Six and eight weeks injection frequencies of bevacizumab are non-inferior to the current four weeks injection frequency for quality of life in neovascular age-related macular degeneration: a randomized controlled trial

This is a secondary analysis of an RCT comparing three treatment regimens of bevacizumab (Avastin) for the treatment of ARMD on visual acuity and central retinal thickness [26]. A total of 191 patients were enrolled in a 1-year, prospective, open-label RCT which investigated the optimal injection frequency of bevacizumab injection for ARMD treatment at the Rotterdam Eye Hospital from June 2008 to March 2010 (Fig. 1). To be eligible, patients had to be at least 65 years old, have a best-corrected visual acuity of 20/200 to 20/20 (Snellen equivalent) in the study eye as assessed using Early Treatment Diabetic Retinopathy Charts (ETDRS), no previous ARMD treatment and active leakage. Patients were only treated in one eye. Fluorescein angiography (FA) and indocyanine green (ICG) angiography were used to observe leakage, and optical coherence tomography (OCT) was used to observe the presence of fluid [26]. Patients who had other significant ocular disorders, had allergies to either FA or ICG dye injections, were immunocompromised, using coumarin-derivatives, had experienced a clinically significant cerebrovascular accident or myocardial infarction or had a planned ocular surgery during the 1-year follow-up, were excluded. Written informed consent was obtained from all participants. After baseline measurements were completed, all eligible patients were randomized to an injection frequency of every 4, 6, or 8 weeks using a computer-based 1:1:1 ratio block randomization procedure.

Apart from the difference in frequency, treatment regimens were comparable among the three groups. At each outpatient visit, a dose of 1.25 mg bevacizumab was administered intravitreally. On top of the measures during regular outpatient visits, patients were assessed every 12 weeks by best-corrected visual acuity, spectral-domain OCT and funduscopy. Monthly checks for adverse events took place by questioning patients. Treatment was continuous for 1 year, independent of visual acuity change, spectral-domain OCT measures, or funduscopy findings. The 4 weeks, 6 weeks, and 8 weeks bevacizumab treatment regimens resulted in totals of 13, 9, and 7 injections and visits a year, respectively.

At baseline and at the final follow-up visit, patients were asked to complete the National Eye Institute 39-Item Visual Function Questionnaire (NEI VFQ-39) [27] and the 36-item Medical Outcomes Study Short-Form General Health Survey (SF-36) [28, 29]. The NEI VFQ-39 assesses vision-related QoL, while the SF-36 evaluates general QoL. Given the nature of the disease, both questionnaires were presented in a larger font size and often administered in the presence and sometimes with support of a caregiver and/or family member.

The primary outcome was vision-related QoL, measured as the composite score on the NEI VFQ-39 [27]. The NEI VFQ-39 consists of a 25-item base set of questions and 14 supplemental items. All items use a Likert-type scaling and five response categories, with occasionally a sixth category to opt out, except for two items that have 10 response options. Responses are converted into 12 vision-targeted multi-item subscales (0–100): general health, general vision, ocular pain, near activities, distant activities, social functioning, mental health, role limitations, dependency, driving, color vision, and peripheral vision. These 12 subscales can be summarized as a single composite score. A 10-point difference in either the sub-scales or the composite score of the NEI VFQ-39 is deemed clinically important, and thus considered a clinically meaningful change [30, 31]. The reliability of the NEI VFQ-39 in age-related macular degeneration varies from a Cronbach’s alpha of 0.86 to 0.96 [32, 33].

Another outcome measure was general QoL measured by the SF-36 [29]. This is a self-report questionnaire comprising 36 questions measuring different aspects of general health. All items use a Likert based scaling and use two to six response options. The responses are converted into eight multi-item subscales: physical functioning, role functioning physical, bodily pain, general health, vitality, social functioning, role functioning emotional, and mental health. These scales can be summarized as a psychometrically based ‘physical component summary’ (PCS), in which the first four scales are most heavily weighted, and a ‘mental component summary’ (MCS), in which the last four scales are most heavily weighted [34]. These summaries are transformed into T-scores with a mean of 50 and standard deviation of 10. Higher scores on SF-36 scales indicate a better quality of life. The UK version reliability of the physical subscale is 0.92, and the mental subscale is 0.89 [34]. Following the approach provided by Jacobson & Truax, the clinical significant change is 7.84 and 9.19 for the respective subscales [35].

Differences between dropouts and retained patients were analyzed with Student’s t- and chi square-tests. Baseline differences for continuous variables between the three groups were analyzed with One-way ANOVA with Bonferroni correction for pairwise differences. Chi square-tests were applied for binary variables and when significant, standardized residuals were evaluated to determine the deviating groups. The non-inferiority limit for the 6 weeks and 8 weeks groups comparison with the 4 weeks group was based on the 10-point clinical significant difference of the NEI VFQ, composite score and the subscales near vision, distance vision and role limitations. This negative 10-point difference indicated the lower end of the ‘region of therapeutic equivalence’ and, together with the maximum possible difference, enclosed the ‘region of non-inferiority’ [36]. The region of non-inferiority ranged from − 10 to 100. Non-inferiority was assumed whenever the 95% confidence interval of the difference in change fell entirely within this region [36]. Note that only the right-hand side of the distribution was relevant, Fig. 2.

In addition, differences between treatment groups were tested for the secondary SF-36 subscales. We applied multilevel linear regression analyses to evaluate differences in change in QoL between the three randomization groups. The patients formed the upper level, their repeated measures the lower level. These analyses can handle data with missing time points efficiently, i.e. data of patients without a follow-up can be included, without a need for imputation. For each outcome we applied a separate model. The random parts of the models only included the intercept. The fixed parts of the models included time (follow-up vs. baseline), centered baseline score, 6-weeks and 8-week frequencies and the interaction of time with baseline, six and eight weeks frequencies. The four-week frequency group served as reference group. In all analyses, gender and age were included as control variables.

The study was originally designed to detect differences in visual acuity, and subsequently powered with a non-inferiority limit of seven letters [26]. When testing QOL, a power analysis for non-inferiority was performed on the NEI VFQ-39 composite score. The clinical important difference for the NEI VFQ-39 is 10 and the standard deviation is 20, the one-sided alpha was set at 0.05 and power at 0.80, for which a sample size of 50 persons per group is needed. This implies that the sample size of 63–64 is sufficient.

All other analyses were performed with IBM SPSS version 24.0 “IBM Corp. Released 2016. IBM SPSS Statistics for Windows, Version 21.0. Armonk, NY: IBM Corp.”

This study was approved by the Erasmus Medical Research Ethics Committee (MEC-2007-254) in accordance with the Code of Ethics of the World Medical Association (Declaration of Helsinki) and was registered in the Dutch Trial Register (NTR 1174).

After randomization, 64 patients were treated in the 4 weeks group, 63 in the 6 weeks group, and 64 in the 8 weeks group. Treatment arms were well balanced with regard to baseline demographic characteristics, visual acuity, and other characteristics of the affected eye (Table 1). However, significant baseline differences were present for the NEI VFQ-39 as the 8 weeks group had lower scores than the 4 weeks group.

Patients lost to follow-up were subdivided based on their exit reasons (Table 1). The highest drop-out rate in the 4 weeks treatment group (29.7%) and the lowest in the 6 weeks group (9.5%) significantly differed, p = 0.004. Patients who dropped out had significantly worse baseline scores than retained patients on the physical component summary of the SF-36: t(176) =  − 2.95, p = 0.004 (not in Table 1). No other statistical significant differences were found.

The changes and differences estimated by the multilevel models are presented in Table 2, the total models are presented in Table 3. Observed differences are presented in Appendix 1 and the observed means and standard deviations in Appendix 2. The 95% confidence intervals of the difference in change scores showed that the composite score interval was well inside the [− 10, 100] point difference interval that represented the non-inferiority region for the three treatment comparisons (Fig. 2). For the subscales near activities, distant activities, role limitations, visual functioning and socio-emotional functioning the 95% confidence intervals of the differences were also entirely within the region of non-inferiority. This also barely holds for the near activities estimate for gain within 6 weeks (10.26) compared to gain within 8 weeks (6.43). This 95% confidence interval of − 9.91 to 2.24 is just within the limit.

The treatment did not significantly affect the SF-36 component summaries. All treatment effects of different injection frequencies were well within the non-inferiority limits (Table 2).