Data source and patient population

This retrospective cohort study was conducted at Yale-New Haven Health, a large healthcare network in the United States, encompassing 5 acute care hospitals in academic and community settings and over 120 outpatient clinics at satellite locations throughout Connecticut [13]. The system-wide electronic medical record database was queried to identify patients age ≥ 18 years who received complete (as opposed to focused exam) transthoracic echocardiogram for any indication between July 1, 2013, and October 31, 2018, either during inpatient or outpatient encounters. These included echocardiograms obtained from 32 sites.

Structured data fields on echocardiographic parameters and full text of the cardiology attendings’ echocardiogram reads were extracted. We identified patients with isolated atrial dysfunction or ventricular dysfunction using the following definitions: isolated atrial dysfunction was defined as enlarged left atrium (LA dimension index >34 mm/m²) [14] with LVEF≥55% or history of atrial fibrillation or flutter with LVEF ≥55% based on prior studies [15]. Ventricular dysfunction was defined as those with LVEF <55% regardless of atrial dimension or atrial arrhythmia histories. We adopted this definition while recognizing that prior studies have variably defined atrial dysfunction [16]. LVEF was estimated using 2D or 3D volumetric tracing. We refer to SMR as MR of non-primary etiology encompassing ventricular or atrial dysfunction.

We restricted the cohort to those without significant concomitant valvular pathologies and MR of primary etiologies by the following exclusion criteria: those with primary MR of any severity (rheumatic, prolapsed, degenerative, or myxomatous findings), mitral stenosis of any severity, greater-than-mild aortic stenosis or aortic regurgitation, those with findings of vegetation or intracardiac mass or thrombus. Patients with prosthetic valves were also excluded (Fig 1). Additionally, patients with missing ejection fraction or left atrial dimension index were excluded. In patients with multiple echocardiograms, the study with the earliest date was analyzed. These criteria yielded 11,987 echocardiograms obtained on unique patients with either atrial or ventricular dysfunction, with secondary MR grades ranging from none to severe.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. CONSORT-style diagram.

The figure shows exclusion criteria and the definition of exposure groups. LA = left atrial; EF = ejection fraction; MR = mitral regurgitation.

https://doi.org/10.1371/journal.pone.0277385.g001

The Institutional Review Board at Yale University approved this study, and individual consent was waived for the minimal risk nature of the study.

Outcome

The outcome of the study was cumulative survival since the time of echocardiogram. Patient data were linked to the Connecticut Vital Statistics data curated by the Connecticut State Department of Public Health to ascertain the date of death. The state’s vital statistics record captures deaths of all Connecticut residents, whether the death occurred in or out of the state. The last date of follow-up available via the vital statistics data was October 15, 2019.

Echocardiographic and patient characteristics, missing values

Demographic variables extracted from the medical records system were age at the time of echocardiogram, sex, and race. Comorbidities were selected based on prior studies and identified via the patient’s past medical history that were grouped based on the corresponding International Classification of Diseases 10^th revision, Clinical Modification (ICD-10-CM) codes [17]. The extracted comorbidities were updated and verified by the care provider at each clinical encounter and were distinct from those used for billing purposes. The comorbidities were: heart failure, diabetes, hypertension, myocardial infarction, peripheral vascular disease, prior coronary artery bypass graft surgery, cerebrovascular disease, chronic pulmonary obstructive disease, liver disease, renal failure, iron-deficiency anemia, rheumatoid disease, peptic ulcer disease, dementia, depression, cancer, substance use, and acquired immunodeficiency syndrome (S1 Table) [17]. In addition to the severity of the valvular disease, the left ventricular ejection fraction and left atrial volume index was extracted from the echocardiogram report. This was missing in 1,234 patients, who were excluded. Patient age, sex, and race had no missing values. Comorbidity variables defined via ICD-10 codes did not have any missing values as the absence of ICD-10 codes corresponding to the condition of interest was defined as the absence of a condition consistent with methodologies adopted by existing studies.

Interval mitral valve surgery, transcatheter mitral valve interventions, and ablations

We linked the echocardiogram database with the institutional Society of Thoracic Surgeons Adult Cardiac Surgery Database and the list of cardiovascular procedures that the patient underwent within our health system since the time of the index echocardiogram and October 15, 2019. Mitral valve surgery (mitral valve replacement or repair) or mitral valve intervention (transcatheter edge-to-edge repair or transcatheter mitral valve replacement) was treated as the time-varying covariate in the Cox proportional hazard models. Similarly, catheter-based ablations performed for atrial fibrillation or flutter were identified and matched.

Statistical analysis

Continuous variables were reported as medians and interquartile ranges (IQR). Categorical variables were expressed as frequencies and percentages. Differences in demographic variables between patients with atrial and dysfunction were compared using Wilcoxon rank-sum tests for non-normally distributed continuous variables and the chi-squared tests for categorical variables.

Kaplan-Meier plots and corresponding log-rank tests were calculated to evaluate unadjusted and matched survival estimates. The study aimed to compare the potential impact of the degree of SMR on survival between patients with isolated atrial dysfunction and patients with ventricular dysfunction. To account for potential confounding between survival and atrial or ventricular dysfunction, matching between isolated atrial and ventricular dysfunction groups was performed by first estimating propensity scores from a multivariable logistic regression model using the covariates listed below. The groups were then matched with a 1:1 coarsened exact nearest-neighbor matching method without replacement using the propensity score and matching the groups exactly on the severity of SMRs [18]. Observations without nearest neighbor match were discarded. This method was used to balance the observed patient characteristics between the two groups across different severity strata of SMRs. The quality of matching was evaluated using the Love plot depicting the standard mean difference of the groups in each covariate and the density plot of the propensity score. We defined the standard mean difference <0.1 as the threshold for balanced matching [19]. We fitted a bivariate Cox proportional hazard model on the matched dataset, including both ventricular and atrial dysfunction patients, to assess the hazard of time-to-death associated with the severity of SMR occurring with atrial or ventricular dysfunction. Hazard ratios for each combination of ventricular versus atrial dysfunction and SMR severities were calculated using the group with ventricular dysfunction without SMR as the reference group.

The propensity score was calculated from the following covariates: atrial or ventricular dysfunction, age, sex, race, inpatient status (in vs. outpatient), heart failure, diabetes, hypertension, myocardial infarction, peripheral vascular disease, prior coronary artery bypass graft surgery, cerebrovascular disease, chronic pulmonary obstructive disease, liver disease, renal failure, iron-deficiency anemia, rheumatoid disease, peptic ulcer disease, dementia, depression, cancer, substance use, and acquired immunodeficiency syndrome. Additionally, mitral valve surgery or intervention and ablation for atrial arrhythmias were matched on time from echo to intervention.

As a sensitivity analysis to assess the differential impact of SMR on the hazard of death between patients with atrial and ventricular dysfunction, Cox models were fitted separately for atrial and ventricular dysfunction groups using the demographic, comorbidity, and SMR severity as time-independent covariates and mitral valve surgery or intervention as a time-dependent covariate. The proportional hazard assumption was evaluated for the MR severity and atrial or ventricular dysfunction using the survival plot, which satisfied the assumption. Because the coefficients of different models cannot be directly compared for the potentially varying residuals [20], we estimated the number of deaths during the constant follow-up duration under the simulated condition of every patient having no SMR, mild SMR, and moderate or severe SMR. This was done separately for patients with atrial dysfunction and ventricular dysfunction. By comparing the percent changes in the number of deaths under all patients having no SMR to all patients having severe SMR, we estimated the impact of increasing severity of SMR on the hazard of death. We characterized the differential impact of SMR between patients with atrial and ventricular dysfunction by comparing the estimated percent increase in the number of deaths as the severity of SMR increased.

A two-tailed p < 0.05 was considered significant. All analyses were conducted with RStudio 1.3.1073 (Rstudio, PBC, Boston, MA) with packages dplyr, gtsummary, matchit, survival, and Python (Python Software Foundation) packages pandas and regex.

This study was conducted under IRB approval (protocol ID: 2000028791, approved by Yale IRES/IRB on date: 9/16/2020).

Limitations

The small sample size of the patients with severe SMR precluded further stratification between moderate and severe MR. We demonstrated the incrementally worse survival across increasing SMR severities, and the finding likely extrapolates to the difference in moderate and severe MR with a larger sample. However, beyond prognostic significance, we could not infer differential therapeutic approach for different types of SMR. The population included those who underwent echocardiograms related to a health-system encounter and may not generalize to the general population in the community. Nevertheless, our inclusion of academic and community sites within the health-system network enabled us to capture individuals with a broad range of characteristics and likely minimized center-specific patient selection. Measurements important to further characterizing the ventricular function, such as left-ventricular end-diastolic dimension, were not consistently available and were not used in adjustment or matching in the main analysis but were descriptively reported. Inter-rater variability may exist among cardiologists who interpreted the echocardiograms. Although likely small in number, patients who relocated to outside of Connecticut and died would not have been captured by the vital statistics used in this study. We were not able to delineate the etiology of atrial dilation. This is an important limitation, as physiologic and pathologic enlargement of the left atrium may have prognostic differences [31].