Individualized prevention in cardiovascular disease and diabetes mellitus: Risk factors, risk prediction, and treatment effects

Cardiovascular disease (CVD) remains the leading cause of morbidity and mortality worldwide. Preventive (pharmacological) measures such as cholesterol and blood pressure lowering, or anticoagulants, are a burden on global health systems and have potential downsides. Therefore, it is important to identify patients who benefit most from these measures. To be able to identify which patients are at high risk of CVD, there needs to be an understanding of what are risk factors and risk predictors. In this thesis, we have shown that traditional risk factors blood pressure, cholesterol, and smoking, are risk factors for multiple events during long-term follow-up even in patients who have already developed CVD, in whom international guidelines already advise intensive risk factor treatment. We further show that thyroid function is a new risk factor for the risk of stroke, but not the risk for other types of CVD, in patients with diabetes type 2. In this thesis, we further show how statistical prediction models can be used to improve clinical decision making in CVD prevention. Models predicting 10-year CVD risk for healthy middle-aged persons have been in international guidelines for years; preventive treatment is given to high risk individuals. Until recently, all older persons (>65 years) were regarded as 'high risk' and treated accordingly. Using a risk prediction score for older persons, we have shown that not all older people have the same high risk of CVD; therefore, they do not benefit equally from preventive strategies. Furthermore, we have looked at predictions from a lifetime perspective (i.e. CVD-free life expectancy) and how these relate to 10-year risks for making treatment decisions. Using lifetime predictions instead of 10-year risks leads to treatment of younger persons with on average more lifetime benefit. The main goal of risk prediction in CVD in clinical practice is to make informed treatment decisions for individual patients. The assumption is made that ‘the higher the risk, the higher the benefit’. However, that assumption does not always have to ring true. Therefore, various statistical methods are examined to predict individualized expected treatment effects of different preventive strategies, including blood pressure lowering, anticoagulation and lifestyle interventions. One possible method is to make a model including an interaction term between the baseline risk (estimated with a multivariate risk model) and treatment allocation in a trial to see if the relative treatment effect is uniform across a range of baseline risks. Another method shown in this thesis is to directly model treatment effects in a trial by including treatment-covariate interaction terms for all predictors. Treatment effect predictions can be made for both treatment benefits and possible adverse effects, such as bleeding with anticoagulation. This way, clinician and patient together can make individualized treatment decisions about initiation or intensification of risk factor treatment, allowing us to choose the right treatment for the right patient. Most of the models discussed in this thesis are available online at www.U-Prevent.com.