I read with great interest the paper titled “Collateral vessels on optical coherence tomography (OCT) angiography in eyes with branch retinal vein occlusion (BRVO)” by Suzuki et al.1
The authors defined collateral vessels as dilated and tortuous capillaries occurring in pre-existing capillary beds and linking the obstructed vessel with the nearest patent vessel, according to previous reports.2-4 The authors demonstrated that collaterals were detected in 23 out of 28 (82%) eyes, all of which already existed at mean 0.95 months after the onset, and that all of the collaterals were observed in both the retinal superficial and the deep layers.
However, some of the vessels which are pointed out as collaterals in the study1 look like simply dilated/tortuous vessels, because they don’t seem to connect obstructed to non-obstructed adjacent vessels nor by-pass obstructions. In a previous report, the authors found collateral vessels in 18 out of 28 (64%) eyes at mean 25.1 months from the onset, while superficial and deep capillary telangiectasias were detected in 13 and 28 out of 28 eyes, respectively.4 Therefore, I suppose that some of the vessels defined as collaterals in this study1 may be simply telangiectasias.
Fruend et al.5 defined collateral vessels as the authors did. After excluding collaterals involving the perifoveal vascular ring, they demonstrated that collaterals were found in 23 out of 23 eyes (100%) at median time of 3.79 years from RVO diagnosis or initial visit, and that all collaterals were primarily located within the deep vascular complex. Indeed there are differences in patients’ and ocular backgrounds between the two studies.1,5 But the findings about the location of collaterals by Fruend et al5 may support my supposition that some of the vessels defined as collaterals in this study1 are simply telangiectasias.
I hope that the authors will comment on this point.

References
1. Suzuki N, Hirano Y, Tomiyasu T, et al. Collateral vessels on optical coherence tomography angiography in eyes with branch retinal vein occlusion. Br J Ophthalmol (in press).
2. Klein R, Klein B, Henkind P, et al. Retinal collateral vessel formation. Invest Ophthalmol 1971;10:471-80.
3. Christoffersen NL, Larsen M. Pathophysiology and hemodynamics of branch retinal vein occlusion. Ophthalmology 1999;106:2054-62.
4. Suzuki N, Hirano Y, Yoshida M, et al. Microvascular abnormalities on optical coherence tomography angiography in macular edema associated with branch retinal vein occlusion. Am J Ophthalmol 2016;161:126-32.
5. Freund KB, Sarraf D, Leong BCS, et al. Association of optical coherence tomography angiography of collaterals in retinal vein occlusion with major venous outflow through the deep vascular complex. JAMA Ophthalmology 2018;136:1262-70.

Dear Editor,

We thank Drs Sabherwal and Sood for their interest in our article.(1) We would like to respond to the interesting points they raise.

Table 3 presents our analyses of potential predictors of the correct diagnosis by rural doctors of diabetic retinopathy (DR) requiring treatment. Details on a number of the characteristics assessed in this table are presented in the first paragraph of the Results section, but not, as Drs Sabherwal and Sood point out, the proportion having received didactic training. Among the 28 rural doctors, 13 (46.4%) received such training and 15 (53.6%) did not.

In the Methods, we describe in detail the training received by ophthalmologists in the CREST (Comprehensive Rural Eye Service and Training) program. As described there, only two doctors per hospital (not all of whom examined patients in the current study) could attend the didactic phase of training at the Zhongshan Ophthalmic Center (ZOC). This is due to the limited number of ophthalmologists at a typical rural Chinese county hospital, and the heavy load of clinical duties. For more doctors to have left their facilities for the two-month didactic training would not have been practical. However, all ophthalmologists participating in the CREST network and in the current study received intensive hands-on training by medical retina experts from ZOC at their own facilities, which included the diagnosis and laser treatment of diabetic retinopathy (DR) as well as the use of the United Kingdom National Health Service Diabetic Eye Screening Program (UK NHS DESP) DR grading system assessed in our study. The median number (inter-quartile range) of hands-on training sessions received in managing DR was 3 (1, 9). As mentioned in our article, CREST is a fully-funded program by Orbis International. It provides DR training well beyond what the typical rural county-level ophthalmologist in China would receive. Training in the UK NHS DESP system was comparable between non-medical graders and doctors. For these reasons, the authors would respectfully disagree with Drs Sabherwal and Sood’s suggestion that our study was biased in favour of non-medical graders. These graders achieved a level of accuracy consistent with the standards of the UK NHS DESP, and the use of non-medical graders has been recommended by bodies such as the UK Royal College of Ophthalmologists.(2)

Drs Sabherwal and Sood also indicate that our use of an arbiter may have improved the accuracy of non-medical graders and added to the expense of the program. Use of an arbiter is standard practice in the UK NHS DESP, whose protocols we follow in CREST, and in most DR screening programs of which we are aware. As noted in our article, performance of graders on those images where the grade was unchanged by arbitration remained good: the median sensitivity was 0.80, specificity was 0.98 and kappa ranged from 0.78-0.88. Use of arbitration is an inherent cost of most such programs, and will be included in our forthcoming paper on cost-effectiveness for DR screening of non-medical graders versus local ophthalmologists. It should be noted that task-shifting approaches such as the use of non-medical graders, by reducing even when they do not eliminate the role of more expensive ophthalmologists, are very likely to be more cost-effective than traditional ophthalmologist-driven approaches.

Finally, Drs Sabherwal and Sood point out 33% of images being deemed of inadequate quality as a short-coming of our program. As clearly described in the Results section of our paper, the majority of such images were “inadequate” only in the sense that a single image, rather than two as required under UK NHS DESP protocols, had been made of the eye. In fact, as we reported, the proportion of eyes to which non-medical graders were unable to assign DR grades was 3.28% (24/732).

The authors would again like to thank Drs Sabherwal and Sood for their interest in our work, and the Editors for the opportunity to clarify the interesting and important issues they have raised.

Best regards,

Helen McAneney, Martha McKenna, Tingting Chen, Miguel Angel Vazquez Membrillo, Ling Jin, Wei Xiao, Tunde Peto, Mingguang He, Ruth Hogg, Nathan Congdon

References

1. Sabherwal S, Sood I. Comments on: "Accuracy of trained rural ophthalmologists versus non-medical image graders in the diagnosis of diabetic retinopathy in rural China". Br J Ophthalmol 2018. https://bjo.bmj.com/content/102/11/1471.responses

2. The Royal College of Ophthalmologists. Improving eye health and reducing sight loss – A ‘Call to Action’ Response. Available at https://www.rcophth.ac.uk/wp-content/uploads/2016/10/RCOphth-Call-to-Act... (Accessed: 8th November 2016)

Dear Editor,

We read the article published by McKenna, et al (1) with great interest and laud them on the quality and design of their study. Screening for diabetic retinopathy in rural, low resource settings is the need of the hour, however models which are cost effective, yet provide intensive screening and continuum of care are limited. Keeping this in mind, we feel that there are a few points requiring further clarity in this article.

The odds-ratio calculated in table 3 displays the significant effect of didactic training on correct diagnosis by rural doctors. However, for the odds-ratio to be calculated, there would have been a comparison group of rural doctors who were not provided didactic training. The numbers of these doctors have not been mentioned, and no details have been provided as to whether they were given any basic level of training related to the program. In the results provided for comparison between rural doctors and the non-medical graders, it has not been made clear whether doctors who had not been provided didactic training were included. In that case, results presented in the study may have been biased towards the non-medical graders.

In the study, the arbitrator changed the grade for a high percentage of the cases, moreover, 33% of the images were not found to be of adequate quality. Hiring an arbitrator, re-checking the grading and assuring high quality images (2) through standard equipment and trained personnel would drive up the cost of the programme. The telemedicine model by itself has already been proved as cost effective for diabetic retinopathy screening. (3) Thus, as mentioned by the authors, calculation of cost-effectiveness of the model mentioned in the study, should be carried out assessing its’ scalability.

We appreciate the opportunity to be able to discuss our views on the subject and the article in question.

References

1. McKenna M, Chen T, McAneney H, et al.
Br J Ophthalmol 2018;102:1471–1476.
2. Taylor DJ, et al. Image-quality standardization for diabetic retinopathy screening. Expert Rev Ophthalmol. 2009;4(5):469–76.
3. Sharafeldin N, Kawaguchi A, Sundaram A, et al. Br J Ophthalmol 2018;102:1485–1491.