Methods

Study design

This is a multicentre, prospective, individually randomised, single-masked, active-controlled non-inferiority trial with concurrent economic evaluation that compared the clinical efficacy and cost-effectiveness of intravitreal aflibercept versus PRP in patients with treatment-naive PDR or post-PRP active retinal NV at 52 weeks. A subset of the participants also took part in a mechanistic evaluation substudy.

The study was registered on 10 July 2014 (as ISRCTN 32207582). Clinical trial authorisation was given by the Medicines and Healthcare products Regulatory Agency (MHRA): European Clinical Trials Database (EudraCT) 2013-003272-12. Ethics approval was granted by the National Research Ethics Service Committee London – South East; 14/LO/0203. Recruitment commenced in August 2014 and was completed to time and target in November 2015.

The trial protocol is available at www.journalslibrary.nihr.ac.uk/programmes/eme/126615. The trial steering and data monitoring committees provided independent oversight.

Setting and locations

The study was conducted at the 22 NHS clinical sites listed below. The sites were chosen based on previous clinical trial experience or by the estimated volume of potentially eligible patients. Interested sites completed a site feasibility questionnaire.

• Birmingham and Midland Eye Centre, Sandwell and West Birmingham NHS Foundation Trust, Birmingham.
• Brighton and Sussex University Hospitals NHS Trust, Brighton.
• Bristol Eye Hospital, Bristol.
• Essex County Hospital, Colchester.
• Frimley Park Hospital NHS Foundation Trust, Surrey.
• Hillingdon Hospitals NHS Foundation Trust, London.
• James Paget University Hospital, Great Yarmouth.
• King’s College Hospital, London.
• Leicester Royal Infirmary, Leicester.
• Maidstone & Tunbridge Wells NHS Trust, Kent.
• Moorfields Eye Hospital NHS Foundation Trust, London.
• Princess Alexandra Hospital, Harlow.
• Royal Bolton Hospital NHS Trust, Greater Manchester.
• Royal Liverpool & Broadgreen University Hospitals NHS Trust, Liverpool.
• Royal Victoria Hospital and Queen’s University, Belfast.
• Royal Victoria Infirmary, Newcastle upon Tyne.
• St James’s University Hospital, Leeds.
• Sunderland Eye Infirmary, Sunderland.
• Torbay Hospital, South Devon.
• University Hospital Southampton NHS Foundation Trust, Southampton.
• Wolverhampton & Midland Counties Eye Infirmary, Wolverhampton.
• York Hospital NHS Trust, York.

Participants

Inclusion criteria

1. Subjects of either sex aged ≥ 18 years.
2. Diagnosis of diabetes mellitus (type 1 or type 2).
3. BCVA in the study eye better than or equal to 54 ETDRS letters (Snellen visual acuity 6/18).
4. PDR with no evidence of previous PRP, or the presence of new or persistent retinal NV despite prior PRP that (1) requires treatment in the opinion of the investigator and (2) has sufficient space in the peripheral retina to perform more PRP treatment. In patients with both eye involvement, the eye with no PRP or the least number of PRP burns will be randomised as the study eye. If both eyes have had no previous PRP, the eye with the better visual acuity will be randomised as the study eye.
5. Media clarity, pupillary dilatation and subject co-operation sufficient for adequate fundus photographs. Eyes with mild preretinal haemorrhage or mild vitreous haemorrhage that does not interfere with clear visualisation of the macula and optic disc are considered eligible for this study.
6. Ability to give informed consent.
7. Female subjects should use effective contraception, be post-menopausal for at least 12 months prior to trial entry, or be surgically sterile.

Exclusion criteria

The following exclusions apply to the study eye only (i.e. they may be present for the non-study eye):

1. coexistent ocular disease that may interfere with visual outcome
2. moderate or dense vitreous haemorrhage that prevents clear visualisation of the macula and/or optic disc or prevents PRP treatment
3. significant fibrovascular proliferation or tractional retinal detachment in the posterior pole
4. prior vitrectomy
5. presence of macular oedema at baseline confirmed by SD-OCT as a CST of > 320 µm [SPECTRALIS optical coherence tomography (OCT) (Heidelberg Engineering, Germany)] because of the presence of morphological evidence of diffuse or cystoid oedema
6. other causes of retinal NV
7. iris or angle NV and NVG
8. anticipated need for cataract extraction or vitrectomy within the next 12 months
9. known allergy to fluorescein or any components of aflibercept formulation
10. previous intravitreal anti-VEGF or steroid treatment for DMO in the last 4 months
11. PRP in the last 8 weeks
12. aphakia
13. uncontrolled glaucoma as per investigator’s judgement
14. severe external ocular infection.

Exclusion criteria also apply to systemic conditions as follows:

1. The participant should not have a glycated haemoglobin (HbA_1C) level of > 12%. As a precautionary measure, normal health-care providers will be informed if any patient with a HbA_1C level of more than 8% is identified using a standard letter, directing the provider to the current National Institute for Health and Care Excellence (NICE) guidelines32 on the management of diabetes mellitus to ensure optimal follow-up.
2. The subject should not have a blood pressure (BP) of > 170/110 mmHg. As a precautionary measure, normal health-care providers will be informed if any patient with a BP of > 150/90 mmHg is identified. A standard letter will be provided directing the provider to the current NICE guidelines on the management of hypertension in patients with diabetes mellitus to ensure optimal follow-up of these patients.
3. A medical condition that, in the opinion of the investigator, would preclude participation in the study.
4. Myocardial infarction, stroke, transient ischaemic attack, acute congestive cardiac failure or any acute coronary event within 6 months of randomisation.
5. Dialysis or renal transplant.
6. Pregnant women.
7. Women of child-bearing potential who do not agree to use effective contraception during the study and for at least 3 months after the study has finished. Effective contraception is defined as one of the following:
   1. barrier method – condoms or occlusive cap with spermicides
   2. true abstinence – when it is in line with the preferred and usual lifestyle of the subject. Periodic abstinence (e.g. calendar, ovulation, symptothermal, post-ovulation methods) and withdrawal are not acceptable methods of contraception
   3. female sterilisation – have had tubal ligation or bilateral oophorectomy (with or without hysterectomy)
   4. male partner sterilisation – the vasectomised male partner should be the only partner for the female participant
   5. use of established oral, injected or implanted hormonal methods of contraception and intrauterine device
   6. Breastfeeding women.
8. Males who do not agree to use an effective form of contraception for the duration of the study and for 3 months after the study has finished. Effective contraception is defined as one of the following:
   1. Barrier method – condoms or occlusive cap with spermicides.
   2. True abstinence – when it is in line with the preferred and usual lifestyle of the subject. Periodic abstinence and withdrawal are not acceptable methods of contraception.
   3. Male sterilisation (vasectomy).
9. Participation in an investigational trial involving an investigational medicinal product (IMP) within 30 days of randomisation.

Randomisation

Randomisation was completed via a bespoke web-based randomisation system hosted at the King’s Clinical Trials Unit (KCTU) on a secure server (www.ctu.co.uk). Patients were randomised 1 : 1 at the level of the individual using the method of minimisation incorporating a random element. The minimisation factors were PDR status (naive PDR and non-naive PDR), HbA_1c level (< 8%, 8–10%, > 10%), diastolic BP (> 90 vs. ≤ 90 mmHg), BCVA (54–69 vs. ≥ 70 letters) and trial site. After informed consent was signed, a patient identification number (PIN) was generated by registering the patient on the MACRO electronic case report form (eCRF) system (InferMed Elsevier Macro v4.0, London, UK) and this PIN was used in the randomisation system. This unique PIN was also recorded on all source data worksheets and used to identify the patient throughout the study. Only authorised staff were allowed to log in to the randomisation system. Once a patient was randomised, the system automatically generated e-mails to key staff within the study. E-mails were also sent to site pharmacies to alert them to a patient’s treatment group – aflibercept or PRP therapy. The pharmacy department used this alert to cross-check the trial prescription to ensure that aflibercept was being dispensed for the correct patient. Additional e-mails were also generated from the randomisation system and sent to key trials staff, with or without treatment allocation information, depending on their role in the study.

Masking

The research optometrists were the primary outcome assessors and they conducted the visual acuity tests at screening, 12 and 52 weeks. They were masked to treatment allocation throughout the study. The optometrists received the participants into the visual acuity lanes with a visual acuity-specific source data worksheet that included the PIN and details of the study eye and non-study eye to be refracted, but with no previous records or case report forms by which the patient’s treatment arm could be identified. The optometrists also assessed the secondary outcome measure of contrast sensitivity and low-luminance visual acuity using the same technique of masking as above. At all other visits, visual acuity examiners were provided with a copy of the refraction log to conduct an open aperture visual acuity test in both eyes with the previous refraction. At these time points, visual acuity tests were also conducted by unmasked professionals and these data were not used for the primary outcome analysis. The other tests of secondary outcome measures, including visual fields and OCT scans, were performed by masked technicians. The technicians received the patients into the visual field and OCT room using the specific source data worksheet that provided details of the patient’s PIN and eye to be examined. After every visit, the completed source data worksheets were kept with the principal investigator’s team. The participants were also advised at enrolment that they must not discuss the study arm they were in with these assessors. The retinal photographs at screening, and at 12 and 52 weeks and FFA at screening and 52 weeks were graded by masked graders in the independent reading centres within the Network of Ophthalmic Reading Centres UK (NetwORC UK). The photographers were trained to take the photographs as per the standard operating procedures (SOPs) for this study. The graders in the Reading Centre were trained and quality assured to grade diabetic retinopathy based on the ETDRS grading system as required for this study. These masking procedures avoided both performance and detection bias.

Intervention

Intravitreal aflibercept (Bayer plc, Reading, UK) dosed at 2 mg/0.05 ml is approved by the US Food and Drug Administration (FDA) and European Medicines Agency (EMA) for wet age-related macular degeneration, DMO, retinal vein occlusion related macular oedema and myopic choroidal NV. Study eyes assigned to the aflibercept arm received an intravitreal injection of aflibercept 2 mg/0.05 ml at baseline, 4- and 8-week visits.

Patterns of regression

The treatment response was defined by degree of regression of retinal NV as assessed by colour photographs as per predefined SOP. The categories of regression patterns are summarised in Figure 1.

FIGURE 1 Classification of patterns of regression. NVD, neovascularisation of the disc; NVE, neovascularisation elsewhere. (Figure not reproduced here; see the original article.)

Week-12 retreatment

At 12 weeks, the patients were reviewed and categorised into three groups depending on treatment response to the first three injections. Figure 2 shows the retreatment plan at week 12 following first review of regression pattern.

FIGURE 2 Retreatment plan at week 12 following the first review of regression pattern. NR, no regression; PR, partial regression; TR, total regression. Reproduced from Sivaprasad et al.33 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) licence, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/. (Figure not reproduced here; see the original article.)

If an eye had experienced adverse effects from prior intravitreal injection, retreatment with intravitreal aflibercept was at the discretion of the investigator. In addition, if any future treatment with aflibercept was contraindicated based on a previous adverse reaction, treatment with PRP for PDR was at the investigator’s discretion after discussion with the chief investigator or his/her designee.

Week-16 to week-48 retreatment

All patients in the aflibercept arm were reviewed on a 4-weekly basis. From week 16, further treatment was determined by both regression and reactivation of NV on clinical examination with adequate visualisation of the entire retina and by comparing the four-field colour photographs or wide-field imaging carried out in the previous visit. The treatment responses were categorised into four groups (no regression, partial regression, reactivation and total regression) and the retreatment was based on the protocols shown in Figure 3.

FIGURE 3 Summary of retreatment plan from week 16 to 48. NR, no regression; PR, partial regression; RAc, reactivation; RP, regression pattern; TR, total regression. Reproduced from Sivaprasad et al.33 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) licence, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/. (Figure not reproduced here; see the original article.)

Further fields of colour retinal photographs or FFA could be performed at any visit if there was any doubt that a clinical feature represented retinal NV.

Aflibercept injections were deferred if an eye had experienced adverse effects from prior intravitreal injection. In this circumstance retreatment with intravitreal aflibercept was at the discretion of the investigator. In addition, if any future treatment with aflibercept was contraindicated based on a previous adverse reaction, treatment with PRP for PDR was also at the investigator’s discretion. Treatment with aflibercept or PRP could also be deferred in cases of total vitreous haemorrhage with no clear view of the fundus until the fundus was sufficiently well visualised to permit subsequent intraocular injection, in an eye that developed a rhegmatogenous retinal detachment or required surgical intervention for tractional retinal detachment threatening the fovea. Aflibercept injections could be resumed following surgical repair. Aflibercept injections could also be deferred if the interval between visits was < 4 weeks or if the intraocular pressure remained > 30 mmHg despite apraclonidine (Iopidine 1%) eye drops. In such scenarios, the participant could be prescribed Iopidine eye drops for a week and rescheduled for aflibercept injection within 1 week if the intraocular pressure was < 30 mmHg. PRP was deferred in the aflibercept arm in the ‘no regression’ category if the medium was too hazy to allow PRP or if, in the opinion of the investigator, PRP was not deemed necessary at that visit.

Comparator

The comparator was PRP, the current standard of care. Patients in the PRP group were initiated on PRP, which was completed in fractionated 2-weekly sessions. From week 12, all patients in the PRP arm were assessed for treatment response every 8 weeks and categorised using exactly the same categories as the intervention arm (as shown in Figure 1). Figures 2 and 3 give the summary of further treatment in the PRP arm. PRP treatment could be carried out using any PRP delivery system including indirect PRP. If PRP had to be performed as a day-case it was not recorded as a serious adverse event (SAE) despite hospitalisation of the patient. PRP was deferred in the PRP arm if the medium was too hazy to perform the procedure or if, in the opinion of the investigator, the eye had received adequate PRP and there was insufficient space for further fill-in PRP.

Assessments

The study assessments for the aflibercept arm of the study are shown in Table 1 and for the PRP arm in Table 2.

The flow of patients in both arms of the study at the initiation of the trial are shown in Figure 4.

FIGURE 4 Summary of flow of patients at screening/baseline. a, The number of sites was increased to 22 from 17 during the conduct of the trial as shown in the list of protocol amendments (see Table 4). CLARITY, clinical efficacy and mechanistic evaluation of aflibercept for proliferative diabetic retinopathy. (Figure not reproduced here; see the original article.)

Participant demographics and other baseline characteristics

This information was retrieved from the participant, hospital medical records or the general practitioner. Data included age, sex and ethnic background. Data were also collected on diabetic history and management, ocular history and treatment, other clinically relevant medical history and their management in the last 12 months, and concomitant medication.

Visual acuity tests

The visual acuity tests were performed using the validated ETDRS vision charts in accordance with SOPs.34,35 Refracted visual acuity was carried out in both eyes at screening, weeks 12 and 52 and at the point of withdrawal. For all other visits, visual acuity was tested with the previous protocol refraction and not included in the primary outcome analysis. Binocular visual acuity and low-luminance acuity were tested at screening and 52 weeks and at the point of withdrawal. The visual acuity scores were recorded in full in the eCRF.

Contrast sensitivity tests

The Pelli–Robson chart was used to test contrast sensitivity in both eyes at screening, week 52 and at the point of withdrawal as per SOP.36 The total contrast sensitivity scores in both eyes were recorded in the eCRF.

Driving fields tests

Uniocular (study eye) and binocular Esterman fields were carried out at screening, week 52 and at the point of withdrawal.37

Standard ophthalmic examination

A standard ophthalmic examination using slit-lamp biomicroscopy, tonometry and dilated fundus examination was carried out in both eyes at all visits. Gonioscopy was performed if neovascularisation of the angle (NVA), neovascularisation of the iris (NVI) or NVG was suspected. The grade of diabetic retinopathy and the presence or absence of macular oedema as assessed by the investigator were recorded in the eCRF.

Spectral-domain optical coherence tomography

The CST in both eyes was recorded from SD-OCT thickness map at screening, 12 and 52 weeks and at the point of withdrawal. This test was repeated at any visit at the investigator’s discretion. If treatment of DMO was planned, OCT was carried out for confirmation of DMO and monitoring treatment. Any SD-OCT machine was used for the study but the same model of SD-OCT was used for each individual throughout the period of the study.

Colour fundus photography and fundus fluorescein angiography

Seven-field colour fundus photography (CFP) was performed to assess the severity level of diabetic retinopathy and area of retinal NV at screening, weeks 12 and 52 and at the point of withdrawal. FFA was carried out at screening, week 52 and at the point of withdrawal. Four-field photography or Optos wide-field imaging was performed at all other visits to evaluate regression and reactivation patterns. In the PRP arm, these included visits at weeks 20, 28, 36 and 44 and in the aflibercept arm at weeks 4, 8,16, 20, 24, 28, 32, 36, 40, 44 and 48. Additional fields for colour photographs and FFA may have been performed to determine the presence or absence of NV in either eye at any of these visits as per local practice or investigator discretion. The 7-field photographs performed at screening, weeks 12 and 52 and withdrawal were read by masked graders at the Independent Reading Centres in NetwORC UK.

Blood pressure

Blood pressure was measured at each study visit except baseline and week 4 for both arms and week 8 for the aflibercept arm. Patients were not considered eligible for the study if the BP was recorded as > 170/110 mmHg at screening.

If the BP was > 170/110 mmHg, the patient was allowed to be rescreened, at least 1 month after the last screening visit, if the parameter was brought under control and the other inclusion/exclusion criteria were still met.

If the BP was > 150/90 mmHg but ≤ 170/110, the participant was eligible but the normal health-care provider was informed via a standard letter directing the provider to the NICE guidance of management of BP in patients with diabetes mellitus. All randomised patients continued to be followed up by their normal health-care provider and remained in the study and underwent all study assessments and treatment as per protocol.

Glycated haemoglobin

The blood samples for HbA_1c level assessment were processed at local laboratories or in accordance with local practice. The HbA_1c assessment was performed at screening and at the final visit. If the test was performed within 3 months of the visit, it was not repeated. At screening, participants who had a HbA_1c level of > 12% at the start of the study were not eligible and could be rescreened after 3 months and randomised if the parameter was brought under control and the other inclusion/exclusion criteria were met. Any patients identified with a HbA_1c level of > 8% at either screening or final visit were referred for follow-up with their normal health-care provider. A standard letter directing the health-care provider to the NICE guidelines for management of diabetes mellitus was provided. If in the opinion of the principal investigator the result was classed as clinically and significantly abnormal this was recorded as an adverse event and was followed up accordingly.

Questionnaires

The following participant satisfaction, generic health-related and vision-specific quality-of-life questionnaires were administered at screening, week 52 and where necessary at the point of withdrawal: RetTSQ,38 NEI-VFQ-25,39 RetDQoL,40 EQ-5D-3L,41 ICECAP-A42 and CSRI. Please see Chapter 4 for details.

Mechanistic tests (Moorfield’s Eye Hospital only)

Forty participants (20 in each arm) who consented for the mechanistic evaluation underwent oximetry tests in both eyes at baseline, weeks 12 and 52 and at the point of withdrawal. A within-visit flexibility of +10 days was allowed for patients to complete these tests. Please see Chapter 5 for details.

Independent Reading Centres in NetwORC UK

The NetwORC provided each site with a manual giving instructions and guidance on how to acquire and transfer the colour retinal photographs completed at screening, weeks 12 and 52 and at the point of withdrawal. The images were anonymised to study PIN and included the time point at which the image was collected. The images were transferred to the reading centre via compact disc, secure file transfer protocol or another suitably secure medium agreed by the reading centre and the chief investigator. The images were accompanied by a transmittal log that required the patient’s date of birth as an identifier. Sites ensured that all PINs and dates on images complement the information recorded on the transmittal log and that all images were captured, exported and submitted in accordance with the requirements of the study imaging protocol. The reading centre sent reports regularly to the KCTU throughout the study with an overview of what had been received and what was outstanding from each of the sites. The reading centres evaluated the images and the results were transferred to the chief investigator and trial manager in the KCTU. The trial manager transferred the data to the study statistician.

Data collection and management

It was the responsibility of the principal investigator at each site to identify the personnel responsible for data collection and handling, including those who had access to the trial database, and to ensure the completeness of the delegation log. All source documents were prepared in advance by the trial team and these forms were signed by the researcher who completed the assessments. The principal investigator at each site was also responsible for the accuracy of all data entered in the worksheets in accordance with good clinical practice and the Data Protection Act 199843, ensuring that the source data worksheets were filed in a suitably secure location to ensure that source data verification could be undertaken throughout the study. Source data worksheets were reconciled at the end of the trial with the patients’ NHS medical notes in the recruiting centres. During the trial, critical clinical information was written into patients’ medical notes to ensure that informed medical decisions could be made in the absence of the study team. Trial-related clinical letters were copied to the medical notes during the trial.

Randomisation and collection of data on completion at each visit were performed by the unmasked research staff. The randomisation details were e-mailed to the pharmacy, the delegated unmasked researcher, the principal investigator at that site and the UK trial manager. The study data were entered into a study-specific eCRF created in collaboration with the trial statistician and the chief investigator, and maintained by the KCTU using the InferMed MACRO database system hosted on a dedicated secure server within King’s College London [MACRO electronic data capture (EDC) V4 at www.ctu.co.uk]. This system was regulatory compliant, supported real-time data cleaning and reporting, and had data discrepancy functionality and database lock functionality. A series of logic and range checks were built into the system to reduce the possibility of erroneous data being entered. The system also contained an audit trail of all notable events associated with the trial (including inserts, updates and deletions) and this provided a clear and complete audit trail throughout the trial. The trial manager was responsible for providing usernames and passwords to permitted local study personnel. Only those authorised by the trial manager were able to use the system.

Monitoring and site visits

The first site visit was a prerecruitment visit for training of the protocol, adverse event reporting, MACRO and randomisation and data entry training and outline of good clinical practice. The optometrists were certified at Moorfields Eye Hospital. The trial photographers were certified by the Central Angiographic Resource Facility (CARF) or by Optos. The trial manager ensured that all relevant documentation and certifications were complete before a greenlight letter was sent to each site to initiate the study. All procedures were detailed in an operation manual. On-site monitoring visits were conducted routinely; more monitoring visits were arranged if problems were identified (e.g. poor data collection or under-reporting of primary end point data) or when a site requested a visit to discuss specific issues (including data collection, screening patients, recruitment and staff training). The main areas of focus included consent, SAEs, essential documents in study site files, and source data verification on selected outcomes including the primary outcome measure. A monitoring report was prepared after each monitoring visit and the points raised in the monitoring report were then addressed with sites remotely or at the next onsite monitoring visit. A data management plan and a monitoring plan were designed to ensure that the relevant data were monitored methodologically. In addition, the unblinded statisticians performed central statistical monitoring by reviewing event rates, unusual trends and data anomalies.

In addition, all adverse events and SAEs were monitored at each data monitoring committee meeting. All SAEs were also sent to the drug manufacturer (Bayer Plc, Reading, UK) and to the sponsor. The SAE data were collected on paper SAE report forms and faxed to the KCTU. Summary details of SAEs were also transcribed to the adverse event section of the eCRF.

The principal investigator provided an electronic signature for each patient case report form once all queries were resolved and immediately prior to database lock.

Multiple systematic approaches were therefore instituted to ensure that all outcome data were as accurate and complete as possible.

Outcomes

Primary outcomes

The primary outcome was change in BCVA from screening to 52 weeks in the study eye measured in ETDRS letter score at 4 metres: difference in means.

Secondary outcomes

To measure the effect of intravitreal aflibercept therapy, relative to PRP on additional visual function and quality-of-life outcomes including:

• percentage of uniocular and binocular Esterman efficiency scores at 52 weeks – difference in proportions
• binocular visual acuity and low-luminance visual acuity at 52 weeks – difference in means
• visual acuity outcomes in terms of visual gain or loss – difference in proportions
• contrast sensitivity measured using the Pelli–Robson chart at 52 weeks – difference in means
• change from baseline in vision-related quality of life measured using NEI-VFQ-25 and RetDQol at 52 weeks – difference in means
• change from baseline in diabetic RetTSQ scores at 52 weeks – difference in means
• gain in HRQoL from baseline over the 52 weeks study period – difference in means.

Cost-effectiveness

From a public sector multiagency perspective, covering health and social care services, we assessed the cost-effectiveness of intravitreal aflibercept compared with PRP. The incremental costs and effects of alternative arms were compared in the primary cost-effectiveness analysis, with effectiveness measured in terms of BCVA. We also undertook a secondary exploratory cost–utility analysis (i.e. cost per QALY analysis). Evidence is mixed but points to the EQ-5D-3L not being sufficiently sensitive to be useful in studies of visual impairment.44,45 Published sources of national unit costs at 2016 price year were used to calculate the total cost of health and social care service use and ophthalmic-related drug use over 52 weeks.46 The costs of aflibercept injection treatment in the intervention arm and PRP laser treatment in the comparator arm were calculated using published national unit costs from the Department of Health46 and the British National Formulary (BNF)47 at 2016 price year. We employed ‘bootstrapping’ to overcome the skewed data and produced cost-effectiveness acceptability curves (CEACs) to quantify uncertainty.48

We used sensitivity analysis to explore whether or not the estimated effect (‘change in BCVA score’) and cost of aflibercept injection treatment relative to PRP laser treatment were sensitive to the key variable factor of our analysis – the pricing of aflibercept. In our opinion, the pricing of aflibercept is the most important variable because the NHS Patient Access Scheme (PAS) could provide the drug at a range of prices.

We undertook a subgroup analysis to explore the potential effect of heterogeneity defined in terms of DMO in the fellow eye. The subgroup analysis was performed by the presence of DMO at baseline (no DMO in both eyes vs. DMO presence in at least one eye). In this study, DMO was an exclusion criterion for the study eye so DMO in at least one eye is defined as presence of DMO in the fellow eye only. The incremental costs and effects of alternative arms were reported within each of the categories of the subgroup variable, and the incremental cost-effectiveness ratio (ICER) (i.e. cost per change in BCVA) was calculated for both of these categories of the subgroup variable.

Measurement of other outcomes

From a methodological perspective, we compared the performance of the EQ-5D-3L (a generic, preference-based HRQoL measure) with the vision-specific HRQoL measures (NEI-VFQ-25 and RetDQoL). We included the ICECAP-A measure as an alternative to EQ-5D-3L focusing on capability to see how this measure compared in terms of sensitivity in this patient group.

Anatomical outcomes

1. To compare the regression patterns of new vessels at 12 weeks and the regression and reactivation patterns between treatment arms at 52 weeks: means and proportions.
2. To compare the proportion of patients with one-step and three-step improvement or worsening of diabetic retinopathy between treatment arms at 12 and 52 weeks: difference in proportions.49

Treatment-related outcomes

To determine the proportions of treatment-naive and post-PRP-treated eyes that do not require PRP through 52 weeks after basic treatment of three loading doses of aflibercept in the aflibercept arm and after initial completion of PRP in the PRP arm: difference in proportions.

Safety profile

To explore the difference in safety profiles between intravitreal aflibercept and PRP at 52 weeks, in terms of proportion of patients developing macular oedema (defined as a CST of > 300 µm on SD-OCT because of clinical evidence of macular oedema), any de novo or increase in existing vitreous haemorrhage, de novo or increasing tractional retinal detachment, NVG, and the requirement for vitrectomy for various indications: difference in proportions.

To determine the systemic safety events including the Anti-Platelet Triallists’ Collaboration (APTC) events between arms.50

Mechanistic evaluation

1. To explore whether intravitreal aflibercept, compared with PRP, causes measurable regression of retinal NV at 12 and 52 weeks in terms of decimal disc area units in 4-field colour photographs and FFA: difference in means.
2. To explore differences in the mean change in retinal vessel calibre and oxygen saturation in eyes treated with intravitreal aflibercept compared with PRP at 12 and 52 weeks: difference in means.
3. To explore whether intravitreal aflibercept reduces angiographically quantifiable areas of retinal non-perfusion compared with PRP through 52 weeks: means and proportions.

Sample size calculation

The sample size calculation was performed using nQuery Advisor v4.0 software (Statistical Solutions, Saugus, MA, USA). The primary outcome is the change in BCVA ETDRS letter score from baseline to 52 weeks. Based on the objectives of this study and the potential deleterious effects on visual function of PRP, a non-inferiority margin of 5 letters was judged to be clinically acceptable.17,33,34,51–53 In addition, this margin is less than the lower limit of the 95% confidence interval (CI) for the comparison of PRP with observation. This helped to ensure that aflibercept is superior to observation alone in the event that it was found to be non-inferior to PRP. In the wider patient population, if aflibercept was no more than 5 letters worse than PRP, it would therefore be defined as being non-inferior. The sample size was based on providing a 95% CI for the between-arm difference in mean change in visual acuity that would be sufficiently narrow to detect non-inferiority (by the CI lying entirely above the margin) with high power, while keeping a false declaration of non-inferiority to 5% through use of a statistical test applied at the two-sided 5% level of significance.

The standard deviation (SD) of the change in visual acuity, after adjustment for baseline, was estimated to be 10.3, based on a relevant trial.54 There was no issue of clustering of outcomes from eyes within subject effects because only one eye per subject was selected for the study.

With 110 patients (one eye per subject) randomised per group (total 220 patients), 182 had to be followed up to 52-week outcome [allowing for 17% dropout or per protocol (PP) exclusion]. This provided 90% power to detect non-inferiority using a two-sided 95% CI from an analysis of covariance (ANCOVA) test with adjustment for baseline visual acuity.

For a continuous secondary outcome, with 182 subjects followed up, we can detect effects of size 0.42 SDs difference between means with 80% power using a two-sided t-test at the 5% significance level. For binary outcomes, we had at least 77% power to detect a difference in proportions of 0.2 using a chi-squared test at the 5% significance level.

Collection and analysis of outcome data

Baseline comparability of randomised groups

Baseline descriptions of participants by treatment and overall were summarised and included age, sex, diabetes mellitus history and management, ocular history and treatment, other clinically relevant medical history and their management in the last 12 months, and concomitant medication. No significance testing was carried out, as any differences found may have been generated by chance and not for hypothesised reasons.

Continuous variables such as contrast sensitivity and macular thickness were summarised using means and SD and/or medians and IQR for variables presenting a skewed distribution. Categorical variables such as proportion of patients with one- and three-step improvement or worsening of diabetic retinopathy were described using numbers and percentages.

Comparison of rates of adherence and follow-up

Compliance rates and attrition rates were compared and reported by arm. High compliance and low attrition rates were anticipated for this study on the basis of previous clinical trial experience.

Analysis covariates

Stratifiers

Covariates were predefined according to the International Conference on Harmonisation (ICH) E9 guideline, which recommends considering factors on which randomisation has been stratified as these factors tend to be predictive of outcome, and therefore the need for adjustment of the minimisation stratifiers.55 Randomisation stratifiers as shown in Figure 5, as covariates were planned to improve the precision of the estimated treatment effects.

FIGURE 5 Randomisation stratifiers. (Figure not reproduced here; see the original article.)

Baseline

The corresponding baseline measure for a continuous outcome is often predictive of the outcome at follow-up and so retinal thickness and BCVA at baseline were included as additional covariates when modelling these continuous outcomes.

Primary outcome analysis

Statistical model

The primary efficacy measure was the change from baseline in refracted BCVA in the study eye, using the ETDRS letter score at 52 weeks. As the analysis approach for continuous outcomes takes advantage of covariate adjustment for the baseline of the outcome, the primary end point can equivalently be regarded to be each participant’s 52-week measurement. This is convenient because then those with a 52-week outcome, but whose baseline measurement is missing, are not regarded to be missing the end point. The primary outcome may therefore be referred to below as the 52-week visual acuity, rather than the change in visual acuity from baseline to 52 weeks.

The primary outcome was analysed using a linear mixed-effects (LME) model, with an unstructured variance–covariance matrix which incorporated the 12-week and 52-week post-baseline measurements of the refracted BCVA outcome. This mixed model had a mix of random- and fixed-effect terms. The random effects in the model were participant, represented as a random intercept at each follow-up time point, allowing for within-participant correlation in the two adjusted post-baseline outcomes. The fixed effects in the model were the main effect terms for arm, baseline PDR status (naive vs. non-naive PDR), HbA_1c (< 8%, ≥ 8% to ≤ 10% and > 10%), and diastolic BP (≤ 90 mmHg vs. > 90 mmHg), ‘time’ (12 weeks vs. 52 weeks), the baseline of the outcome and its missing indicator required for the missing indicator method.56,57 The other fixed effects included in the model were the interactions between ‘time’ and each of the other fixed effects in the model. This model allowed the treatment effect to be estimated at 12 weeks and at the primary time point of 52 weeks, adjusting for time-specific effects of baseline BCVA and of other baseline clinical covariates that were chosen for minimisation stratifiers.

Intention-to-treat strategy

Outcome data were valid and included if the BCVA measure was refracted. All randomised subjects who provide at least one post-baseline valid measurement were included.

The achieved trial sample comprised those study participants who consented to participate and were actually randomised into this trial. This randomised trial sample was also the trial intention-to-treat (ITT) population. The ITT principle states that every subject will be analysed according to the treatment group to which they were randomised. In this trial, subjects’ data were analysed in accordance with Intention to Treat Strategy,58 under which at least one analysis is recommended to be based on the ITT population.

The trial ITT population comprised all randomised participants, regardless of eligibility (inclusion/exclusion) error, post-randomisation withdrawal, and whether the correct study treatments or other interventions were received. On 20 November 2015, The Data Monitoring and Ethics Committee (DMEC) discussed the circumstances under which the BCVA score at either 12 or 52 weeks would not reflect the underlying visual status of a participant. In particular, recent vitreous haemorrhages that might cause low BCVA scores which would then return to normal for the patient, either spontaneously or through appropriate clinical management (vitrectomy), were discussed. The challenge was that any such measurements occurring at week 12 or 52 could artificially induce very large negative changes in BCVA which would have enormous influence in statistical analysis – specifically by leading to very large inflations in the standard deviation for the change from baseline. This could have profound implications for the ability of this non-inferiority trial to achieve its objectives, which rely on the 95% CI for the difference between randomised groups in the change from baseline falling within prespecified bounds (the non-inferiority margin). As such values do not intrinsically reflect the underlying visual status of the patient, the DMEC proposed that the trial steering committee (TSC) consider amending the primary analysis population to exclude from analysis any BCVA measurement at 12 or 52 weeks, that is both > 3SD below the mean at that timepoint (including all measurements) and taken within 3 months of occurrence of a vitreous haemorrhage. Endophthalmitis was also considered as a comparable cause (to vitreous haemorrhage), but, as this is rare and unlikely to occur in a trial of this size, the definition of primary outcome analysis measurements had not encompassed this. The absolute number of measurements excluded was expected to be small.

Per-protocol analysis

For the analysis of the primary outcome, the mixed-effects models were refitted in a reduced PP population, excluding patients found to be ineligible at entry, those patients who received the alternative treatment to that allocated up to the end point, and those not receiving at least the minimal randomised treatment up to and including the 8-week visit (whether because of discontinuation, exclusion or another reason for missing a randomised treatment in this period). This was prior to the point of treatment stratification from the 12-week visit, which included a stratum for patients who required no further treatment.

Rationale

The main reason for having a PP set came from the fact that this was a non-inferiority trial and so the use of the full analysis set is generally not conservative. As Lesaffre59 states, ‘dropouts and a poor conduct of the study might direct the results of the two arms towards each other’. Although this can be interpreted as an indication that the PP analysis is the conservative choice for non-inferiority studies, Garrett60 states that ‘The perceived conservative nature of the PP population appears to be much more a reflection of reduced patient numbers than the presence of bias, while bias can be in either direction depending on the pattern of violations’. Moreover, with two active treatments it may be more likely that any bias affecting both treatments would be reduced in comparison with a placebo-controlled trial.

Prominence

Non-inferiority was planned to be declared only if both ITT and the PP analysis were supportive of a non-inferiority conclusion at 52 weeks. This is supported by the Committee on Proprietary Medical Products61 and in several other papers.59,62

The requirement to declare non-inferiority in both the ITT and the PP analyses promotes the adherence to treatment protocol and the minimisation of exclusions, maintaining power. Non-inferiority was also assessed secondarily in ITT and PP populations at 12 weeks from the same models. Non-inferiority was declared if the estimated 95% CI for the difference in means was wholly above the margin of –5 letters in both ITT and PP analysis models primarily at 52 weeks and secondarily at 12 weeks.

Superiority

If non-inferiority was concluded, superiority was planned to be assessed from the ITT LME model by reporting the p-value from the two-sided test of the hypothesis of a zero difference in population means using a 5% significance level without need for correction for multiple testing.

Subgroup analysis

Subgroup analysis was performed by baseline retinopathy status (naive and non-naive PDR), HbA_1c (< 8%, ≥ 8% to ≤ 10% and > 10%), diastolic BP (≤ 90 mmHg vs. > 90 mmHg) and BCVA (54–69 vs. ≥ 70 letters). The p-value for the between-subgroup comparison of effectiveness was obtained from the ITT LME model from the interaction between arm and the subgroup variable at 52 weeks. The treatment effect and associated 95% CI were reported within each of the categories of the subgroup variable.

Sensitivity to missing data

An expert missing-data group concluded that, rather than statisticians reacting to missing data at the end of a trial, there should be comprehensive, proactive planning for handling missing data at the stage of designing trials.63 The group recommended that there should be consideration of missing data mechanisms (e.g. missing at random) and, if the missing data may be informative, that appropriate sensitivity analyses should be undertaken to investigate the robustness of the inferences to the different assumptions made by the main analysis. It has also been recommended that analyses allowing for non-response and low intervention uptake (or compliance) are best specified in advance and included in the analysis plan.64 As it was expected that compliance would be high from the fear of loss of sight, and as non-inferiority is concluded only when declared in both a compliant PP population and a less compliant ITT population, the focus was on the handling of missing data.

The main reasons for dropout in this study were thought to be related to patients’ comorbidities, such as hypertension and hyperlipidaemia, and less connected with underlying visual acuity. Such reasons mean that dropout may not depend on visual acuity and, as data on BP and HbA_1c were collected, it was possible to explore the association between these data and dropout to improve the interpretation of the sensitivity analysis results. The primary outcome of refracted BCVA was collected only at two post-baseline measurement points, and there were limited serial data. Nevertheless, as described in Sensitivity analysis to use of concomitant treatments, a sensitivity analysis was undertaken to assess the possibility of alternative plausible values of treatment effect arising from potential mishandling of missing data in the primary analysis model.

The LME model for the primary outcome analysis described above was the first of a two-part approach called the intention-to-treat strategy,64 in which a second analysis examined the sensitivity of the results to missing data in the full randomised ITT population. This met the ideal objectives of ITT. The approach to missing data taken for this study followed the recently published implementation paper of the ITT strategy.62 The few cases with values that were affected substantially to become unrepresentative temporarily because of vitreous haemorrhage were included among the measurements with missing visual acuity data. This approach was then also applied again to the PP population so that the non-inferiority conclusion could be reassessed under the sensitivity analysis.

For the sensitivity analysis, we prespecified a range for BCVA from –20 to +20 letters over which the mean of the ‘unobserved outcome data’ might depart (or be different) from the mean of the ‘observed outcome data’.62 In other words, this range can be thought of as how much a typical subject with missing data may on average have had a different estimated treatment effect from the corresponding subject with the outcome data observed (given the same baseline covariates and follow-up data in the LME model). The range (–20 to +20) was chosen to represent both negative and positive departures that could potentially arise as the ‘net effect’ of alternative reasons that may be unknown, such as dropout because of no anticipated further improvement or dropout because of no improvement so far together with no anticipated achievable improvement.

This range of 40 letters (from –20 to +20) was generously wide for exploring sensitivity of the main results to departures from the missing at random (MAR) assumption because 20 letters (as the maximum departure in either direction) is larger than the detectable between-arm treatment effect of 3 lines (15 letters) seen in superiority trials (difference in means), a sizeable shift in the mean of the distribution for dropouts than completers.

At the end of the trial, the fractions of individuals with missing data for visual acuity at 52 weeks were available in each arm: f_i (for intervention) and f_c (for control). The parameter representing excess visual acuity in those missing compared with those observed, δ, took values by passing across the range –20 to +20. Three scenarios were undertaken within the sensitivity analysis.61,62 These reflected whether departures from the MAR assumption applied within the intervention arm only (aflibercept), within the comparator arm only (PRP) or within both arms equally and in the same direction (thereby potentially cancelling out across the sensitivity range, if the dropout rate were to be the same in both arms).

Scenario 1: the treatment effect from the LME model will be increased by f_iδ.

Scenario 2: the treatment effect from the LME model will be increased by –f_cδ.

Scenario 3: the treatment effect from the LME model will be increased by (f_i – f_c)δ.

Sensitivity analysis to use of concomitant treatments

The use of concomitant treatments was monitored by the DMEC. A sensitivity analysis was not undertaken to examine the robustness of the 52-week PP analysis to the use of concomitant treatments as there were none.

Sensitivity analysis to adjust for site in the primary outcome model

Sites were added as a covariate in the primary outcome analysis model.

Secondary outcome analysis

Analysis of continuous outcomes

The analyses of continuous secondary outcomes were compared between arms at 52 weeks using ANCOVA or the LME model, as the primary outcome, adjusting for baseline PDR status (naive PDR and non-naive PDR), HbA_1c (< 8%, ≥ 8% to ≤ 10% and > 10%), diastolic BP (≤ 90 mmHg vs. > 90 mmHg), and BCVA (54–69 vs. ≥ 70 letters), and where collected, the baseline of BCVA with the associated missing indicator. ‘Time’ (12 weeks vs. 52 weeks) was represented as a categorical contrast in main effect form and in interaction with all other fixed effects. For those outcomes where there were no post-baseline measures taken on the outcome before 52 weeks, ANCOVA was used instead.

Analysis of binary outcomes

For the binary outcomes, such as the proportion of patients with a three-step improvement of diabetic retinopathy, chi-squared tests were used. Safety outcomes were also reported as unadjusted patient proportions and rates within and between arms with 95% CIs using exact methods where appropriate (Wilson’s method with no continuity correction). Furthermore, CST and macular volume were corrected in the analysis plan to the use of the LME model instead of ANCOVA as there were two time points (12 weeks and 52 weeks) to take into account. In a DMEC meeting on 14 October 2016, it was decided that the number of patients with, and number of events of, vitreous haemorrhage in the study eye and in the non-study eye would be analysed and reported as a separate table from the adverse events tables.

There was a list of outcomes that required derivations:

1. The NEI-VFQ-25 questionnaire is a validated tool for vision-related quality of life. It consists of a base set of 25 vision-targeted questions representing 11 vision-related subscales, plus an additional single-item general health rating question. The overall composite score is computed as the simple average of the vision-targeted subscale scores, excluding the general health rating question. The overall score can range from 0 (worst possible score) to 100 (best).
2. The RetTSQ is a diabetic retinopathy treatment satisfaction questionnaire consisting of 13 items asking participants to rate different aspects of treatment. It can be scored as total score or as two subscales (one covering negative experiences and the other one positive aspects of treatment). The total score for treatment satisfaction was calculated by summing the scores for items 1 to 13, with a possible range of 0 to 78, where higher scores represent more satisfaction. For the positive and negative subscales, the possible ranges are from 0 to 42 and from 0 to 36, respectively.
3. The RetDQoL is a questionnaire designed to measure individualised quality of life in people with diabetic retinopathy. It consists of two overview items using a 7-point scale and 26 domain-specific items. The first overview item asks participants to complete the statement ‘In general, my present quality of life is – ’. Answers to this question range from ‘excellent,’ scored as 3, through ‘neither good nor bad,’ scored as 0, to ‘extremely bad,’ scored as –3. The second item asks how quality of life is affected by diabetic eye problems; answers include ‘very much better’ (scored –3), ‘much better’ (–2), ‘better’ (–1), ‘the same’ (0) and ‘worse’ (1). For the domain-specific items, the participant indicates where items are not applicable to them and, for the items that are applicable, they first rate the impact of diabetic eye problems on each aspect of life and then rate the importance of each aspect of life to their quality of life. The impact and importance ratings for each applicable item are then multiplied to obtain a weighted impact score with a range from –9 to 3. A more negative average weighted impact (AWI) score indicates a more negative impact of diabetic retinopathy on quality of life; a positive score would indicate a positive impact of diabetic retinopathy on quality of life. A total score, the AWI, can be obtained by summing the weighted impact scores of all applicable domain-specific items and dividing the result by the number of applicable domains.
4. The EQ-5D-3L is a generic HRQoL measure that consists of two pages. The EQ-5D-3L descriptive system comprises five dimensions (mobility, self-care, usual activities, pain/discomfort, anxiety/depression) and the EuroQol Visual Analogue Scale (EQ VAS), which records the respondent’s self-rated health on a vertical visual analogue scale. Each of the five dimensions has three levels (no problems, some problems and extreme problems). The health state preference values (utilities) for EQ-5D-3L profiles were based on time trade-off valuations by members of the UK general public.
5. The ICECAP-A42 is a brief questionnaire that measures the ability of an individual to carry out activities and is a measure of adult capability. The original website has Stata^® (StataCorp LP, College Station, TX, USA) code that was used for the scoring of the scale. The questionnaire consists of only five items classified from 1 to 4.

If there was existing syntax code to derive a variable within the KCTU then this was used. Otherwise new code was developed by the trial statistician and verified by the senior statistician.

Missing items in scale and subscales

The number (%) of patients with complete data for each scale was reported. If scales provide, missing value guidance was used.

Use of data transformation

It was not anticipated that any continuous outcomes would need to be considered for transformation, because the sample size was reasonably large for group comparisons in the main trial analyses. Assumptions of normality and constant variance required by the models were examined using residual and other diagnostic plots. If relevant, and necessary (e.g. in the mechanistic evaluation where sample size is reduced) a log-transformation could be considered, because this retains a sensible interpretation for inferences in relative terms between arms. If an absolute interpretation was needed then data transformation would not be undertaken; instead, a non-parametric bootstrap method for obtaining CIs was considered.65 For the mechanistic evaluation, non-parametric methods were also considered.

Defining outliers

Outliers are observations that have extreme values relative to other observations detected under the same conditions. An outlier was defined here as a data point at least 4 SDs from the mean of its distribution of values observed across other patients. This definition was applied to the transformed scale for those outcomes that were log-transformed.

A ‘bivariate outlier’ was defined here as the difference between successive serial data points of the same measure being at least 4 SDs from the mean of these differences. Simple plots of successive pairs of serial measures were used through the 24-month period to assist in identifying outliers.

Handling outliers

Outliers were identified for further investigation by looking at the distributions of the data through histograms, scatterplots or box plots. Univariate tests for the compatibility of the distribution with a normal distribution were not undertaken since they can be too sensitive to departures that are often not relevant for the comparison of means (the central limit theorem).

Once an outlier was found, a blinded member of the team with sufficient clinical experience was involved in the decision as to whether a data value was impossible versus implausible versus plausible. If the outlier was impossible it would be set to missing and a list of these occurrences then appended to the statistical analysis plan. If an outlier was clinically plausible, the outlier remained. If an outlier was clinically implausible (but possible), it was not ignored or deleted and was retained for ITT analysis.

If outliers remained in the distribution of a variable then data transformations or non-parametric methods of analysis were considered.

Sensitivity analyses were undertaken to check whether the outlier was influential by obtaining results with and then without the inclusion of the outlier. If the conclusions were changed, these were documented.

Analysis methods for secondary outcomes

All study analyses were based on tests that are two-sided, including the two-sided 95% CIs assessed at the 5% significance level. Table 3 shows the method of analysis chosen for each secondary measure.

Handling multiple comparisons

Significance tests were used sparingly and were restricted, where possible, to addressing stated hypotheses. Secondary outcomes, as well as the primary outcome, were summarised using an effect size with a 95% CI. Interpretation for those secondary outcomes that did not directly address the stated study hypotheses was more cautious.

Ethics approval and monitoring

The study was granted approval by the National Research Ethics Committee Service London – South East (14/LO/0203). Clinical Trials Authorisation was given by the MHRA (11518/0013/001-0001) and the European Union Drug Regulating Authorities Clinical Trials (EudraCT) number was 2013-003272-12. The trial was run using the SOPs of the sponsor, Moorfields Eye Hospital NHS Foundation Trust. The sponsor provided the oversight of the study and the KCTU collaborated with the sponsor to ensure efficient delivery of the study. The trial had two committees overseeing its conduct: the TSC and the Trial Management Group (TMG). In addition, there was an independent DMEC to ensure the safety of patients in the trial and review operational issues such as recruitment. The membership of the DMEC is provided in the Acknowledgements.

The DMEC was the only group to review interim analyses broken down by treatment groups during recruitment and follow-up of patients in the trial. The DMEC performed interim safety analyses. The interim reports contained details of patient recruitment, demographic and baseline characteristics, the intervention, primary safety end points, the primary efficacy end point and other end points identified by the DMEC including adverse events and SAEs.

The TSC consisted of four independent members, two lay members and key members of the TMG, in addition to clinical and methodological experts. The membership of the TSC is provided in the Acknowledgements. The TSC had overall responsibility for the scientific integrity and quality of the trial. This involved conducting the trial to the standards set out in the guidelines for good clinical practice, adhering to the protocol as far as possible and holding responsibility for overall patient safety, as well as considering new relevant information as it arose throughout the duration of the trial. The TSC was also responsible for considering any recommendations made by the DMEC. The TSC met throughout the trial to monitor the progress and qualities of the trial (review the recruitment rate and consider protocol amendments). The TMG was responsible for the day-to-day running of the trial, meeting fortnightly during the setting up of the clinical efficacy and mechanistic evaluation of aflibercept for proliferative diabetic retinopathy (CLARITY) trial and the early stages of recruitment and then approximately monthly for the remainder of the trial. This group consisted of the chief investigator, co-lead, trial manager, research fellow, statisticians, sponsor representative, data manager and the KCTU operations manager. The role of the TMG was to monitor the conduct and ensure progress of the trial according to the study protocol and to take appropriate action to safeguard participants and the trial itself.

Patient and public involvement

The study design, patient information sheets and consent forms for the main trial and mechanistic substudy were discussed with the North East London Diabetes Research Network patient and public involvement group, who provided feedback and support for the study. A user representative was a grant co-applicant and two other user representatives were members of the TSC and as part of this role made suggestions for the use of laser instead of PRP to describe the comparator.

Reporting

The trial was reported in accordance with the Consolidated Standards of Reporting Trials (CONSORT) statement.

Summary of changes made to protocol

Table 4 summarises the changes made to the protocol. In addition, the statistical analysis plan was amended after publication of the protocol. Version 5.0 was amended to version 5.1 as a result of the DMEC meeting held on 3 November 2015, in open session, and the DMEC recommendation to the TSC, accepted at the TSC meeting on 20 November 2015.

The TSC requested confirmation that vitreous haemorrhage will be reported as an outcome by arm, and this has been confirmed to be present in Table 3; the number of measurements excluded from the analysis, by arm, will be reported.

Version 5.1 of the protocol was amended to version 5.2 as a result of the DMEC meeting held on 4 October 2016.

It was decided that study site should be taken out of the primary and secondary outcome LME and ANCOVA models as a covariate since the models would be parameter-heavy with as many as 22 sites having recruited, and where some sites have only included one or a few patients who, at follow-up, could contribute to estimating site rather than to treatment effect. For transparency, sites will then be added as a covariate in a sensitivity analysis. In response to a query by the principal investigator, it was clarified, in the open session of the DMEC meeting, that the actual categorised baseline values of stratifying covariates, rather than those used in the randomisation, which included errors, will be used in the outcome models, so that any baseline confounding by these is more fully adjusted for, and analyses are consistent with subgroup analyses using the same categorisations of these covariates. There was agreement for this approach as the trial employs minimisation. The outcome ‘A small number of summary measures will be calculated and reported to represent the pattern and frequency of randomised treatment received over time’ was added as a process outcome.