Association between metabolic risk factors and optic disc cupping identified by deep learning method.

PURPOSE: This study aims to investigate correlation between metabolic risk factors and optic disc cupping and the development of glaucoma.
METHODS: This study is a retrospective, cross-sectional study with over 20-year-old patients that underwent health screening examinations. Intraocular pressure (IOP), fundus photographs, Body Mass Index (BMI), waist circumference (WC), serum triglycerides, serum HDL cholesterol (HDL-C), serum LDL cholesterol (LDL-C), systolic blood pressure (BP), diastolic BP, and serum HbA1c were obtained to analyse correlation between metabolic risk factors and glaucoma. Eye with glaucomatous optic neuropathy(GON) was defined as having an optic disc with either vertical cup-to-disc ratio(VCDR) ≥ 0.7 or a VCDR difference ≥ 0.2 between the right and left eyes by measuring VCDR with deep learning approach.
RESULTS: The study comprised 15,585 subjects and 877 subjects were diagnosed as GON. In univariate analyses, age, BMI, systolic BP, diastolic BP, WC, triglyceride, LDL-C, HbA1c, and IOP were significantly and positively correlated with VCDR in the optic nerve head. In linear regression analysis as independent variables, stepwise multiple regression analyses revealed that age, BMI, systolic BP, HbA1c, and IOP showed positive correlation with VCDR. In multivariate logistic analyses of risk factors and GON, higher age (odds ratio [OR], 1.054; 95% confidence interval [CI], 1.046-1.063), male gender (OR, 0.730; 95% CI, 0.609-0.876), more obese (OR, 1.267; 95% CI, 1.065-1.507), and diabetes (OR, 1.575; 95% CI, 1.214-2.043) remained statistically significant correlation with GON.
CONCLUSIONS: Among the metabolic risk factors, obesity and diabetes as well as older age and male gender are risk factors of developing GON. The glaucoma screening examinations should be considered in the populations with these indicated risk factors.

Introduction

Glaucoma is defined as a progressive optic neuropathy in which specific damage to the optic nerve and visual field (VF) defects occur. Although the most important risk factor for glaucoma is elevated intraocular pressure (IOP), the exact mechanisms underlying the anatomic and functional damage that occur in glaucoma remain unknown [1, 2]. Some risk for glaucoma may be due to systemic risk factors, as has been demonstrated by several previous studies on glaucoma pathogenesis [3, 4].

Metabolic syndrome (MS) is defined as comorbid obesity, hypertension, hyperglycaemia, and hyperlipidaemia. All of these conditions are themselves risk factors for cardiovascular disease and diabetes mellitus [5, 6], which can lead to ischemic vascular abnormalities and may influence glaucoma’s pathogenesis [7, 8]. Despite this, previous studies have demonstrated conflicting results with regard to the association between metabolic syndrome and glaucoma [9, 10]. For instance, some large population-based studies have revealed that hypertension and diabetes mellitus are positively correlated with open angle glaucoma (OAG). However, other studies have reported no association between hypertension and OAG or diabetic mellitus and OAG. Previous studies have also demonstrated that higher BMI is associated with a lower risk of OAG [11, 12].

Because the prevalence of primary OAG with an IOP ≤ 21 mmHg in East Asia, including South Korea and Japan, is higher than that in most previous worldwide reports [13], health care centres have begun to use optic disc evaluation and IOP assessments for glaucoma screening in healthy adults. Then, this makes it possible to investigate the relationship between optic disc cupping and metabolic risk factors in data from the subjects visiting the health care centre.

The present study aims to clarify whether metabolic risk factors are associated with a higher or lower risk of OAG, an association that has remained unclear considering prior findings. Using data from the health care centre, we investigated whether metabolic risk factors are associated with optic disc cupping and the occurrence of glaucomatous optic neuropathy (GON).

Materials and methods

This retrospective, cross-sectional study was performed in accordance with the tenets of the Declaration of Helsinki. The present study was approved by Institutional Review Board of Pusan National University Yangsan Hospital, South Korea. The requirement for informed consent was waived, as only data collected as part of routine health screening examinations was used.

Subjects

Subjects who visited the health promotion centre at Pusan National University Yangsan Hospital from March 2009 to December 2018 and were 20 years or older were enrolled. All subjects underwent IOP measurements and fundus photography to screen for glaucoma. IOP was measured with a non-contact tonometer (Canon T-2, Canon, Tokyo, Japan), and the mean value of IOP which was measured consecutively three times was calculated. Fundus photographs were taken using a digital non-mydriatic fundus camera (TRC-NW200; Topcon, Tokyo, Japan). In healthy subjects, the right eye of each subject was used for statistical analyses and in subjects with GON, data from eyes with glaucoma were obtained.

Measurement of optic disc cupping and definition of glaucomatous optic neuropathy

All fundus photographs were reviewed by deep learning system. We trained the state-of-the-art object-detection deep learning architecture, YOLO V3 [14]. The script codes for these architectures and darknet C source codes were directly downloaded from the homepage of the darknet and compiled in a Windows console application using Microsoft Visual Studio 2015 [15]. The hardware used included an Intel 8th generation central processing unit (CPU) (i5-8400, 2.81 GHz, 32 GB main memory) and an NVIDIA Titan Xp (12 GB; Santa Clara, CA, USA). Deep learning system was trained to find the location of the optic nerve head (ONH) and determine its vertical cup-to-disc ratio (VCDR) (S1 Fig). VCDR is a continuous number from 0 to 1 and, to label it, the number was binned by 0.1 starting from 0.3. A VCDR < 0.3 was labeled as ‘0.3’ (which actually meant ≤0.3) because physicians also had greater disagreement when determining a VCDR < 0.3 [16]. Continuous measurements of the VCDR parameter were recorded, and eyes were classified with or without GON or using the following criteria: Eyes with GON were defined as having an optic disc with either VCDR ≥ 0.7 or a VCDR difference ≥ 0.2 between the right and left eyes.

Definition of metabolic risk factors

MS was defined according to the National Cholesterol Education Program Adult Treatment Panel III (NCEP ATP III), with Korean-specific guidelines for waist circumference. The components of MS according to these guidelines were waist circumference (WC), serum triglycerides, serum HDL-cholesterol, systolic and diastolic blood pressure (BP), and fasting glucose. In addition, given that obesity is a cause of metabolic abnormalities such as dyslipidaemia, hypertension, and diabetes mellitus, Body Mass Index (BMI) was also analysed in the present study.

The present study defined metabolic risk factors including BMI, WC, serum triglycerides, serum HDL cholesterol (HDL-C), serum LDL cholesterol (LDL-C), systolic BP, diastolic BP, and serum HbA1c. The Measurements of height, weight, and waist circumference (WC) were also taken for all subjects, and BMI was calculated as weight (kg) divided by height (m) squared. Blood samples were obtained and analysed for serum triglycerides, HDL-C, LDL-C, and HbA1c. Systolic and diastolic BP were measured with a brachial Riva-Rocci sphygmomanometer on the upper left arm with the patient in a seated position.

Statistical analyses

Data were collected and analysed separately for males and females. Statistical analyses were performed using SPSS version 20.0 (SPSS, Chicago, IL). Student’s t-test and chi-square tests were used to compare demographic and clinical data between healthy eyes and those with GON. Pearson correlation coefficients were calculated to evaluate correlations between VCDR and age, BMI, systolic BP, diastolic PB, WC, serum triglyceride, HDL-C, LDL-C, HbA1c, and IOP. A multiple regression analysis with stepwise variable selection was conducted to assess the effects of metabolic risk factors on enlarging VCDR in the optic nerve head.

In risk factor analyses, the metabolic risk factors were categorized as follows. BMI was categorized into BMI less than 25 kg/m2, and BMI of 25 kg/m2 or more. BP was categorized into SBP / DBP < 130 / 80 mmHg, and SBP / DBP ≥ 130 / 80 mmHg. Participants were divided into 2 groups by WC: WC less than 90 cm for men and less than 85 cm for women, and WC of 90 cm or more for men and 85 cm or more for women. TG was categorized into TG less than 150 mg/dL, and TG of 150 mg/dL or more. HDL was categorized as follows: HDL of 40 mg/dL or more for men and 50 mg/dL or more for women, and HDL less than 40 mg/dL for men and less than 50 mg/dL for women. Subjects were divided into 2 groups by HbA1c: HbA1c less than 6.5%, and HbA1c of 6.5% or more. For evaluating the potential risk factor for GON, univariate logistic regression analysis was conducted, and only variables with a P value of less than 0.10 were selected as independent variables in multivariate logistic regression model. Odds ratios with 95% CI values were calculated in all of the regression analyses. Statistical P value < 0.05 was considered statistically significant.

Results

Among the 16,153 participants who underwent health screening examinations, 568 were excluded due to incomplete medical check-up records (n = 122) or poor fundus photography quality (n = 446). Of the remaining 15,585 subjects, 877 had GON defined with a VCDR ≥ 0.7 or a VCDR difference between the right and left eyes of ≥ 0.2 (Fig 1).

Fig 1

Flow chart of eligible participants recruited from the Pusan National University Yangsan Hospital Health Study.

Among 16,153 participants underwent health care examinations, 15,585 participants were enrolled in study and 877 participants were diagnosed with glaucomatous optic neuropathy.

Table 1 compares the clinical characteristics of eyes with and without GON. Subjects with GON were more likely to be older (P < 0.001), male (P < 0.001), and had significantly higher BMI (P = 0.002), systolic BP (P < 0.001), diastolic BP (P < 0.001), WC (P < 0.001), HbA1c (P < 0.001), and lower LDL-C (P = 0.019) than those without GON.

Table 1

Subject characteristics based on the presence of glaucomatous optic neuropathy.

Characteristics	Eyes with GON	Eyes without GON	P value
	(n = 877)	(n = 14708)
Age (years)	54.74 ± 11.22	48.57 ± 10.84	< 0.001a
Sex (male /female)	584 / 293	8432 / 6276	< 0.001b
Body mass index (kg/m2)	24.32 ± 3.39	23.95 ± 3.25	0.002 a
Systolic blood pressure (mmHg)	122.29 ± 14.71	119.31 ± 13.50	< 0.001 a
Diastolic blood pressure (mmHg)	79.44 ± 10.45	77.98 ± 10.08	< 0.001 a
Waist circumference (cm)	85.08 ± 8.88	83.67 ± 9.48	< 0.001 a
Triglyceride (mg/dL)	125.90 ± 76.90	128.70 ± 92.79	0.302 a
HDL cholesterol (mg/dL)	53.66 ± 13.25	54.19 ± 12.84	0.249 a
LDL cholesterol (mg/dL)	124.98 ± 34.07	127.75 ± 34.11	0.019 a
HbA1c (%)	5.32 ± 1.65	5.03 ± 1.72	< 0.001 a
Intraocular pressure (mmHg)	12.34 ± 3.25	12.21 ± 2.95	0.22 a

a By Student t test,

b By chi square test.

Comparison of VCDR between eyes with GON and without GON area shown in Fig 2. In 877 eyes with GON, there was 673 eyes in VCDR 0.7, and 204 eyes in VCDR 0.8. 14708 eyes without GON were classified into 3891 eyes as VCDR 0.3, 2068 eyes as VCDR 0.4, 4619 eyes as VCDR 0.5, and 4130 eyes as VCDR 0.6. The mean VCDR value (0.723 ± 0.042) in eyes with GON were significantly higher than that (0.462 ± 0.115) in eyes without GON (P < 0.001) (Fig 2).

Fig 2

Comparison of vertical cup disc ratio (VCDR) between eyes with glaucomatous optic neuropathy (GON) and without GON.

The independent t-test showed statistically significantly large VCDR in eyes with GON compared with eyes without GON (P < 0.001).

Correlation coefficients were calculated to evaluate the effects of age, BMI, systolic BP, diastolic BP, WC, triglyceride, HDL-C, LDL-C, HbA1c, and IOP on VCDR. Per univariate analyses, age, BMI, systolic BP, diastolic BP, WC, triglyceride, LDL-C, HbA1c, and IOP were significantly and positively correlated with VCDR (P < 0.001, 0.006, < 0.001, < 0.001, < 0.001, 0.007, < 0.001, < 0.001, < 0.001, respectively). HDL-C was negatively correlated with VCDR (P < 0.001) (Table 2).

Table 2

Pearson correlation coefficients for vertical-cup-disc-ratio and age, metabolic risk factors, and intraocular pressure.

	Vertical-cup-disc-ratio, P valuea
Age	r = 0.182, P < 0.001
Body mass index	r = 0.013, P = 0.006
Systolic blood pressure	r = 0.086, P < 0.001
Diastolic blood pressure	r = 0.074, P < 0.001
Waist circumference	r = 0.047, P < 0.001
Triglyceride	r = 0.028, P = 0.007
HDL cholesterol	r = -0.025, P < 0.001
LDL cholesterol	r = 0.017, P < 0.001
HbA1c	r = 0.032, P < 0.001
Intraocular pressure	r = 0.044, P < 0.001

a By Pearson correlation coefficients.

When age, BMI, systolic BP, diastolic BP, WC, triglyceride, HDL-C, LDL-C, HbA1c, and IOP were entered into a linear regression analysis as independent variables, stepwise multiple regression analyses revealed that older age, higher BMI, higher systolic BP, higher HbA1c, and higher IOP were associated with larger VCDR in the optic nerve head (P < 0.001, 0.022, 0.037, 0.024, < 0.001, respectively) (Table 3) (Fig 3).

Table 3

Results of multivariate regression analyses of enlarged vertical cup-disc ratio.

	Vertical-cup-disc-ratio
	Β (95%)	P valuea
Age	0.002 (0.0019–0.0023)	< 0.001
Body mass index	0.001 (0.0003, 0.002)	0.022
Systolic blood pressure	0.001 (0.0008, 0.0011)	0.037
HbA1c	0.004 (0.002, 0.005)	0.024
Intraocular pressure	0.002 (0.001, 0.003)	< 0.001

a By stepwise method.

Fig 3

Scatter plots showing the relationship between various factors and vertical cup-disc ratio (VCDR) in total subjects.

(A) Age and VCDR, (B) Body mass index and VCDR, (C) Systolic blood pressure and VCDR, (D) HbA1c and VCDR, and (E) intraocular pressure and VCDR. The dashed lines represent the 95% confidence intervals for the solid trend lines.

Univariate logistic analyses revealed that GON was significantly associated with older age, male gender, higher BMI, higher BP, and higher HbA1c (P < 0.001, < 0.001, 0.007, < 0.001, < 0.001, respectively). Those variables were evaluated in multivariable models with the covariates, using backward stepwise selection to eliminate those with P values > 0.05. Higher age, male gender, higher BMI, and higher HbA1c remained statistically significant via multivariable analyses (P < 0.001, < 0.001, 0.007, < 0.001, respectively) (Table 4). Moreover, the result of the predicting the GON has shown the 79.4% accuracy from multivariate logistic regression model.

Table 4

Logistic regression analyses of metabolic risk factors and their associations with glaucomatous optic neuropathy.

	Univariate		Mutivariatea
	OR (95% CI)	P	OR (95%)	P
Age	1.053 (1.046, 1.060)	<0.001	1.054 (1.046, 1.063)	<0.001
Gender (reference: Male)	0.674 (0.584,0.779)	<0.001	0.708 (0.587, 0.854)	<0.001
Body mass index (reference < 25 kg/m2)	1.214 (1.055, 1.397)	0.007	1.267 (1.065, 1.507)	0.007
Blood pressure (SBP / DBP) (reference < 130 / 80 mmHg)	1.376 (1.174, 1.613)	<0.001	1.068 (0.867, 1.317)	0.534
Waist circumference (reference < 90 cm or men, <85 for women)	1.099 (0.957, 1.263)	0.082	1.025 (0.856, 1.227)	0.790
Triglyceride (reference < 150 mg/dL)	0.933 (0.799, 1.089)	0.378	0.845 (0.694, 1.030)	0.086
HDL cholesterol (reference ≥ 40mg/dL for men, ≥ 50 mg/dL for women)	0.997 (0.991, 1.002)	0.916	0.981 (0.787, 1.224)	0.867
HbA1c (reference < 6.5%)	2.274 (1.769, 2.922)	<0.001	1.606 (1.237, 2.086)	< 0.001
Intraocular pressure	1.016 (0.993, 1.039)	0.099	1.016 (0.986, 1.048)	0.299

a By backward stepwise logistic regression model.

Discussion

In the present study, older age, higher IOP, higher BMI, higher systolic BP, and higher HbA1c were correlated with higher VCDR. In addition, of multiple metabolic risk factors, more obese and diabetes as well as older age and male gender were associated with an increased prevalence of GON. BMI had positively correlated with VCDR, and was also a risk factor for GON.

Given that BMI is a major indicator of obesity, previous studies have evaluated the association between BMI and glaucoma or increased IOP [8]. However, such prior work has revealed contradictory results on the relationship between BMI and glaucoma or BMI and IOP. Interestingly, some earlier studies demonstrated a negative association between BMI and glaucoma. This may result from impaired endothelium-dependent vasodilatation or a higher translaminar cribrosa pressure gradient induced by lower cerebrospinal fluid pressure, which is protective against glaucoma [17, 18]. Other studies reported that a higher BMI is a risk factor for glaucoma and further suggested that the positive association between BMI and glaucoma might be related to the presence of excess intraorbital fat tissue and increased resistance to outflow in the episcleral veins given increased blood viscosity with obesity [19, 20]. In agreement with the present study, a meta-analysis by Liu et al. revealed that adiposity (including BMI, WC, and waist-to-hip ratio) is associated with increased risk for elevated IOP and glaucoma [21]. The discrepancies in these results may be related to differences in their measurement methods, eligibility criteria, ethnicity, and participant composition. Additional studies are needed to illustrate more definitively, the relationship between obesity and glaucoma.

Higher HbA1c, an indicator of glycaemic control, is also positively correlated with VCDR and is a risk factor for glaucoma [3, 22]. Evidence from previous studies has supported the positive association found here between diabetes and glaucoma [3, 22]. Various contributing mechanisms have been proposed, including structural and functional abnormalities in the small vessels that feed the optic nerve, as well as increased susceptibility of retinal ganglion cells to apoptosis [23, 24]. In addition, Welinder et al. reported a positive association between HbA1c levels and glaucoma [25]. Zhao et al. further reported a significant relationship between HbA1c and glaucoma, suggesting that those with higher HbA1c’s may be at greater glaucoma risk [23].

The present study also found a positive correlation between systolic BP and VCDR, despite the fact that elevated systolic BP was not a risk factor for glaucoma. The relationship between blood pressure and glaucoma remains controversial. A meta-analysis of the association between hypertension and open angle glaucoma revealed that systemic hypertension increases the risk of developing open angle glaucoma [26, 27]. Previous studies have also demonstrated that systemic hypertension may directly contribute to impaired microvasculature in the anterior optic nerve and increased IOP via overproduction or decreased outflow of aqueous humour due to abnormal autoregulation of the ciliary circulation [28, 29]. However, earlier studies also revealed that low blood pressure was a risk factor of developing glaucoma [30, 31], potentially leading to low perfusion pressure and subsequent glaucomatous changes in the optic nerve head (e.g., decreased rim area, increased cup area, and greater VCDR) [30, 32]. In addition, Jonas et al. reported that the neuroretinal rim area in healthy eyes without arterial hypertension was thicker than that in individuals with arterial hypertension, and that disc haemorrhage risk was also significantly different between glaucomatous eyes with and without arterial hypertension [18]. Though systolic BP may be positively associated with VCDR in the optic nerve head, various BP-related parameters including lower perfusion pressure, large BP fluctuations at night, systemic hypotension, antihypertensive treatments, and systemic cardiovascular disease with hypertension may also contribute to glaucoma risk. Thus, further investigation is necessary to determine more comprehensively, the influence of BP changes on vascular insufficiency in glaucoma.

In the present study, males had a higher prevalence of GON than females. In agreement with this, a large number of population-based studies using multivariate analysis approaches have indicated that males have a higher odds of developing glaucoma [2, 33–36]. This suggests that the low incidence of glaucoma in females may be related to the protective influence of endogenous oestrogen or exogenous hormone replacement following menopause [37, 38].

To the best of our knowledge, the present is the first report to evaluate continuously, the association between various metabolic risk factors and optic cup-to-disc ratio, a clinical marker of glaucomatous damage. Previous studies have simply evaluated anthropometric parameters (BP, glucose level, BMI etc) as risk factors for the prevalence of glaucoma. Measuring VCDR in the optic nerve head is essential for diagnosing glaucoma, and thus the continuous analysis of associated metabolic risk factors for increased VCDR in the optic nerve head is necessary to enhance our understanding of glaucoma risk factors. Despite its strengths, the present study does have several limitations, which warrant some discussion. First, it was not population-based but rather health centre-based, introducing a potential source of selection bias. Second, subjects did not undergo a complete glaucoma examination, including retinal nerve fibre layer photography, visual field examination, and optical coherence tomography, in the present study. We defined GON configuration as category 1, per the International Society of Geographical and Epidemiological Ophthalmology guidelines. However, optic nerve heads with GON cannot be diagnosed as glaucoma because category 1 requires visual field defects that correspond with glaucomatous damage. Third, we were unable to evaluate subjects for other ocular factors, such as axial length, refractive errors, or corneal thickness, which can affect optic disc measurements. Fourth, because the present study utilized a cross-sectional design, a causal relationship between glaucoma and associated metabolic risk factors cannot be established. Fifth, since the participants to undergo health screening examination enrolled in the health-care centre, the imbalance between normal subjects and study group was inevitably occurred in the present study. Finally, this study may have limitations from the inclusion of eyes with tilted discs, for which the optic disc analysis using deep learning system may have been erroneous. However, these ambiguous images were reviewed by an experienced ophthalmologist in an attempt to minimize any limitations.

In conclusion, our findings reveal that subjects with higher BMI, SBP, and HbA1c have greater VCDR in the optic nerve head. In a health centre-based Korean population, we also identified old age, male gender, obesity, and diabetes as significant GON risk factors. Given these findings, populations with the indicated risk factors should be targeted for glaucoma screening examinations, and population-based screening approaches should be implemented to reduce the under or misdiagnosis of glaucoma, which, if untreated, can result in severe outcomes including blindness.

A diagram of deep learning architecture, YOLO V3 algorithm flow.

(PNG)

Click here for additional data file.

12 Jun 2020

PONE-D-20-07735

Association between metabolic risk factors and optic disc cupping identified by deep learning method

PLOS ONE

Dear Dr. Lee,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

The two reviewers have offered constructive criticisms that need to be addressed during revision.

Please submit your revised manuscript by Jul 27 2020 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols

We look forward to receiving your revised manuscript.

Kind regards,

Sanjoy Bhattacharya

Academic Editor

PLOS ONE

Journal Requirements:

When submitting your revision, we need you to address these additional requirements.

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at

https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and

https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

2. We note that you have indicated that data from this study are available upon request. PLOS only allows data to be available upon request if there are legal or ethical restrictions on sharing data publicly. For information on unacceptable data access restrictions, please see http://journals.plos.org/plosone/s/data-availability#loc-unacceptable-data-access-restrictions.

In your revised cover letter, please address the following prompts:

a) If there are ethical or legal restrictions on sharing a de-identified data set, please explain them in detail (e.g., data contain potentially identifying or sensitive patient information) and who has imposed them (e.g., an ethics committee). Please also provide contact information for a data access committee, ethics committee, or other institutional body to which data requests may be sent.

b) If there are no restrictions, please upload the minimal anonymized data set necessary to replicate your study findings as either Supporting Information files or to a stable, public repository and provide us with the relevant URLs, DOIs, or accession numbers. Please see http://www.bmj.com/content/340/bmj.c181.long for guidelines on how to de-identify and prepare clinical data for publication. For a list of acceptable repositories, please see http://journals.plos.org/plosone/s/data-availability#loc-recommended-repositories.

We will update your Data Availability statement on your behalf to reflect the information you provide.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

Reviewer #2: Partly

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: No

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: No

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: No

Reviewer #2: Yes

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: There are only a few grammatical errors to correct.

In the abstract line 23 and 24, reword it to say 'with over 20 year-old patients THAT underwent health screening examinations'

Line 134: Should it read 'HDL was categorized as follows' with an 's' at the end of 'follow' ?

Line 136: divided into 2 'groups' instead of 'group'

Line 203: Add 'a' to 'BMI had positively correlated with VCDR, and was also a risk factor for GON.'

Reviewer #2: This paper did a great job in presenting a statistical analysis of correlations between various metabolic factors and Optic Disc Cupping, there are some major areas that need improvement. Some highlights: The discussion was clear, and the authors did a great job contextualizing their results with prior work; The paper was also organized nicely and there was a logical flow to the writing. Below are recommendations for revision:

The one major challenge with this paper was the obvious class imbalance. There was an almost 17:1 ratio of healthy subjects to GON subjects used in the statistical analysis. The authors should address this obvious class imbalance, and perhaps discuss the limitations of their models’ predictions given this fact. Other attempts to resolve this could include incorporating a focal loss cost function into their statistical methods.

To allow readers to better understand what the authors meant by the YOLO deep learning model, a system diagram detailing the chosen deep learning model architecture is required. The diagram should include abbreviated number of layers, feature maps, input and output. The trained model with weights should be included as a supplemental file or a link that can be accessed (I know models tend to be massive in size).

An additional system diagram detailing the flow of the entire process, from raw to final predictions would also be helpful for the reader to visualize the authors’ transition from raw clinical data to mathematical and clinical insights.

In regard to the statistical methods, there need to be clear mathematical equations, especially for the logistic/linear regressions. Describe reasons for using each statistical model. It was unclear if authors used a one tailed or two tailed t-test.

For results, the authors should not only include correlation coefficients, but also include the specificity and sensitivity (in the form of an F1 score perhaps). A confusion matrix would help readers visualize how univariate or multivariate logistic models performed in predicting GON.

There is much work to be done to improve this paper. Luckily, a lot of it is mostly computational, so I am confident the authors have the opportunity to improve upon the work.

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: Yes: Jada Morris

Reviewer #2: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

12 Jul 2020

Dear Editor,

We appreciate the reviewers for their efforts and the constructive comments. A point-by-point response to each of reviewers’ comments follows. The authors wish to thank the Editorial Board for their thoughtful consideration and recommendations, and hope that this revised manuscript now meets your requirements for publication.

There are only a few grammatical errors to correct.

In the abstract line 23 and 24, reword it to say 'with over 20 year-old patients THAT underwent health screening examinations'

Line 134: Should it read 'HDL was categorized as follows' with an 's' at the end of 'follow' ?

Line 136: divided into 2 'groups' instead of 'group'

Line 203: Add 'a' to 'BMI had positively correlated with VCDR, and was also a risk factor for GON.'

A) As your comments, we corrected grammatical errors in manuscript.

The one major challenge with this paper was the obvious class imbalance. There was an almost 17:1 ratio of healthy subjects to GON subjects used in the statistical analysis. The authors should address this obvious class imbalance, and perhaps discuss the limitations of their models’ predictions given this fact. Other attempts to resolve this could include incorporating a focal loss cost function into their statistical methods.

A) We fully agreed with your comments. Unfortunately, the statistical consultant can’t suggest the definite solution about the obvious imbalance, so we described this imbalance between two groups in the limitation as following sentences;

“Fifth, since the participants to undergo health screening examination enrolled in the health-care centre, the imbalance between normal subjects and study group was inevitably occurred in the present study.”

To allow readers to better understand what the authors meant by the YOLO deep learning model, a system diagram detailing the chosen deep learning model architecture is required. The diagram should include abbreviated number of layers, feature maps, input and output. The trained model with weights should be included as a supplemental file or a link that can be accessed (I know models tend to be massive in size).

A) As your comments, we attached the trained model as a supplemental file. After submitting this paper, the study about automatically measuring cup-to-disc ratio using deep-learning approach was published in Scientific Reports. (Park K, Kim J, Lee J. Automatic optic nerve head localization and cup-to-disc ratio detection using state-of-the-art deep-learning architectures. Scientific Reports. 2020.10:5025. doi.org/10.1038/s41598-020-62022-x). We can’t suggest more detailed algorithm of measuring CDR because of commercial availability.

An additional system diagram detailing the flow of the entire process, from raw to final predictions would also be helpful for the reader to visualize the authors’ transition from raw clinical data to mathematical and clinical insights.

A) We fully agreed with your suggestion. Because the previous flow chart didn't include the detailed contents of metabolic risk factors, we added and revised the flow chart in Figure 1.

In regard to the statistical methods, there need to be clear mathematical equations, especially for the logistic/linear regressions. Describe reasons for using each statistical model. It was unclear if authors used a one tailed or two tailed t-test.

A) As your comments, we clearly explained the mathematical equations for logistic regression model in the table 4 and manuscripts as following sentences.

“Univariate logistic analyses revealed that GON was significantly associated with older age, male gender, higher BMI, higher BP, and higher HbA1c (P < 0.001, < 0.001, 0.007, < 0.001, < 0.001, respectively). Those variables were evaluated in multivariable models with the covariates, using backward stepwise selection to eliminate those with P values > 0.05.”

For results, the authors should not only include correlation coefficients, but also include the specificity and sensitivity (in the form of an F1 score perhaps). A confusion matrix would help readers visualize how univariate or multivariate logistic models performed in predicting GON.

A) We fully agreed with your comments. So, we added the specificity and sensitivity of the predicted probability as following sentence.

“ in logistic regression results, the predicted probability to discriminate between normal and GON showed 89.2 % specificity and 27.2 % sensitivity.”

Submitted filename: response_to_reviewers.docx

Click here for additional data file.

11 Aug 2020

PONE-D-20-07735R1

Association between metabolic risk factors and optic disc cupping identified by deep learning method

PLOS ONE

Dear Dr. Lee,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

A reviewer have raised a few criticisms. They need to be addressed to the extent possible by incorporating changes in the manuscript. A revised manuscript will be worthy of further consideration.

Please submit your revised manuscript by Sep 25 2020 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols

We look forward to receiving your revised manuscript.

Kind regards,

Sanjoy Bhattacharya

Academic Editor

PLOS ONE

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #1: All comments have been addressed

Reviewer #2: All comments have been addressed

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

Reviewer #2: Yes

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: Yes

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: (No Response)

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: (No Response)

**********

6. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: (No Response)

Reviewer #2: Overall, I'm glad the researchers made significant improvements to the paper and it reads as more rigorous and technically sound.

Some additional comments:

1. Please also include the email of the IRB for data access (most researchers don't have access to international calls)

2. Figure 2 has "vertical cup disk ratio" on the y axis. Is this correct, or is it the number of samples?

3. Please explain the significance of a 89.2 % specificity and 27.2 % sensitivity. There is a stark contrast between the true positive and true negative rate; this needs to be discussed.

Thank you so much for the prior revisions; these are minor edits to an overall insightful study.

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: Yes: Jada Morris

Reviewer #2: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

21 Aug 2020

Dear Editor,

We appreciate the reviewers for their efforts and the constructive comments. A point-by-point response to each of reviewers’ comments follows. The authors wish to thank the Editorial Board for their thoughtful consideration and recommendations, and hope that this revised manuscript now meets your requirements for publication.

1. Please also include the email of the IRB for data access (most researchers don't have access to international calls)

A) The study protocol was approved by the Pusan National University Yangsan Hospital Institutional Review Board and you can access for data through below email address pnuyhirb@gmail.com.

2. Figure 2 has "vertical cup disk ratio" on the y axis. Is this correct, or is it the number of samples?

A) We make a mistake indexing on the y axis. As you would expect, index on the y axis was the number of samples. So, we changed the figure 2 as below.

3. Please explain the significance of a 89.2 % specificity and 27.2 % sensitivity. There is a stark contrast between the true positive and true negative rate; this needs to be discussed.

A) We fully agreed with your comments. As your comment in first revision, we previously analyzed the specificity and sensitivity for predicting GON by using multivariate logistic model, and described the 89.2% specificity and 27.2% sensitivity in the manuscript. Since cut-off value was set up as 0.5 automatically in SPSS program, we did not try calculating the various predicted probability as cut-off value was ranged from 0 to 1.0. According to the statistical consultant’s advices (Mi Sook Yoon PhD., msyun@pusan.ac.kr), we analyzed the all available parameters (accuracy, sensitivity, specificity, precision) showing the predicted probability in the range from 0.3 to 1.0 of cut-off value as below.

The low value of sensitivity in our study may be influenced by the parameters analyzed in logistic regression method, and could be improved as adding the useful diagnostic factors as the independent parameters in logistic regression analysis. In fact, the statistical results showed that the value of sensitivity was increased as the cut-off value was lower than 0.5, and then we should investigate the further evaluation including the ROC analysis.

In some studies of ophthalmogy (Sharma, et al. Post penetrating keratoplasty glaucoma: Cumulative effect of quantifiable risk factors. DOI: 10.4103/0301-4738.129790 / Lu et al. Comparison of Ocular Biomechanical Machine Learning Classifiers for Glaucoma Diagnosis. DOI: 10.1109/BIBM.2018.8621238), the goodness-of-fit of logistic regression models uses the accuracy of the classification table. Moreover, as your comments, there was outstanding gap between the sensitivity and specificity rate. Then, we and statistical consultant determined that the accuracy rate could be the appropriate parameter as probability for predicting GON.

So, we described these results as following sentences. (page 11, line 1)

“The result of the predicting the GON has shown the 79.4% accuracy from multivariate logistic regression model.”

Submitted filename: response_to_reviewers_2nd_revision.docx

Click here for additional data file.

31 Aug 2020

Association between metabolic risk factors and optic disc cupping identified by deep learning method

PONE-D-20-07735R2

Dear Dr. Lee,

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

Kind regards,

Sanjoy Bhattacharya

Academic Editor

PLOS ONE

Additional Editor Comments (optional):

Reviewers' comments:

3 Sep 2020

PONE-D-20-07735R2

Association between metabolic risk factors and optic disc cupping identified by deep learning method

Dear Dr. Lee:

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

If we can help with anything else, please email us at plosone@plos.org.

Thank you for submitting your work to PLOS ONE and supporting open access.

Kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Dr. Sanjoy Bhattacharya

Academic Editor

PLOS ONE