Ratio of early transmitral inflow velocity to early diastolic strain rate predicts atrial fibrillation following acute myocardial infarction

This was a prospective study of acute ST-segment elevation myocardial infarction (STEMI) patients treated with primary percutaneous coronary intervention (pPCI) at Gentofte Hospital, Denmark during the period of September 2006 to December 2008. A total of 391 patients were included and underwent a detailed echocardiographic examination. Four patients were excluded due to inadequate quality of the echocardiographic examination, 14 due to known AF, and 4 due to missing transmitral Doppler flow. This left 369 patients eligible for analysis in this study. The study population has previously been described in detail [17‐19].

Baseline data including demographic and clinical parameters relevant for this study was obtained prospectively when the patients were enrolled in the study. Hypertension was defined as the use of blood pressure-lowering drugs on admission. Diabetes mellitus was defined as non-fasting plasma glucose concentration ≥ 11.1 mmol/L, fasting plasma glucose concentration ≥ 7 mmol/L, or the use of glucose-lowering drugs at admission. Troponin I was measured immediately upon admission.

Echocardiography was performed using Vivid 7 ultrasound system (GE Healthcare, Horten Norway) using a 3.5-MHz transducer by trained sonographers. The echocardiographic examinations were performed within 5 days (median 2 days (IQR: 1–3)) after pPCI [20, 21]. All participants were examined with conventional two-dimensional echocardiography, pulsed-wave Doppler, and color TDI according to standardized protocols. The echocardiography was then stored and analyzed offline using commercially available software (EchoPAC, GE Healthcare, Horten Norway) by a single investigator who was blinded to the patients’ clinical characteristics and outcomes.

Chamber quantifications were performed according to guidelines [22]. Left ventricular mass index (LVMI) was calculated by left ventricular mass divided by body surface area. LV ejection fraction (LVEF) was measured using the Simpson’s biplane method. Left atrial (LA) end-systolic volume was measured using the biplane area-length method in the apical 2- or 4- chamber view at end-systole. Two-dimensional parasternal long axis images were used to measure LV internal dimension at end-diastole (LVIDd) and LV end-systolic volume and LV end-diastolic volume were measured in the 2- and 4- chamber view at end systole and end diastole respectively. Peak velocities of transmitral early filling (E) and late filling (A) along with E/A-ratio and deceleration time (DT) of the E-wave were assessed using pulsed-wave Doppler in the apical 4- chamber view. Pulsed-wave tissue Doppler was used to measure peak longitudinal early diastolic tissue velocity (e’) with sample volumes placed at the lateral and septal mitral annular level in the apical 4-chamber view. The E/e’ was calculated from these Doppler measurements, and a high E/e’ was defined as above 14.

2D speckle tracking was performed in the 3 apical projections (4-, 2- and 3-chamber view) with the highest available frame rate (mean 86 ± 23 Hz). The region of interest (ROI) was defined by a semiautomated function and could be manually adjusted to avoid inaccurate tracking of the myocardial wall. A total of 18 segments, 6 per projection, were included. The global value was calculated as a mean of all LV segments. Segments were excluded if deemed untraceable due to shadow/artefacts. Global longitudinal stain (GLS) was calculated as the mean peak systolic longitudinal strain of all LV segments. The global SRe was calculated from the mean values of all LV segments. E/SRE was defined as E velocity divided by global SRe (Fig. 1). A total of 368 patients had 4-chamber views adequate for speckle tracking analysis, 362 subjects had adequate 2-chamber views and 354 had adequate 3-chamber view images for speckle tracking analysis. LA speckle tracking was performed in the apical 4-chamber and 2-chamber view by tracing the endocardial border manually using LV dedicated strain software to acquire peak atrial longitudinal strain.

The feasibility of E/SRe was 99% for the study population as we could not measure E/SRe in 4 of 373 patients. Previously our lab has shown good intra-variability and interobserver variability of both SRe with a small bias (for SRe: mean difference ± 1.96 SDs was 4 ± 23 m for the intraobserver analysis and 6 ± 37 m for the interobserver analysis [23].

The endpoint was the development of new-onset AF. The endpoint was obtained by International Classification of Disease (ICD-10) codes from the Danish Board of Health's National Patient Registry in 2013. The competing endpoint of all-cause death was obtained from the National Causes of Death Registry. Follow-up was 100%. The ICD-10 codes were obtained from all hospital contacts, both in- and outpatient contact for all the patients.

Normal distribution for continuous variables was assessed by histograms and box plots. Continuous variables showing normal distribution are presented as mean±standard deviation, whereas those not showing normal distribution are presented as median with interquartile ranges (IQR). Categorical variables are presented as total numbers and percentages. Groupwise comparison was made with the Student’s T-test for normally distributed continuous variables, and with the rank-sum test for variables not showing normal distribution. Categorical variables were compared with the Chi²-test. Comparisons were made according to the endpoint of AF and according to high vs. low E/SRe (defined by the median of 97 and confirmed by ROC curves as the optimal cut-off value with the highest AUC). Of note, among healthy participants from the Copenhagen City Heart Study, 14 of 1623 (0.9%) have an E/SRe above this threshold. Univariable proportional hazards Cox regression was used to investigate the association between both E/SRe and E/e’ and AF. Harrell’s C-statistics were calculated from the univariable regressions. Multivariable Cox regressions were made to adjust for confounders and obtain adjusted hazard ratios. Adjustments were made for: age, sex, hypertension, diabetes mellitus, peak Troponin I, LVMI, LVEF, heart rate, significant valve disease and left anterior descending culprit lesion. The multivariable model was extended to include LA strain in the subgroup of patients with feasible LA speckle tracking (n: 299, events: 18). Proportional hazards were assessed by Schoenfeld residuals. To avoid a potential issue of collinearity, E/SRe and E/e’ were investigated in separate multivariable models adjusted for the same confounders but were not included in the same model. Linearity was assessed by linear and quadratic effects. Test for interaction was made with sex and GLS as these have previously shown to modify the association between E/SRe and outcome. To investigate whether LVEF impacted the association between E/LVSRe and AF a test for interaction against LVEF was applied the association between E/SRe and the outcomes were depicted using restricted cubic splines based on a Poisson model, with the number of knots based on the lowest Akaike information criterion.

Cox regression was also applied in a subgroup analysis of patients with normal filling pressure defined as an E/e’ < 14. Competing risk regression analyses were performed with similar multivariable adjustments as noted above by the Fine and Gray method to account for death as a competing event (number of competing events: 52). Estimates are presented as subdistributional hazard ratios (SHR). Kaplan–Meier curves were created to show the association between high vs. low E/SRe and the combined outcome throughout the follow-up period. Statistical difference was tested with the log-rank test. All statistical analyses were performed with STATA Statistics/Data analysis, SE 15.1 (StataCorp, Texas, USA). A p-value < 0.05 was considered significant in all tests.

Compared to patients with low E/SRe, those with high E/SRe had an almost four-fold increased risk of AF during follow-up (HR = 3.85 (1.43–10.35), p = 0.004) (Fig. 2). The incidence rate was significantly higher in patients with high E/SRe as compared to those with low E/SRe (high E/SRe: 20 (95% CI: 13–32) events per 1,000 patient-years; low E/SRe: 5 (95% CI: 2–12) events per 1000 patient-years).

In univariable Cox regression analyses, both E/SRe and E/e’ were significantly associated with AF (E/SRe: HR = 1.06; (1.03–1.10); p < 0.001, per 10 increase; E/e’: HR = 1.11 (1.05–1.17); p < 0.001, per 1 increase). Both measures had a Harrell’s C-statistic of 0.71. Of note, a non-linear association between E/SRe and AF was observed. Showing that an increasing E/SRe was associated with an increased risk of AF as shown in Fig. 3. Furthermore, decreasing LA strain, increasing age, increasing peak TnI, and history of diabetes were also significantly associated with an increased risk of AF (Table 2). However, after multivariable adjustments, E/SRe was the only parameter to remain independently associated with AF (HR: 1.06 (1.00–1.11); p = 0.044, per 10 increase) (Table 2). When applying the same multivariable adjustments for E/e’, no parameter was associated with AF, including E/e’ (HR 1.08 (1.00–1.16), p = 0.06). If we extended the multivariable model to include LA strain (available in a subset, n: 299), E/SRe was still significantly associated with AF (HR 1.09 (1.02–1.16), p = 0.009, per 10 increase), and E/e’ was not (HR: 1.09 (0.98–1.21), p = 0.10, per 1 increase). No effect-modification was observed by sex nor GLS (p for interaction > 0.05) for the association between E/SRe and AF. Moreover, LVEF did not significantly modify the association between E/LVSRe and AF (p for interaction = 0.15). All of the findings were unchanged in multivariable competing risk regression analyses when accounting for all-cause death as a competing event (E/SRe: SHR = 1.05 (1.00–1.11), p = 0.038, per 10 increase; E/e’: SHR = 1.08 (0.98–1.20), p = 0.12, per 1 increase).

E/SRe was also significantly associated with AF in patients with normal filling pressure (E/e’ < 14) (n = 301) in unadjusted analysis (HR = 1.09 (1.01–1.17); p = 0.030, per 10 increase). After multivariable adjustments E/SRe remained associated with AF (HR = 1.12 (1.00–1.26); p = 0.048, per 10 increase).

When looking at well-established echocardiographic predictors of AF, abnormal E/e (> 14) and TR gradient (> 2.8 m/s), were significantly associated with an increased risk of AF, whereas abnormal LVEF, GLS and LAVi were not (Supplementary table 2). However, on a continuous scale, only E/SRe remained significantly associated with AF after multivariable adjustments as described previously.

As previously mentioned, AF is the most common cardiac arrhythmia following AMI. Several factors increase the risk of developing AF after AMI, particularly elevated LV filling pressure associated with diastolic dysfunction. As LV filling pressure increases—as often observed after AMI–an increased afterload pressure gradient develops across the LV and LA chambers, causing LA remodeling, and therefore increases the risk of developing AF [24]. E/e’ has long been used to assess LV filling pressure in a non-invasive approach and is one of the key guideline-based methods for estimating filling pressure [19, 25]. However, E/e’ is subjected to several technical limitations (as outlined below), and other markers have therefore been proposed, including E/SRe. Several studies have demonstrated an association between invasively measured LV end diastolic pressure and E/SRe [14].

The increasing usage of two-dimensional myocardial speckle tracking, which is a fully automatic technique, along with its implementation into clinical guidelines, strain rate has a potential to play a key role in echocardiographic assessments in the future. E/SRe is therefore a fast, non-invasive, and accurate way of estimating the LV filling pressure in patients undergoing echocardiography. Recently, the measure has been benchmarked even further for clinical use as normal values have been defined based on 1623 healthy individuals from the Copenhagen City Heart study, with an overall upper reference limit of normality for E/SRe of 80 [26].

In our study, we found that both E/SRe and E/e’ were significantly associated with AF and had similar predictive value by Harrell’s C-statistics. However, after adjusting for confounders and further accounting for all-cause death and AF as a competing event, only E/SRe remained associated with the outcome. In addition to this, several technical circumstances favor the use of E/SRe. It is well-established that E/e’ often lacks accuracy due to several technical limitations that challenge its interpretation, including susceptibility to sampling location, myocardial tethering and significant errors with angulation > 20 degrees [27]. In contrast, strain rate derived from speckle-tracking is not angle dependent nor is it affected by myocardial tethering [20]. Furthermore, it may better reflect the LV relaxation as it is a global measurement taking the entire LV myocardium into account, in contrast to e’ witch only measures velocities at the basal segments near the mitral annulus of the LV. Furthermore, regional wall motion abnormalities- which is frequently observed in MI—near the sampling region along with mitral annular calcifications, may underestimate values of e’ and thereby falsely classify patients as having diastolic dysfunction despite overall normal LV relaxation. Consequently, E/SRe may facilitate diastolic function assessment when E/e’ is technically unreliable. However, our findings suggest that its value is not restricted to these settings since we also observed that E/SRe was a predictor of AF in patients without elevated LV filling pressure by E/e’. By conventional standards these patients may have either been classified as having normal diastolic function or indeterminate diastolic function, and E/SRe could help properly classify and risk stratify these patients.

One former small study conducted by Bonapace et al. have investigated the relationship between E/SRe and the development of new-onset AF. They investigated a cohort of 180 type-2 diabetic patients and found an increase in E/SRe in those who developed AF, similarly to our findings. However, E/SRe was not significantly associated with AF in the multivariable logistic regression. However, they had a low number of events (14 events) and their findings may therefore have been influenced by low power [16]. While there is sparse data regarding the association between E/SRe and AF, the prognostic value of E/SRe has been investigated in multiple other settings. Ersbøll et al. showed in a cohort of AMI patients (n = 1100), that E/SRe had a better prognostic value than E/e’ and was a stronger predictor of a composite endpoint HF, AF, stroke, and all-cause mortality. Remarkably there was no association between E/SRe and AF, due to the fact that this study was driven by HF and all cause death [15] Lassen et al. have further demonstrated E/SRe as an important and reliable prognostic measurement superior to E/e’ in a number of studies in specific patient populations including type 2 diabetes, heart failure with reduced ejection fraction, acute coronary syndrome, and the overall general population in relation to outcomes including AMI, HF, AF, and mortality [21‐23, 28]. Similar to our findings, these studies also showed a non-linear relationship between E/SRe and outcome. Lassen et al. [21, 22] further found an effect modification with systolic function such that E/SRe was a particularly strong predictor of outcome in patients with preserved systolic function. In our study, we did not observe any effect-modification, which could reflect the relatively few numbers of events.

The LA is directly exposed to pressure from the LV during diastole and will therefore enlarge as LV filling pressure increases. Accordingly, elevated LV filling pressure and abnormal LA volume is associated with AF [24]. No difference in LA volume by outcome was observed in our population, which may be because LA remodeling develops as a consequence of chronic pressure overload, and therefore remodeling of the LA may not appear in the acute settings of MI within the brief time period from infarct to echocardiography (median of 2 days in this study).

Besides E/SRe and LA volume, LA strain has proven valuable. A recent study found that LA strain did not add additional prognostic value, when both LA volume and LV filling pressure was already obtained [29]. Intrinsic LA remodeling and dysfunction may be a key factor in the pathogenesis of AF. In our study, however, we found that E/SRe remained an independent predictor of outcome even after adjusting for LA strain, emphasizing that the association is not driven exclusively by changes in the LA. Rather, LV abnormalities may be the first change of clinical significance that eventually leads to LA functional impairment. Our findings further indicate that employing E/SRe in the diastolic dysfunction assessment was useful for identifying patients at-risk of outcome, who would not otherwise be perceived as having elevated filling pressure. The results emphasize that E/SRe could be a useful diastolic measure to be used in concert with conventional measures of diastolic function.

Although strain rate has shown potential in risk stratification across several studies, the technique itself has its limitations. The technique relies–to a larger extent than GLS–on adequate temporal resolution, since a low frame rate may lead to erroneous sampling of peak amplitudes [30]. Consequently, operators should keep this technical aspect in mind. This issue may be resolved with future technical developments as high-frame rate applications become available [31]. Our results should be interpreted in light of this technical limitation, which also emphasizes that our findings are hypothesis-generating and should be validated externally and further validated once high-frame rate techniques become widely available.