Risk prediction models in atrial fibrillation: from theory to practice

This editorial refers to ‘Utility of risk prediction models to detect atrial fibrillation in screened participants’, by M.H.F. Poorthuis et al., pp. 586–595. Atrial fibrillation (AF) is the most common sustained arrhythmia all over the world and is burdened by high morbidity and mortality as well as a significantly negative impact on healthcare costs.1 The widely known high risk of stroke linked to AF seems to be related not only to the atrial rhythm, but also to remodelling of the atrium, now defined atrial cardiomyopathy.2 Although the CHA2DS2-VASC risk score is nowadays widely accepted as practical and helpful to stratify the risk of stroke in AF and even non-AF populations,3,4 no widespread consensus exists regarding the available non-invasive tools able to screen AF in the general population. Moreover, as proper anticoagulation can reduce the risk of stroke and long-term mortality,5 it seems reasonable to screen high-risk individuals in order to intervene early and reduce AF-associated consequences. In the present issue of the European Journal of Cardiovascular Prevention, Poorthuis et al. present a systematic review of available and established AF prediction models and assess their performance in a large validation cohort.6 Out of a total of 6961 reports screened, 14 studies were included with a number of predictors varying from 4 to 13. The validation cohort consisted of 2 541 702 participants, of whom 10 464 (0.4%) had AF. All models showed a statistically significant better discrimination compared with the age threshold of 65 years or older as suggested for opportunistic screening in current guidelines.1 All models taken into account also had a statistically significant better discrimination than CHADS2 and CHA2DS2-VASc, which were, we must take into account, never intended to predict AF. Among all the studied models, the CHARGE-AF7 and the MHS8 were those with the highest prevalence of AF in the highest decile, and the number needed to screen was quite low (63). These were therefore seen as more reliable and might help with the task of identifying undetected AF. Moreover, they found that age is not the best discriminator of AF risk. The work presented is methodologically very sound and gives solid key messages, while opening some interesting practical issues. The first is the real elephant in the room and is represented by the target population. It has been demonstrated that untargeted AF screening delivers very little in terms of positive results and carries the potential risk of many false alarms triggered in an otherwise healthy population, thus leading to unnecessary visits and increased health costs.9 According to the recent guidelines, age (and particularly patients ≥65 years old) could represent a good criterion, as previous studies found better discrimination values in older populations.10 By excluding those younger than 45 years old, Poorthuis et al. were more inclusive, but still able to target increasing age as a known risk factor for AF. They were also able to deliver better results in terms of concordance between predicted and observed risk when compared with a similar meta-analysis.11 Nonetheless, having all the population over 64 years old screened implies a massive effort by the national health care systems, and leads to the second big question, how? As already mentioned, the CHARGE-AF7 and the MHS8 come out as the best candidates with the former already confirmed by a similar study11 and both with variables easy to assess during routine visits. The problem remains: how to implement such screening in clinical practice? Is the general practitioner the activator of such a screening chain or should all people be gifted a routine cardiology visit along with their 65th birthday presents? And how to manage those with high risk of AF? Is a yearly out-of-hospital follow-up with a 12-lead electrocardiogram (ECG) enough or should we summon the power of modern technology in order to equip our high-risk patients with long-term ECG monitoring, ECG-capable wearable or even implantable loop recorders? It is clear how there is a plethora of potential strategies, and how all of them differ in term of pros and cons. The third, and probably most important, problem is the management of such patients once asymptomatic AF has been detected. In other words, do these patients have the same thromboembolic risk of clinically relevant AF or should they be considered as silent?12 There is still debate on which the optimal treatment should be, as two large clinical trials on device-detected atrial high-rate episodes are struggling to finish the enrolment.13 In conclusion, the paper by Poorthuis et al. demonstrates the importance to establish solid models able to detect early AF. How to implement such models in an effective and efficient way is still a matter of debate. Conflict of interest: none declared. The opinions expressed in this article are not necessarily those of the Editors of the European Journal of Preventive Cardiology or of the European Society of Cardiology.