|Name||Sander van Doorn|
|Date defended||September 28, 2017|
|(Co-) Supervisors||Prof. Dr K.G.M. Moons, Prof. Dr A.W. Hoes, Dr G.J. Geersing, Dr F.H. Rutten|
|Title of thesis||Risk management of patients with atrial fibrillation in general practice|
Read the thesis online
Atrial fibrillation (AF) is the most common cardiac arrhythmia and poses patients at many risks. In this thesis we address the management of these risks in general practice including the prediction and prevention of ischaemic stroke, and other adverse events including hospitalisation, mortality and heart failure. AF is a well-known risk factor for ischaemic stroke. Anticoagulants can effectively prevent stroke, but only in patients at high enough risk of stroke does their benefits outweigh the risk of bleeding complications. Practice guidelines recommend the CHA2DS2-VASc model for stroke risk prediction but external validation studies show ambiguous and conflicting results. In a meta-analysis of studies validating CHA2DS2-VASc we observed consistently higher risks of stroke for each CHA2DS2-VASc score in studies recruiting patients in hospitals as compared to studies recruiting in the open general population, and substantial heterogeneity that could not be fully explained by meta-regression. While their benefits are evident, anticoagulants are often underused. We evaluated the reasons for non-adherence to practice guidelines on stroke prevention in AF in general practice. General practitioners (GPs) most often mentioned sustained sinus rhythm after an episode of AF, and the cardiologist considered responsible for the anticoagulation as reasons for non-adherence. To subsequently improve the anticoagulant management of AF patients, we undertook a cluster-randomised trial assessing the effectiveness of automated CHA2DS2-VASc based decision support on stroke prevention. In index practices, GPs were provided an automatically generated CHA2DS2-VASc based anticoagulant treatment recommendation. GPs in the reference practices provided care as usual. Underuse of anticoagulants in our study was low and the decision support did not result in a reduction in stroke incidence, nor did it affect bleeding risk or anticoagulant treatment. Besides stroke, patients with AF are at increased risk of many other adverse events. We observed a high incidence of hospitalisation and mortality, predominantly for non-cardiac causes. Existing stroke prediction models, particularly the ATRIA score, may be useful in identifying patients at high risk of such events. Heart failure (HF) is another adverse condition that frequently co-occurs in patients with AF. Combiningindividual participant data from four HF-screening studies, we observed a high prevalence of HF in older community-dwelling AF patients and an impaired diagnostic accuracy of NTproBNP for HF screening. Medical research nowadays often relies on data collected during routine clinical practice. We assessed possible misclassification and its effect on the validation of a prediction model, using CHA2DS2-VASc as a case study. Misclassification was substantial in some predictors, though overall model performance was not affected. We conclude that routinel care data may be a useful source when validating prognostic models. This thesis concludes by discussing the two pathways along which more accurate prediction of stroke may be achieved: i) updating the predictors and their weights in the CHA2DS2-VASc model, and evaluating interaction terms; ii) evaluating additional predictors including clinical parameters, biomarkers, and echocardiographic measurements. In the end, better stroke prediction then should lead to better anticoagulant treatment and ultimately better prevention of stroke in patients with atrial fibrillation.