 1 mmHg) 0.0241 1.296 1.037 1.620 Mean no. of medications during follow-up 0.3523 1.227 0.789 1.909 No. of glaucoma surgeries 0.0650 1.690 0.966 2.959 Length of follow-up (yrs) 0.0439 1.213 1.006 1.462 Previous cataract surgery Age (per 5 yrs) 0.1063 1.265 0.943 1.697 Gender (reference, female) 0.0101 3.011 1.305 6.949 Race (reference, white) 0.8894 0.940 0.390 2.263 Education (reference, less than grade 12) 0.8429 1.088 0.472 2.506 Presence of hypertension (yes) 0.1972 1.696 0.758 3.797 Presence of diabetes (yes) 0.4735 0.680 0.236 1.961 Refractive error (reference, less than Ϫ4.0 D) 1.000 Ͼ1.0 D 0.8851 1.139 0.127 10.247 Ϫ1.0–1.0 D 0.5206 0.663 0.143 3.067 Ϫ4.0 to less than 1.0 D 0.3580 0.624 0.189 2.066 Visual acuity score (no. of letters) 0.3890 1.022 0.970 1.077 Vertical cup/disc ratio (reference, Ͻ0.7) 0.4056 0.606 0.136 2.709 Baseline AGIS visual field score 0.6358 0.977 0.881 1.084 Baseline IOP (per 1 mmHg) 0.6343 0.984 0.918 1.056 No. of medications at baseline 0.8314 0.949 0.586 1.537 Intervention sequence (reference, ATT) 0.7852 0.888 0.341 2.312 Mean IOP during follow-up (per 1 mmHg) 0.5064 1.062 0.877 1.285 IOP fluctuation during follow-up (per 1 mmHg) 0.0409 1.390 1.016 1.902 Mean no. of medications during follow-up 0.5267 1.229 0.617 2.449 No. of glaucoma surgeries 0.2392 1.582 0.706 3.545 Length of follow-up (yrs) 0.1184 1.264 0.933 1.713

AGIS ϭ Advanced Glaucoma Intervention Study; ATT ϭ argon laser trabeculoplasty–trabeculectomy–trabeculectomy; CI ϭ confidence interval; D ϭ diopters; IOP ϭ intraocular pressure. subset of AGIS patients (Table 2). This alone could have led when the AGIS criteria were used as the outcome measure to a higher IOP fluctuation. However, IOP fluctuation re- supports the latter hypothesis. mained a significant predictor of VF worsening despite We did not find any relationship between mean IOP inclusion of the number of glaucoma interventions as an during follow-up and worsening of the VF, which is in independent covariate in the regression models. contrast to the conclusion reached by the original analyses The length of follow-up proved to be the next important of the AGIS data9 and other randomized controlled tri- variable associated with worsening VFs. However, the als21,35 and retrospective studies.17,18,36 However, some length of follow-up could be an artifact of using PLR to other studies have similarly failed to show such a relation- define VF outcomes. Linear regression models are more ship.27,37–39 Intraocular pressure fluctuation had a weak likely to show a significant slope as the number of obser- correlation with mean IOP during follow-up (r ϭ 0.22, vations increases.34 Another explanation is that glaucoma- PϽ0.001) in this investigation. When multivariate logistic tous eyes are more likely to change the longer they are regression was repeated, excluding IOP fluctuation, the followed over time. Finding the length of follow-up as a mean IOP reached statistical significance (P ϭ 0.045, OR ϭ statistically significant risk factor on logistic regression 1.099) along with age, number of glaucoma surgeries, and

1633 Ophthalmology Volume 111, Number 9, September 2004

Figure 1. This graph compares the change of Advanced Glaucoma Intervention Study (AGIS) score over follow-up in eyes with an intraocular pressure (IOP) fluctuation of Ͻ3 mmHg with that of those with an IOP fluctuation of Ն3 mmHg. In eyes with a lower IOP fluctuation, the AGIS score remained stable during follow-up, whereas eyes with a higher IOP fluctuation had a significant field progression (regression slope, 0.026/year, P ϭ 0.0006). male gender (data not shown). The low correlation between was low. Although addressing a different question, the IOP fluctuation and mean IOP during follow-up and the less Study Group also failed to confirm significant P value for the latter after exclusion of IOP African American race as a risk factor for development of fluctuation from the multivariate analysis suggest that IOP VF defects in multivariate analyses.45 fluctuation is an independent and stronger predictor than A shortcoming of the current investigation and many mean IOP for VF progression in the AGIS. One plausible others is the lack of an effective approach for taking into explanation has been suggested for the controversial find- account the effect of cataract progression or its removal on ings regarding the role of IOP control in glaucoma progres- the VF. One of the advantages of PLR is that it considers all sion in different studies.40 It is possible that tailoring man- the data during follow-up for detection of change. This agement of glaucoma to glaucoma severity is the main should theoretically reduce the confounding effects of cat- reason some investigators have not found IOP control to be aract extraction. However, inaccurate diagnosis of glauco- a risk factor for glaucoma progression. matous VF progression may have occurred due to unoper- The baseline AGIS VF score was not predictive of VF ated advancing cataracts. This notwithstanding, analysis of progression when PLR was used for definition of VF out- different subgroups consistently showed IOP fluctuation to comes but was confirmed when AGIS criteria were used for be associated with worsening VF. this purpose. The finding of a relationship between a better In conclusion, using PLR analysis for evaluation of VF baseline VF status and risk of subsequent VF worsening outcomes in AGIS, the 2 parameters consistently associated remains unexplained. Although a few smaller studies have with VF progression were greater IOP fluctuation and older reached similar conclusions,30,41,42 other reports have found age at first glaucoma intervention. The fact that we found the either a positive correlation between the degree of field loss same variables with both PLR and AGIS criteria strongly at baseline and VF worsening19,21,33,43 or no correlation at suggests that they are significant risk factors for predicting VF all.26,44 The above finding has been attributed, at least progression. Longer follow-up and a higher number of glau- partially, to the nonlinear nature of the AGIS scoring sys- coma interventions were other parameters that were, not un- tem.15 It is also possible that the same criteria for worsening expectedly, found to be risk factors for worsening of VF. of VFs are not as sensitive for eyes already demonstrating Acknowledgments. The authors thank all the AGIS investiga- advanced field loss at randomization. Detection of VF tors for their contributions. change may be easier when field loss is minimal at the onset. Whether or not the severity of baseline glaucomatous damage has any predictive power for subsequent progres- References sion of glaucoma remains debatable. The AGIS is the only multicenter randomized study on 1. Werner EB, Drance SM, Schulzer M. Trabeculectomy and the glaucoma in which more than half of the enrolled patients progression of glaucomatous visual field loss. Arch Ophthal- are African American. However, none of the analyses re- mol 1977;95:1374–7. ported so far, including the present one, have reported 2. The Advanced Glaucoma Intervention Study (AGIS): 1. Study African American race as a significant risk factor for pro- design and methods and baseline characteristics of study pa- gression of VFs. This is in contrast to the Collaborative tients. Control Clin Trials 1994;15:299–325. Initial Glaucoma Treatment Study, in which non-Caucasian 3. Advanced Glaucoma Intervention Study. 2. Visual field test race was found to be a predictor for VF worsening.19 scoring and reliability. Ophthalmology 1994;101:1445–55. 4. Gardiner SK, Crabb DP. Examination of different pointwise Thirty-eight percent of the patients recruited in that study linear regression methods for determining visual field progres- were African American. The Collaborative Normal-Tension sion. Invest Ophthalmol Vis Sci 2002;43:1400–7. Glaucoma Study did not reach a definitive conclusion re- 5. Noureddin BN, Poinoosawmy D, Fietzke FW, Hitchings RA. 28 garding the effect of race on the risk of VF progression. Regression analysis of visual-field progression in low tension However, the number of non-Caucasian patients enrolled glaucoma. Br J Ophthalmol 1991;75:493–5.

1634 Nouri-Mahdavi et al Predictive Factors for Visual Field Progression in the AGIS

6. Membrey WL, Poinoosawmy DP, Bunce C, et al. Comparison 26. Wilson R, Walker AM, Dueker DK, Crick RP. Risk factors for of visual field progression in patients with normal pressure rate of progression of glaucomatous visual field loss: a com- glaucoma between eyes with and without visual field loss that puter-based analysis. Arch Ophthalmol 1982;100:737–41. threatens fixation. Br J Ophthalmol 2000;84:1154–8. 27. Tezel G, Siegmund KD, Trinkaus K, et al. Clinical factors 7. Membrey WL, Bunce C, Poinoosawmy DP, et al. Glaucoma associated with progression of glaucomatous optic disc dam- surgery with or without adjunctive antiproliferatives in normal age in treated patients. Arch Ophthalmol 2001;119:813–8. tension glaucoma: 2. Visual field progression. Br J Ophthal- 28. Drance S, Anderson DR, Schulzer M, Collaborative Normal- mol 2001;85:696–701. Tension Glaucoma Study Group. Risk factors for progression 8. Chen PP. Correlation of visual field progression between eyes of visual field abnormalities in normal-tension glaucoma. in patients with open-angle glaucoma. Ophthalmology 2002; Am J Ophthalmol 2001;131:699–708. 109:2093–9. 29. Niesel P, Flammer J. Correlations between intraocular pressure, 9. AGIS Investigators. The Advanced Glaucoma Intervention visual field and visual acuity, based on 11 years of observa- Study (AGIS): 7. The relationship between control of intraoc- tions of treated chronic . Int Ophthalmol 1980;3:31–5. ular pressure and visual field deterioration. Am J Ophthalmol 30. O’Brien C, Schwartz B, Takamoto T, Wu DC. Intraocular 2000;130:429–40. pressure and the rate of visual field loss in chronic open-angle 10. Katz J, Congdon N, Friedman DS. Methodological variations in glaucoma. Am J Ophthalmol 1991;111:491–500. estimating apparent progressive visual field loss in clinical trials 31. Stewart WC, Chorak RP, Hunt HH, Sethuraman G. Factors of glaucoma treatment. Arch Ophthalmol 1999;117:1137–42. associated with visual loss in patients with advanced glauco- 11. Katz J. Scoring systems for measuring progression of visual matous changes in the head. Am J Ophthalmol field loss in clinical trials of glaucoma treatment. Ophthalmol- 1993;116:176–81. ogy 1999;106:391–5. 32. Bergea B, Bodin L, Svedbergh B. Impact of intraocular pres- 12. Vesti E, Johnson CA, Chauhan BC. Comparison of different sure regulation on visual fields in open-angle glaucoma. Oph- methods for detecting glaucomatous visual field progression. thalmology 1999;106:997–1004, discussion 1004–5. Invest Ophthalmol Vis Sci 2003;44:3873–9. 33. Asrani S, Zeimer R, Wilensky J, et al. Large diurnal fluctua- 13. Katz J, Gilbert D, Quigley HA, Sommer A. Estimating pro- tions in intraocular pressure are an independent risk factor in gression of visual field loss in glaucoma. Ophthalmology patients with glaucoma. J Glaucoma 2000;9:134–42. 1997;104:1017–25. 34. Gardiner SK, Crabb DP. Frequency of testing for detecting 14. Spry PG, Bates AB, Johnson CA, Chauhan BC. Simulation of visual field progression. Br J Ophthalmol 2002;86:560–4. longitudinal threshold visual field data. Invest Ophthalmol Vis 35. Collaborative Normal-Tension Glaucoma Study Group. Com- Sci 2000;41:2192–200. parison of glaucomatous progression between untreated pa- 15. AGIS Investigators. The Advanced Glaucoma Intervention tients with normal-tension glaucoma and patients with thera- Study (AGIS): 12. Baseline risk factors for sustained loss of peutically reduced intraocular pressures. Am J Ophthalmol visual field and visual acuity in patients with advanced glau- 1998;126:487–97. coma. Am J Ophthalmol 2002;134:499–512. 36. Vogel R, Crick RP, Newson RB, et al. Association between 16. Armaly MF, Krueger DE, Maunder L, et al. Biostatistical intraocular pressure and loss of visual field in chronic simple analysis of the collaborative glaucoma study. I. Summary glaucoma. Br J Ophthalmol 1990;74:3–6. report of the risk factors for glaucomatous visual-field defects. 37. Schulzer M, Drance SM, Douglas GR. A comparison of Arch Ophthalmol 1980;98:2163–71. treated and untreated glaucoma suspects. Ophthalmology 17. Mao LK, Stewart WC, Shields MB. Correlation between intraoc- 1991;98:301–7. ular pressure control and progressive glaucomatous damage in 38. Martinez-Bello C, Chauhan BC, Nicolela MT, et al. Intraoc- primary open-angle glaucoma. Am J Ophthalmol 1991;111:51–5. ular pressure and progression of glaucomatous visual field 18. Stewart WC, Kolker AE, Sharpe ED, et al. Factors associated loss. Am J Ophthalmol 2000;129:302–8. with long-term progression or stability in primary open-angle 39. Oliver JE, Hattenhauer MG, Herman D, et al. Blindness and glaucoma. Am J Ophthalmol 2000;130:274–9. glaucoma: a comparison of patients progressing to blindness 19. Lichter PR, Musch DC, Gillespie BW, et al. Interim clinical from glaucoma with patients maintaining vision. Am J Oph- outcomes in the Collaborative Initial Glaucoma Treatment thalmol 2002;133:764–72. Study comparing initial treatment randomized to medications 40. Palmberg P. Risk factors for glaucoma progression: where does or surgery. Ophthalmology 2001;108:1943–53. intraocular pressure fit in? Arch Ophthalmol 2001;119:897–8. 20. Landers J, Goldberg I, Graham SL. Analysis of risk factors 41. Holmin C, Krakau CE. Trabeculoplasty and visual field decay: that may be associated with progression from ocular hyper- a follow-up study using computerized perimetry. Curr Eye tension to primary open angle glaucoma. Clin Experiment Res 1984;3:1101–5. Ophthalmol 2002;30:242–7. 42. Suzuki Y, Shirato S, Adachi M, Hamada C. Risk factors for 21. Leske MC, Heijl A, Hussein M, et al. Factors for glaucoma the progression of treated primary open-angle glaucoma: a progression and the effect of treatment: the Early Manifest multivariate life-table analysis. Graefes Arch Clin Exp Oph- Glaucoma Trial. Arch Ophthalmol 2003;121:48–56. thalmol 1999;237:463–7. 22. Nouri-Mahdavi K, Coleman AL, Hoffman D, et al. Pointwise 43. Grant WM, Burke JF Jr. Why do some people go blind from linear regression for evaluation of visual field outcomes and glaucoma? Ophthalmology 1982;89:991–8. comparison to the AGIS methodology. Arch Ophthalmol. In 44. Wilson MR, Kosoko O, Cowan CL Jr, et al. Progression of press. visual field loss in untreated glaucoma patients and glaucoma 23. The Advanced Glaucoma Intervention Study. Manual of Op- suspects in St. Lucia, West Indies. Am J Ophthalmol 2002; erations, 1993. Accession no. PB93-220192. Springfield, VA: 134:399–405. National Technical Information Service; 1993. 45. Gordon MO, Beiser JA, Brandt JD, et al, Ocular Hypertension 24. SPSS [computer program]. Version 11.5. Chicago: SPSS Inc.; Treatment Study Group. The Ocular Hypertension Treatment 2003. Study: baseline factors that predict the onset of primary open- 25. SAS [computer program]. Version 8.2. Cary, NC: SAS Insti- angle glaucoma. Arch Ophthalmol 2002;120:714–20, discus- tute; 2001. sion 829–30.

1635