Available via license: CC BY-NC-ND 4.0
Content may be subject to copyright.
Retinal Age as a Predictive Biomarker for Mortality Risk
Running title: Retinal age predicts mortality
Authors
Zhuoting Zhu, MD PhD1
Danli Shi, MD2
Guankai Peng, BS3
Zachary Tan, MBBS, MMed, MMSc4
Xianwen Shang, PhD1
Wenyi Hu, MBBS1
Huan Liao, MD5
Xueli Zhang, PhD1
Yu Huang, MD PhD1
Honghua Yu, MD PhD1
Wei Meng, BS3
Wei Wang, MD PhD2
Xiaohong Yang, MD PhD1
Mingguang He, MD PhD1,2,4,6
Affiliations
1. Department of Ophthalmology, Guangdong Academy of Medical Sciences,
Guangdong Provincial People's Hospital, Guangzhou, China.
2. State Key Laboratory of Ophthalmology, Zhongshan Ophthalmic Center, Sun
Yat-sen University, Guangzhou, China.
3. Guangzhou Vision Tech Medical Technology Co., Ltd.
4. Centre for Eye Research Australia; Ophthalmology, University of Melbourne,
Melbourne, Australia.
5. Neural Regeneration Group, Institute of Reconstructive Neurobiology, University
of Bonn, Bonn, Germany.
. CC-BY-NC-ND 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint The copyright holder for thisthis version posted December 30, 2020. ; https://doi.org/10.1101/2020.12.24.20248817doi: medRxiv preprint
NOTE: This preprint reports new research that has not been certified by peer review and should not be used to guide clinical practice.
6. Ophthalmology, Department of Surgery, University of Melbourne, Melbourne,
Australia
Word count: Summary: 249; Research in context: 286; Whole paper: 2549.
Tables: 2; Figures: 5.
Corresponding author
Mingguang He, MD PhD
Email: mingguang.he@unimelb.edu.au
Xiaohong Yang, MD PhD
Email: syyangxh@scut.edu.cn
Wei Wang, MD PhD
Email: zoc_wangwei@yahoo.com
. CC-BY-NC-ND 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint The copyright holder for thisthis version posted December 30, 2020. ; https://doi.org/10.1101/2020.12.24.20248817doi: medRxiv preprint
Summary
Background
Ageing varies substantially, thus an accurate quantification of ageing is important. We
developed a deep learning (DL) model that predicted age from fundus images (retinal
age). We investigated the association between retinal age gap (retinal
age-chronological age) and mortality risk in a population-based sample of
middle-aged and elderly adults.
Methods
The DL model was trained, validated and tested on 46,834, 15,612 and 8,212 fundus
images respectively from participants of the UK Biobank study alive on 28th February
2018. Retinal age gap was calculated for participants in the test (n=8,212) and death
(n=1,117) datasets. Cox regression models were used to assess association between
retinal age gap and mortality risk. A restricted cubic spline analyses was conducted to
investigate possible non-linear association between retinal age gap and mortality risk.
Findings
The DL model achieved a strong correlation of 0·83 (P<0·001) between retinal age
and chronological age, and an overall mean absolute error of 3·50 years. Cox
regression models showed that each one-year increase in the retinal age gap was
associated with a 2% increase in mortality risk (hazard ratio=1·02, 95% confidence
interval:1·00-1·04, P=0·021). Restricted cubic spline analyses showed a non-linear
relationship between retinal age gap and mortality (Pnon-linear=0·001). Higher retinal
age gaps were associated with substantially increased risks of mortality, but only if
the gap exceeded 3·71 years.
Interpretation
Our findings indicate that retinal age gap is a robust biomarker of ageing that is
closely related to risk of mortality.
Funding
National Health and Medical Research Council Investigator Grant, Science and
Technology Program of Guangzhou.
. CC-BY-NC-ND 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint The copyright holder for thisthis version posted December 30, 2020. ; https://doi.org/10.1101/2020.12.24.20248817doi: medRxiv preprint
Key words: retinal age, mortality, prediction
. CC-BY-NC-ND 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint The copyright holder for thisthis version posted December 30, 2020. ; https://doi.org/10.1101/2020.12.24.20248817doi: medRxiv preprint
Research in context
Evidence before this study
Ageing at an individual level is heterogeneous. An accurate quantification of the
biological ageing process is significant for risk stratification and delivery of tailored
interventions. To date, cell-, molecular-, and imaging-based biomarkers have been
developed, such as epigenetic clock, brain age and facial age. While the invasiveness
of cellular and molecular ageing biomarkers, high cost and time-consuming nature of
neuroimaging and facial ages, as well as ethical and privacy concerns of facial
imaging, have limited their utilities. The retina is considered a window to the whole
body, implying that the retina could provide clues for ageing.
Added value of this study
We developed a deep learning (DL) model that can detect footprints of aging in
fundus images and predict age with high accuracy for the UK population between 40
and 69 years old. Further, we have been the first to demonstrate that each one-year
increase in retinal age gap (retinal age-chronological age) was significantly associated
with a 2% increase in mortality risk. Evidence of a non-linear association between
retinal age gap and mortality risk was observed. Higher retinal age gaps were
associated with substantially increased risks of mortality, but only if the retinal age
gap exceeded 3·71 years.
Implications of all the available evidence
This is the first study to link the retinal age gap and mortality risk, implying that
retinal age is a clinically significant biomarker of ageing. Our findings show the
potential of retinal images as a screening tool for risk stratification and delivery of
tailored interventions. Further, the capability to use fundus imaging in predicting
ageing may improve the potential health benefits of eye disease screening, beyond the
detection of sight-threatening eye diseases.
. CC-BY-NC-ND 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint The copyright holder for thisthis version posted December 30, 2020. ; https://doi.org/10.1101/2020.12.24.20248817doi: medRxiv preprint
Introduction
Globally, the population aged 60 and over is estimated to reach 2·1 billion in 2050.1
Ageing populations place tremendous pressure on health-care systems.2 The rate of
ageing at an individual level is heterogeneous. An accurate quantification of the
biological ageing process is significant for risk stratification and the delivery of
tailored interventions.3
To date, several tissue-, cell-, molecular-, and imaging-based biomarkers have been
developed, such as DNA-methylation status, brain age and three dimensional (3D)
facial age.4-7 While the invasiveness of cellular and molecular ageing biomarkers,
high cost and time-consuming nature of neuroimaging and facial ages, and ethical and
privacy concerns of facial imaging, have limited their utilities.
The retina is considered a window to the whole body.8-12 In addition, the retina is
amenable to rapid, non-invasive, and cost-effective assessments. The advent of deep
learning (DL) has greatly improved the accuracy of image classification and
processing. Recent studies have demonstrated successful applications of DL models
in the prediction of age using clinical images.5,6,13 Taken together, this raises the
potential that biological age can be predicted by applying DL to retinal images. For
optimal utility, viable biomarkers of ageing must also relate to the risk of age-related
morbidity and mortality.
We therefore developed a DL model that can predict age from fundus images, known
as retinal age. Using a large population-based sample of middle-aged and elderly
adults, we investigated the association between retinal age gap, defined as the
difference between retinal age and chronological age, and all-cause mortality.
Methods
Study population
. CC-BY-NC-ND 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint The copyright holder for thisthis version posted December 30, 2020. ; https://doi.org/10.1101/2020.12.24.20248817doi: medRxiv preprint
The UK Biobank is a large-scale, population-based cohort of more than 500,000 UK
residents aged 40-69 years. Participants were recruited between 2006 and 2010, with
all participants completing comprehensive health-care questionnaires, detailed
physical measurements, and biological sample collections. Health-related events were
ascertained via data linkage to hospital admission records and mortality registry.
Ophthalmic examinations were introduced in 2009. The overall study protocol and
protocols for each test have been described in extensive details elsewhere.14
The National Information Governance Board for Health and Social Care and the NHS
North West Multicenter Research Ethics Committee approved the UK Biobank study
(11/NW/0382) in accordance with the principles of the Declaration of Helsinki, with
all participants providing informed consent. The present analysis operates under UK
Biobank application 62525.
Fundus photography
Ophthalmic measurements including LogMAR visual acuity, autorefraction and
keratometry (Tomey RC5000, Tomey GmbH, Nuremberg, Germany), intraocular
pressure (IOP, Ocular Response Analyzer, Reichert, New York, USA), and paired
retinal fundus and optical coherence tomography imaging (OCT, Topcon 3D OCT
1000 Mk2, Topcon Corp, Tokyo, Japan) were collected. A 45-degree non-mydriatic
and non-stereo fundus image centered to include both the optic disc and macula was
taken for each eye. A total of 131,238 images from 66,500 participants were obtained
from the UK Biobank study, among which 80,170 images from 46,970 participants
passed the image quality check.
Deep learning model for age prediction
To build the DL model for age prediction, participants from the UK Biobank study
alive on 28th February 2018 (Nsubj=46,970) were randomly split into three datasets –
training (Nsubj=27,424, 60% of participants), validation (Nsubj=9,142, 20%), and test
(Nsubj=9,142, 20%). For the training and validation datasets, fundus images from both
. CC-BY-NC-ND 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint The copyright holder for thisthis version posted December 30, 2020. ; https://doi.org/10.1101/2020.12.24.20248817doi: medRxiv preprint
eyes (if available) were used to maximise the volume of data available (Nimg=46,834
and 15,612 respectively). For the test dataset, fundus images from right eyes were
selected for primary analyses (Nsubj=8,212), while fundus images from left eyes were
selected for sensitivity analyses.
The development and validation of the DL model for age prediction are outlined in
Figure 1. Briefly, all fundus images were preprocessed by subtracting average color,15
resized to a resolution of 299*299 pixels, and pixel values rescaled to 0~1. After
preprocessing, images were fed into a DL model using a Xception architecture.
During training, data augmentation was performed using random horizontal or
vertical flips and the algorithm optimised using stochastic gradient descent. To
prevent overfitting, we implemented a dropout of 0·5, and carried out early stopping
when validation performance did not improve for 10 epochs. The selection of DL
models was based on performance in the validation set. The performance of the DL
model, including mean absolute error (MAE) and correlation between predicted
retinal age and chronological age, was calculated. We then retrieved attention maps
from the DL models using guided Grad-CAM,16 which highlights pixels in the input
image based on their contributions to the final evaluation.
Retinal age gap definition
The difference between retinal age predicted by the DL model and chronological age
was defined as the retinal age gap. A positive retinal age gap indicated an ‘older’
appearing retina, while a negative retinal age gap indicated a ‘younger’ appearing
retina.
Mortality ascertainment
Mortality status and date of death were ascertained via data linkage to the National
Health Service central mortality registry. Participants who had died from all causes
during the follow-up period (Nsubj=1,117) were included in the death dataset. Duration
of follow-up for each participant (person-year) was calculated as the length of time
. CC-BY-NC-ND 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint The copyright holder for thisthis version posted December 30, 2020. ; https://doi.org/10.1101/2020.12.24.20248817doi: medRxiv preprint
between baseline age and date of death, loss to follow-up, or complete follow-up (28th
February 2018), whichever came first.
Covariates
Factors previously known to be associated with mortality17 were included as potential
confounders in the present analyses. These variables included baseline age, sex,
ethnicity (recorded as white and non-white), Townsend deprivation indices (an
area-based proxy measure for socioeconomic status), education attainment (recorded
as college or university degree, and others), smoking status (recorded as
current/previous and never), physical activity level (recorded as above
moderate/vigorous/walking recommendation and not), general health status (recorded
as excellent/good and fair/poor), and comorbidities (obesity, diabetes mellitus,
hypertension, history of heart diseases, and history of stroke).
Body mass index (BMI) was calculated as body weight in kilograms divided by
height squared. Obesity was defined as BMI >30 kg/m2. Diabetes mellitus was
defined as self-reported or doctor-diagnosed diabetes mellitus, the use of
anti-hyperglycaemic medications or insulin, or a glycosylated haemoglobin level
of >6·5%. Hypertension was defined as self-reported, or doctor-diagnosed
hypertension, the use of antihypertensive drugs, an average systolic blood pressure of
at least 130mmHg or an average diastolic blood pressure of at least 80mmHg.
Self-reported history of angina and heart attack was used to classify history of heart
diseases.
Statistical analyses
Descriptive statistics, including means and standard deviations (SDs), numbers and
percentages, were used to report baseline characteristics of study participants. The
retinal age gap was calculated for participants in the test and death datasets, and
further used to explore the association between retinal age gap and mortality risk. Cox
proportional hazards regression models considering retinal age gap as a continuous
. CC-BY-NC-ND 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint The copyright holder for thisthis version posted December 30, 2020. ; https://doi.org/10.1101/2020.12.24.20248817doi: medRxiv preprint
linear term were fitted to estimate the effect of a one-year increase in retinal age gap
on mortality risk. We then investigated associations between retinal age gaps at
different quantiles with mortality. In addition, a restricted cubic spline analyses of
possible non-linear associations between retinal age gap and mortality status was
performed, with 5 knots placed at equal percentiles of the retinal age gap, and retinal
age gap of zero years used as the reference value. We adjusted Cox models for the
following covariates – baseline age, sex, ethnicity, and Townsend deprivation indices
(model I); additional educational level, obesity, smoking status, physical activity level,
diabetes mellitus, hypertension, general health status, history of heart diseases, and
history of stroke (model II).
The proportional hazards assumption for each variable included in the Cox
proportional hazards regression models were graphically assessed. All variables were
found to meet the assumption. A two-sided p value of < 0·05 indicated statistical
significance. Analyses were performed using R (version 3.3.0, R Foundation for
Statistical Computing, www.R-project.org, Vienna, Austria) and Stata (version 13,
StataCorp, Texas, USA).
Role of the funding source
The funders had no role in study design, data collection, data analyses, data
interpretation, preparation of the manuscript, and decision to publish. The
corresponding author had full access to all data and final responsibility for the
decision to submit for publication.
Results
Study sample
The study population characteristics are described in Table 1. The DL model was
trained and validated on subsets of participants with mean ages of 55·6 ± 8·21 and
55·7 ± 8·19 years; and with 55·9% and 55·2% female, respectively. For the test and
. CC-BY-NC-ND 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint The copyright holder for thisthis version posted December 30, 2020. ; https://doi.org/10.1101/2020.12.24.20248817doi: medRxiv preprint
death datasets, participants had mean ages of 55·5 ± 8·22 and 61·0 ± 6·67 years; and
were 55·1% and 42·2% female, respectively.
Deep learning model performance for age prediction
Figure 2A shows the performance of the DL model on the test dataset. The trained DL
model was able to achieve a strong correlation of 0·83 (P<0·001) between predicted
retinal age and chronological age, with an overall MAE of 3·50 years. Two
representative examples of fundus images with corresponding attention maps for age
prediction are shown in Figure 3. Regions around retinal vessels are highlighted by
the DL model for age prediction.
Retinal age gap
The distribution of the retinal age gap followed a nearly normal distribution (Figure
2B). The mean (SD) and median (interquartile range) of the retinal age gap were -0.16
(4.54) and -0.19 (-2.99, 2.60). The proportions of fast agers with retinal age gaps more
than 3, 5 and 10 years were 22.0%, 12.0% and 1.67%, respectively.
Retinal age gap and mortality
Considering linear effects only and following adjustment for all confounding factors,
each one-year increase in retinal age gap was associated with a 2% increase in
mortality risk (hazard ratio [HR] = 1·02, 95% confidence interval [CI]: 1·00-1·04, P =
0·021; Table 2). Compared to participants with retinal age gaps in the lowest quantile,
mortality risk was comparable for those in the second and the third quantiles (HR =
1·05, 95% CI: 0·88-1·24, P = 0·602; HR = 0·89, 95% CI: 0·73-1·09, P = 0·261,
respectively). Mortality risk was significantly increased for participants with retinal
age gaps in the fourth quantile (HR = 1·33, 95% CI: 1·06-1·67, P = 0·012; Table 2).
Allowing for non-linearity, Figure 5 illustrates the estimated association between
retinal age gap and mortality risk. Evidence of an overall and non-linear association
between retinal age gap and mortality risk was observed (Poverall < 0·001; Pnon-linear =
. CC-BY-NC-ND 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint The copyright holder for thisthis version posted December 30, 2020. ; https://doi.org/10.1101/2020.12.24.20248817doi: medRxiv preprint
0·001). Higher retinal age gaps were associated with substantially increased risks of
mortality, but only if the retinal age gap exceeded 3·71 years.
Sensitivity analyses
In order to verify the robustness of our findings, fundus images from left eyes were
chosen for the statistical analyses. Similar results were observed for left eyes (data not
shown).
Discussion
Using a large population-based sample of middle-aged and elderly adults, we
developed a DL model that could predict age from fundus images with high accuracy.
Further, we found that the retinal age gap, defined as the difference between predicted
retinal age and chronological age, independently predicted the risk of mortality. Our
findings have demonstrated that retinal age is a robust biomarker of ageing that can
predict all-cause mortality.
To the best of our knowledge, this is the first study that has proposed retinal age as a
biomarker of ageing. Our trained DL model achieved excellent performance with a
MAE of 3·5, outperforming most existing biomarkers in the prediction of age.
Previous studies have demonstrated MAEs of 3·3-5·2 years for DNA methylation
clock,18,19 5·5-5·9 years MAEs for blood profiles,20,21 and 6·2-7·8 years MAEs for the
transcriptome ageing clock.22,23 Neuroimaging and 3D facial imaging have achieved
accurate performances in age prediction with MAEs between 4·3 and 7·3,7,24 and 2·8
and 6·4 years,6,25 respectively. Despite these reasonable accuracies, the invasiveness
of cellular and molecular ageing biomarkers, high cost and time-consuming nature of
neuroimaging and 3D facial ages, and ethical and privacy concerns of facial imaging,
have limited their utilities. In addition to excellent performance in age prediction,
determining retinal age using fundus images is fast, safe, cost-effective and
user-friendly, thus offering great potential for use in a large number of people.
. CC-BY-NC-ND 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint The copyright holder for thisthis version posted December 30, 2020. ; https://doi.org/10.1101/2020.12.24.20248817doi: medRxiv preprint
Beyond age prediction, our study has extended the application of retinal age to the
prediction of survival. Our novel findings have determined that the retinal age gap is
an independent predictor of increased mortality risk, further suggesting that retinal
age is a clinically significant biomarker of ageing. The relevance of retinal age for
general health is intuitive, given that the retina is the only organ that is amenable to in
vivo visualisation of the microvasculature and neural tissue. The retina offers a unique,
accessible ‘window’ to evaluate underlying pathological processes of systemic
vascular and neurological diseases that are associated with increased risks of mortality.
This hypothesis is supported by previous studies which have suggested that retinal
imaging contains information about cardiovascular risk factors,26 chronic kidney
diseases27 and systemic biomarkers.28 In addition, this hypothesis is also consistent
with previously reported qualitative and quantitative studies that have found that
ocular imaging measures (e.g. retinal-vessel calibre) and retinal diseases (e.g.
glaucoma) are significantly associated with mortality.29,30 This body of work supports
the hypothesis that the retina plays an important role in the ageing process and is
sensitive to the cumulative damages of ageing which increase the mortality risk.
Our findings have several important clinical implications. Firstly, the fast,
non-invasive, and cost-effective nature of fundus imaging enables it to be an
accessible screening tool to identify individuals at an increased risk of mortality. This
risk stratification will assist tailored health-care decision-making, as well as targeting
and monitoring of interventions. Given the rising burden of non-communicable
diseases and population ageing globally, the early identification and delivery of
personalised health-care may have tremendous public health benefits. Further, the
recent development of smartphone-based retinal cameras, together with the
integration of DL algorithms, may in the future provide point-of-care assessments of
ageing and improve accessibility to tailored risk assessments. Secondly, the capability
to use fundus images in predicting ageing may improve potential health benefits of
eye disease screening, beyond the diagnosis of sight-threatening eye diseases. This
may improve the health economic cost-effectiveness of programs such as diabetic
. CC-BY-NC-ND 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint The copyright holder for thisthis version posted December 30, 2020. ; https://doi.org/10.1101/2020.12.24.20248817doi: medRxiv preprint
retinopathy screening, thus increasing the impact and access to eye disease screening
programs.
The large-scale sample size, long-term follow-up, standardised protocol in capturing
fundus images, validity of mortality data, and adjustment for a wide range of
confounding factors in the statistical models of this study support the robustness of
our findings. Despite these promising results, our study has several limitations. Firstly,
these current analyses are limited by retinal images that were captured at a particular
cross-section in time, with trajectories in retinal ageing potentially being a better
indicator of mortality. Secondly, participants involved in the UK Biobank study were
volunteers, who might not be representative of the population from which they were
drawn. Of note, the potential healthy effect might underestimate effects of retinal age
gap on mortality, as individuals with extremely poor health were less likely to
participate in this study. Thirdly, the lack of external datasets might limit the
generalisability of our DL algorithms and findings. Lastly, we were unable to fully
exclude the possibility of residual confounders between retinal age gap and mortality.
Conclusion
In summary, we have developed a DL algorithm that can detect footprints of ageing in
fundus images and predict age with high accuracy. Further, we have been the first to
demonstrate that the retinal age gap is significantly associated with an increased risk
of mortality. Our findings suggest that retinal age is a robust biomarker of ageing.
Lastly, our work calls for future research into applications of the retinal age gap, and
whether retinal age can be used to better understand processes underpinning ageing.
Contributors
ZZ and SD conceptualised and designed the study with WW, HM, and YX. ZZ and
SD did the literature search and wrote the first draft of the manuscript. SD, PG and
MW did the deep learning modelling, ZZ, SX and WW did the statistical analysis.
. CC-BY-NC-ND 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint The copyright holder for thisthis version posted December 30, 2020. ; https://doi.org/10.1101/2020.12.24.20248817doi: medRxiv preprint
WW, HM and YX had full access to all of the data. All authors commented on the
manuscript.
Declaration of interests
We declare no competing interests.
Acknowledgments
This present work was supported by the NHMRC Investigator Grant (APP1175405),
Fundamental Research Funds of the State Key Laboratory of Ophthalmology,
National Natural Science Foundation of China (82000901), Project of Investigation
on Health Status of Employees in Financial Industry in Guangzhou, China
(Z012014075), Science and Technology Program of Guangzhou, China
(202002020049). Professor Mingguang He receives support from the University of
Melbourne through its Research Accelerator Program and the CERA Foundation. The
Centre for Eye Research Australia (CERA) receives Operational Infrastructure
Support from the Victorian State Government.
References:
1. United Nations, Department of Economic and Social Affairs, Population Division
(2017). World Population Ageing 2017 Available at:
https://www.un.org/en/development/desa/population/publications/pdf/ageing/WPA20
17_Highlights.pdf [Accessed November 28, 2020].
2. MacNee W, Rabinovich RA, Choudhury G. Ageing and the border between
health and disease. Eur Respir J 2014; 44(5): 1332-52.
3. Lopez-Otin C, Blasco MA, Partridge L, Serrano M, Kroemer G. The hallmarks of
aging. Cell 2013; 153(6): 1194-217.
4. Horvath S, Raj K. DNA methylation-based biomarkers and the epigenetic clock
theory of ageing. Nat Rev Genet 2018; 19(6): 371-84.
5. Wang J, Knol MJ, Tiulpin A, et al. Gray Matter Age Prediction as a Biomarker
for Risk of Dementia. Proc Natl Acad Sci U S A 2019; 116(42): 21213-8.
6. Xia X, Chen X, Wu G, et al. Three-dimensional facial-image analysis to predict
heterogeneity of the human ageing rate and the impact of lifestyle. Nat Metab 2020;
2(9): 946-57.
7. Cole JH, Ritchie SJ, Bastin ME, et al. Brain age predicts mortality. Mol
Psychiatry 2018; 23(5): 1385-92.
. CC-BY-NC-ND 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint The copyright holder for thisthis version posted December 30, 2020. ; https://doi.org/10.1101/2020.12.24.20248817doi: medRxiv preprint
8. Flammer J, Konieczka K, Bruno RM, Virdis A, Flammer AJ, Taddei S. The eye
and the heart. Eur Heart J 2013; 34(17): 1270-8.
9. Wong CW, Wong TY, Cheng CY, Sabanayagam C. Kidney and eye diseases:
common risk factors, etiological mechanisms, and pathways. Kidney Int 2014; 85(6):
1290-302.
10. London A, Benhar I, Schwartz M. The retina as a window to the brain-from eye
research to CNS disorders. Nature reviews Neurology 2013; 9(1): 44-53.
11. Pavkov ME, Harding JL, Chou CF, Saaddine JB. Prevalence of Diabetic
Retinopathy and Associated Mortality Among Diabetic Adults With and Without
Chronic Kidney Disease. Am J Ophthalmol 2019; 198: 200-8.
12. Kawasaki R, Xie J, Cheung N, et al. Retinal microvascular signs and risk of
stroke: the Multi-Ethnic Study of Atherosclerosis (MESA). Stroke 2012; 43(12):
3245-51.
13. Cole JH, Franke K. Predicting Age Using Neuroimaging: Innovative Brain
Ageing Biomarkers. Trends Neurosci 2017; 40(12): 681-90.
14. Sudlow C, Gallacher J, Allen N, et al. UK biobank: an open access resource for
identifying the causes of a wide range of complex diseases of middle and old age.
PLoS Med 2015; 12(3): e1001779.
15. Ebner M. Color constancy based on local space average color. Machine Vision
and Applications 2008; 20(5): 283-301.
16. Xu K, Ba J, Kiros R, et al. Show, attend and tell: neural image caption generation
with visual attention. arXiv. 2016; (published online April 19.) (preprint)
https://arxiv.org/abs/1502.03044 [Accessed November 28, 2020].
17. Ganna A, Ingelsson E. 5 year mortality predictors in 498,103 UK Biobank
participants: a prospective population-based study. Lancet 2015; 386(9993): 533-40.
18. Weidner CI, Lin Q, Koch CM, et al. Aging of blood can be tracked by DNA
methylation changes at just three CpG sites. Genome Biol 2014; 15(2): R24.
19. Hannum G, Guinney J, Zhao L, et al. Genome-wide methylation profiles reveal
quantitative views of human aging rates. Mol Cell 2013; 49(2): 359-67.
20. Putin E, Mamoshina P, Aliper A, et al. Deep biomarkers of human aging:
Application of deep neural networks to biomarker development. Aging (Albany NY)
2016; 8(5): 1021-33.
21. Mamoshina P, Kochetov K, Putin E, et al. Population Specific Biomarkers of
Human Aging: A Big Data Study Using South Korean, Canadian, and Eastern
European Patient Populations. J Gerontol A Biol Sci Med Sci 2018; 73(11): 1482-90.
22. Peters MJ, Joehanes R, Pilling LC, et al. The transcriptional landscape of age in
human peripheral blood. Nat Commun 2015; 6: 8570.
23. Fleischer JG, Schulte R, Tsai HH, et al. Predicting age from the transcriptome of
human dermal fibroblasts. Genome Biol 2018; 19(1): 221.
24. Liem F, Varoquaux G, Kynast J, et al. Predicting brain-age from multimodal
imaging data captures cognitive impairment. Neuroimage 2017; 148: 179-88.
25. Chen W, Qian W, Wu G, et al. Three-dimensional human facial morphologies as
robust aging markers. Cell Res 2015; 25(5): 574-87.
. CC-BY-NC-ND 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint The copyright holder for thisthis version posted December 30, 2020. ; https://doi.org/10.1101/2020.12.24.20248817doi: medRxiv preprint
26. Poplin R, Varadarajan AV, Blumer K, et al. Prediction of cardiovascular risk
factors from retinal fundus photographs via deep learning. Nat Biomed Eng 2018; 2(3):
158-64.
27. Sabanayagam C, Xu D, Ting DSW, et al. A deep learning algorithm to detect
chronic kidney disease from retinal photographs in community-based populations.
The Lancet Digital Health 2020; 2(6): e295-e302.
28. Rim TH, Lee G, Kim Y, et al. Prediction of systemic biomarkers from retinal
photographs: development and validation of deep-learning algorithms. The Lancet
Digital Health 2020; 2(10): e526-e36.
29. Wang JJ, Liew G, Klein R, et al. Retinal vessel diameter and cardiovascular
mortality: pooled data analysis from two older populations. Eur Heart J 2007; 28(16):
1984-92.
30. Khanna RC, Murthy GVS, Giridhar P, et al. Glaucoma-associated long-term
mortality in a rural cohort from India: the Andhra Pradesh Eye Disease Study. Br J
Ophthalmol 2018; 102(11): 1477-82.
Figure 1. Overview of the study workflow
Figure legend: Figures showing the study workflow used to calculate retinal age gaps
from fundus images. Fundus images were preprocessed and fed into the DL model. (A)
The Xception architecture was used to train fundus images, with chronological age as
the outcome variable; (B) The selection of DL models was based on performance in
the validation set, where predicted retinal and chronological ages were compared; (C)
The selected trained DL model was then applied to make retinal age predictions from
. CC-BY-NC-ND 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint The copyright holder for thisthis version posted December 30, 2020. ; https://doi.org/10.1101/2020.12.24.20248817doi: medRxiv preprint
fundus images for participants in the test and death datasets; (D) The difference
between predicted retinal age and chronological age was defined as the retinal age gap.
A positive retinal age gap indicated an ‘older’ appearing retina, while a negative
retinal age gap indicated a ‘younger’ appearing retina. This figure was created with
BioRender.com.
Figure 2. Performance of the deep learning model on the test dataset
Figure legend: (A) Scatterplot depicting correlation of predicted age (y-axis) with
chronological age (x-axis); (B) Histogram showing the nearly normal distribution of
the retinal age gap.
. CC-BY-NC-ND 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint The copyright holder for thisthis version posted December 30, 2020. ; https://doi.org/10.1101/2020.12.24.20248817doi: medRxiv preprint
Figure 3. Attention maps for age prediction
Figure legend: Figures showing representative examples of fundus images with
corresponding attention maps for age prediction. Regions highlighted with a brighter
colour indicate areas that are used by the DL model for age prediction. Regions
around the retinal vessels are highlighted.
. CC-BY-NC-ND 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint The copyright holder for thisthis version posted December 30, 2020. ; https://doi.org/10.1101/2020.12.24.20248817doi: medRxiv preprint
Figure 4. Adjusted survival curves for mortality risk by retinal age gap quantiles
Figure legend: Mortality risk is shown over time for participants in different retinal
age gap quantiles. Lower quantiles corresponded to participants who had
chronological ages greater than predicted retinal age, whereas higher quantiles
corresponded to those with chronological ages lower than predicted retinal age. Plots
were based on Cox proportional hazards regression models, adjusted for age, sex,
ethnicity, Townsend deprivation indices, educational level, obesity, smoking status,
physical activity level, diabetes mellitus, hypertension, general health status, history
of heart diseases, and history of stroke. Compared to participants with retinal age gaps
in the lowest quantile, mortality risk was comparable for those in the second and the
third quantiles (hazard ratio [HR] = 1·05, 95% confidence interval [CI]: 0·88-1·24, P
= 0·602; HR = 0·89, 95% CI: 0·73-1·09, P = 0·261, respectively). Mortality risk was
significantly increased for participants with retinal age gaps in the fourth quantile (HR
= 1·33, 95% CI: 1·06-1·67, P=0·012).
. CC-BY-NC-ND 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint The copyright holder for thisthis version posted December 30, 2020. ; https://doi.org/10.1101/2020.12.24.20248817doi: medRxiv preprint
Figure 5. Association between retina age gap and mortality risk, allowing for
non-linear effects
Figure legend: The reference retinal age gap for this plot (with hazard ratio [HR]
fixed as 1·0) was 0 years. The model was fitted with a restricted cubic spline for
retinal age gap (knots placed at equal percentiles of retina age gap), adjusted for age,
sex, ethnicity, Townsend deprivation indices, educational level, obesity, smoking
status, physical activity level, diabetes mellitus, hypertension, general health status,
history of heart diseases, and history of stroke. Evidence of an overall and non-linear
association between retinal age gap and all-cause mortality was observed (P
overall
<
0·001; P
non-linear
= 0·001). Higher retinal age gaps were associated with substantially
increased risks of mortality, but only if the retinal age gap exceeded 3·71 years.
. CC-BY-NC-ND 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint The copyright holder for thisthis version posted December 30, 2020. ; https://doi.org/10.1101/2020.12.24.20248817doi: medRxiv preprint
Table 1. Characteristics of datasets derived from the UK Biobank study
Training a Validation
a Test Death
Nsubj 27424 9142 8212 1117
Nimg 46834 15612 8212 1117
Mean age (mean+SD, yrs) b 55·6+8·21 55·7+8·19 55·5+8·22 61·0+6·67
Female, N (%) b 15,341 (55·9) 5,045 (55·2) 4,526 (55·1) 471 (42·2)
White ethnicity, N (%) b 25,400 (92·6) 8,468 (92·6) 7,618 (92·8) 1,035 (92·7)
Townsend index (mean+SD) b -1·09+2·94 -1·10+2·96 -1·08+2·95 -0·73+3·12
College or university degree, N (%) b 9,981 (36·4) 3,344 (36·6) 3,035 (37·0) 299 (26·8)
Current/previous smoker, N (%) b 11,671 (42·8) 3,905 (43·0) 3,532 (43·1) 659 (59·4)
Above physical activity recommendation, N (%) b 18,746 (82·7) 6,219 (82·4) 5,682 (83·4) 681 (77·4)
Excellent/good health status, N (%) b 20,290 (74·5) 6,738 (74·0) 6,149 (75·2) 618 (55·8)
Obesity, N (%) b 6,239 (22·9) 2,112 (23·2) 1,840 (22·5) 312 (28·1)
Diabetes mellitus, N (%) b 1,332 (4·86) 443 (4·85) 428 (5·21) 133 (11·9)
Hypertension, N (%) b 19,632 (71·6) 6633 (72·6) 5,956 (72·5) 928 (83·1)
History of heart diseases, N (%) b 848 (3·09) 296 (3·24) 250 (3·04) 94 (8·42)
History of stroke, N (%) b 309 (1·13) 83 (0·91) 103 (1·25) 35 (3·13)
Nsubj = number of subjects; Nimg = number of images; yrs = years; SD = standard deviation.
a Selection of images of both eyes if available.
b Values are based on Nsubj.
. CC-BY-NC-ND 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint The copyright holder for thisthis version posted December 30, 2020. ; https://doi.org/10.1101/2020.12.24.20248817doi: medRxiv preprint
Table 2. Association between retinal age gap with mortality using Cox proportional
hazards regression models
Retinal age gap
N
Mean+SD (yrs)
Model I Model II
HR (95% CI) P value HR (95% CI) P value
Retinal age gap, per one age (yrs) 9,329 -0·31+4·59 1·03 (1·01-1·05) <0·001 1·02 (1·00-1·04) 0·021
Retinal age gap
Quantile 1 2,333 -5·98+2·69 Reference - Reference -
Quantile 2 2,332 -1·72+0·82 1·11 (0·96-1·29) 0·168 1·05 (0·88-1·24) 0·602
Quantile 3 2,332 1·01+0·81 0·96 (0·80-1·15) 0·666 0·89 (0·73-1·09) 0·261
Quantile 4 2,332 5·43+2·59 1·46 (1·20-1·78) <0·001 1·33 (1·06-1·67) 0·012
HR = hazard ratio; CI = confidence interval.
Model I adjusted for age, sex, ethnicity, and Townsend deprivation indices.
Model II adjusted for covariates in Model I + educational level, obesity, smoking status, physical activity level, diabetes mellitus,
hypertension, general health status, history of heart diseases, and history of stroke.
. CC-BY-NC-ND 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.(which was not certified by peer review)preprint The copyright holder for thisthis version posted December 30, 2020. ; https://doi.org/10.1101/2020.12.24.20248817doi: medRxiv preprint
