The reward of understanding risk in atrial fibrillation

Atrial fibrillation is a growing epidemic with rapidly escalating rates of hospitalizations and healthcare expenditures.1 Hospitalizations account for the majority of the costs of atrial fibrillation care and are also associated with an increased risk of mortality.2,3 While understanding national trends in hospitalization rates and care patterns can help inform policy decisions, significant heterogeneity in the clinical course of atrial fibrillation introduces challenges in understanding risk on an individual level.4 Prior studies have shown that demographics, environmental factors, the presence of comorbidities, and the use of certain medication are associated with an increased risk of hospitalization in patients with atrial fibrillation;5–7 however, risk scores for atrial fibrillation hospitalization are not routinely implemented nor used in routine clinical practice. In this issue of the European Journal of Preventive Cardiology, Meyre et al. have developed and validated a risk score for all-cause hospitalization in patients with atrial fibrillation from two independent multicenter cohorts from Switzerland.8 They evaluated patients 65 years or older with a documented history of atrial fibrillation who were enrolled in either the Swiss-AF cohort study (derivation cohort),9 or the BEAT-AF cohort study (validation cohort).10 Potential predictors including demographics, comorbidities, prior procedures, medications, and family history were selected with a least absolute shrinkage and selection operator (LASSO) model. Of the 41 candidate variables, 11 were included in the final model including age, prior pulmonary vein isolation (PVI), and several comorbidities. These variables were entered into a Cox-proportional hazards model for the time to first unplanned, overnight admission. The regression coefficients from this model were transformed into a clinical risk score (Admit-AF score) and applied to the 2387 patient Swiss-AF cohort. The Admit-AF score demonstrated a C-index of 0.64 for one-year outcomes and 0.63 for three-year outcomes. Validation of the Admit-AF score in the 1300 patient BEAT-AF cohort yielded a C-index of 0.59 for both one-year and three-year outcomes. The authors concluded that the Admit-AF risk score translates readily available clinical information into valuable information about hospitalization risk and could be used to inform preventive strategies. The Admit-AF investigators addressed an important, yet challenging, issue. The authors leveraged two large datasets to develop and validate their model. While the predictive accuracy of the model is moderate at best, it does have the advantage of having an interpretable structure that follows clinical gestalt and demonstrates good calibration in the derivation and validation cohorts. Factors such as increasing age, more comorbidities, and a history of previous falls all convey a higher risk of hospital admission. Interestingly, previous PVI was associated with a 28% lower risk of admission. This is consistent with other studies which have shown that atrial fibrillation ablation is associated with reduced all-cause hospitalizations.11 Given the observational nature of this study, it is possible that patients undergoing PVI had less overall comorbidity and/or had less severe atrial fibrillation and thus prior PVI is simply an indirect marker of relative health. However, the authors used a LASSO model that included a large list of comorbidities as well as atrial fibrillation type in part to help mitigate issues of multi-collinearity. The model’s structure corroborates prior studies and helps solidify our understanding of the drivers of risk in atrial fibrillation. The observed and predicted rates of hospitalization in the study by Meyre and colleagues are also informative. It is notable that even the lowest Admit-AF score (–3) is associated with a 12% risk of admission at one year and the “low risk” group had a 50% risk of hospitalization over four years of follow-up. While this may have been “low” relative to the rest of the cohort, it is hard to consider these hospitalization rates negligible nor deserving of risk-reduction interventions. This paradox highlights the magnitude of the problem of risk-stratifying hospitalization in patients with such elevated absolute risk. While the Admit-AF study is highly informative, it also has some notable limitations. The C-statistic of the model was modest at best (0.59 in validation). This is similar to prior models of hospitalization in atrial fibrillation patients,6 though the present model has the added advantage of being formulated as a calculable clinical score. The model was developed and validated in two Swiss populations. Countries with different health system structures, such as the USA, may not follow the same hospitalization patterns and may have different drivers of risk. The model also targets all-cause hospitalization in patients with atrial fibrillation. Given the dramatic variation in causes of hospitalization in these patients, a cause-specific model of hospitalization (i.e. atrial fibrillation-related hospitalization or cardiovascular hospitalization) might be more actionable in clinical practice. Would more aggressive treatment of comorbidities decrease the risk of hospitalization in these patients? Use of guideline-directed therapies for comorbidities in atrial fibrillation patients is low, and is associated with improved outcomes with some conditions, but not others.12 Without understanding the impact of risk factor interventions on outcomes, clinicians may not have a clear course of action when treating a patient with a high Admit-AF risk score. While the Admit-AF score does not dictate a particular treatment strategy, understanding risk among atrial fibrillation patients has value. Being able to assess an individual’s risk profile may help inform discussions about treatments and prognosis. Clinical trials of new therapies for atrial fibrillation may also take advantage of risk scores such as Admit-AF to guide enrollment and target high-risk patient cohorts. Risk scores may also guide allocation and prioritization of treatments in settings where resources are scarce. Studies of treatment heterogeneity patterns across systems could also benefit from a quantification of risk to understand how treatments vary across both a system and a spectrum of risk. Clinical risk scores such as Admit-AF might also be integrated with other diagnostic modalities (such as hemodynamic and imaging data)13 to develop a more robust picture of a patient’s condition which may result in more accurate models. The Admit-AF score provides insight into the drivers of hospitalization risk in atrial fibrillation patients and may be a useful tool in clinical care. Questions remain about how to best leverage information about risk to guide decision making. Further challenges lie ahead as to whether clinicians will integrate risk tools such as the Admit-AF score into their routine clinical practice. Providing tailored care based on a patient’s individualized risk is not easy, but worth the reward. The author(s) declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: ZL is supported by NIH T32 training grant #5T32HL069749 and receives grant support from Boston Scientific. JPP is supported by R01HL128595 from the National Heart, Lung and Blood Institute. He receives grants for clinical research from Abbott, American Heart Association, Association for the Advancement of Medical Instrumentation, Bayer, Boston Scientific, and Philips and serves as a consultant to Abbott, Allergan, ARCA Biopharma, Biotronik, Boston Scientific, LivaNova, Medtronic, Milestone, Myokardia, Sanofi, Philips, and Up-to-Date. The author(s) received no financial support for the research, authorship, and/or publication of this article.