Association of right atrial structure with incident atrial fibrillation: a longitudinal cohort cardiovascular magnetic resonance study from the Multi-Ethnic Study of Atherosclerosis (MESA)

MESA is a population-based cohort of participants free of clinically-evident cardiovascular disease at baseline. The MESA protocol has previously been described in detail [23]. In brief, 6814 participants were recruited from six US communities between July 2000 and Aug 2002, aged between 45 and 84 years from four different, self-reported racial-ethnic backgrounds. The protocol was approved by the institutional review board at each study site and all participants provided written informed consent. Approximately every 9 months, participants were contacted to inquire as to cardiovascular diagnoses and events, hospital admissions, and mortality. Medical records and information were successfully obtained on an estimated 98% of reported hospitalized cardiovascular events and 95% of reported outpatient cardiovascular diagnostic encounters. Study inclusion and exclusion are shown in the flowchart (Fig. 1).

Incident cases of AF were identified through December 2014. Cases were identified through MESA hospitalization surveillance, study electrocardiograms (ECGs), and for participants enrolled in fee-for-service Medicare, from inpatient and outpatient Medicare claims data. We excluded individuals who developed AF after coronary artery bypass or valve surgery or those who were found to have an AF claim occurring before their baseline MESA examination.

The CMR protocol has previously been described in detail and was performed at time of enrollment [24]. In brief, baseline images were acquired with 1.5 T scanners (Signa, General Electric Healthcare, Waukesha, Wisconsin, USA; Symphony or Sonata, Siemens Healthineers, Erlangen, Germany). Long-axis cine images were obtained in the 4-chamber view using ECG-gated breath hold fast gradient-echo pulse sequence. This view was acquired as one slice of thickness 6 mm and the matrix size was 256 × 160. Temporal resolution varied, averaging 50 ms.

RA function was assessed using multi-modality tissue tracking software (MTT; Toshiba,

Tokyo, Japan), which has been previously described at length [25]. This software has been extensively validated in the LA with excellent reproducibility [26]. We included participants from the MESA cohort who had adequate views of the RA for assessment. Four-chamber cine CMR images were used for analysis. The study was conducted by a single experienced operator blinded to the cardiovascular disease (CVD) outcomes of participants. Endocardial borders were manually traced along the RA at the end-systolic frame and the epicardial border was automatically defined by software at a fixed distance of 3 mm from the manually-traced endocardial border. The borders were then propagated across frames of the cardiac cycle via multimodality tissue-tracking software (MTT) and manually verified for quality control (Fig. 2). The software then generated volume vs time curves from which maximum, minimum, and pre-atrial kick RA volumes were extracted (Fig. 3). The RA volume metrics included in this study are as follows:

RA EF was calculated based on RA volumes at different time points in the cardiac cycle, as follows:

Strain measurement in 4-chamber cine images was automated within the software with segmentation of the borders into free-wall and septal distributions. Strain is computed as an average across the entire manually-defined cardiac contour. The zero-strain reference was set at RA end-diastole. Peak global strain refers to the maximum calculated strain during the cardiac cycle within the RA, which in these baseline measurements of a CVD-free population reflect RA reservoir strain. Free-wall strain refers to the RA lateral wall and RA roof, excluding the septal wall. Strain curves were computed for both peak global and free-wall longitudinal segments during the cardiac cycle and visually assessed for quality.

Study quality was defined categorically from 1 to 4. A single operator was responsible for assigning quality across all studies. Definitions are as follows:

Standardized questionnaires were used at baseline to collect information from participants about demographics and cardiac risk factors as previously described [21]. We screened the following covariates for our analysis: age, sex, race/ethnicity, height, weight, body mass index (BMI), body surface area (BSA), cigarette smoking status, alcohol consumption, diastolic blood pressure (BP), systolic BP, fasting glucose, diabetes, use of hypertension medication, total cholesterol, high density lipoprotein (HDL) cholesterol. Baseline adjudication of COPD, deep venous thrombosus (DVT), PE, or OSA were not available in MESA.

Our multivariate Cox models adjusted for these covariates as follows. In model 1, we included demographic characteristics (age, race, sex) and traditional cardiovascular risk factors (BMI, smoking status, HDL, lipid-lowering medication, systolic BP, hypertension medication, fasting glucose, alcohol use). RA volume index (RAVI_max) and RAVI_min remained significantly associated with incident AF. RA EF, peak global and free-wall strain were not significant after this adjustment. In model 2, we made additional adjustments for LA maximum volume index, peak LA strain, and LA EF as obtained from previous studies [27].

Additionally, we adjusted for LA, left ventricular (LV), and RV variables as obtained from and described in previous studies in MESA [5, 7, 21]. In brief, the LA variables most closely associated with incident AF were: elevated LA maximum volume index (adjusted HR 1.38, p = 0.04), passive LA EF (adjusted HR 0.55, p < 0.001), and peak LA strain (adjusted HR = 0.68, p = 0.031). In the LV, LV mass was the strongest association with AF (adjusted HR = 1.45, p < 0.001). For the RV, EF (adjusted HR 1.15, p = 0.04) and RV mass (adjusted HR = 1.16, p = 0.07) were the most significant associations for AF.

Statistical analysis was performed with Stata software (Stata Corp, College Station, Texas, USA). Continuous variables are denoted here as mean ± standard deviation while categorical variables are denoted as numbers and percentages. Baseline participant characteristics and RA metrics were evaluated using the Student’s T-test for continuous variables and Chi-square test for categorical variables. ANOVA with Tukey’s post-hoc test was used to assess for variation in RA metrics by image quality. Cox proportional hazard models were used to assess the association of RA structure and function parameters with time to incident AF, with death, end of follow-up, and withdrawals treated as censored. Cox models were adjusted for significant predictors of AF selected from among the covariates listed above. Propensity score matching was used for validation of our results. We modeled the odds of AF for participants with a below-median RAVI_max compared with above-median RAVI_max, with matching on demographic and traditional risk factor covariates. P-value < 0.05 was considered significant. GraphPad Prism (GraphPad Software, La Jolla, California, USA) was used to create the additional figure.

We evaluated the reproducibility of both RA volume and function metrics in MTT. Interobserver and intraobserver variability was assessed with 20 randomly-chosen cases. RA analysis was independently performed by two operators (E.X. and E.C., a cardiac technologist with 10 years of CMR experience) while blinded to AF incidence and other clinical parameters. Intraclass correlation coefficient analysis was performed to evaluate variability.

RA volume was indexed by BSA yielding maximum volume indexed (RAVI_max) and minimum volume indexed (RAVI_min). RA structure and function in patients who did and did not develop AF are summarized in Table 2. The baseline RAVI_max was 6% higher and RAVI_min is 9% higher in patients with incident AF (23.7 ± 9.2 vs 22.4 ± 8.0 mL/m², p = 0.002, 13.2 ± 6.7 vs 12.1 ± 5.8 mL/m², p < 0.001, respectively). The RA EF was 4% lower in AF cases (p = 0.02). In parallel, baseline RA peak global strain and free-wall strain were 6% lower in AF cases (p < 0.001 and p = 0.049 respectively).

Multivariable Cox regression models were used to assess the association of baseline RA structure and function metrics with the development of AF (Table 3). Results are presented as HR per SD. Unadjusted, all RA measures were significant.

Larger RA structure as assessed by higher RAVI_max and RAVI_min remained positively associated with incident AF (HR per SD 1.13 and 1.12, respectively). Comparatively, the effects of LA metrics in the model were: LA maximum volume index (HR per SD: 1.46), LA passive EF (HR per SD: 0.79), LA peak strain (HR per SD: 0.82). Propensity score results yielded a similar direction. The top quantile of RAVI_max was associated with a 3.5% higher incident AF (p = 0.007) at 1:1 matching for AF and non-AF participants. In adjusted quantile analyses of RAVI_max, there was a notable graded increase in risk with increasing quantiles. Compared to the middle quintile, the 4th and 5th quintiles had, respectively: HR 1.4, p = 0.04 and HR = 1.7, p = 0.002. In contrast, risk did not appear to be mitigated with lower RA volumes and the lowest quintiles were not significantly different than the middle quintile.

Additionally, correlation between AF risk factors (including those in CHA₂DS₂-VASc) and RA volumes is shown (Additional file 1).

We additionally created models adjusting for demographics, traditional risk factors, and RV variables previously identified as being associated with incident AF [7] (Additional file 2). RAVI_max was significant in fully adjusted models including RV EF and mass (HR per SD: 1.15, p = 0.01). RAVI_min was not significant after adjustment (HR per SD: 1.11, p = 0.09). Further adjustment for LV mass had no effect on the model. Neither RAVI_max nor RAVI_min were significant after adjustment for demographics, traditional risk factors, and both LA and RV variables within the same model.

Reproducibility of RA was evaluated in 20 randomly selected subjects. Intraclass correlation coefficients (ICC) for intraobserver reproducibility were 0.93 for RAV_max, 0.89 for RAV_min, 0.91 for global strain, and 0.89 for free-wall strain. ICC for interobserver reproducibility were 0.93 for RAV_max, 0.85 for RAV_min, 0.85 for global strain, and 0.82 for free-wall strain.

The sub-population of MESA who underwent CMR examination, as expected, was healthier than the sub-cohort who did not (Additional file 3). Within the sub-population who underwent CMR and entered the analysis, measured parameters were significantly different at different image/tracking qualities (Additional file 4). To account for this, interaction between image/tracking quality and RA parameters were incorporated into the preliminary assessment of the Cox models as covariates. They were not found to affect the association with AF for any RA measure and there was no evidence for interaction. Therefore, image quality variables were not included in the final models.

As structural changes in the LA are well-understood to precede the development of AF [6, 36, 37], we had hypothesized that there would be comparable findings in the RA. Our study corroborated our hypothesis, as RA volume was significantly associated with incident AF in all models. We suspect RA volume serves as an intermediate marker between pathologic processes and AF. It has been demonstrated that in pulmonary arterial hypertension and systolic heart failure, RA volume is significantly increased [30, 38]. An enlarged RA at baseline may reflect subclinical fibrosis, pulmonary insult, or other etiologies preceding detectable disease. In particular, the pathophysiology of AF associated with pulmonary disease may be mediated through the RV and in this study, we demonstrate that RA alterations remain as a significant predictor of AF after adjustment for RV variables. Moreover, early RA remodeling may be a reversible process as has been shown of early LA remodeling [39]. Clinical studies and computational models have indicated that re-entrant drivers of AF in both RA and LA may be associated with fibrotic substrate [40, 41]. In turn, small studies have suggested that markers for increased fibrosis in the LA may be correlated with increased volume index, a potential pathophysiologic link which merits further study in the RA [42]. As the natural history of AF is marked by both structural and electrical remodeling becoming increasingly challenging to treat with increasing burden of disease, appropriate therapies are necessary in both atria to address time-critical mechanisms including development of fibrosis [41].

Existing literature on the association of RA volume with AF is limited. We could not find studies in participants free of CVD. A previous study of RA remodeling after ablation of paroxysmal AF suggested that larger volumes were associated with increased risk of recurrence [43]. However, a similar study of RA volume before ablation found no significant association with AF recurrence [44]. In patients with paroxysmal AF on pharmacological control, RA volume was not associated with maintenance of sinus rhythm [45]. In patients with atrial septal defects, RA volume was not associated with development of AF [46]. However, as these studies followed patients for less than a year, it is also possible that RA volume was a less powerful biomarker in the short follow-up timescale.

The absolute differences in baseline RA volume in patients who did and did not develop AF may be underrepresented as the majority of AF originates from sources in the LA [12, 13]. Though our models suggest a contribution of RA volume to AF risk, the magnitude of the unindexed baseline difference in simple comparison tests, on the order of 4 mL, may be challenging to clinically appreciate. Though most patients have multiple mechanisms of AF development, sometimes involving both atria, ablation of LA sources is often sufficient for rhythm control in upwards of 70% of patients [14]. It is thus likely we had a smaller subset of cases of AF arising predominantly from the RA. This is further supported in our multivariate models which suggest that LA parameters have a larger effect size than RA parameters, though both are independent and significant predictors. However, since it is not infrequent to find RA mechanisms for AF among patients who fail LA ablation procedures to treat AF, recognition of these differences may improve medical management and guide procedural intervention in otherwise refractory cases. It is important to note that when RV variables were incorporated into our models alongside RA and LA variables, RA and RV variables lost significance while LA variables remained significant, a manifestation of the collinearity between RA and RV which makes it challenging to claim in our study that the RA is a predictor entirely independent of other chambers, unlike the LA.

The healthy RA modulates systemic venous return to ventricular filling in three phases: as a reservoir during systole, as a conduit in early diastole, and as a pump in late diastole. Peak global strain occurs during the second phase, when the ventricle is being passively filled, and serves as a marker for RA compliance and function. In this study, we also measured the free-wall strain, which excludes the atrial septum at this phase. Large studies of CMR-derived strain and LA emptying fraction (LA EF) have highlighted their potential role as predictors of AF [21].

As with RA volume, literature on the relationship between RA function and incident AF is limited, the few existing studies having been conducted in symptomatic patients with cardiac disease. RA emptying fraction was not associated with development of AF in patients with a pathological atrial septal defect undergoing repair [46]. In patients with paroxysmal AF, total peak strain was not associated with greater rate of maintenance of sinus rhythm at 1 year [45]. Our results are consistent with these findings as we did not find associations of peak global strain or RA EF with AF after adjustment for risk factors and subclinical CV disease. This is first study assessing RA function as a predictor of AF in a CVD-free multi-ethnic population. Interestingly, groups have investigated LA EF across different phases (total, passive, and active) though report differing results as to associations with AF [21, 47, 48]. This may be another area of potential future study.

Our study has several limitations. A single reader was responsible for post-processing though reproducibility studies were excellent with multiple readers. Use of a single reader reduces variability in research as in this study but does not reflect general clinical usage. Our study had a higher rate of unusable or uninterpretable images than comparable studies featuring LA wall tracking. Examples of poor quality images which were excluded are shown (Additional file 5). We posit that these differences result from less emphasis on RA than LA imaging during the original CMR acquisition process. Though our quantification methods were rigorously validated in the LA, we do not have standalone validation in the RA. As the MTT software used one slice to compute volume, the calculated volumes may slightly underestimate true volumes [49, 50]. As noted in the Results, there were differences in our measurements in association with RA image quality, which we suspect are due to limitations of feature tracking in noise and artifact that become more prominent in low-quality images. Additionally, some CMR sequences had suboptimal temporal resolution for strain assessment and we suspect that this could have introduced measurement error and reduced statistical power. Future prospective cohort studies with improved spatial and temporal resolution and additional views of the RA would be beneficial to overcoming these limitations. Similarly, incorporation of clinical comorbidities such as DVT/PE, OSA, and COPD, which were not available to us, would be beneficial in future studies. As of the writing of this, use of CMR in patients without clinical CVD is generally limited to research and exploration of the RA with echocardiography could further validate our results for clinical applicability.