This thesis investigates the use of different research methods, varying from traditional statistics towards artificial intelligence (AI), on regular care data. Chapter 2 introduces the CCN database; a regular care database. This database encompasses all patients suspected of cardiovascular disease who visited one of the Cardiology Centers of the Netherlands (CCN) between 2007 and 2018 (n = 109,151) and their measurements. Long-term follow-up was available after linkage with the personal registry data of Statistics Netherlands. Chapter 3 used survival analysis and the CCN database to evaluate the effects of computed tomography (CT) for calcium scoring and angiography (CT-first strategy) in patients with chest pain. This chapter used methods to make a regular care database fit for research purposes and showed that a CT-first strategy reduces all-cause mortality, but not cardiovascular mortality. Chapter 4 focussed on the prognostic value of the calcium score and race-, age- and sex-specific percentiles of this score. Both measures predict mortality equally well, in men and women. Addition of stenosis degree showed increased discrimination, although non-significant, in women compared to men. This finding supports a sex-specific view on coronary artery disease (CAD). Chapter 5A and 5B evaluated the New York Heart Association classification (NYHA) for risk assessment in patients visiting CCN. Chapter 5A showed that the use and value of the NYHA classification can be extended beyond complaints related to heart failure in men and women. As NYHA classification is a subjective measure of the experienced daily disability by the patient, Chapter 5B explains the NYHA class with objective measures in men and women. We studied the causality of stress electrocardiogram (ECG) variables on the relation between NYHA classification and all-cause mortality with mediation analysis. Proportional workload (i.e. maximum work load during exercise as a proportion of the predicted work load) appeared to be the largest mediator in the association between NYHA classification and mortality, but the majority of the association remains unexplained. In Chapter 6 a pipeline was developed for the identification of adverse drug reactions in clinical notes of the cardiologist. We trained word embedding models with clinical notes in the CCN database. Overall performance and set-up of the pipeline was good. The pipeline facilitated interpretation and can be easily tuned for other applications. Chapter 7 gives a narrative overview of the available opportunities that apply AI to cardiovascular image analysis and showed that AI will impact all steps of the imaging chain. Chapter 8 studies the performance and improvement of the pre-test probability, recommended by the 2019 European Guidelines, of CAD in patients with chest pain or dyspnoea. This probability misclassified approximately 20% of women with a diagnosis of CAD in the CCN database, hampering referral for further diagnostic testing. The prediction of CAD significantly improved when all available data was used with Lasso Logistic Regression. In Chapter 9, a deep neural network was developed to classify sex based on the ECG. Misclassified individuals had worse survival than their correctly classified biological peers. Mediation analysis revealed a sex-specific relation between shortened QRS duration and increased mortality risk, emphasizing the importance of sex-stratified research. Chapter 10 reflects on the use of traditional methods and AI for cardiovascular research with regular care data. AI has the potential to automatically analyse large datasets and to pick up associations that have not been previously discovered. AI is specifically useful when data is unstructured, as these types of data cannot be (automatically) analysed with traditional methods. Nonetheless, clinical implementation of AI is still limited, due to a lack of validation studies, reduced generalizability of AI models, reproducibility and replicability issues and a lack of knowledge on efficacy.