This study presents a deep learning approach to predict cardiovascular risk factors from retinal fundus photographs. Using data from 284,335 patients, the models were trained and validated on two independent datasets of 12,026 and 999 patients. The models successfully predicted several cardiovascular risk factors, including age (within 3.26 years), gender (AUC 0.97), smoking status (AUC 0.71), HbA1c (within 1.39%), systolic blood pressure (within 11.23 mmHg), and major adverse cardiac events (AUC 0.70). The models used distinct anatomical features of the retina, such as the optic disc and blood vessels, to generate predictions, suggesting potential for further research. The study used retinal fundus images from two datasets: UK Biobank (500,000 participants) and EyePACS (over 300 clinics worldwide). The models were trained on images from 48,101 UK Biobank patients and 236,234 EyePACS patients, and validated on images from 12,026 UK Biobank patients and 999 EyePACS patients. The models achieved high accuracy in predicting age, systolic blood pressure, and other risk factors, with mean absolute errors of 3.26 years and 3.42 years for UK Biobank and EyePACS datasets, respectively. The models also demonstrated the ability to infer ethnicity, another potential cardiovascular risk factor, with a kappa score of 0.60 in the UK Biobank validation set and 0.75 in the EyePACS-2K validation set. The study further used soft attention maps to identify the anatomical regions used by the models to make predictions, showing that blood vessels were highlighted in models predicting age, smoking, and systolic blood pressure, while perivascular surroundings were highlighted in models predicting HbA1c, and the optic disc in models predicting gender. The study's results indicate that deep learning can predict multiple cardiovascular risk factors from retinal fundus images, potentially allowing for better cardiovascular risk stratification. However, the study has limitations, including a relatively small dataset size and the need for further validation on other datasets. The study suggests that retinal fundus images may be able to augment or replace some of the other markers, such as lipid panels, to yield more accurate scores. The study provides evidence that deep learning may uncover additional novel signals in retinal images that will allow for better cardiovascular risk stratification.This study presents a deep learning approach to predict cardiovascular risk factors from retinal fundus photographs. Using data from 284,335 patients, the models were trained and validated on two independent datasets of 12,026 and 999 patients. The models successfully predicted several cardiovascular risk factors, including age (within 3.26 years), gender (AUC 0.97), smoking status (AUC 0.71), HbA1c (within 1.39%), systolic blood pressure (within 11.23 mmHg), and major adverse cardiac events (AUC 0.70). The models used distinct anatomical features of the retina, such as the optic disc and blood vessels, to generate predictions, suggesting potential for further research. The study used retinal fundus images from two datasets: UK Biobank (500,000 participants) and EyePACS (over 300 clinics worldwide). The models were trained on images from 48,101 UK Biobank patients and 236,234 EyePACS patients, and validated on images from 12,026 UK Biobank patients and 999 EyePACS patients. The models achieved high accuracy in predicting age, systolic blood pressure, and other risk factors, with mean absolute errors of 3.26 years and 3.42 years for UK Biobank and EyePACS datasets, respectively. The models also demonstrated the ability to infer ethnicity, another potential cardiovascular risk factor, with a kappa score of 0.60 in the UK Biobank validation set and 0.75 in the EyePACS-2K validation set. The study further used soft attention maps to identify the anatomical regions used by the models to make predictions, showing that blood vessels were highlighted in models predicting age, smoking, and systolic blood pressure, while perivascular surroundings were highlighted in models predicting HbA1c, and the optic disc in models predicting gender. The study's results indicate that deep learning can predict multiple cardiovascular risk factors from retinal fundus images, potentially allowing for better cardiovascular risk stratification. However, the study has limitations, including a relatively small dataset size and the need for further validation on other datasets. The study suggests that retinal fundus images may be able to augment or replace some of the other markers, such as lipid panels, to yield more accurate scores. The study provides evidence that deep learning may uncover additional novel signals in retinal images that will allow for better cardiovascular risk stratification.