Methods

From January 1986 to January 2007, 3,031 patients underwent primary isolated MV surgery for degenerative MR at Cleveland Clinic. Those undergoing concomitant ablation procedures for atrial fibrillation (except a full cut-and-sew Maze procedure) or tricuspid valve procedures for functional regurgitation were considered to have secondary consequences of degenerative MV disease and were included in the present study. However, those with a history of previous cardiac surgery or concomitant coronary artery bypass or aortic valve procedures were excluded. Patients with epicardial coronary artery stenosis ≥50% were also excluded, as were patients with a history of coronary intervention.

To assess the postoperative changes in LV morphology and function, we required ≥1 postoperative transthoracic echocardiogram. Using the Cleveland Clinic echocardiographic database, we determined that 2,778 (92%) of the 3,031 patients had undergone such an examination, and they formed the final study group. Of these, 2,607 (94%) underwent MV repair and 171 (6%) MV replacement. A summary of the patient demographics and clinical and echocardiographic characteristics is provided in Table 1 . Clinical data were retrieved from the prospective Cardiovascular Information Registry. This registry has been approved for use in research by the institutional review board, with patient consent waived.

Transthoracic echocardiograms were performed routinely before discharge and at the discretion of referring physicians during follow-up. Intraoperative transesophageal echocardiograms were not used in the present report. The interpretation of follow-up echocardiograms was obtained at as many postoperative points as available for each patient. A total of 5,336 transthoracic echocardiographic records were available for the 2,778 patients. The distribution of the postoperative echocardiograms permitted assessment of temporal trends for ≤5 years ( eFigure 1 ).

Chamber measurements were derived from 2-dimensional images. The morphologic data recorded and analyzed for temporal responses included left atrial (LA) diameter (from which the LA volume was calculated ), end-diastolic and end-systolic LV diameters, and septal and posterior wall thickness (from which the LV mass was calculated). The LV end-diastolic volume, end-systolic volume, and ejection fraction were measured from multiple 2-dimensional projections by planimetry and edited by visual interpretation of all available views.

The interobserver variability in our echocardiographic laboratory is −0.1 mL (95% confidence interval [CI] −42.8 to 42.5) for the LV end-diastolic volume, 5.9 mL (95% CI −21.6 to 33.3) for the LV end-systolic volume, and −4% (95% CI −16.1% to 8.0%) for the LV ejection fraction. The corresponding intraobserver variability for these same 3 measurements was 7.1 mL (95% CI, −38.8 to 50), 4.4 mL (95% CI −27.9 to 36.7), and 0.0% (95% CI −13.3% to 13.3%).

To determine the temporal response to mitral surgery, the trends of repeated postoperative transthoracic echocardiographic measurements were analyzed longitudinally using a generalized nonlinear mixed model regression for continuous variables (SAS PROC NLMIXED). The temporal patterns of values were characterized by temporal decomposition for each transthoracic echocardiographic variable.

Multivariate analysis, using the variables listed in eAppendix 1 , was performed to identify the preoperative factors that modulated each temporal phase for the postoperative LV mass index and ejection fraction. We initially screened the variables using ordinary multivariate linear regression (PROC REG). Bootstrap bagging methods were used to identify the possible predictors with random resampling and automated stepwise selection. Variables or clusters of variables that entered >50% of 1,000 models then underwent refined generalized nonlinear mixed model regression analysis. Specific variables of interest (including tricuspid regurgitation, MV procedure [repair or replacement], and New York Heart Association functional classification) were forced into the analysis.

Sporadic missing values were imputed using fivefold multiple imputation (PROC MI). For each imputed complete data set, we estimated the regression coefficients and their variance–covariance matrix. We then combined the estimates from the 5 models using PROC MIANALYZE.

Continuous variables are summarized as the mean ± SD and as the 15th, 50th (median), and 85th percentiles. They were compared using Wilcoxon rank-sum nonparametric tests. Categorical data are summarized by frequencies and percentages and compared using chi-square tests. All analyses were performed using SAS statistical software, version 9.1 (SAS, Cary, North Carolina). Uncertainty is expressed by confidence limits equivalent to ±1 standard error (68%).