In total, 2,017,445 eyes from 1,555,917 patients were identified as having initiated anti-VEGF between 01 January 2015 and 31 December 2020, with no evidence of injections in the preceding 12 months. Of these, 217,696 eyes from 166,868 patients met the eligibility criteria and were included in the study (Table 1).

The initial dose cohort included 77,769 eyes from 65,867 patients, and the non-initial dose cohort contained 139,927 eyes from 113,383 patients. Some patients had one eye meeting the criteria for the initial dose cohort and another eye that did not; thus, some patients are counted in both groups.

The baseline demographic and clinical characteristics are shown in Table 2 and supplementary Table S1.

The mean (SD) length of follow-up was differed slightly between cohorts: 2.87 (1.61) years for the initial dose cohort and 2.94 (1.67) years for the non-initial dose cohort.

The mean age was slightly higher in the initial dose cohort, and the cohorts had slightly different distributions of males vs. females. Most eyes were from White/Caucasian patients, and these were slightly more numerous in the initial than non-initial dose cohort (67.8% [n = 52,754] initial; 62.3% [n = 87,147] non-initial). Compared with those in the non-initial dose cohort, fewer eyes in the initial dose cohort were from patients of Hispanic ethnicity (14.5% [n = 11,242] initial; 17.5% [n = 24,447] non-initial).

Beginning treatment with or without initial doses was similarly common in all census regions, but overall, the largest number of eyes were enrolled from patients in the South (42.7% [n = 33,180] initial; 42.5% [n = 59,441] non-initial]). Just over half of patients had Medicare insurance, and approximately a further third had commercial insurance, with the proportions slightly different between cohorts. Over 86% of patients were treated by a retina specialist (Table 2). Eyes with proliferative diabetic retinopathy were less commonly treated with initial doses than without initial doses (39.1% [n = 30,382] initial; 50.7% [n = 70,944] non-initial; supplementary Table S1.

Over the average follow-up period, the number of intravitreal injections (including initial doses) of anti-VEGF agents was higher in eyes that received initial doses than in eyes that received no initial doses (mean [SD] intravitreal injections: 11.6 [8.9] initial; 6.1 [6.8] non-initial). Most patients had ≤ 3 years of follow up (56.9%, n = 34,878).

At Year 1 of follow-up, the mean (SD) number of intravitreal injections was 6.3 (2.5) for eyes in the initial dose cohort and 2.9 (2.1) (Fig. 1A) for eyes in the non-initial dose cohort. 30% (n = 23,435) of eyes in the initial dose cohort had received only 3–4 injections in Year 1, compared with 22.8% (n = 31,913) of those in the non-initial dose cohort. Similarly, 37.4% (n = 29,059) of eyes in the initial dose cohort received only 5–7 injections in Year 1, compared with 15.8% (n = 22,075) in the non-initial dose cohort.

A Non-annualized number of injections and (B) injection interval at Years 1–6, stratified by initial dose. Error bars are standard deviation. I, initial dose cohort; NI, non-initial dose cohort

At Year 6 of follow-up, the mean (SD) number of intravitreal injections was 3.5 (2.5) for eyes in the initial dose cohort and 3.0 (2.1) for eyes in the non-initial dose cohort. In Year 6, 26.6% (n = 980) eyes in the initial dose cohort received 3–4 injections compared with 28.3% (n = 1679) in the non-initial dose cohort. The numbers that received 5–7 injections in Year 6 were lower: 21.6% (n = 797) in the initial dose cohort and 16.5% (n = 979) in the non-initial dose cohort.

Over the average follow-up period, mean (SD) injection intervals were shorter for the eyes that received initial doses (7.6 [2.8] weeks) compared with those that did not (12.6 [7.7] weeks; Fig. 1B). Most eyes in the initial dose cohort (59.5%; n = 46,279) experienced injections at intervals > 5 weeks to ≤ 9 weeks, compared with 30,599 (44.3%) in the non-initial dose cohort. Intervals of > 13 to ≤ 18 weeks, and of > 18 weeks were far more numerous among eyes in the non-initial dose cohort than in the initial dose cohort (> 13 to ≤ 18 weeks: 4.7% [n = 3656] initial vs. 12.6% [n = 18,233] non-initial; and > 18 weeks: 0.6% [n = 504] initial versus 10.8% [n = 17,384] non-initial; supplementary Table S2.

Mean (SD) baseline VA was similar between eyes in the initial dose cohort and in the non-initial dose cohort (63.0 [18.1] vs. 62.5 [19.8] ETDRS letters, respectively; Fig. 2; supplementary Table S3. After 100 days, mean (SD) VA was slightly higher for eyes in the initial dose cohort than in the non-initial dose cohort (67.4 [15.9] vs. 65.94 [18.0] ETDRS letters). This trend continued across 1–5 years of follow-up, although the difference in mean VA reduced with each additional year of follow-up (mean [SD] VA in initial vs. non-initial doses, respectively, was 68.0 [16.3] vs. 66.2 [18.4] at Year 1 [P < 0.0001]; 67.6 [16.9] vs. 65.9 [18.7] at Year 2 [P < 0.0001]; 67.2 [17.3] vs. 65.8 [18.9] at Year 3 [P < 0.0001]; 66.5 [17.9] vs. 65.6 [18.9] at Year 4 [P < 0.0001], 66.0[18.1] vs. 65.2 [19.1] at Year 5 [P = 0.0041], and 65.3 [18.2] vs. 65.9 [18.2] at Year 6 [P < 0.0001]).

Visual acuity outcomes in initial and non-initial dose cohorts. ETDRS, Early Treatment Diabetic Retinopathy Study; I, initial dose cohort; NI, non-initial dose cohort; VA, visual acuity

At any point in the study, VA improvement of ≥ 15 letters was more common among eyes that received initial doses than those that did not (37.7% [n = 21,204] initial; 31.6% [n = 31,979] non-initial; Fig. 3).

Eyes with 15-letter improvement at (A) any point during the study, and (B) at the end of the study, stratified by receipt of initial dose. I, initial dose cohort; NI, non-initial dose cohort; VA, visual acuity

By the end of the study, 20.3% (n = 11,395) of eyes in the initial dose cohort, and 18.2% (n = 18,408) in the non-initial dose cohort had achieved 15-letter improvements. Conversely, 10.3% (n = 5766) and 12.1% (n = 12,221) of eyes in the initial and non-initial dose cohorts, respectively, had experienced 15-letter declines by the end of the study. In both cohorts, VA improvement was most prevalent among eyes that initiated anti-VEGF treatment with worse than 20/40 vision, while 15-letter declines did not correlate with baseline VA.

Switching treatments was less common among eyes that received initial doses than eyes that did not receive initial doses. Overall, one quarter (n = 17,028) of eyes in the initial dose cohort and just under a third (n = 22,481) of eyes in the non-initial dose cohort underwent a treatment switch. The median time to the first switch was 39 and 38 weeks, respectively (Fig. 4).

Treatment switches among eyes with ≥ 4 injections in initial and non-initial dose cohorts (A). Change in VA from baseline at time of switch (B). I, initial dose cohort; NI, non-initial dose cohort

In both groups, patients who discontinued did so after experiencing improvements in VA. The mean (SD) improvement from baseline to discontinuation was + 3.87 (13.9) letters in the initial-dose cohort and + 1.14 (10.9) letters in the non-initial dose cohort. Those who re-initiated did so after VA had worsened, by a mean (SD) 6.73 (17.6) and 4.79 (19.0) letters in the initial and non-initial dose cohorts, respectively, from the time of discontinuation, and by a mean (SD) 2.83 (19.4) and 3.62 (19.9) letters, respectively, from baseline (Fig. 5).

Treatment discontinuation, re-initiation, and switching in initial and non-initial dose cohorts. Mean (SD) VA change, baseline to discontinuation: I: +3.87 (13.9) letters, NI: +1.14 (10.9) letters. Mean (SD) VA change, baseline to re-initiation: I: −2.83 (19.4) letters, NI: −3.62 (19.9) letters. Discontinuation to re-initiation: I: −6.73 (17.6) letters, NI: −4.79 (19.0) letters. Discontinuation defined as ≥ 365 days of contribution to the IRIS® Registry without an anti-VEGF injection. The mean (SD) number of injections received from index to discontinuation was 6.83 (4.8) among patients in the initial dose cohort and 2.82 (3.2) among patients in the non-initial dose cohort. I, initial dose cohort; NI, non-initial dose cohort; SD, standard deviation; VA, visual acuity; VEGF, vascular endothelial growth factor

Switching away from bevacizumab to FDA-approved agent(s) was more common than the other way around for both cohorts. In the initial and non-initial dose cohorts, respectively, 18.4% (n = 12,762) and 22.9% (n = 16,310) switched away from bevacizumab, while 1.8% (n = 1251) and 4.2% (n = 3003) switched to bevacizumab.

Both treatment discontinuation and subsequent re-initiation were less common among eyes in the initial dose cohort. Less than half (45.7%; n = 35,558) of the eyes in the initial dose cohort discontinued treatment, compared with nearly two thirds (63.8%; n = 89,285) of those in the non-initial dose cohort. A similar trend was evident with eyes re-initiating treatment: 17.2% (corresponding to 37.7% of the patients discontinuing, n = 13,392) and 25.0% (corresponding to 39.2% of patients discontinuing, n = 35,015) of eyes in the initial and non-initial dose cohorts re-initiated treatment, after a median treatment-free interval of 81.7 (range: 52.3–354.1) and 84.0 (range: 52.3–384.9) weeks (Fig. 5).

After controlling for baseline characteristics, receiving initial dosing was associated with a slight increase in VA from baseline to Year 1 (+ 1.23 letters), Year 3 (+ 1.47 letters) and Year 5 (+ 1.23 letters) compared with eyes that did not receive the initial dosing regimen. Similarly, initial doses were associated numerically with VA improvements from Year 1 to Year 3 (+ 0.51 letters) and to Year 5 (+ 0.53 letters). All baseline VA statuses were significantly associated with vision gains following anti-VEGF initiation (P < 0.01). Across all models, having Medicaid insurance was significantly associated with vision loss following anti-VEGF initiation (P < 0.01), as was starting bevacizumab as a first agent and receiving multiple anti-VEGF agents during the follow-up period (P < 0.05; Table 3).

GEE modelling revealed that a patient’s race/ethnicity and insurance status significantly impacted the likelihood of receiving initial doses. Asian (− 0.17), Black (− 0.23), and patients of other (− 0.11) and unknown race (− 0.20) had a significantly lower expected likelihood (P < 0.01) of receiving initial doses than White patients, adjusted for other baseline demographic and clinical factors. Similarly, patients of Hispanic ethnicity were significantly less likely to receive initial doses than their non-Hispanic counterparts (− 0.17; P < 0.01). Furthermore, patients insured with Medicare (− 0.05) or Medicaid (− 0.12) were significantly less likely (P < 0.01) to receive initial doses than patients with commercial insurance. Compared with aflibercept, initial doses were significantly less likely when bevacizumab was given as the first treatment (− 0.20; P < 0.01), and significantly more likely when ranibizumab was given as the first treatment (+ 0.13; P < 0.01). Patients with baseline VA of worse than 20/25 to better than 20/200 were also more likely to receive initial doses than those starting treatment with vision 20/25 or better (Table 4).

In multivariate analyses, the strongest predictor of treatment discontinuation was starting with bevacizumab (HR: 1.7; 95% confidence interval [CI]: 1.67–1.73; P < 0.005). Interestingly, receiving < 3 initial doses was the strongest predictor of treatment continuation (HR: 0.51; 95% CI: 0.51–0.52; P < 0.005), followed by receiving multiple agents (HR: 0.62; 95% CI: 0.6–0.63; P < 0.005). Treatment re-initiation was less likely for patients aged 70–79 years (HR: 0.88; 95% CI: 0.82–0.94; P < 0.005), or 80–89 years (HR: 0.79; 95% CI: 0.73–0.85; P < 0.005), and those with baseline VA worse than 20/80 but better than 20/200 (HR: 0.92; 95% CI: 0.88–0.96; P < 0.005), or 20/200 or worse (HR: 0.79; 95% CI: 0.76–0.82; P < 0.005), compared with patients in other age or VA categories. Ages 60–69 and 50–59 at baseline were also strong predictors of switch (for ages 60–69, HR: 1.23; 95% CI: 1.15–1.33; P < 0.005, and ages 50–59, HR: 1.21; 95% CI: 1.12–1.3; P < 0.005). Receiving < 3 initial doses also predicted treatment switch (HR: 1.17; 95% CI: 1.15–1.2; P < 0.005; Fig. 6), although this may correlate with starting bevacizumab treatment.

Multivariate Cox proportional-hazard modelling of likelihood of treatment (A) discontinuation, (B) re-initiation, or (C) switch. VA, visual acuity