Diastolic dysfunction (DD), a condition of impaired left ventricular (LV) relaxation, is a risk factor for heart failure (HF) and cardiac mortality irrespective of LV ejection fraction (LVEF)1,2,3. The associated increase in LV filling pressures results in dilation of left atrial (LA) size and loss of LA compliance and contractility4,5, which can be assessed using LA wall strain and LA emptying fraction (LAEF). Therefore, features of the LA are regarded as the barometer of DD. To date, many publications have addressed some LA features and have established compelling evidence that LA parameters are essential in the evaluation of DD6,7,8,9. However, few have compared the relative importance of their associations with DD and DD-related adverse outcomes. In addition, most of the published work is based on echo parameters. As Cardiovascular Magnetic Resonance Imaging (CMR) becomes increasingly utilized in the evaluation of cardiomyopathy or heart failure there is a need to better understand CMR based LA indices. In this study we sought to compare the associations of four LA parameters: minimum and maximum LA volume (LAVmin and LAVmax), LAEF, and phasic LA strain (reservoir, conduit, booster pump) assessed by CMR, with DD, DD severity, and hazards of HF and mortality. We have also compared the hazards of LA features with that of LV longitudinal strain to demonstrate the relative importance of LA parameters.

Methods

Participants

This is a sub-study of a previously published larger study where participants underwent CMR and echocardiography within 7 days, between January 2007 and December 2015 at a single center10. The patient cohort (N = 132) was selected if there was evidence of DD, which was characterized by echocardiography based on the current ASE guidelines11. Of those, 66 were prospectively recruited for research and 66 were retrospectively analyzed from clinical studies. Patients were excluded if they were found to have history of atrial fibrillation, mitral stenosis, more than moderate mitral regurgitation, or prosthetic valve in the mitral position, which are conditions not supported by ASE guidelines for DD evaluation11. We also excluded subjects with indeterminate diastolic dysfunction grade. In addition to DD patients, normal controls were prospectively recruited who were free of cardiovascular history, major risk factors, and had both normal ECG and echocardiography. The vital signs and body surface area (BSA) for all subjects were collected at the time of CMR. Outcome data of hospitalized heart failure was extracted from electronic medical records of the health system consisting of 6 hospitals. The all-cause mortality information was obtained from National Death Index.

Ethics approval and informed consent

This study was approved by the St. Francis Hospital Institutional Review Board. Informed written consent was obtained from all participants and research was performed in accordance with relevant guidelines and regulations.

Imaging acquisition

Transthoracic echocardiography

Patients were scanned with a multi-frequency transducer ultrasound system (Philips IE 33, Andover, MA, USA). Comprehensive echocardiographic examination was performed using standard views. From the apical window, pulsed wave Doppler was used to record mitral inflow for 3–5 cardiac cycles at the level of the mitral valve annulus and at the mitral leaflets’ tips. Tissue Doppler was applied to record mitral annular velocities at the septal and lateral corners of the annulus. The resulting annular velocities by pulsed wave Doppler were recorded for 3 to 5 cardiac cycles at a sweep speed of 100 mm/s. Tricuspid regurgitation signals were recorded by continuous wave Doppler from multiple windows. Two-dimensional measurements were performed according to recommendations of the American Society of Echocardiography11 and indexed to body surface area. DD was assessed according to the algorithm from the recent ASE/EACI guidelines where mitral inflow early (E) and late (A) peak velocities, early diastolic annular myocardial longitudinal velocity (e′), tricuspid regurgitation and LA volume index were measured in order to determine the DD grade12.

CMR Imaging acquisition

All subjects underwent CMR on a 1.5 T scanner (Avanto, Siemens, Malvern, PA, USA) with an 8-element phased array surface coil. Balanced ECG-gated cine imaging of the long axis planes (2-, 3- and 4-chamber views) and a stack of 8–12 short axis planes (8 mm slice thickness, with 0 mm skip), starting from the mitral annulus, was acquired using balanced steady state free precession sequence with 30 phases per cardiac cycle. The average temporal resolution was 50 ms, with a field of view of 240 mm, flip angle of 70 degrees, repetition time (TR) of 3.1 ms and echo time (TE) 1.3 ms. All patients were in normal sinus rhythm at the time of imaging.

Post processing

LA volume was analyzed following the area and length method13 using 2- and 4-chamber long axis cine images with commercial software (Circle Cardiovascular Imaging Inc, Calgary, Canada) and indexed to BSA. LA maximum volume (LAVmax) was assessed at LV end systole and LA minimum volume (LAVmin) at LV end diastole. Phasic LA strain was analyzed using feature tracking software (TomTec, Germany). LA endocardial contours were first drawn manually and propagated through cardiac phases along the 2-, 3-, and 4- chamber views. Manual adjustment was made when needed. Peak LA strains were assessed as the average peak strain values of the 3 longitudinal planes (Fig. 1). LV longitudinal strain was analyzed by feature tracking (Circle Cardiovascular Imaging Inc, Calgary, Canada) in order to capture the strain of full myocardial thickness. The epi- and endocardial contours were drawn manually on the end diastolic phase of the 2-, 3-, and 4-chamber cine images and propagated by the software to calculate 2D peak systolic longitudinal strain. All image analysis was performed by experienced operators.

Figure 1
figure 1

The left panel represents the respective 2-, 3-, 4-, chamber cine views from a normal control subject with the green feature tracking contours overlaying the left atrial wall. The right panels are the corresponding left atrial strain graphs obtained from each view as well as representative strain graphs from subjects with diastolic dysfunction grades 1, 2, and 3.

Full size image

Statistical analysis

Continuous variables were described as mean (standard deviation) and count data as frequency (percent). Our primary exposures of interest were phasic LA strain (reservoir, conduit, and booster pump), LAEF, LAVmin, and LAVmax. Intra-observer reliability was assessed 6 months between the repeated measurements and reported with intra-class correlation coefficients (ICC). The association between our predictors and DD was assessed using area under the ROC curve (AUC) comparisons, and multinomial logistic regression. In these models, our covariates included age, gender, BSA, prevalent diabetes, hypertension, and smoking status chosen based on a disjunctive cause criterion14.

The association between our predictors of interest and a composite outcome of all-cause mortality and HF admission was assessed using Cox proportional hazards models. All-cause mortality data was obtained from the National Death Index and HF admissions based on electronic medical record review of the entire health care network. HF admissions were defined by an inpatient stay in a hospital setting with ICD 9 or ICD 10 diagnosis codes associated with congestive heart failure at the time of discharge. Individual Cox proportional hazards models were created for each primary exposure variable adjusting for age, gender, BSA (except for LAVmin and LAVmax), diabetes, hypertension, and smoking status. The associations of LA parameters with the composite outcome were also examined using AUC comparisons. In addition, the association between E/A and E/a′ with composite outcome was assessed using AUC curves. Statistical analyses were performed using SAS v. 9.4 (Cary, NC, USA).

Results

LA indices and diastolic dysfunction

Our study subjects (n = 164) consisted of four groups: normal controls (n = 32), grade I DD (n = 69), grade II DD (n = 42), and grade III DD (n = 21). On average, patients with DD were 60 ± 14 years old and about half (N = 63) had advanced DD (grades II or III). Normal LVEF (> 50%) was present in 18 (14%) patients. Baseline characteristics of each DD grade are displayed in Table 1. Examples of phasic LA strains are shown in Fig. 1 where graded decreases of LA strains are shown from normal to grade 1, II and III diastolic dysfunction. Phasic LA strains and LAEF were progressively lower with increasing severity of DD (p < 0.001) (Fig. 2A–C, F; Table 2). Similarly, LV longitudinal strain was also progressively lower (Fig. 2G). Conversely, LAVmin and LAVmax were significantly higher with increasing severity of DD (p < 0.001) (Fig. 2D, E).

Table 1 Baseline characteristics* describing subjects with and without left ventricular diastolic dysfunction.
Full size table
Figure 2
figure 2

The graded change of left atrial reservoir strain (A), left atrial conduit strain (B), left atrial booster pump strain (C), left atrial minimum volume (D), left atrial maximum volume (E), left atrial emptying fraction (F), and left ventricular global longitudinal strain (G) in relation to the severity of left ventricular diastolic dysfunction. ANOVA with post-hoc Dunnett’s test yielded statistically significant differences between the means (p < 0.05) in each panel.

Full size image
Table 2 Distribution of left atrial strains and volumes across diastolic dysfunction categories.
Full size table

A reproducibility analysis was performed on 15 randomly selected subjects. The Intra-observer reliability by ICC for LA indices were 0.99 (0.96, 1.00), 0.96 (0.89, 0.99), 0.92 (0.75, 0.98), and 0.6 (0.29, 0.79) for LAVmin, LAVmax, LAEF and LA reservoir strain respectively.

In order to examine the dose–response relationship, we created an adjusted, generalized multinomial logistic regression model to estimate the log-odds of the association between LA parameters of each DD grade compared with normal controls (Table 3). We observed significant dose–response relationship in LAVmin, LAVmax, LAEF, LA reservoir strain, and booster pump strain, but not in LA conduit strain. Using LA reservoir strain as an example, for every 1% decrease in LA strain the log-odds of a person having a higher grade of DD compared with normal control were 1.11 (CI 1.03–1.19), 1.25 (CI 1.14–1.38), and 1.50 (1.30–1.73) for grades I, II and III, respectively. In attempting to remove the potential confounding effect of LV strain on DD, we added LV longitudinal strain as a covariate to the previously described model and found that all exposures except LA conduit strain remained associated with worsening categories of DD (Table 3).

Table 3 The association between change in LA features and worsening categories of left ventricular diastolic dysfunction.
Full size table

LA indices and composite clinical outcome

LA parameters were then examined as predictors of a composite outcome of HF admission and all-cause mortality. After a mean follow-up time of 38 ± 37 months, 61 (37%) patients reached the composite outcome due to 33 HF admissions and 28 deaths. No outcome events occurred in the normal controls. The outcome incidence was higher in more advanced DD: 45%, 63%, and 65% for grades I, II, and III, respectively (p < 0.001). After adjusting for age, BSA, gender, prevalent diabetes, hypertension, and smoking status, all LA parameters with the exception of LA conduit strain were independently associated with the composite outcome, and remained independently associated when the models were additionally adjusted for LV longitudinal strain (Table 4).

Table 4 The associations between LA features and increased hazards of clinical composite outcome.
Full size table

In a post-hoc analysis with an intention to remove potential confounding from LV abnormalities we tested the independent associations of LA parameters with the composite outcome in models adjusting for LVEF and indexed LV end diastolic volume in addition to age, gender, BSA, diabetes and hypertension. We found that LA parameters remained to be independent predictors of outcome with HR 1.05 (1.00–1.10) for LAVmin, 1.06 (1.00–1.11) for LAVmax, 1.05 (1.00–1.10) for LA reservoir strain, 1.00 (0.95–1.05) for LA conduit strain and 1.07 (1.01–1.13) for LA booster pump strain. Overall, the results were similar to previously adjusted models.

The associations of LA parameters with the composite outcome were additionally examined using AUC comparisons. The AUC was highest in LAEF (0.760) followed by LA reservoir strain (0.733) and LAVmin (0.725) (Fig. 3A). For reference purpose the AUC of LV longitudinal strain was 0.747. Additionally, we examined markers of diastolic dysfunction measured by echocardiography for reference including E/A ratio and averaged E/e′ ratio (Fig. 3B). The performance of E/e′ ratio (AUC 0.76) was similar to that of LAEF and reservoir LA strain but not E/A ratio (AUC 0.54). When comparing the AUC of LAVmax alone, the AUC of combined LAVmax and LAEF made an improvement in outcome prediction (AUC 0.77, p = 0.046) but combined LAVmax and LA reservoir strain did not (AUC 0.73, p = 0.28). Furthermore, combined LAVmin and LAEF (0.76) or LAVmin and LA reservoir strain (0.73) did not improve outcome prediction from LAVmin alone (p = 0.33 and p > 0.99, respectively).

Figure 3
figure 3

Receiver operating characteristic analyses of left atrial parameters (A) for predicting the composite clinical outcome as well as standard echocardiographic parameters for diastolic dysfunction (B) for comparison.

Full size image

Discussion

In this study, we compared the relationships of LA parameters assessed by CMR including LAVmin, LAVmax, LAEF, and phasic LA strains with DD. Each LA parameter except LA conduit strain was independently associated with DD in a dose–response manner. Furthermore, each of the LA parameters except LA conduit strain was independently associated with the composite outcome of HF admission and all-cause mortality in the adjusted model including LV longitudinal strain or LVEF as a covariate.

Consistent with prior observations15,16,17,18, we found dose-responses between LAV (both LAVmax and LAVmin) and the severity of DD. We have also confirmed prior reports that impaired LA function is closely associated with DD19,20,21,22 in a dose–response relationship in our study. Similarly, phasic LA strains also demonstrated similar dose–response relationships, confirming prior observations made from an echocardiography speckle-tracking study23.

While LA reservoir and booster pump strain are valuable, we found it is comparable at best to LAVmin and LAEF in predicting adverse clinical outcomes. The outcome prediction by LAVmax alone is not as strong as LAVmin and LAEF. But by combining it with LAEF, the prediction is significantly enhanced. Our findings highlights the strength of LAVmin in DD evaluation which is superior to widely used LAVmax. Physiologically, LAVmin is directly modulated by LV filling pressure during LV end diastole as opposed to LAVmax which is predominantly influenced by the descending of mitral annulus thereby related to LV systolic function and LA compliance. Our collective findings suggest that LAVmin evaluation should be considered more favorably in the assessment of DD. It is also important to understand the relative importance of LA parameters in the context of other well-established imaging markers such as E/e′ and LV longitudinal strain in outcome prediction. We found that LAEF, the strongest outcome predictor of all LA indices including LA strain, has a comparable AUC from E/e′ and that from LV longitudinal strain. Given that LAEF correlates well with LV end diastolic pressure18 it is biologically plausible for it to predict clinical outcome effectively. Future research should examine the potential role of LAEF as an essential parameter in routine clinical evaluation.

It should be recognized that both LAV and LA strain evaluation are modality dependent. For instance, CMR has better spatial resolution than echocardiography rendering more accurate and reproducible LAV evaluation. In contrast, echocardiography has higher temporal resolution than CMR, hence providing a more favorable LA strain assessment. The technical differences may yield discrepancies in observations. That is why CMR specific investigations for LA in DD is essential despite the large body of literature using echocardiography. A recently published review article by Thomas et al.24 included a comprehensive list of studies that had investigated LAV and LA strain individually or collectively for their associations with DD. Most, if not all of the studies are echocardiography based and few, if any, use CMR. Additionally, we delineated the relative value of LAVmax, LAVmin, LAEF and phasic LA strain in DD and outcome prediction which has not been reported before. We have also established the associations of these LA indices with outcomes independent of LV longitudinal strain or LVEF. Of the LA parameters studied, LAVmin is easily attainable which allows LAEF to be calculated  from routine CMR or echocardiographic examination without the demand of additional post processing such as for LA strain. It is foreseeable to incorporate LAVmin and LAEF into future clinical algorithms for DD evaluation.

We recognize the limitations of our study. This is a pooled sample of both prospective and retrospective cohorts. The demographic data from the retrospective cohort was collected from electronic medical records and thereby subject to availability. CMR and echocardiography were performed within the same day for half of the DD patients but within 7 days for the other half. As DD diagnosis and grading were based on echocardiographic criteria, CMR evaluation that followed echocardiography was subject to the interval variation of hemodynamics, although there was no significant change reported during the interval. The DD patients consisted of both reduced and preserved EF. While the sample size is powered for outcome risk assessment we acknowledge the relatively small sample size originating from a single center clinical cohort. In addition, female gender is under represented. Similar to LV strain assessment, the value of LA strain is likely vendor dependent25. Therefore, the LA strain values generated from our normal subjects or from patients with DD cannot be directly compared with published studies using different vendors. Nonetheless, the relative importance of LA strain in our study that was compared with the normal controls is not subject to the variability of absolute value. It should also be noted that most of the feature tracking software is designed for LV strain evaluation and not for LA specifically. That is probably why the reproducibility of LA strain is only modest at best. However, our experience showed that the modern feature tracking program tracks the LA very well despite the thin wall. Tracking of the mobile mitral annulus is not always reproducible, and that seems to contribute to the variability. While the spatial resolution is excellent the temporal resolution of the CMR cine imaging is relatively low (about 50 ms) which may have compromised the accuracy of assessing the cardiac motion occurring within a short time frame such as LA boost pump strain. LA fibrosis evaluation by CMR can be valuable in patients with DD, but it was not performed in this study. While our electronic medical record allows us to capture HF admissions in the entire health system consisting of 6 hospitals, it is still possible that patients were admitted elsewhere, and therefore we may underestimate the scope of the outcomes. Lastly, this is a single center study and most of our patients are Caucasian. Future multicenter studies with diverse patient populations are warranted.

Conclusion

LA features including LAV, emptying fraction, as well as LA reservoir and booster pump strains have dose–response relationships with DD severity. They are also independent predictors of HF admission and all-cause mortality. LAVmax is not a strong outcome predictor but the prediction can be enhanced significantly by combining it with LAEF. While LA strains are valuable, conventional parameters such as LAEF and LAVmin remain highly effective in outcome prediction with comparable performance.