Personalized medicine and atrial fibrillation: will it ever happen?

Many medications used for AF treatment have narrow therapeutic margins. Antiarrhythmic efficacy is only about 50% at 12 months [26], and prohibitive adverse effects such as proarrhythmia or other reactions may occur [27, 28]. Thromboembolism prophylaxis with warfarin is limited by increased risks of disabling stroke among subtherapeutic patients, and bleeding among those with supratherapeutic doses [29].

Pharmacogenetics refers to how genetic variation accounts for differences in drug responses among individuals. Measurable differences in pharmacodynamics (that is, the interaction of drugs with their target molecules) and pharmacokinetics (that is, properties relating to drug uptake, distribution, duration of effect, metabolism, excretion, and so on) are likely to be associated with genetic variation. In a recent proof-of-concept paper, whole genome sequencing in a single individual identified over 60 previously reported pharmacogenetic variants affecting drug response [30].

A high-profile example of a pharmacogenetic effect relevant to AF management relates to warfarin. Warfarin dose requirements have been linked to variation in CYP2C9 [31] and VKORC1 [32], genes encoding products involved in warfarin pharmacokinetics and pharmacodynamics, respectively. Dosing algorithms that include genetic variation in these genes explain a striking one-third to one-half of the variability in warfarin dose requirements [33]. Accounting for variants in these genes may be particularly beneficial (1) during initiation of warfarin therapy when complications relating to excessive or underdosing are particularly prevalent, and (2) for identifying individuals with unexpectedly high or low therapeutic dose requirements [34] who may therefore be at elevated risk for thromboembolism or bleeding, respectively. The US Food and Drug Administration has advised physicians that genetic testing may improve initial dosing estimates among patients taking warfarin [35] and dosing prediction algorithms are available.

Nevertheless, routine warfarin testing has not been widely adopted for several reasons. Novel anticoagulants [36] and procedures [37] are emerging as potential alternatives to warfarin, rigorous surveillance may improve time spent in the therapeutic range thereby improving safety and efficacy [38], and logistical considerations such as costs and genetic literacy pose challenges to the adoption of genetic testing. Currently, the safety and efficacy of genotype-guided warfarin dosing is being compared to routine dosing in the ongoing Clarification of Optimal Warfarin Dosing through Genetics (COAG) trial (clinicaltrials.gov identifier NCT00839657).

Regardless of the outcome of this trial, the striking associations between genetic variants and warfarin doses have revealed the clinical potential for pharmacogenetics. Emerging reports demonstrate other associations between genetic variants and drug response. Genetic variation has been reported to associate with response to β-adrenergic antagonism in patients with AF [39], and genetic variation associates with drug-induced torsade de pointes [40, 41]. Given the heterogeneity in the molecular mechanisms underlying AF [42], as well as variability in medication efficacy and adverse effects, there may be a practical role for pharmacogenetics in the management of patients with this arrhythmia. It is worth noting that genetic variation has also been reported to associate with AF recurrence after catheter ablation [43]; however, the role of genetic testing for a procedure in which success is dependent on many technical factors remains undefined.

The next steps for assessing the clinical utility of pharmacogenetics will involve performing systematic phenotyping, genotyping, and association testing of established and novel pharmacological agents used for AF management. Candidate genes that ought to be prioritized for screening include those encoding products involved in drug uptake or metabolism, and those encoding drug targets.

Risk prediction encompasses estimation of AF probability and prediction of AF-related complications such as stroke, heart failure, and death. The goal of prediction algorithms is to identify individuals prior to the onset of AF or AF-related morbidity, and thereby help target prevention efforts. Estimation of risk may also assist with clinical trial development by enabling enrollment of individuals in a particular risk stratum.

Clinical prediction algorithms for new-onset AF have been developed in community-based cohorts of European and African American ancestry [44‐46]. These algorithms have good but not excellent discriminative properties. In a given sample of individuals without AF, they correctly assigned a higher predicted risk to those who ultimately developed AF only about 60% to 70% of the time [44‐46]. Simply flipping a coin would correctly assign a higher risk to those developing AF 50% of the time.

We and other investigators from the Framingham Heart Study tested whether accounting for the occurrence of AF in a first-degree relative (familial AF) improved AF risk prediction beyond clinical risk factors [7]. AF prediction improved slightly after accounting for familial AF. In a separate study from Sweden, adding genotypes for two common polymorphisms at the chromosome 4q25 and 16q22 AF susceptibility loci to clinical risk factors did not significantly improve AF risk prediction [47].

For most complex conditions, genetic risk scores have not achieved substantial improvements in discrimination over scores comprised of clinical risk factors. Yet enhancements in the understanding of genetic risk modeling reveal insights that may improve the utility of genetic risk scores in the future. For example, variants included in genetic risk scores to date have small effect sizes (for example, relative risks of 1.1 to 1.5). Single risk factors with such modest relative risks are unlikely to have meaningful impacts on discrimination alone [48]. Large-effect variants, which have greater potential to improve discrimination, are rare and not captured by genotyping panels used in genome-wide association studies.

Fine mapping and sequencing offer promise for the discovery of such large-effect variants [49]. For example in a fine mapping analysis we previously identified independent genetic variants at the chromosome 4q25 locus associated with AF [50]. A multimarker risk score comprised of genotypes tagging these three independent signals demonstrated that AF risk increased with an increasing number of AF risk alleles. The multimarker score identified about 12% of individuals with a twofold or greater relative risk for AF, and 1% with an estimated sixfold elevated risk of AF compared to those with the most common genotypes. This is one of the largest relative risks to date for a qualitative trait in the era of genome-wide association studies.

Another potential reason that genetic risk scores have not substantially improved prediction models is that few variants were tested. Accounting for additional genetic variants associated with conditions at far less stringent significance thresholds can substantially increase the proportion of variance explained by genetic factors [51‐53]. In simulations, genetic profiling using large numbers of genetic variants significantly improved risk discrimination [54].

Novel metrics have been developed to assess clinical utility of prediction models, with increased recognition that measures of discrimination alone can be insensitive [55]. The role of novel prediction metrics with respect to genetic profiling has not been extensively explored.

Furthermore, although a widespread heritable component underlies AF, the arrhythmia may result from a variety of different pathological processes. Whether genetic profiling will successfully predict incident AF may depend on the extent to which the 'type' of AF studied is influenced by genetic factors. The importance of refining AF classification has been highlighted elsewhere [56, 57], and may overcome the potential challenges that phenotypic heterogeneity presents to risk prediction efforts.

The absence of data demonstrating clinical utility of routinely using genotype information to guide clinical management in AF should not deter hope at this stage, but rather should be viewed as an opportunity for systematic testing of the clinical role of genetic information. Nevertheless, there is a risk that hype can obscure the monumental insights gained from the current era of genetic discovery. As an example, companies such as deCODE and 23andMe are performing direct to consumer genetic testing, and purport to calculate one's genetic risk of AF based on one or a few polymorphisms associated with modestly increased relative risks of AF. These calculations are not based on well-defined genetic effect estimates, and do not consider other genetic risk factors, interactions between genes, the predictive utility of genetic risk markers, or competing risks. Moreover, in the absence of data regarding clinical risk factors and environmental exposures, how are these results to be interpreted? Problems with direct to consumer genetic testing are reinforced by recent literature highlighting the facts that patients often have misperceptions about the results genetic testing [58], and practitioners frequently lack confidence in explaining direct to consumer genetic testing results [59].