Development and Internal Validation of a Prediction Model to Estimate the Probability of Left-Ventricular Diastolic Dysfunction in Stable Maintenance Hemodialysis Patients without Clinical Heart Failure

Objective: To develop and internally validate a multivariable prediction model based on simple clinical record data and laboratory tests and to estimate the probability of left-ventricular diastolic dysfunction (LVDD) in stable maintenance hemodialysis (MHD) patients without clinical heart failure. Design: Cross-sectional study. Setting and Participants: In all, 181 patients on MHD without clinical heart failure were eligible from the dialysis center of Tongren Hospital, Shanghai Jiao Tong University School of Medicine between October 2017 and September 2018. Outcomes: LVDD detected by echocardiography. Measurements: We performed multivariable logistic regression using demographic, clinical, and laboratory data to identify predictors of LVDD with internal validation by using 200 bootstrap replications. Results: Overall, 78 out of 181 (43.1%) patients were affected by LVDD. Predictors included in the LVDD Model were high-sensitive cardiac troponin T (OR 5.91 per 1-ln unit; 95% CI 2.17–16.13), B-type natriuretic peptide (OR 2.35 per 1-ln unit; 95% CI 1.14–4.86), and dialysis vintage (OR 1.04 per 1-month; 95% CI 1.01–1.07). The area (AUC) under the receiver operating characteristic curve of the Model was 0.871 (95% CI 0.811–0.932, p < 0.001). The calibration plot showed good agreement between predicted and observed probabilities with a calibration slope of 1.023 and intercept of –0.010. After internal validation, the Model maintained excellent discrimination (AUC 0.858, 95% CI 0.798–0.919) and good calibration (slope of 1.079, 95% CI 1.058–1.097 and intercept of –0.120, 95% CI –0.142 to –0.093). LVDD Model <0.23 can be used to rule out (sensitivity = 87.3%, negative likelihood ratio = 12.7%) and ≥0.44 can be used to rule in (specificity = 87.6%, positive likelihood ratio = 12.4%) echocardiography diagnosed LVDD. Limitation: External validation of the Model will be required. Conclusions: A model using routinely available simple clinical record data and laboratory tests can accurately predict the risk of LVDD in stable MHD patients. The Model and its cutoff values may be useful for early diagnosis and intervention of LVDD.