Effectiveness of anti-vascular endothelial growth factors in neovascular age-related macular degeneration and variables associated with visual acuity outcomes: Results from the EAGLE study

Objective To assess the overall effectiveness of anti-vascular endothelial growth factor (VEGF) therapy in treatment-naïve patients with neovascular age-related macular degeneration (nAMD) in a clinical practice setting. Study design EAGLE was a retrospective, 2-year, cohort observational, multicenter study conducted in Italy that analyzed secondary data of treatment-naïve patients with nAMD. The primary endpoint evaluated the mean annualized number of anti-VEGF injections at Years 1 and 2. The main secondary endpoints analyzed the mean change in visual acuity (VA) from baseline and variables associated with visual outcomes at Years 1 and 2. Results Of the 752 patients enrolled, 745 (99.07%) received the first dose of anti-VEGF in 2016. Overall, 429 (57.05%) and 335 (44.5%) patients completed the 1- and 2-year follow-ups, respectively. At baseline, mean (standard deviation, SD) age was 75.6 (8.8) years and the mean (SD) VA was 53.43 (22.8) letters. The mean (SD) number of injections performed over the 2 years was 8.2 (4.1) resulting in a mean (SD) change in VA of 2.45 (19.36) (P = 0.0005) letters at Year 1 and −1.34 (20.85) (P = 0.3984) letters at Year 2. Linear regression models showed that age, baseline VA, number of injections, and early fluid resolution were the variables independently associated with visual outcomes at Years 1 and 2. Conclusions The EAGLE study analyzed the routine clinical practice management of patients with nAMD in Italy. The study suggested that visual outcomes in clinical practice may be improved with earlier diagnosis, higher number of injections, and accurate fluid resolution targeting during treatment induction.

The study protocol was reviewed and approved by Institutional Review Boards or Independent Ethics Committees. The complete list of investigational sites and related Ethics Committees are listed in S1 Table. All required local approvals from Ethics Committees were obtained before commencing data collection at each site.

Key eligibility criteria
EAGLE enrolled treatment-naïve patients with a confirmed nAMD diagnosis, who started onlabel anti-VEGF therapy between 1 st January 2016 and 31 st December 2016. Patients provided written and signed informed consent for study inclusion and reviewing of charts.

Study objectives and endpoints
The primary objective was to evaluate anti-VEGF injections performed in clinical practice in patients with nAMD, treated for the first time with an anti-VEGF licensed for intraocular use, with respect to the mean (annualized) total number of anti-VEGF injections at Year 1, Year 2, and over 2 years.
The key secondary objectives were to (1) evaluate changes in VA from baseline at Years 1 and 2 in the treated eye, (2) evaluate factors associated with VA outcomes in the treated eye at Years 1 and 2 (age, gender, baseline VA, time from diagnosis to treatment, baseline type of macular neovascularization [MNV]; [14]), loading phase [LP], number of injections in the first year of treatment, and bilateral diagnosis), (3) evaluate factors associated with VA outcomes in the treated eye at Years 1 and 2 in the subgroup of patients who completed the LP, defined as patients receiving at least the first 3 injections in 90 days, and (4) estimate the median survival time of observation (overall exposure) from the index date to specific time points (6,12,18, and 24 months) stratified by baseline VA.

Assessments
Effectiveness assessments included annualized number of anti-VEGF injections to evaluate mean values and absolute changes in VA (Early Treatment Diabetic Retinopathy Study [ETDRS]) at Years 1 and 2 after the start of anti-VEGF therapy in the treated eye (compared with baseline). The association of variables such as age, gender, baseline VA, time from diagnosis to treatment, number of injections in the first year of treatment, LP, bilateral diagnosis and baseline type of MNV lesion on VA outcomes of the treated eye were assessed by means of linear regression models at Years 1 and 2 in the whole population and in the subgroup of patients who completed the LP. LP patients were classified as 'wet' or 'dry' based on the presence or absence of fluid in the treated eye at the corresponding optical coherence tomography (OCT) evaluation and by investigator's judgment.

Statistical analysis
Owing to the descriptive nature of the study, the statistical analyses associated with the primary endpoint (number of injections) and the secondary endpoint (change in VA) are descriptive; therefore, no formal statistical hypotheses have been stated. Sample size calculations were estimated using MedCalc Statistical Software version 18.5 (MedCalc Software bvba, Ostend, Belgium) and all statistical analyses were performed using software R, version 3.6.3. Sample size calculations referred to the desired precision for the main outcome estimate (i.e., the average number of injections per year). The main secondary endpoint (i.e., the estimated mean change in VA) was also taken into account. With respect to the primary outcome, it was calculated that 668 patients were required to estimate the mean number of injections per year with a 99% probability of obtaining a confidence interval (CI) with a width of not more than 1, assuming a standard deviation (SD) of 5 for the mean number of injections. Regarding the main secondary endpoint, it was calculated that 668 patients were also required to estimate the mean change from a baseline score in VA based on letter count with a 99% probability of obtaining a CI with a width of not more than 2 letters, assuming a SD for the difference distribution of 10. Assuming that 10% of enrolled patients will have only one measure, at least 742 patients were expected to be enrolled. During the enrollment period, this size was respected considering a substantial loss because some patients refused to give their informed consent or due to difficulty in reaching them 3 years after the start of the therapy; thus, around 1300 patients' charts were needed to be screened.
The primary endpoint of the study was estimated in terms of mean (annualized) total number of injections calculated at the end of the first year (Month 12), the second year (i.e., between Months 13 and 24), and overall, at Year 2 (Month 24) and reported with the corresponding 95% CI. The following analysis populations/groups were considered: Overall Exposed (OE, defined as all enrolled patients who received at least one dose of anti-VEGF treatment during the enrollment period i.e., from 1 st January to 31 st December 2016) and Efficacy Analysis (EA, defined as all patients in the OE who had a baseline and at least one postbaseline VA assessment) sets categorized by i) type of MNV lesion at baseline, ii) baseline VA and visual impairment classes for the treated eye, iii) LP versus no loading phase (NLP) sets and iv) 'dry' versus 'wet' patients after the LP.
To evaluate factors associated with VA outcomes, linear regression models were estimated at 1 and 2 years in the First-year Completer Analysis set (1stCA_EA) and Second-year Completer Analysis set (2ndCA_EA) populations, using the following covariates: age, gender, baseline VA, time from diagnosis to treatment, baseline type of MNV lesion, bilateral diagnosis, number of injections in the first year of treatment, and LP completed. Retinal fluid at the end of LP was evaluated as a covariate (instead of LP completed) in similar regression models on VA outcomes at 1 and 2 years in the groups of wet and dry patients with an available VA assessment at 1 and 2 years, respectively.
The subset of independent variables associated with the mean change in VA was estimated using a multivariable analysis approach, starting from a full model and retaining variables with P<0.05. As a supplementary analysis, to describe changes in VA during the follow-up using all available measurements per patient, a regression Linear Mixed Effects Model (LMM) with a random intercept for each patient was implemented using time as covariate (modeled as a restricted cubic spline).

Patient disposition, demographics, and baseline ocular characteristics
Patient disposition. Of the 1336 patients screened for study enrollment, 752 were deemed eligible and signed informed consent, and were therefore included as the Enrolled Population (EP). Of these, 745 (99.07%) were in the OE set and 617 (82.05%) were included in the EA set. In the EA set, 429 patients had an available follow-up evaluation of VA at least 1 year after the first injection and were categorized as First-year Completer Analysis set (1stCA_EA), while 335 had an available follow-up evaluation of VA at least 2 years after the first injection (2ndCA_EA). Furthermore, 452 patients from the EA set completed the LP (LP population). Of the 366 patients in the LP who had an available OCT performed between the end of LP and the 4 th injection, 191 were classified as 'wet' nAMD and 175 were classified as 'dry' nAMD as deemed by the investigator based on the presence or absence of fluid after the LP (Fig 1). Table 1, the EA population was nearly all Caucasian (612 [99.19%]) and more than half were female (EA: 335 [54.29%]). The mean (SD) age at the index date was 75.5 (8.7) years. The baseline mean VA, Screened population (SCR): All patients who were screened, including those who did not give consent, but were contacted by the investigators; Enrolled population (EP): All screened patients who were eligible (i.e. fulfilled all inclusion and exclusion criteria) and who gave consent to participate in the study or were dead; Overall Exposed population (OE): All enrolled patients who received at least one dose of an anti-VEGF treatment during the enrollment period; Effectiveness Analysis set (EA): All patients in the OE who had a baseline and at least one post-baseline assessment of VA; Excluded: All patients included in the OE but not in the EA; Loading Phase population (LP): All patients in the EA who received at least 3 injections within 3 months (90 days) from the index date, with the date of the end of the LP for each patient being the date of the third injection in this time window; Not complete Loading Phase (NLP): All patients included in the EA but not in the LP; First-year Completer Analysis set (1stCA _EA): All patients in the EA with an available follow-up evaluation of VA at least 1 year after the first injection; Second-year Completer Analysis set (2ndCA _EA): All patients in the EA with an available follow-up evaluation of VA at least 2 years after the first injection; First-year OCT completer analysis set (1stCA_OCT): All patients in the LP with an available follow-up OCT assessment at least 1 year after the first injection; Second-year OCT completer analysis set (2ndCA _OCT): All patients in the LP with an available follow-up OCT assessment at least 2 years after the first injection. Wet and Dry classification (patients in the LP with an available OCT evaluation performed after the end of LP and before the date of the subsequent injection): A patient was classified as "dry" if at the corresponding OCT evaluation there was no presence of fluid in the treated eye based on investigator's judgment, while if at the corresponding OCT evaluation there was presence of fluid in the treated eye based on investigator's judgment, he/she was classified as "wet".

Annualized number of anti-VEGF injections
The mean (SD) number of anti-VEGF injections in the EA set was 5.6 (± 2.5) at Year 1, 3.0 (± 3.1) during Year 2 and 8.2 (± 4.1) during the overall 2-year study period (Fig 2). The median (Q1;Q3) time from diagnosis to treatment was 16 (7;33) days in the EA set ( Table 1). The number of injections and time from diagnosis were comparable between the EA and OE sets (S2 Fig; S2

VA outcomes over time after anti-VEGF therapy
At Year 1, the mean (SD) VA increased by 2.45 (±19.36) letters compared with baseline; the change was statistically significant (P = 0.0005). Patients who were kept on treatment throughout the observation period maintained their VA with a mean (SD) VA change at Year 2 from baseline that was -1.34 (±20.85) letters with P = 0.3984 (S4 Table).
Furthermore, to describe changes in VA through follow-up using all available measurements per patient, a LMM regression model with a random intercept for each patient was implemented using time as a covariate (modelled as a restricted cubic spline). As shown in Fig  3, when using all available VA measures per patient, a significant positive time effect was detected followed by a decrease in VA after the first 200 days post-baseline VA assessment.
To evaluate factors associated with VA outcomes, linear regression models were estimated at Years 1 and 2 in the 1stCA_EA and 2ndCA_EA populations, using the following covariates: age, gender, baseline VA, time from diagnosis to treatment, baseline type of MNV lesion, number of injections during the first year, bilateral diagnosis, and loading phase completion. The subset of independent factors associated with VA outcomes at 1 and 2 years were age  Table 3).
The persistence of retinal fluid at the end of the LP was also evaluated as a covariate in similar regression analysis on VA outcomes in the subgroup of LP patients with available VA assessments at 1 and 2 years. Results indicated that the presence of retinal fluid after the LP was significantly associated with VA outcomes: VA at Years 1 and 2 decreased on average by 4.7143 (P = 0.0364) and 5.0244 (P = 0.0536) letters, respectively, for patients with persistent fluid compared with patients with 'dry' retina ( Table 4).

Safety outcomes
No systemic safety data were collected owing to the retrospective nature of the study.

Discussion
EAGLE was a 2-year, cohort observational, retrospective, multicenter study in Italy that evaluated the effectiveness of anti-VEGF therapy conducted on secondary data retrieved directly from hospital charts. It is also the first study in Italy to provide insights about the management of treatment-naïve patients with nAMD in clinical practice and to evaluate factors affecting responses to anti-VEGF treatment after 1 and 2 years. The mean number of anti-VEGF injections in the EAGLE study at Year 1 was 5.6, an improvement on the previous routine clinical practice study, AURA (5.2 in 2 years for Italy) [15]. Despite this improvement, the mean number of injections clearly shows that patients in the EAGLE study were undertreated with respect to pivotal randomized controlled trials [10][11][12]16] and other routine clinical practice studies, particularly those adopting a proactive and customized treatment approach, with the goal of preventing disease recurrence [17,18].
The completion of a loading scheme was important for receiving anti-VEGF injections at a higher frequency during both the first and second year of follow-up, as well as overall (overall period: LP, 8.7; NLP, 6.7; P<0.0001) and data demonstrates that more injections correlates with better VA outcomes. Other than completion of the LP, a better VA at baseline was the factor associated with a higher number of injections. In particular, data suggests that patients with better baseline VA were able to follow more appropriate treatment in terms of number of injections and treatment persistence. The reasons for this could be two-fold; lower baseline VA could be associated with increased fibrosis and disorganization of neurosensory layers at baseline and thus, a more rapid evolution of scar. Furthermore, physician might be reluctant to treat patients with a lower chance of improving VA based on the risk-benefit ratio assessment and the way patients with better baseline VA might be more motivated to continue therapy [19]. Although the reasons that confer a lower baseline VA at diagnosis may be multiple (such as age, other ocular pathologies, presence of geographic atrophy), data suggest the need for earlier diagnosis, earlier treatment, and better awareness on disease chronicity [20], the improvement of which might contribute to reduction in the high number of patients lost to follow-up in the first year of therapy (~43% (n = 323)). With respect to functional results, mean VA increased during Year 1 of treatment (2.45 letters; P = 0.0005) and was maintained in Year 2 (-1.34 letters; P = 0.398). In line with reports in other real-world evidence studies, the extent of VA gain and its maintenance over time is associated with injection frequency [21][22][23]; receiving <7 injections in the first year does not guarantee a significant gain in letters with current available therapies [15,24]. A possible explanation for such an observation in clinical practice is the application of a flexible pro re nata regimen of treatment based on disease reactivation. This therapeutic scheme does not allow disease activity to be promptly detected and treated because monthly monitoring is challenging in clinical practice, resulting in the recurrence of exudative changes which culminate in unsatisfactory clinical outcomes [25]. In contrast, patients receiving proactive regimens (such as treat-and-extend or fixed dosing) are more likely to receive an adequate number of injections allowing VA gains in the first year that are maintained during follow-up years [22]. Apart from number of injections, the literature indicates that older age and a higher proportion of follow-up visits with active MNV lesions are associated with poor VA outcomes in patients with nAMD [26].
To better understand the variables associated with VA outcomes in Italian clinical practice, a multiple regression analysis was conducted in this study. As mentioned previously, the number of injections given in Year 1 was one of the principal factors determining a better VA at Years 1 and 2. In addition, the analysis revealed that age and baseline VA are the variables that had the most impact on visual gains supporting available evidences. In EAGLE, the completion of the LP per se had no significant effect on VA gains at 1 and 2 years compared with NLP subgroup; bigger real-world studies, such as AURA [15] and LUMINOUS [24], reported a tendency but did not demonstrate a strong association. These data suggest that in routine clinical practice, patients might take longer than the mandated number of days to complete the LP (3 injections) which could possibly result in milder VA outcomes [27]. Remarkably, EAGLE data showed that~52% (191 out of 366) of patients who were assessed for fluid status after the LP, have unresolved fluid (wet) indicating this as one of the hurdles for better visual outcomes in current clinical practice, of which inappropriate LP might be one of the potential contributors.
Multivariable analysis in patients who completed the LP at Years 1 and 2, revealed that the presence of retinal fluid at the end of LP correlated with worse VA outcomes at Years 1 and 2. This even more highlights the importance of achieving an early dry condition to maintain a better VA in the long term. This observation from a clinical practice setting further confirmed what has already been demonstrated in post hoc analyses of pivotal studies [28,29], where early fluid control determined successful management of patients with nAMD, thereby decreasing the burden associated with number of injections. Thus, achieving a 'dry' retina is deemed to be more important for achieving better or desired long-term VA outcomes.
The results of EAGLE are representative of patients enrolled in an observational real-life setting and may not necessarily apply to all patients with nAMD. This study was subject to the limitations inherent to medical chart reviews. Moreover, this observational, retrospective study presented some methodological limitations, such as different clinical centers using different OCT machines and different retreatment criteria, which were not standardized. Estimations of VA measurements are possible owing to conversions (decimals to ETDRS). MNV lesion type was classified based on the data entered into CRFs from different centers or OCT assessments. Furthermore, the study might have unmeasured confounders (i.e., latent characteristics not taken into account in the regression models such as reactive or proactive individualized strategy).
To conclude, EAGLE represents a comprehensive database of nAMD patients, showing a reliable picture of management with approved anti-VEGF drugs in Italy from 2016-2018. As observed in other routine clinical practices, the present analysis showed a reduced number of injections and a great loss of patients at follow-up, suggesting that treatment centers could deliver a limited number of injections that are carried out for a limited period and varied from one patient to another. The data analyzed from this chart review confirmed that absolute VA gains were always higher in patients with better VA at baseline, strengthening the need of prompt treatment with anti-VEGF drugs following diagnosis of nAMD. Patients with lower baseline VA (<58 letters) are usually lost early to follow-up and received an average of one less injection in the first year and 2 fewer injections in the second year compared with the rest of the study population. Moreover, the study demonstrated that in the clinical practice setting, early fluid resolution is a critical factor for achieving a better functional outcome. This study identified that in Italy, like many other countries, there is a critical unmet medical need linked to under-treatment of patients. More effective therapies targeting retinal fluid along with individualized proactive treatment strategies and stricter follow-ups are needed to achieve appropriate patient treatment in terms of efficacy, despite the limited treatment capability of the Italian healthcare system.