Atrial fibrillation accounts for a great proportion of strokes in older people, and as populations age its importance increases. Stroke prevention with anticoagulants has long been known to be highly effective and new drugs are also available. Historically many patients with atrial fibrillation were not treated.

For several years we have investigated the management of atrial fibrillation, analysing a large database of UK electronic primary care records. Such databases are an unparalleled resource for understanding how the health service delivers care. We found confusion in understanding the importance of contraindications to anticoagulants and that stroke prevention generally seemed to be better in patients who are on chronic disease registers and therefore followed up regularly. [1,2,3,4]

A big question for the health service is whether we should focus efforts on finding more people with undiagnosed atrial fibrillation or on treating more of those with a diagnosis. It was thinking about this that led us to another quite different question.

We initially wanted to describe trends in the prevalence of diagnosed atrial fibrillation and the proportion of patients with a diagnosis who were treated. In our primary care records database we identified those with a documented diagnosis, excluding those in whom the diagnosis was subsequently recorded as resolved. We counted the number of patients with a diagnosis and the number of those with a diagnosis who were then treated. Year on year there were increases in recorded prevalence and in the proportion of patients treated. [5]

But our method prompted a series of questions. Did we exclude many participants when we excluded those in whom atrial fibrillation seemed to have resolved? Yes, about 8%. How can a clinician really know if atrial fibrillation has resolved? We already know it can be intermittent. Can you really tell if it has gone completely? And doesn’t it often recur? Would it be possible to determine from patient records if resolved atrial fibrillation had really resolved? Then we hit on the right question. What is the risk of stroke in those whose condition is documented as resolved? Treatment of atrial fibrillation is determined by stroke risk. Whether the arrhythmia is intermittent, recurs silently, or has gone for good are distractions from what should inform clinical decisions. If stroke risk in resolved atrial fibrillation falls to the levels of those without atrial fibrillation then most anticoagulants aren’t necessary, but if risk remains as high as those with atrial fibrillation then they probably are.

We found that stroke risk was higher in those with resolved atrial fibrillation than those without atrial fibrillation. A further analysis strengthened the conclusions. In terms of their stroke risk, people with resolved atrial fibrillation were becoming more like those with atrial fibrillation. In recent years the risk of stroke in those with resolved atrial fibrillation was the same as in those with atrial fibrillation. This is important because the frequency of resolved atrial fibrillation seems to have increased markedly after 2005. It was also clear that once atrial fibrillation is considered resolved, the prescriptions of anticoagulants almost always stop. We had discovered a group of patients at high risk of stroke who aren’t on the chronic disease register. Because they aren’t counted as at high risk, they won’t appear as untreated in any analysis of anticoagulant use. Ironically this has happened at about the same time as dramatic improvements in the identification and management of atrial fibrillation. 

Without databases of electronic patient records it is impossible to know what is really happening to patients. Our analysis did not rely on any innovative methods and we have to acknowledge constructive input from reviewers. But in the end we also asked the right question.

Tom Marshall is professor of public health and primary care at the University of Birmingham.