Preoperative Prediction Models for 30-Day All-Cause Mortality After Mitral Valve Repair in Dogs: A Single-Center Retrospective Cohort Study

Abstract

Background: Mitral valve repair (MVR) has emerged as a novel surgical intervention for dogs with myxomatous mitral valve disease (MMVD). However, no objective risk assessment method has been established for these cases.

Objectives: The primary aim of this study was to develop and evaluate preoperative prediction models for 30-day postoperative mortality in dogs undergoing MVR. The secondary aim was to assess the association between short-term predictive risk and long-term mortality following MVR.

Animals: A total of 2089 client-owned dogs with MMVD that underwent MVR between 2016 and 2023 were included.

Methods: This was a single-center retrospective cohort study. Preoperative variables including demographic data, routine blood test results, diagnostic imaging examination data, and medication history were selected as predictor candidates. Prediction models for 30-day all-cause mortality were developed using these variables and shrinkage estimation methods, and the model performances were evaluated. The association between the predicted probabilities and 2-year cumulative all-cause mortality was assessed using Cox proportional hazards analysis.

Results: The 30-day all-cause mortality rate after MVR was 4.9% (102/2089). The best preoperative prediction model for 30-day all-cause death demonstrated low-to-moderate discrimination abilities (c-statistics, 0.654) and good calibration performance (slope = 1.003; intercept = 0.007; E_avg = 0.002) in internal validation. The quartile grouping of the predicted 30-day all-cause mortality risk was associated with 2-year mortality.

Conclusions and clinical importance: The preoperative prediction model for short-term mortality in dogs undergoing MVR demonstrated acceptable predictive performance. The prediction model may provide an objective preoperative risk assessment in dogs undergoing MVR at this center.

Keywords: canine; cardiac surgery; clinical prediction; myxomatous mitral valve disease; prognosis.

FIGURE 1

Graphical presentation of the overall analysis flow.

FIGURE 2

Predictive contributions of all predictor candidates estimated as standardized odds ratios using a multivariable binary logistic regression model with ridge estimation for 30‐day postoperative mortality. ACEi, angiotensin‐converting enzyme inhibitor; Alb, serum albumin; AOV, peak aortic jet velocity; APTT, activated partial thromboplastin time; ACT, activated clotting time; BUN, blood urea nitrogen; BW, body weight; CKCS, Cavalier King Charles Spaniel; Cl, chloride; Cre, serum creatinine; CRP, C‐reactive protein; Emax, peak velocity of early diastolic transmitral flow; Fib, fibrinogen; FS, fractional shortening percent; Glu, serum glucose; K, potassium; LA/Ao, left‐atrium to aorta ratio; LVIDDn, left ventricular end diastolic diameter normalized for body weight; Na, sodium; PT, prothrombin time; Stage, myxomatous mitral valve disease stage; Tbil, total bilirubin; TP, total protein; TRV > 3.7 m/s, tricuspid regurgitation velocity greater than 3.7 m/s; VHS, vertebral heart score; WBC, white blood cell. Breeds refer to Chihuahua. Stage refers to B1 and B2.

FIGURE 3

Apparent calibration plot of the prediction model (final model) using lasso estimation for 30‐day all‐cause mortality in dogs after MVR. The plot compares the observed proportions of 30‐day mortality with the predicted probabilities. The red line represents perfect calibration (ideal line), while the black line shows the flexible calibration using a Loess smoother. The shaded area represents the 95% confidence interval (CI) for the calibration curve. Apparent calibration metrics include an intercept of 0.00 (95% CI: −0.20 to 0.20) and a slope of 1.14 (95% CI: 0.86–1.43). The apparent discrimination ability of the model is reflected by a c‐statistic of 0.70 (95% CI: 0.64–0.75).

FIGURE 4

Kaplan–Meier survival curves stratified by risk quartiles for 2‐year cumulative all‐cause mortality in dogs who survived more than 30 days after MVR. The quartiles were based on predicted 30‐day mortality risk, with ranges of first quartile: 0.53%–2.77%, second quartile: 2.77%–3.92%, third quartile: 3.92%–5.71%, and fourth quartile: 5.71%–82.31%. The shaded areas represent the 95% confidence intervals for each quartile. A difference was observed between the quartiles (log‐rank p = 0.023). The vertical dotted line represents the 30‐day time point after MVR.