Five-Year Pooled Data Analysis of the Ahmed Baerveldt Comparison Study and the Ahmed Versus Baerveldt Study

PANOS G. CHRISTAKIS, DONGYU ZHANG, DONALD L. BUDENZ, KEITH BARTON, JAMES C. TSAI, AND IQBAL I.K. AHMED, FOR THE ABC-AVB STUDY GROUPS

PURPOSE: To determine the relative efficacy of the lower mean IOP on fewer medications than the Ahmed Ahmed-FP7 and Baerveldt BG101-350 implants. group. Baerveldt implantation carried a higher risk of DESIGN: Pooled analysis of 2 multicenter, randomized hypotony. (Am J Ophthalmol 2017;176:118–126. Ó clinical trials. 2017 Elsevier Inc. All rights reserved.) METHODS: A total of 514 patients aged 18 or older with uncontrolled that had failed or were at high risk of failing were randomized to receive an EDUCING (IOP) IS THE Ahmed implant (n [ 267) or Baerveldt implant (n [ only known modifiable risk factor to prevent vision 1,2 247). Cumulative failure rates (using an intraocular pres- R loss in glaucoma. Treatment begins with the use sure [IOP] target of 6–18 mm Hg inclusive), de novo of topical antiglaucoma eye drops or with laser rates, mean IOP, mean glaucoma medi- trabeculoplasty, and surgery is reserved for cases cation use, and visual acuity were compared. refractory to or at high risk of failing these treatment 3 RESULTS: Baseline characteristics were similar be- modalities. Trabeculectomy has historically been the tween groups. Mean preoperative IOP of the study popu- most common filtration surgery, and works by removing lation was 31.5 ± 11.3 mm Hg on an average of 3.3 ± 1.1 limbal tissue to allow aqueous to flow through a scleral glaucoma medications. At 5 years, mean IOP was 15.8 ± flap to an adjacent subconjunctival bleb. However, failure 5.2 mm Hg in the Ahmed group and 13.2 ± 4.7 mm Hg in rates of up to 10% per year have been reported, even with the Baerveldt group (P < .001). Mean glaucoma medica- the use of intraoperative (eg, mitomycin C, 5-fluorouracil) 4 tion use was 1.9 ± 1.5 in the Ahmed group and 1.5 ± 1.4 to prevent bleb fibrosis. Risk factors for failure include in the Baerveldt group (P [ .007). The cumulative fail- young age, black race, neovascular or uveitic glaucoma, 5 ure rate at 5 years was 49% in the Ahmed group and 37% and prior failed surgery. Furthermore, there is a risk for in the Baerveldt group (P [ .007). High IOP was the hypotony with trabeculectomy given the unpredictable na- most common reason for failure in both groups, and de ture of flow, and a lifetime risk for the development of 6 novo glaucoma surgery was required in 16% of the blebitis. For these reasons, there has been a significant Ahmed group and 8% of the Baerveldt group (P [ shift in practice patterns between 1996 and 2008, with rates .006). Failure owing to hypotony occurred in 0.4% of of trabeculectomy decreasing from 81% to 46% and rates of the Ahmed group and 4.5% of the Baerveldt group glaucoma drainage device implantation increasing from 7 (P [ .002). Visual outcomes were similar between 18% to 51%. This shift was validated by the Tube Versus groups (P [ .90). Trabeculectomy (TVT) Study, a multicenter randomized CONCLUSIONS: The Baerveldt group had a lower fail- study found that the Baerveldt implant had a higher success ure rate, lower rate of de novo glaucoma surgery, and rate at 5 years with fewer complications compared with trabeculectomy with mitomycin C.6,8 Furthermore, a recent meta-analysis showed that Ahmed valve implanta- tion had a similar success rate and IOP control trabeculec- Supplemental Material available at AJO.com. 9 Accepted for publication Jan 6, 2017. tomy, with fewer adverse events. From the Department of Ophthalmology and Vision Sciences, Glaucoma drainage devices function by shunting University of Toronto Faculty of Medicine, Toronto, Canada (P.G.C., aqueous humor from the anterior chamber to an equatorial I.I.K.A.); Department of Epidemiology, University of North Carolina at Chapel Hill Gillings School of Global Public Health (D.Z., D.L.B.), and subconjunctival endplate using a long tube, which flows Department of Ophthalmology, University of North Carolina at Chapel into an overlying bleb and drains into the venous plexus. Hill (D.L.B.), Chapel Hill, North Carolina; NIHR Biomedical Research The 2 most commonly used glaucoma drainage devices Centre for Ophthalmology, Moorfields Eye Hospital, London, United Kingdom (K.B.); and Department of Ophthalmology, Icahn School of are the Ahmed valve implant (New World Medical Inc, Medicine at Mount Sinai, New York Eye and Ear Infirmary of Mount Rancho Cucamonga, California, USA) and the Baerveldt Sinai, New York, New York (J.C.T.). implant (Advanced Medical Optics, Santa Ana, Califor- Inquiries to Donald L. Budenz, Department of Ophthalmology, University of North Carolina, 5151 Bioinformatics CB7040, Chapel nia, USA). The principal difference between these devices Hill, NC 27599-7040; e-mail: [email protected] is that the Ahmed valve has a build-in, venturi-based valve

118 © 2017 ELSEVIER INC.ALL RIGHTS RESERVED. 0002-9394/$36.00 http://dx.doi.org/10.1016/j.ajo.2017.01.003 that regulates flow and serves to prevent hypotony. The target despite maximum tolerated medical therapy; (3) pa- Baerveldt implant requires the surgeon to temporarily tients who had previously failed trabeculectomy were ligate the tube to prevent postoperative filtration until included; (4) high-risk glaucoma, including neovascular there is adequate healing around the endplate to regulate and uveitic glaucoma, were included; (5) only 1 eye flow. As a result, the Ahmed valve implant achieves imme- eligible for enrollment. diate predictable IOP control, although exposure to post- operative inflammatory mediators has been implicated in Exclusion Criteria. The following comprised criteria for bleb encapsulation and long-term failure.10,11 The exclusion from the study: (1) scheduled to undergo an addi- Baerveldt implant takes 4–6 weeks for the ligature to tional procedure at the time of device implantation (eg, dissolve or be released and for flow to commence. Given cataract surgery, corneal transplant); (2) prior cyclodes- that flow is regulated by the degree of healing around the tructive procedure; (3) no light perception vision; (4) un- endplate, which may vary between patients, there is a able to provide informed consent or adhere to the risk for hypotony.12,13 If adequate healing is present, the requirements of the study. large size of the Baerveldt implant and flat smooth profile In addition to the above criteria, the ABC Study has been reported to provide good long-term IOP control excluded patients who had uveitis secondary to a systemic and reduce the need for glaucoma medications.8 condition (eg, juvenile rheumatoid arthritis), patients with Retrospective studies comparing the Ahmed valve glaucoma associated with elevated episcleral venous pres- implant and the Baerveldt implant have been inconclusive sure, and patients who had silicone oil in the operative eye. and difficult to interpret on account of their small sample sizes, use of different device models and success criteria, TREATMENT RANDOMIZATION: The ABC and AVB and susceptibility to a device selection bias.10,11,14–16 The studies were open-label trials; both the surgeon and patient Ahmed Versus Baerveldt Comparison (ABC) Study and were aware of the treatment arm. The ABC Study used a the Ahmed Versus Baerveldt (AVB) Study are 2 permuted variable block randomization design with strati- independent randomized clinical trials that have recently fication by surgeon and glaucoma subtype. The AVB Study reported their 5-year outcomes and provide prospective used simple randomization without stratification. data to answer this question.17,18 Their study design, patient population, and outcome criteria are remarkably SURGICAL METHODS: Patients were randomized to similar, and the purpose of this study is to report the receive either an Ahmed valve (Model AGV FP7) implant results of the pooled data.19–24 or a Baerveldt implant (Model 101-350 BGI) by an experi- enced surgeon using standardized technique. Detailed sur- gical methodology has been reported in the ABC and AVB baseline papers.19,20 Briefly, implants were sutured METHODS at 8–10 mm posterior to the limbus in the superotemporal quadrant. Baerveldt implants were ligated STUDY DESIGN: Data were pooled from 2 independent intraoperatively using the preferred method of the randomized clinical trials, the ABC and AVB studies. surgeon; either a peritubular polyglactin ligation or a rip The AVB Study commenced in July 2005, and the ABC cord technique was employed. Fenestration of the tube Study commenced in October 2006. Patients were anterior to the ligation to allow early postoperative recruited from 22 clinical centers in the United States, filtration was left to the discretion of the surgeon. Canada, the United Kingdom, Chile, Singapore, and Brazil (Appendix; Supplemental Material available at AJO.com). FOLLOW-UP AND OUTCOME CRITERIA: Baseline ocular Institutional Review Board approval was obtained at each and demographic characteristics were collected preoper- clinical center and the study protocol conformed to the atively. Postoperative visits were scheduled at 1 day, principles of the Declaration of Helsinki. Written informed 1 week, 1 month, 3 months, 6 months, 12 months, consent was obtained from all patients, and both 18 months, 2 years, 3 years, 4 years, and 5 years. The studies were registered on ClinicalTrials.gov (ABC: AVB Study had additional scheduled visits at 2 weeks, NCT00376363, AVB: NCT00940823). 2 months, and 9 months. Additional visits were made as clinically indicated by the treating physician in STUDY POPULATION AND ELIGIBILITY CRITERIA: The both studies. Data collection at each follow-up visit ABC and AVB studies recruited patients with refractory included best-corrected Snellen visual acuity (BCVA); or high-risk glaucoma. Eligibility criteria were similar in IOP using Goldmann applanation tonometry, or a the 2 studies and are summarized below. Tono-Pen (Reichert Ophthalmic Instruments, Depew, New York, USA) if severe corneal pathology was Inclusion Criteria. The following comprised criteria for present; glaucoma medication use (both topical and sys- study inclusion: (1) 18 years of age or older; (2) uncon- temic); complications; and interventions (both slit-lamp trolled glaucoma, defined as IOP greater than clinical and surgical).

VOL. 176 POOLED DATA ANALYSIS OF THE ABC STUDY AND AVB STUDY 119 DATA HETEROGENEITY AND POOLING STRATEGY: The ABC and AVB Study data was obtained from the respec- TABLE 1. Ahmed Baerveldt Comparison Study and Ahmed tive principal investigators (D.L.B., I.I.K.A.). Underlying Versus Baerveldt Study: Baseline Patient Characteristics data heterogeneity was addressed by recategorizing data us- ing a common methodology. Baseline variables recatego- Ahmed Baerveldt (N ¼ 267) (N ¼ 247) P Value rized were race (white, black, and other) and glaucoma a diagnosis (primary open-angle, neovascular, uveitic, Age (y 6 SD) 65 6 15 64 6 14 .60 Sex: female 129 (48%) 137 (55%) .11b chronic angle closure, and other). A new composite crite- b rion for failure was created given differences in the primary Laterality: right 159 (60%) 128 (52%) .078 Race IOP criteria. The ABC Study used a primary IOP target of b White 157 (59%) 147 (60%) .10 6–21 mm Hg inclusive, and performed secondary analysis Black 58 (22%) 38 (15%) using IOP targets of 6–18 mm Hg and 6–15 mm Hg as Other 52 (19%) 62 (25%) 25 per the World Glaucoma Association guidelines. The Prior trabeculectomy 81 (30%) 87 (35%) .24b AVB Study used a primary IOP target of 5–18 mm Hg in- Prior laser trabeculoplasty 78 (29%) 81 (33%) .38b clusive and secondary IOP targets of 5–21 mm Hg and Prior cataract surgery 167 (63%) 153 (62%) .89b 5–15 mm Hg. Glaucoma subtype .82b Given data suggesting that lower target IOPs may be Open-angle glaucoma 122 (51%) 108 (49%) required to prevent visual field progression, a primary Neovascular glaucoma 69 (29%) 61 (28%) IOP target of 6–18 mm Hg inclusive was selected, and sec- Glaucoma associated 21 (9%) 20 (9%) ondary analysis was performed using IOP targets of with uveitis Chronic angle closure 17 (7%) 15 (7%) 6–21 mm Hg and 6–15 mm Hg.2 To achieve data homoge- glaucoma neity, failure criteria from the AVB Study were reclassified Other 11 (5%) 16 (7%) such that the lower limit of IOP was increased from 5 mm Hg to 6 mm Hg, and the upper limit of IOP for secondary aStudent t test. analysis was increased from 14 mm Hg to 15 mm Hg. bx2 test.

OUTCOME CRITERIA: The primary outcome criterion for both studies was failure, defined as (1) IOP > 18 mm Hg or IOP < 6 mmHg or IOP reduced less than 20% from base- graphic and ocular characteristics comparing the 2 treat- line, for 2 consecutive visits after 3 months; (2) de novo ment groups. Categorical variables were analyzed using a x2 > glaucoma procedure (eg, cyclodestruction, additional test, unless 25% of cells had an expected value of < tube shunt); (3) loss of light perception or severe vision 5 and Fisher exact test was used. Continuous variables loss*; or (4) removal of the implant. were analyzed with a 2-sample Student t test to compare Additional analysis using alternate IOP targets of groups at defined time points. BCVA was converted to logMAR values for the purpose of analysis using the 6–15 mm Hg and 6–21 mm Hg was performed. Secondary 1 outcome criteria compared groups based on mean IOP, following formula: Log10(BCVA ). Treatment failure mean glaucoma medication use, and BCVA at each post- was analyzed with Kaplan-Meier analysis using the log- operative visit. rank test, and hazard ratios (HR) were calculated using a crude Cox proportional hazard model. Patients who under- went de novo glaucoma surgery, developed phthisis bulbi, STATISTICAL METHODS: Statistical analysis was performed using Stata 13.0 (StataCorp, LLP, College Sta- had an enucleation or evisceration, or had care withdrawn tion, Texas, USA) and significance was defined as a 2- were censored from subsequent IOP and glaucoma medica- sided P value of .05 or less. Intention-to-treat analysis tion use analysis, but visual acuity, complications, and was used and patients were analyzed based on their original required interventions were included. As a result of randomization. Four patients in the ABC Study received a censoring confounding data, reported sample sizes when different device than planned; 1 patient in the Ahmed analyzing each variable may vary. group received a different model (AGV-S2), 2 patients in the Baerveldt group received an Ahmed implant owing to lack of availability, and 1 patient in the Baerveldt group RESULTS received an Ahmed implant owing to surgeon error. Descriptive analysis was performed for baseline demo- BASELINE CHARACTERISTICS: A total of 514 patients were randomized in the ABC (n ¼ 276) and AVB studies (n ¼ 238): 267 to the Ahmed group and 247 to the Baer- The AVB Study had a small number of patients who failed owing to severe vision loss from a complication of surgery (eg, central suprachoroi- veldt group. Baseline demographic and ocular characteris- dal hemorrhage, endophthalmitis) or who developed phthisis bulbi or tics were similar between treatment groups (Tables 1–3). required enucleation/evisceration. The average patient age was 65 6 15 years in the Ahmed

120 AMERICAN JOURNAL OF OPHTHALMOLOGY APRIL 2017 TABLE 2. Ahmed Baerveldt Comparison Study and Ahmed TABLE 3. Ahmed Baerveldt Comparison Study and Ahmed Versus Baerveldt Study: Mean Intraocular Pressure and Versus Baerveldt Study: Visual Outcomes Mean Glaucoma Medication Use Ahmed Baerveldt P Ahmed Baerveldt P (N ¼ 267) (N ¼ 247) Value (N ¼ 267) (N ¼ 247) Valuea Mean logMAR Snellen visual Baseline acuity 6 SD Mean IOP (mm Hg) 6 SD 31.2 6 10.9 31.8 6 11.8 .53 Baseline 1.2 6 1.0 1.1 6 1.0 .77a Mean glaucoma medications 3.3 6 1.1 3.3 6 1.1 .60 6 months 1.2 6 1.0 1.3 6 1.1 .31a 6 SD 12 months 1.3 6 1.1 1.3 6 1.2 .75a Number randomized (N) 267 247 18 months 1.3 6 1.2 1.4 6 1.3 .56a 6 months 24 months 1.3 6 1.2 1.3 6 1.2 .83a Mean IOP (mm Hg) 6 SD 16.2 6 5.2 15.0 6 6.6 .039 36 months 1.4 6 1.2 1.4 6 1.3 .87a Mean glaucoma medications 1.7 6 1.4 1.2 6 1.3 <.001 48 months 1.4 6 1.2 1.5 6 1.3 .53a 6 SD 60 months 1.5 6 1.2 1.5 6 1.3 .90a N (% of baseline) 230 (86%) 222 (90%) Visual acuity change between 12 months baseline and 60 months Mean IOP (mm Hg) 6 SD 15.9 6 5.3 13.6 6 5.9 <.001 Within 2 Snellen lines 62 (36%) 71 (42%) .87b Mean glaucoma medications 1.8 6 1.3 1.4 6 1.4 .003 Loss of 2þ Snellen lines 83 (47%) 77 (46%) 6 SD Gain of 2þ Snellen lines 30 (17%) 21 (12%) N (% of baseline) 237 (89%) 218 (88%) a 18 months Student t test. b Mean IOP (mm Hg) 6 SD 15.7 6 5.0 13.7 6 5.2 <.001 Spearman rank correlation coefficients. Mean glaucoma medications 1.7 6 1.4 1.3 6 1.3 .003 6 SD N (% of baseline) 212 (79%) 198 (80%) 2 years and other (4% AGV, 6% BGI). Mean preoperative IOP Mean IOP (mm Hg) 6 SD 15.3 6 6.0 14.2 6 6.0 .069 6 6 6 6 < was 31.2 10.9 mm Hg in the Ahmed group and 31.8 Mean glaucoma medications 1.9 1.4 1.2 1.4 .001 ¼ 6 11.8 mm Hg in the Baerveldt group (P .53) on an average SD 6 ¼ N (% of baseline) 221 (83%) 203 (82%) of 3.3 1.1 glaucoma medications in both groups (P 6 3 years .60). Mean preoperative logMAR acuity was 1.2 1.0 in Mean IOP (mm Hg) 6 SD 15.3 6 4.8 13.8 6 4.8 .007 the Ahmed group and 1.1 6 1.0 in the Baerveldt group Mean glaucoma medications 1.9 6 1.4 1.3 6 1.4 <.001 (P ¼ .77). 6 SD N (% of baseline) 204 (76%) 190 (77%) RETENTION RATES: Median follow-up time was 4 years 37 months (range, 1–66 months), with 66% of Ahmed pa- Mean IOP (mm Hg) 6 SD 15.8 6 5.4 13.9 6 4.6 .002 tients and 68% of Baerveldt patients completing 5-year 6 6 < Mean glaucoma medications 2.1 1.5 1.5 1.4 .001 follow-up (P ¼ .483). During the course of the study, 20 6 SD (7%) patients died in the Ahmed group and 21 (9%) pa- N (% of baseline) 182 (68%) 173 (70%) tients died in the Baerveldt group (Figure 1). 5 years Mean IOP (mm Hg) 6 SD 15.8 6 5.2 13.2 6 4.7 <.001 Mean glaucoma medications 1.9 6 1.5 1.5 6 1.4 .023 FAILURE RATES: At 5 years, the cumulative proportion 6 SD failing (standard error) was 49% (3%) in the Ahmed group N (% of baseline) 175 (66%) 169 (68%) and 37% (3%) in the Baerveldt group (P ¼ .01, Table 4, Figure 2). The relative hazard of treatment failure in the IOP ¼ intraocular pressure. Ahmed group was 1.34 times that of the Baerveldt group a t Student test. (confidence interval [CI]: 1.03–1.74, P ¼ .03). The most common reason for failure was high IOP in both groups, which accounted for 112 (42%) failures in the Ahmed group and 64 6 14 years in the Baerveldt group, and the group and 56 (23%) failure in the Baerveldt group most common races were white (59% AGV, 60% BGI) (Table 4). and black (22% AGV, 15% BGI). The most common The cumulative proportion of patients requiring de novo glaucoma subtype was primary open-angle glaucoma glaucoma surgery (standard error) was 17% (2%) in (51% AGV, 49% BGI), neovascular glaucoma (29% the Ahmed group and 9% (2%) in the Baerveldt group AGV, 28% BGI), uveitic glaucoma (9% AGV, 9% BGI), (P ¼ .01). The relative risk of requiring de novo glaucoma chronic angle closure glaucoma (7% AGV, 7% BGI), surgery in the Ahmed group was 2 times that of the

VOL. 176 POOLED DATA ANALYSIS OF THE ABC STUDY AND AVB STUDY 121 FIGURE 2. Ahmed and Baerveldt cumulative failure rate (6 mm Hg £ intraocular pressure criteria £ 18 mm Hg). Kaplan-Meier analysis using the log-rank test.

FIGURE 1. Ahmed Baerveldt Comparison Study and Ahmed Versus Baerveldt Study: study design and patient retention.

TABLE 4. Ahmed Baerveldt Comparison Study and Ahmed Versus Baerveldt Study: Failure Analysis

Ahmed Baerveldt (N ¼ 267) (N ¼ 247) P Value

Overall outcome Success 137 (51%) 156 (63%) .007a Failure 130 (49%) 91 (37%) Reasons for failure FIGURE 3. Ahmed and Baerveldt cumulative failure rate £ £ Hypotony 1 (0.4%) 11 (4%) .002b (6 mm Hg intraocular pressure criteria 15 mm Hg). High intraocular pressure 112 (42%) 56 (23%) <.001a Kaplan-Meier analysis using the log-rank test. Severe vision loss 13 (4.9%) 18 (7%) .25a Explantation 4 (2%) 6 (2%) .53b

ax2 test. b had a 5-year cumulative failure rate of 61%, compared Fisher exact test. with 48% for Baerveldt implantation (HR ¼ 1.46, 95% CI 1.15–1.85, log-rank P ¼ .002, Figure 3). When using an IOP criterion of 6–21 mm Hg, Ahmed implantation Baerveldt group (CI: 1.2–3.3, P ¼ .010). At the time of de had a 5-year cumulative failure rate of 42% compared novo glaucoma surgery, the mean IOP of the Ahmed group to 35% in the Baerveldt group (HR ¼ 1.23, 95% CI was 29.4 6 9.8 mm Hg compared with 29.1 6 8.7 mm Hg 0.93–1.63, log-rank P ¼ .14, Figure 4). in the Baerveldt group (P ¼ .91). Severe vision loss occurred in 13 (5%) patients in the INTRAOCULAR PRESSURE: Both devices were effective Ahmed group and 18 (7%) patients in the Baerveldt group in reducing IOP (Table 2, Figure 5). The mean IOP in (P ¼ .27). Device explantation was required in 4 (2%) pa- the Ahmed group decreased from 31.2 6 10.9 mm Hg pre- tients in the Ahmed group and 6 (2%) patients in the Baer- operatively to 15.8 6 5.2 mm Hg at 5 years (49% reduction, veldt group (P ¼ .53). Hypotony resulted in failure in 1 P < .001). The mean IOP in the Baerveldt group decreased (0.4%) patient in the Ahmed group and 11 (4%) patients from 31.8 6 11.8 mm Hg preoperatively to 13.2 6 4.8 mm in the Baerveldt group (P ¼ .002). Hg at 5 years (58% reduction, P < .001). When comparing Figures 3 and 4 present the results of Kaplan-Meier ana- the 2 groups, the Baerveldt group had a lower mean IOP at lyses using alternate IOP criteria for success. When using all postoperative visits beginning at 6 months and an IOP criterion of 6–15 mm Hg, Ahmed implantation continuing through 5 years (P < .05).

122 AMERICAN JOURNAL OF OPHTHALMOLOGY APRIL 2017 FIGURE 4. Ahmed and Baerveldt cumulative failure rate (6 mm Hg £ intraocular pressure criteria £ 21 mm Hg). FIGURE 6. Mean glaucoma medication use over 5 years of Kaplan-Meier analysis using the log-rank test. follow-up. Error bars represent standard deviation.

1.5 6 1.3 at 5 years (P < .001). The 2 groups had similar visual acuities at all postoperative visits (P > .05). At 5 years, 36% of each group remained within 2 Snellen lines of the preoperative visit, 47% of the Ahmed group and 46% of the Baerveldt group lost more than 2 Snellen lines, and 17% of the Ahmed group and 18% of the Baerveldt group gained more than 2 Snellen lines (P ¼ .87).

DISCUSSION

APPROPRIATELY POWERED RANDOMIZED CONTROLLED TRI- als are considered the best study design for comparing the efficacy of 2 treatments. However, meta-analysis and FIGURE 5. Mean intraocular pressure (IOP) over 5 years of pooled data analysis of similarly designed randomized pro- follow-up. Error bars represent standard deviation. spective clinical trials provide an even higher level of evi- dence than single trials because of their greater sample size, narrower confidence intervals, and greater generaliz- 26 GLAUCOMA MEDICATIONS: Both devices were effective ability. The ABC and AVB studies recruited glaucoma in reducing the need for glaucoma medications (Table 2, patients who had failed or were at high risk of failing trabe- Figure 6). The mean number of glaucoma medications culectomy, and randomized them to receive an Ahmed- required in the Ahmed group decreased from 3.3 6 1.1 pre- FP7 valve implant or a Baerveldt-350 implant. Baseline pa- operatively to 1.9 6 1.5 at 5 years (42% reduction, tient populations, study design, and outcome criteria were P < .001). The mean number of glaucoma medications similar between studies, allowing for valid pooling of data. required in the Baerveldt group decreased from 3.3 6 1.1 Failure was the primary outcome measure in both studies preoperatively to 1.5 6 1.4 at 5 years (55% reduction, and required adequate IOP control without de novo glau- P < .001). When comparing the 2 groups, the Baerveldt coma surgery, and without severe vision loss related to sur- group required fewer glaucoma medications at all postoper- gery or loss of light perception. The primary IOP target ative visits, beginning at 6 months and continuing through selected was 6–18 mm Hg based on Advanced Glaucoma 5 years (P < .05). Intervention Study data suggesting that higher IOPs may result in progression of visual field loss.2 At 5 years, the cu- VISUAL OUTCOMES: There was a moderate but signifi- mulative failure rate was 49% in the Ahmed group and cant decrease in visual acuity in both groups between base- 37% in the Baerveldt group (P ¼ .007). The most common line and 5 years. Mean logMAR acuity in the Ahmed group reason for failure in both groups was high IOP, and de novo worsened from 1.2 6 1.0 preoperatively to 1.5 6 1.2 at 5 glaucoma surgery was required in 16% of the Ahmed group years (P < .001). Mean logMAR acuity in the Baerveldt and 8% of the Baerveldt group (P ¼ .006). The most com- group worsened from 1.1 6 1.0 preoperatively to mon de novo glaucoma surgeries were placement of an

VOL. 176 POOLED DATA ANALYSIS OF THE ABC STUDY AND AVB STUDY 123 additional glaucoma drainage device and cyclodestructive relative efficacy of these models (ClinicalTrials.gov, procedures. Our results are comparable to the 10% failure NCT01159314). rate per year described in the American Academy of However, the Baerveldt implant has almost twice the Ophthalmology report on aqueous shunts, and the 30% surface area as the Ahmed valve implant (350 mm2 BGI, failure rate and 9% glaucoma reoperation rate for the Baer- 184 mm2 AGV), and there is clinical evidence that may in- veldt implant in the TVT Study.4,8 crease filtration.30 The composition of the bleb may also Both devices were effective in reducing IOP and the play a role in the rate of filtration, and fibrous encapsulation need for glaucoma medications. When comparing the de- has been implicated as the pathogenesis for device fail- vices at 5 years, the Baerveldt group had a 2.6 mm Hg lower ure.11,31 There are several theories as to why the Ahmed mean IOP (13.2 mm Hg BGI, 15.8 mm Hg AGV, P < valve implant may have higher rates of bleb .001), on 0.4 fewer glaucoma medications (1.5 BGI, 1.9 encapsulation than the Baerveldt implant. Firstly, the AGV, P ¼ .02). The greater reduction in IOP and the Ahmed valve implant has early postoperative flow, which need for glaucoma medications in the Baerveldt group is may expose the bleb to inflammatory mediators from noteworthy, as it also coincides with a lesser need for addi- surgery, which stimulates fibrosis.32 Secondly, early flow tional glaucoma surgery compared with the Ahmed group. may result in an imbalance between the fibroproliferative Our results show similarities to the 2 large-scale retro- and fibrodegenerative process required to create a thin, spective studies comparing the Ahmed valve implant and avascular bleb.31,32 Thirdly, although both implants are the Baerveldt implant. Tsai and associates10,11 used a made of silicone, the Ahmed-FP7 endplate may be more in- retrospective comparative design to compare 48 Ahmed- flammatory, given that electron microscopy data shows a S2 valves, 20 Baerveldt-250 implants, and 50 Baerveldt- 10-fold greater roughness than the Baerveldt implant, 350 implants. The groups had differences in baseline char- which correlates in vitro to increased fibroblast adhesion acteristics: the Ahmed group was older, had a higher rate of and fibrous encapsulation.33 inflammatory glaucoma, and had a higher baseline IOP. At Although IOP is the primary outcome criterion for glau- 4 years, the failure rate was 38% in the Ahmed group and coma surgical trials given that it is objective, quantitative, 36% in the Baerveldt group (P ¼ .84). The Ahmed group and used clinically to monitor response to treatment, the had a higher rate of de novo glaucoma surgery (19% main goal of glaucoma therapy is preservation of vision. AGV, 6% BGI, P < .05), had a higher rate of bleb encap- However, visual acuity is affected late in the disease and sulation (AGV 60%, BGI 27%, P < .001), and required a may be confounded by concomitant ocular pathology, mak- greater number of glaucoma medications (AGV 2.3, BGI ing visual field testing important for monitoring progres- 1.2, P < .05). However, the Baerveldt group had a higher sion. Our study population had advanced glaucoma with rate of failure from hypotony and its related complications poor baseline vision, which limited the reliability of (AGV 4%, BGI 13%, P < .05). analyzing visual field results, although they were used clin- Subsequently, Goulet and associates16 used a retrospec- ically to guide treatment decisions. Furthermore, the poor tive case-control design to compare 59 Ahmed-S2 valves baseline vision of the study population may have resulted and 133 Baerveldt-250 implants. At 2 years, the Ahmed in insufficient power to detect subtle changes in vision be- group had a failure rate of 38%, compared with 15% in tween groups, especially with respect to device-specific the Baerveldt group (P ¼ .03). The Ahmed group had a complications such as hypotony. Our primary visual higher mean IOP (AGV 19.8 mm Hg, BGI 15.8 mm Hg, outcome measure was BCVA, which showed a moderate P ¼ .003) and required a greater number of glaucoma med- but similar worsening in both groups. At 5 years, 53% of ications (AGV 1.4, BGI 0.9, P ¼ .008). However, the the Ahmed group and 54% of the Baerveldt group main- applicability of these studies to current practice may be tained within 2 lines or better of their baseline visual acu- limited, as they used the older AGV-S2 model, which ity. However, 47% of the Ahmed group and 46% of the was made of polypropylene rather than flexible silicone, Baerveldt group lost greater than 2 lines of vision. Over as is the AGV-FP7 model. There is evidence that the the course of the study, 16 (6%) patients in the Ahmed AGV-FP7 model has a lower failure rate, has a greater group and 23 (9%) patients in the Baerveldt group lost light IOP reduction, and requires fewer medications than the perception (P ¼ .18). Of the patients who progressed to no AGV-S2 model, owing to lower rates of bleb encapsula- light perception, 31 (80%) patients had neovascular glau- tion.27 Furthermore, both studies included patients coma. receiving a Baerveldt 250 mm2 model, which has a smaller Several factors should be considered when interpreting endplate compared with the Baerveldt 350 mm2 model the results of the ABC and AVB studies. Firstly, our patient used in our studies. There may be an upper limit to the ef- population had high rates of secondary glaucoma, and fect of endplate size, and retrospective comparisons have many patients had previously failed trabeculectomy with found no difference in failure rates, IOP reduction, glau- antimetabolite. Hence, our results may not accurately coma medication use, or complication rates between the depict the role that glaucoma drainage devices play as an BGI-250 and BGI-350 models.28,29 A prospective initial surgical procedure, and the Primary Tube randomized trial is currently underway to assess the Versus Trabeculectomy Study (ClinicalTrials.gov,

124 AMERICAN JOURNAL OF OPHTHALMOLOGY APRIL 2017 NCT00666237) is underway to address this question. Sec- but are censored from analysis to prevent confounding. ondly, our studies did not mask the surgeon to the treat- Hence, the reported mean IOP and medication use may ment arm because of differences in surgical technique. be an underestimation of the true value. This may provide This introduces a potential for bias when making postoper- an advantage to the Ahmed group, which had a higher rate ative clinical decisions, such as when to add glaucoma med- of de novo glaucoma surgery, although this may be counter- ications or perform de novo glaucoma surgery. To evaluate balanced by outliers in the Baerveldt group who failed for bias, the mean IOP and mean medication use of patients owing to hypotony. Finally, although our retention rates booked for de novo glaucoma surgery were compared and were acceptable in comparison to other prospective clinical found to be similar between groups. Thirdly, we used a trials in ophthalmology, there is always the possibility of population-based IOP target (6–18 mm Hg) to allow for attrition bias.4,8 an objective, standardized way of comparing groups, but In conclusion, both devices were effective in this may not accurately reflect the status of individual pa- lowering IOP and the need for glaucoma medications. tients. Each patient has his or her own clinical IOP target When comparing the devices, the Baerveldt implant based on preoperative IOP, glaucoma subtype and severity, had a lower failure rate, had a lower rate of de novo and threshold for progression. By using a single IOP target, glaucoma surgery, and provided a lower IOP on fewer a stable neovascular patient with an IOP of 19 mm Hg on medicationsat5years,buthadahigherriskforhypot- no medications will be characterized as a failure, whereas ony. Ultimately, selecting a device should consider the an end-stage glaucoma patient showing progression at an IOP target, their compliance with glaucoma medica- IOP of 18 mm Hg despite maximum medical therapy will tions, and the urgency for IOP lowering. Surgeons’ fa- be considered a success. Furthermore, averaging IOP across miliarity with each device and their personal a population is susceptible to skew from outliers, and outcomes should also be considered. Additional analysis censoring these patients from analysis creates a selection of risk factors for treatment failure is forthcoming, and bias. In particular, patients who require de novo glaucoma may provide guidance in selecting a device for individ- surgery have the highest IOP and medication requirement, ual patients.

FUNDING/SUPPORT: THE ABC STUDY WAS SUPPORTED BY NATIONAL INSTITUTES OF HEALTH (GRANT P30 EY014801, BETHESDA, Maryland, USA), New World Medical, Inc. (Rancho Cucamonga, California, USA), Abbott Medical Optics (Abbott Park, Illinois, USA), Glaucoma Research Society of Canada (Toronto, Canada), Research to Prevent Blindness (New York, New York, USA). The funding organizations had no role in the design or conduct of this research. Financial Disclosures: Donald L. Budenz: Consulting fees – Alcon Labs, Ft. Worth, Texas, USA; Inotek, Lexing- ton, Massachusetts, USA; ForSight Labs, Menlo Park, California, USA; Alimera, Alpharetta, Georgia, USA; Data Safety Monitoring Board – Ivantis Research, Irvine, California, USA; Research Grant –National Eye Institute (NIH), Bethesda, Maryland, USA; Glaucoma Research Foundation, San Fran- cisco, California, USA. Keith Barton: Consulting fees – Alcon Labs, Ft. Worth, Texas, USA; Ivantis, Irvine, California, USA; Aquesys, Aliso Viejo, Cal- ifornia, USA. Advisory Board fees – Glaukos, Laguna Hills, California, USA; Kowa, Aichi, Japan; Amakem, Diepenbeek, Belgium; Laboratoires Thea, Clermont-Ferrand, France; Travel grant – New World Medical, Rancho Cucamonga, California, USA; Supplied implants for research - New World Med- ical, Rancho Cucamonga, California, USA; Abbott Medical Optics, Abbott Park, Illinois, USA; Research grant – Merck & Co, Kenilworth, New Jersey, USA. Stock and options – Aquesys Inc, Ophthalmic Implants Pte, Vision Futures Ltd, Vision Medical Events Ltd. Iqbal I.K. Ahmed: Consulting fees – Abbott Medical Optics, Abbott Park, Illinois, USA, Accelerated Vision, ACE Vision Group, Ade Therapeutics, Aerie Pharmaceuticals, Alcon Labs, Ft. Worth, Texas, USA; Allergan, Irvine, California, USA; Aquesys Inc, Aliso Viejo, California, USA; Bausch and Lomb, Rochester, New York, USA; Carl Zeiss Meditec, Clarity Medical Systems, Envisia Therapeutics, Morrisville, North Carolina, USA; ForSight Labs, Menlo Park, California, USA; Glaukos, Laguna Hills, California, USA; InnFocus Inc, Miami, Florida, USA; Ivantis Research, Irvine, California, USA; LayerBio, Arlington, Massachusetts, USA; New World Medical, Rancho Cucamonga, California, USA; Omega Ophthalmics, Lexington, Kentucky, USA; Oculus Inc, Arlington, Washington, USA; Ono Pharma, Trenton, New Jersey, USA; PolyActiva, Melbourne, Australia; Sanoculis, Ono, Israel; Science Based Health, Houston, Texas, USA; Stroma Inc, Laguna Beach, California, USA; Transcend Medical, Menlo Park, California, USA; TrueVision Inc, Santa Clara, California, USA. The following authors have no financial disclosures: Panos G. Christakis, Dongyu Zhang, and James C. Tsai. All authors attest that they meet the current ICMJE criteria for authorship.

REFERENCES 4. Minckler DS, Francis BA, Hodapp EA, et al. Aqueous shunts in glaucoma: a report by the American Academy 1. Heijl A, Leske MC, Bengtsson B, et al. Reduction of intraoc- of Ophthalmology. Ophthalmology 2008;115(6): ular pressure and glaucoma progression: results from the Early 1089–1098. Manifest Glaucoma Trial. Arch Ophthalmol 2002;120(10): 5. Broadway DC, Chang LP. Trabeculectomy, risk factors for 1268–1279. failure and the preoperative state of the conjunctiva. J Glau- 2. The Advanced Glaucoma Intervention Study (AGIS): 7. coma 2001;10(3):237–249. The relationship between control of intraocular pressure 6. Gedde SJ, Herndon LW, Brandt JD, et al. Postoperative com- and visual field deterioration. The AGIS Investigators. Am plications in the Tube Versus Trabeculectomy (TVT) study J Ophthalmol 2000;130(4):429–440. during five years of follow-up. Am J Ophthalmol 2012; 3. Prum BE Jr, Rosenberg LF, Gedde SJ, et al. Primary Open- 153(5):804–814e1. Angle Glaucoma Preferred Practice Pattern((R)) Guidelines. 7. Desai MA, Gedde SJ, Feuer WJ, et al. Practice preferences for Ophthalmology 2016;123(1):P41–P111. glaucoma surgery: a survey of the American Glaucoma

VOL. 176 POOLED DATA ANALYSIS OF THE ABC STUDY AND AVB STUDY 125 Society in 2008. Ophthalmic Surg Lasers Imaging 2011;42(3): tics, and intraoperative complications. Ophthalmology 2011; 202–208. 118(11):2172–2179. 8. Gedde SJ, Schiffman JC, Feuer WJ, et al. Treatment out- 21. Christakis PG, Tsai JC, Kalenak JW, et al. The Ahmed versus comes in the Tube Versus Trabeculectomy (TVT) study after Baerveldt study: three-year treatment outcomes. Ophthal- five years of follow-up. Am J Ophthalmol 2012;153(5): mology 2013;120(11):2232–2240. 789–803e2. 22. Christakis PG, Kalenak JW, Zurakowski D, et al. The Ahmed 9. HaiBo T, Xin K, ShiHeng L, Lin L. Comparison of Ahmed Versus Baerveldt study: one-year treatment outcomes. glaucoma valve implantation and trabeculectomy for glau- Ophthalmology 2011;118(11):2180–2189. coma: a systematic review and meta-analysis. PLoS One 23. Barton K, Feuer WJ, Budenz DL, et al. Three-year treatment 2015;10(2):e0118142. outcomes in the Ahmed Baerveldt comparison study. 10. Tsai JC, Johnson CC, Kammer JA, Dietrich MS. The Ahmed Ophthalmology 2014;121(8):1547–1557e1. shunt versus the Baerveldt shunt for refractory glaucoma II: 24. Budenz DL, Barton K, Feuer WJ, et al. Treatment outcomes in longer-term outcomes from a single surgeon. Ophthalmology the Ahmed Baerveldt Comparison Study after 1 year of 2006;113(6):913–917. follow-up. Ophthalmology 2011;118(3):443–452. 11. Tsai JC, Johnson CC, Dietrich MS. The Ahmed shunt versus 25. Heuer D, Barton K, Grehn F, et al. Consensus on definitions the Baerveldt shunt for refractory glaucoma: a single-surgeon of success. In: Shaarawy T, SM, Grehn F, eds. Guidelines on comparison of outcome. Ophthalmology 2003;110(9): Design and Reporting of Surgical Trials. Amsterdam, The 1814–1821. Netherlands: Kugler Publications; 2008:15–24. 12. Nguyen QH, Budenz DL, Parrish RK 2nd. Complications of 26. Centre for Evidence Based Medicine. Available at http:// Baerveldt glaucoma drainage implants. Arch Ophthalmol www.cebm.net. Accessed June 25, 2016. 1998;116(5):571–575. 27. Ishida K, Netland PA, Costa VP, et al. Comparison of poly- 13. Stein JD, McCoy AN, Asrani S, et al. Surgical management propylene and silicone Ahmed Glaucoma Valves. Ophthal- of hypotony owing to overfiltration in eyes receiving glau- mology 2006;113(8):1320–1326. coma drainage devices. J Glaucoma 2009;18(8):638–641. 28. Seah SK, Gazzard G, Aung T. Intermediate-term outcome of 14. Syed HM, Law SK, Nam SH, et al. Baerveldt-350 implant Baerveldt glaucoma implants in Asian eyes. Ophthalmology versus Ahmed valve for refractory glaucoma: a case- 2003;110(5):888–894. controlled comparison. J Glaucoma 2004;13(1):38–45. 29. Allan EJ, Khaimi MA, Jones JM, et al. Long-term efficacy of 15. Wang JC, See JL, Chew PT. Experience with the use of Baer- the Baerveldt 250 mm2 compared with the Baerveldt veldt and Ahmed glaucoma drainage implants in an Asian 350 mm2 implant. Ophthalmology 2015;122(3):486–493. population. Ophthalmology 2004;111(7):1383–1388. 30. Heuer DK, Lloyd MA, Abrams DA, et al. Which is better? 16. Goulet RJ 3rd, Phan AD, Cantor LB, WuDunn D. Efficacy of One or two? A randomized clinical trial of single-plate versus the Ahmed S2 glaucoma valve compared with the Baerveldt double-plate Molteno implantation for in aphakia 250-mm2 glaucoma implant. Ophthalmology 2008;115(7): and pseudophakia. Ophthalmology 1992;99(10):1512–1519. 1141–1147. 31. Molteno AC, Fucik M, Dempster AG, Bevin TH. Otago 17. Budenz DL, Barton K, Gedde SJ, et al. Five-year treatment Glaucoma Surgery Outcome Study: factors controlling outcomes in the Ahmed Baerveldt comparison study. capsule fibrosis around Molteno implants with Ophthalmology 2015;122(2):308–316. histopathological correlation. Ophthalmology 2003; 18. Christakis PG, Kalenak JW, Tsai JC, et al. The Ahmed 110(11):2198–2206. Versus Baerveldt Study: five-year treatment outcomes. 32. Freedman J, Iserovich P. Pro-inflammatory cytokines in glau- Ophthalmology 2016;123(10):2093–2102. comatous aqueous and encysted Molteno implant blebs and 19. Barton K, Gedde SJ, Budenz DL, et al. The Ahmed Baerveldt their relationship to pressure. Invest Ophthalmol Vis Sci Comparison Study methodology, baseline patient character- 2013;54(7):4851–4855. istics, and intraoperative complications. Ophthalmology 33. Choritz L, Koynov K, Renieri G, et al. Surface topographies of 2011;118(3):435–442. glaucoma drainage devices and their influence on human 20. Christakis PG, Tsai JC, Zurakowski D, et al. The Ahmed tenon fibroblast adhesion. Invest Ophthalmol Vis Sci 2010; Versus Baerveldt study: design, baseline patient characteris- 51(8):4047–4053.

126 AMERICAN JOURNAL OF OPHTHALMOLOGY APRIL 2017