Diabetic retinopathy (DR) and cardiovascular disease (CVD) are significant complications in patients with Type 1 diabetes, often indicating systemic vascular damage. This thesis studies the impact of OCTA scan field size on the performance of machine learning models designed to predict DR and CVD risk. Using multimodal retinal imaging, including Fundus Retinography, Optical Coherence Tomography, and OCT Angiography, radiomic features were extracted and combined with demographic, blood analysis, and ocular data to train and evaluate predictive models. The study aims not only to quantify the influence of OCTA scan size on model accuracy, but also to analyze the contribution of different feature groups, including radiomics alone and in combination with clinical variables, to determine which combinations provide the most reliable predictions. Additionally, the impact of including cardiovascular risk factors on model performance is assessed to understand their relative predictive value. Experimental results provide insights into the optimal imaging and feature strategies for predicting DR and CVD risk, highlighting the potential of wide-field OCTA and combined feature strategies to support early detection and personalized risk assessment in Type 1 diabetes patients.