This study presents recent evidence on real-world AF costs. To the best of our knowledge, it is the first to link data-derived patient clusters with cost trajectories. We observed mean annual and AF-adjudicated healthcare costs to remain stable over several years. Furthermore, among three clinically well-differentiated patient clusters cost levels differed but were also stable throughout the observation period.

Mean annual total costs amounted to CHF 19,037 per AF patient (∼23% of the Swiss Gross Domestic Product per capita), whereof CHF 5,679 were attributable to AF and its complications. Across the entire observation period, these costs remained relatively stable. Inpatient and outpatient costs were of similar magnitude; this applied to both, AF-adjudicated and non-adjudicated costs. This finding is in line with existing literature, where inpatient costs incurred at least half of the total costs.^19–22 The observation of a stable proportion across five years, also apparent in AF-adjudicated costs, has not been previously described.

Of the AF-adjudicated costs, 71.8% were directly related to AF, followed by bleeding, heart failure, fall, and stroke / TIA costs. In the literature so far, AF-related costs were mostly assessed through various excess cost calculations²³ and a clinically-based bottom-up approach was missing. Even though our adjudication algorithm was unable to distinguish between the costs of care episodes triggered by AF versus by other causes, we believe that this bottom-up approach contributes to a more realistic understanding of AF and related costs. Also, in contrast to most studies,^19,21,24,25 our observed contribution of AF-adjudicated costs to overall healthcare costs was smaller. One reason may be that we analysed prevalent AF patients while the literature focussed mainly on incident patients.

Three well-differentiated patient clusters resulted from our analysis, with distinct cost trajectories: CV-dominated with rather high costs, HF-dominated with similarly high costs, and IS with substantially lower costs. Some attempts to cluster the heterogeneity of AF patients were reported previously, but not linked with cost trajectories.^12,26–28 To facilitate an understanding of expected patient trajectories, we based our analysis on baseline characteristics.

Our first cluster, the CV-dominated patients, featured many strong risk factors for cardiovascular diseases (e.g., high rates of past smokers, low rate of never smokers, high prevalence of diabetes) and high cardiovascular drug use (i.e. aspirin, statin, and beta-blocker). The high rates of cardiovascular revascularisation procedures, myocardial infarction and stroke/TIA suggested underlying cardiovascular disease. These characteristics are roughly comparable with those of the Atherosclerotic Comorbid AF cluster found in two other cluster studies.^12,26 The CV-dominated patients accrued substantial costs, especially AF-adjudicated costs, which remained stable at a high level across time. This result is strongly consistent with the complementary multivariate approach.

The second cluster, representing IS, showed the lowest costs among the three clusters. Patients were noticeably younger, had fewer comorbidities, but high rates of previous cardioversion, AF symptoms, and paroxysmal AF. This cluster has partially similar characteristics as the Younger Paroxysmal AF cluster identified in a Japanese cohort²⁹ and the Low Comorbidity cluster identified in a US cohort¹². Differences in cultural and clinical settings, and in available descriptors of patient characteristics may have limited the degree of similarity. Considering the results of the multivariable regression-based analysis, this cluster seems to possess very few covariates that drive the costs significantly above a base level, explaining the low cost level.

The third cluster, HF-dominated, featured similarly high costs as the CV-dominated, as well as similarly high rates of heart failure. However, patients had almost no prior myocardial infarction, little diabetes, low aspirin intake, yet high diuretics intake. This may be consistent with heart failure caused by underlying causes other than cardiovascular disease. The patients also featured a relatively low prevalence of covariates associated with higher AF-adjudicated costs but were at the same cost level as the CV-dominated cluster. A possible explanation for this result may be unplanned hospitalisations and complications due to the ongoing challenge to treat patients with heart failure. This specific cluster has not yet been reported by other AF cluster studies.^12,26

The prospective cohort study patients we report on differ from the patient populations underlying recent AF costs studies in four ways, which has implication for the interpretation of the results. Firstly, most studies to-date relied on administrative databases^21,25,30; secondly, they were often retrospective^19,21,25,27; and thirdly, had short follow-up times between 0.5 to 3 years^19,24,29. In combining clinical and health insurance claims data prospectively, the present analysis adds detailed insight into the longer-term healthcare costs of AF patients. Fourthly, the present study examined prevalent patients with varying disease duration (including 13% enrolled within six months after first AF diagnosis). Even though the Swiss-AF study population cannot be regarded as a representative sample of AF patients, it provides insights on the longer-term cost impact of AF in a substantially wider spectrum of AF patients than previous studies.

There are several limitations inherent to our study that require discussion. First, the data structure is highly complex and patients in Swiss-AF are very heterogeneous. We tried to account for the heterogeneity with the clustering method, and covariate selection was based on the current literature as well as discussions with clinicians from the Swiss-AF study centres. Second, as the study population is not truly representative of all Swiss AF patients, a typical ‘average’ AF cost could not be reliably deducted. Further work will approach this topic with a general population-based, matched control population. Third, given deductible levels in the Swiss statutory health insurance system, the possibility that not all claims were handed in for reimbursement exists, which may have led to an underestimation of true costs. However, as AF patients have substantial costs, it is rather unlikely that claims were not submitted. A fourth limitation lies in the adjudication algorithm, which was developed carefully and considered clinical information together with claims data. Still, it may not fully reflect the true causation of costs by AF and its complications, and could not distinguish which specific complications, e.g. strokes, were truly due to AF or not.

In conclusion, this study analysed real-world AF costs, patient clusters and their cost trajectories. We observed mean annual AF-adjudicated costs of roughly CHF 6,000 per patient and stable relative contributions of different cost categories during five years. Three well-differentiated patient clusters were identified using a hierarchical clustering method: CV-dominated, HF-dominated and IS patients. The cost levels for AF-adjudicated and overall healthcare costs depended on patient cluster but remained again stable over time. CV-dominated patients accrued the highest costs and HF-dominated patients only slightly lower ones, while IS patients had substantially lower costs. These insights may help clinicians and payers to develop an early understanding of the magnitude of to be expected AF-related healthcare costs.